UNIVERSITY OF VIRGINIA

Transcription

1 UNIVERSITY OF VIRGINIA ROTATING MACHINERY & CONTROLS LABORATORY ROMAC NEWSLETTER 2009 Issue Special points of interest: New Test Rigs New Rotor Codes New Bearing Codes 2009 Annual Meeting in Myrtle Beach SC Depiction of the Engine Test Rig assembly showing the fluid damper bearings, magnetic exciters, and unbalance disks in the center Inside this issue: Bearings 3 Seals 26 Compressor Surge Test Rig Photo of Three Mass Rotor Test Rig Fluid Flows 31 Rotordynamics 34 Optimization 41 Gears 42 Magnetic Bearings 43 GUI 48 Heartpump 51 Overhead View of the test rig clearly showing the AC motor

2 Message from Director To ROMAC Industrial Members: Pradip N. Sheth: In Memory The newsletter this year is dedicated to Professor Pradip Sheth. Pradip passed away suddenly in December 2008 from a stroke. He graduated from the University of Wisconsin and worked at Ford Motor Company and Allis Chalmers after that. At Ford, he developed a vehicle analysis program. While at Allis Chalmers he was Manager of Engineering Development, the leader of the CAD/CAM implementation group and established a Computer Integrated Manufacturing Group. Professor Sheth was a long time faculty member in Mechanical and Aerospace Engineering, arriving at UVA in He had a knack for matching faculty with industry and served as regional coordinator for the Manufacturing Action Program of the Virginia Center for Innovative Technology. He established the Mechatronics Laboratory in Mechanical and Aerospace Engineering to bring aspects of electrical engineering and computer engineering into undergraduate mechanical and aerospace engineering. That laboratory is now named for Pradip. He loved working with students and came to know many of them as personal friends over the years at UVA. Pradip was a major contributor to ROMAC in recent years as many of you know. He was interested in many research topics including rotordynamics, integration of NASTRAN with the ROMAC codes, modeling of aircraft systems with Boeing. ROMAC is establishing one speaking prize, the Pradip N. Sheth Outstanding Presentation of $500 for the best talk given by a student at the ROMAC Annual Meeting each year. This will be determined by vote of the companies in attendance at the annual meeting each year. Also, RO- MAC is establishing a Pradip N. Sheth Fellowship to be given to an outstanding graduate student nearing his graduation each year, on top of his normal Graduate Research Assistantship. The particular student will determined by the ROMAC faculty members. ROMAC Research Activities Our research efforts continue to expand to better serve our ROMAC member companies. ROMAC continues to grow rapidly. Over the past few years we have had many new members from marine applications of rotating equipment and expansion in the aerospace field. This area couples with our NASTRAN efforts, squeeze film dampers, rolling element bearings, instability, and gears. We also continue to grow in the international area with new companies over the past couple of years. The many research projects outlined in this Newsletter cover most of the topics in dynamics of rotating machinery, from rotor modeling and computer codes to fluid film bearings, seals, gears, rolling element bearings, fluid flows, magnetic bearings, and optimization. The new Graphical User Interface is continuing development with releases to companies in June, 2008 and another one coming up in This goes along with a major improvement in our rotor dynamics codes Comborotor and Matlabrotor. Our association with NASTRAN through MSC Software continues with the integration of ROMAC codes. Our test rigs are coming along extremely well with the compressor surge test rig, the fluid film bearing test, and flexible rotor magnetic bearing/controls test rig leading the way. Several other test rigs on squeeze film dampers, seals, and magnetic bearings are underway. A new area of research for ROMAC is efficient wind turbines. More on that topic later. Paul Allaire, ROMAC Director and Mac Wade Professor Paul Allaire, Wade Professor of Mechanical and Aerospace Engineering and Director of ROMAC Page 2

3 1. BEARINGS 1.1 Fluid Film Bearing/Squeeze Film Damper Test Rig Students: Tim Dimond (ROMAC Test Lab Engineer), Tanner Hall, Mohit Chhabra, Feng He, Ali Gerami, Simon Mushi, Jeff Walls Faculty: Paul Allaire, Zongli Lin, George Gillies, Roland Krauss Research Professionals: Wei Jiang, Robert Rockwell FFBTR Consortium Sponsors Rotating Machinery Technology, Inc. Kingsbury Turbo Components and Engineering Zollern FFBTR Design Reviewers Elliott Compressor Waukesha Bearings The fluid film bearing test rig (FFBTR) continues to be the highest rated project in ROMAC Introduction: The fluid film bearing test rig (FFBTR) continues to be the highest rated project in ROMAC based upon last year s research objectives. As we all know, fluid film bearings are now operating at surface speeds that clearly place them in the turbulent flow range for the lubricant film, both for oil and water lubricant films. This means that both turbulent effects and inertia effects are present in the lubricant films. Extremely little measured data on dynamic properties of bearings in this range have been obtained. The test rig is also capable of testing squeeze film dampers. This testing on a rigid foundation will complement the work on squeeze film dampers being performed on the flexible foundation balancing and optimization rig. Magnetic Bearings Speed Increaser Motor Fluid Film Test Bearing Page 3

4 1. BEARINGS The FFBTR continues to progress towards assembly. We now have four paid sponsors of the FFBTR Group: 1) Rotating Machinery Technology Inc., 2) Kingsbury, 3) Turbo Components and Engineering, and 4) Zollern. All of them are interested in testing their bearings. The FFBTR Group is also extensively supported by several companies, providing materials, components, technical assistance, etc. The rig has been designed by Tim Dimond, several other graduate students, research professionals, and faculty. The motor and drive was delivered by Emerson/US Electric Motors, the gear box was delivered by Lufkin, the magnetic bearings are on the way from Innovative Power Solutions and RMT is constructing many of the test rig components. Other ROMAC companies are kind enough to supply technical expertise as needed. Design reviews have been promised by Waukesha Bearing and by Elliott Compressor. The FFBTR is expected to be operational in late summer or early fall of Magnetic Bearings Test Bearing The ROMAC FFBTR Plan View FFBTR Test Section Cross-Section Major Equipment Delivery Dates: Motor Drive: December 15, 2007 Motor: January 30, 2008 Gearbox: December 18, 2008 Fluid Film Bearing Specifications Test Section The purpose of the ROMAC fluid film bearing test rig is to measure the load capacity, thermal effects, stiffness and damping of oil-lubricated and water lubricated bearings under high speed application conditions. Testing of squeeze film dampers with this test rig is also possible. The technical specifications for FFBTR fluid film bearing tests are provided in Table 1: The FFBTR is expected to be operational in late summer or early fall of 2009 Page 4

5 1. BEARINGS Table 1. FFBTR Characteristics Test Rig Characteristic Values, SI Values, BG Test Bearing Diameter 127 mm 5 in L/D Ratios Pad Pivot Offsets Orientations LBP, LOP Rotational Speed Range rpm Surface Speed Range m/s ft/ s ft/ s Oil lubricated, typical m/ s Water lubricated, typical m/s ft/ s Lubricants ISO VG 32 ISO VG 46 Water Maximum Bearing Unit Load 3.3 MPa 480 psi Dynamic perturbation displacement, 0-p 12 μm 0.5 mil Excitation Frequency Range 60 Hz 515 Hz Notes: 1. Radial linearity of the lubricating film is assumed. The rotor motion of 0.5 mil 0- p results in shaft orbits that are approximately 6 percent of the total bearing clearance. 2. The actuator sizing and bearing static load limits are based on the following: Nominal bearing clearance of 1.5 mil/in Four pad bearing, LBP Preload: m = 0.3 L/D =0.75 Offset: = 0.6 ISO VG 46 lubricant Rotor 0-p displacement at FFB of 0.5 mils Maximum excitation frequency of 1.5x (20,400 rpm) or 1.37x (22,400 rpm) Maximum rotational speed of 22,440 rpm Bearing coefficients calculated with MAXBRG Rigid Rotor Full Bearing coefficients Bearing configurations with lower stiffness require lower dynamic force for rotor displacement and can be tested at higher static load. Evaluation of the test rig includes investigation of potential detrimental natural frequencies in the operating range of the FFBTR. This evaluation included a check of the natural frequencies in the base. A finite element analysis of the original concept of a reinforced concrete base showed that base natural frequencies would occur in the operating range of the FFBTR. Based on this analysis, the base design was changed to all-steel construction. A thorough finite element analysis of the FFBTR steel base concept indicates that Page 5

6 1. BEARINGS there should be no detrimental natural frequencies in the operating range of the FFBTR. The fundamental mode shape for the test section portion of the base from the finite element analysis is shown in Figure X. Fundamental Bending Mode of Test Section Base Motor The motor for the FFBTR has been delivered by Emerson Electric/U.S. Electric Motors in January The motor was sized to overcome the parasitic losses predicted by MAXBRG for a fluid film bearing with preload of 0.3, L/D ratio of 0.75, ISO VG 46 lubricant and maximum bearing static unit load of 480 psi. Additional parasitic losses from magnetic bearing eddy current and the gearbox were also considered. A factor of safety of 1.5 was also applied to allow robust speed control during a bearing test. A summary of the motor specifications follows: Table 2. Motor Specifications Test Rig Characteristic Values, SI Values, BG Output Power Rating 261 kw 350 HP Base Speed (60 Hz) 3600 rpm Maximum Speed (73 Hz) 4400 rpm Emerson Electric Drive Page 6

7 1. BEARINGS Speed Increaser US Electric motor Lufkin Industries Gearbox Speed Increaser The speed increaser for the FFBTR has been delivered by Lufkin Industries, Inc. The gearbox has been designed to deliver the rotational speed required for high-speed fluid film bearing tests and to deliver the input power required to overcome the parasitic losses from the fluid film bearing and the magnetic bearings. Provisions for radial probes and accelerometers were provided for measurement of radial and axial vibrations. RTDs were installed to monitor operating temperatures. The data from these probes will be used to validate combined torsional-lateral-axial finite elements developed by ROMAC by Blake Stringer, and for other experiments involving geared systems. The gear box will be instrumented to study gearbox operational properties with Lufkin. Speed Increaser Horsepower Rating: 350 HP Speed Increaser Gear Ratio: 1:5.026 Active Magnetic Bearings The FFBTR is being constructed with two magnetic bearings employed as the means of applying the static and dynamic loads to the single fluid film bearing. The magnetic bearings will be used as exciters to support the shaft and perturb it with small displacement and velocity motions. The bearings will be manufactured by Innovative Power Solutions, LLC. The active magnetic bearings employ four e-cores separated into quadrants. The design characteristics of the active magnetic bearings are summarized in Table 3. Page 7 The data from these probes will be used to validate combined torsional-lateralaxial finite elements...and for other experiments involving geared systems

8 1. BEARINGS Table 3. Active Magnetic Force Exciter Characteristics Force Exciter Characteristic Values, SI Values, BG Magnetic Bearing Diameter mm 6 in Stator Back Iron Material M-19 Magnetic Steel Rotor Target Material M-19 Magnetic Steel Total Load Capacity, Per Bearing 11 kn, linear operation 13 kn, nonlinear operation 2500 lb f, linear operation 3000 lb f, non-linear operation L/D Ratio 1 Main Pole Width 38.1 mm 1.5 in Auxiliary Pole Width 19.1 mm 0.75 in Active Magnetic Bearing Cross-Section Magnetic Flux Line Plot Static force design curves for the active magnetic bearings were developed for varying levels of bias current to determine when non-linear back-iron saturation effects become important. Based on the analysis, the bearing force profile is essentially linear for bias currents up to 27 A. For higher bias currents, the non-linear effects become increasingly important. Operation in the nonlinear region will be necessary to perform some experiments on heavily loaded fluid film bearings. The maximum static unit load that can be applied while operating with bias current of 27 A is 2.2 MPa (320 psi). The API limit for bearing unit load in gearboxes is 3.4 MPa (500 psi) [13]. For a bias current of 36 A, a bearing unit load of 3.3 MPa (480 psi) can be applied, but the range of linear operation and the available slew are reduced. Page 8

9 1. BEARINGS Magnetic Actuator Static Force Curves 4000 I b =15 Force (lbs) 2000 I b =13 I b =11 I b =10 I b =9 I b =8 I b =7 I b =6 I b = Perturbation Current / Bias Current Static Force Design Curves, FFBTR Active Magnetic Bearings The active magnetic bearings will act both as force application devices and force measurement devices. The theoretical correlation between magnetic flux and input force and magnetostructural strain and input force were developed by Bob Rockwell. This information is used to select either Hall sensors or fiber optic strain gages as the primary input force measurement method. A side-by-side comparison of the two sensors will be done with a separate experiment, which is described in the Force Tester section of the newsletter. Perturbation Current 20.1 amps Actuator Force 9 kn Total Magnetic Flux Total Magnetostructural Strain Magnetic Flux And Magnetostructural Strain Contour Plots Rotordynamics and Control Rotor modeling of the entire FFBTR train is largely complete. Tanner Hall has carried out detailed critical speeds, unbalance response and stability analyses, with supervision by Tim Dimond. The goal is to identify any undesirable modes that could adversely impact FFBTR dynamic measurements and mitigate or eliminate Page 9

10 1. BEARINGS them. The controllers for the active magnetic bearings are under development by Wei Jiang and Mohit Chhabra. The controller concepts are being demonstrated on a rigid rotor to work out implantation issues prior to implementation on the FFBTR. Rigid Rotor Control Implementation Rig Rigid Rotor Magnetic Bearings Rigid Rotor Orbital Plots at 2,000, 4,000, 7,000, and 9,000 rpm Page 10

11 Work has focused on producing non-synchronous sinusoidal orbits of the shaft for perturbation of the rotor, assuming frequency-domain system identification. However, some time-domain system identification methods under consideration require bandwidth-limited white noise input for the mathematics involved to be valid. Generating bandwidth limited white noise will also be considered for the control algorithm. Static and Dynamic Measurements In order to estimate bearing stiffness and damping using an input-output identification routine, we have to have a method of determining the force exerted by the magnetic bearings. We will be using either fiber optic sensors attached to the magnetic bearing radial poles, or Hall sensors placed in the magnetic bearing pole ends, in addition to measuring the currents in the magnetic bearing coils. A special force test rig with a single sided e-core bearing has been built to evaluate and establish the best methods for this. The update on the force test rig is described in detail in a separate section of the newsletter. Separate evaluations of the fiber optic gages with a cantilever beam have also been completed. This project includes two additional topics: identification of experimental data for fluid film bearings and full coefficient dynamic analysis of tilting pad bearings. Several papers on these topics have been written already and are available as ROMAC reports. Shaft and pad motions will be measured using non-contact sensors and/or accelerometers for evaluation of full dynamic stiffness and damping, including pad properties. Pressure measurements will be taken in the rotating shaft to verify the computer calculations. Temperature measurements will be taken in the fluid film bearing and on the journal. The current goal is build and make the test rig operational without the rotating measurements, then add them in after initial grooming and dynamic tests. One of the challenges in obtaining on-journal data is obtaining telemetry at the rotational speeds required to produce full turbulence in the bearing lubricant. Conventional slip rings introduce noise into the measurements and have a top rotational speed of 10,000 rpm. Oil-lubricated bearings will be tested at twice that speed. Wei Jiang and Mohit Chhabra are developing data acquisition methods that can store the data on the rotor during a test for download after a test is completed. Logistics 1. BEARINGS The FFBTR will share space with the ROSTR in the Aerospace Research Laboratory. New lab space has become available and the conversion process is underway. Initial changes to the rooms have been made, with additional electrical upgrades and auxiliary services installations pending by the University of Virginia Facilities Management group. Remote operation will be carried out in a separate office. Future Home of the FFBTR and ROSTR Page 11

12 1. BEARINGS 1.2 Fluid Film Bearing System Identification Modal Analysis of Tilting Pad Journal Bearings One of the key goals of the FFBTR is to characterize or identify the dynamic response of tilting pad journal bearings (TPJBs). There has been much controversy as to the frequency response of journal bearings. The seminal paper by Lund accounted for pad motion and considered synchronous reduction of the bearing coefficients, or reduction at the operating speed of the rotor. However, there is much experimental evidence that there is a frequency response in TPJBs that cannot be fully explained using the synchronously reduced stiffness and damping coefficients. This controversy has extended to the specifications used by the turbomachinery industry. The American Petroleum Institute Specification 617 for axial and centrifugal compressors currently requires the use of synchronously reduced coefficients in dynamic analyses of rotating systems. However, API Standard Practice 684 acknowledges the requirements of API 617 but notes that full coefficients will give a less conservative estimate of the log decrement of a rotor. There is a need to resolve these issues for the turbomachinery community. Most analyses of rotors with TPJB supports have concentrated on the shaft degrees of freedom while either neglecting the pad degrees of freedom or considering them implicitly. This is done for practical reasons in the field. There are generally measurement locations available for the shaft at the journals, but not at the bearing pads themselves. One of the benefits of a laboratory experimental program is that these additional measurements can be taken without compromising the function of the machine. Two distinct non-synchronous approaches to the frequency response of TPJBs have been explored in the literature. Most theoretical studies and some experimental studies have considered the pad dynamics implicitly and determined frequency-dependent stiffness and damping coefficients, or frequency-dependent KC. However, other experimenters have only considered shaft degrees of freedom and have found an experimentally identified second-order nonsynchronous model for bearings with frequency-independent stiffness, damping, and ROSTR added mass coefficients (KCM). After study of the two dynamic reduction schemes, frequency dependent KC and KCM, a new approach that considers the bearing pads explicitly became apparent. The fundamental linearized dynamics of the bearing depend on N p + 2 degrees of freedom, where N p is the number of bearing pads. This approach does not consider any reduction of the pad degrees of freedom in the bearing dynamics. FFBTR FFBTR and ROSTR ARL Arrangement Page 12

13 1. BEARINGS By performing modal analysis of this full coefficient model and the KCM model, direct comparisons can be made to between full-order and reduced-order bearing models. For example, consider this bearing which was analyzed using MAXBRG. For this analysis, an isoviscous lubricant was assumed. Bearing Parameter Values Diameter 127 mm L/D 0.75 Shaft mass 68 kg External Radial 16 kn Load Bearing Specific 689 kpa Load Diameteral Clearance μm Preload 0.3 Pivot Offset 0.5 Static Eccentricity 0.12 Ratio Lubricant dynamic 33 mpa-s viscosity Lubricant temperature 32 C Rotational speed 5,000 rpm Reynolds Number 16 Reduced Reynolds Number Number of Pads 4 Pad arc length 75 Lubricant flow rate 56.8 L/ min Pad Inertia 716 kgmm 2 The full coefficient stiffness and damping matrices calculated using MAXBRG in SI units are One of the benefits of a laboratory experimental program is that these additional measurements can be taken without compromising the function of the machine Page K = C =

14 1. BEARINGS Using dynamic reduction, the real and imaginary part of the frequency response function for a rigid rotor in this bearing was determined. A quadratic curve fit was applied to the real part of the complex impedance, and a linear curve fit was applied to the imaginary part of the complex impedance. The dynamically reduced response and the curve fits are plotted in Figure x. Re(Z): Im(Z): 2 r = r = 0.98 Figure X. Complex Impedance of Example Bearing and Curve Fits Using the KCM identification procedure, the identified dynamic coefficients were found to be K = C = 10 N-s/m M = kg x 10 N/m With the two representations of the bearing dynamics available, modal analysis was conducted on both representations to compare system eigenvalues and eigenvectors. Table x reports the system eigenvalues and eigenvectors for the full coefficient model, and Table y reports the system eigenvalues and eigenvectors for the reduced order KCM model. Page 14

15 1. BEARINGS Table X. Bearing Full Coefficient Eigenvalues and Eigenvectors Mode 1,2 3,4 5,6 7,8 9,10 11,12 Eigenvalue 286 ± 335 j 286 ± 335 j 568 ± 244 j ± 18.6 j ± 9.84 j ± 8.29 j f n, Hz ζ X ± j m j m j j Y ± j ± j m j j ± ± j m m j ± j ± j θ m j ± j ± j j m ± j θ 2.011m j ± j m 6 10 j θ m j ± j ± 0.701j m j 0.261m j θ ± j m j 0.701± j 0.001m j m j 1 Table Y. Bearing Reduced Order (KCM) Bearing Coefficient Eigenvalues and Eigenvectors Mode 1,2 3,4 Eigenvalue 506 ± 521j 506 ± 521 j ω n ζ x 1 1 y ± j m j Two representations of a rigid rotor supported in a TPJB have been considered. The full bearing coefficient model of a TPJB dynamic response has very different modal results when compared to the reduced order identified KCM bearing model. When the natural frequency and mode shape predictions are considered for the two bearing models, several trends become apparent. Both the KCM and full coefficient models result in four (two forward and two backward) underdamped modes that dominate the frequency response up to 90 Hz. However, the KCM model resulting from application of a frequency domain system identification method to the reduced model of shaft frequency response overpredicts the system natural frequency when compared to the full coefficient 6x6 model. The overprediction of natural frequency is in the range of percent. This overprediction indicates that the effective identified stiffness is overpredicted, or the identified effective added mass is underpredicted, or both. The calculated modal damping ratio was higher for the reduced order model as compared to the full coefficient model. The damping ratio estimate from the KCM model was 7 percent higher than the full coefficient modal damping estimate. Page 15

16 1. BEARINGS Inspection of the mode shapes generated by the full bearing model indicates that the bearing pads participate significantly in the four dominant underdamped modes. This behavior cannot be captured by the KCM model. When the natural frequencies of the full coefficient model are considered, there is sufficient separation between the dominant underdamped modes and the overdamped modes dominated by pad motion for the pad modes to not be apparent in a limited frequency sweep of 0-1.1X, where X is the rotor running speed for the bearing design. Added mass terms resulting from the KCM identification method have been attributed to several effects, including pad motion, dynamic reduction, and fluid inertia effects. In some cases, the mass term has been reported to be negative. It is notable that the identified mass for this example laminar bearing designs is very large, 1938 kg. For the bearing cases considered, with reduced Reynolds number of 0.077, the effects due to fluid inertia should be very small. If an experimental goal is to confirm the presences of fluid inertia terms for laminar or turbulent TPJBs, then a different system identification procedure would be required. Only measuring shaft degrees of freedom will not allow differentiation of pad motion, dynamic reduction, and fluid inertia effects. The modal analysis of the KCM results indicates that the KCM identification method is capturing the first two underdamped modes. However, information on tilting pad participation in the modes is lost. Additionally, there is a definite difference in the modal parameters of natural frequency and damping ratio that invites more experimental study. This would be especially important in determining the first bending mode of a rotor-bearing system, which would affect the prediction of the running speed corresponding to onset of instability. It would also affect log decrement calculations, since the different methods give different modal damping ratios. It is more appropriate to evaluate the author s of KCM method papers claims within this modal framework. The first two underdamped modes for all bearing cases from the full coefficient model correspond closely to the modes determined from the KCM bearing model, with the exception of a 67 percent difference between the calculated natural frequencies. Additionally, there is a large frequency separation between shaft modes and pad modes. This lends credence to the theory that the KCM model is a model-reducing approach to TPJB behavior. Another limitation of the KCM identification approach is the lack of explicit information on the bearing pads. It has been shown through the analysis in this paper that the bearing pads are significant contributors to the TPJB-rotor system modes. Experimental identification procedures should include pad information to properly capture TPJB dynamics. The difference in estimates of damping ratio and natural frequency with the same bearing also has implications for rotordynamic analysis. With differing estimates of modal natural frequency and modal damping for a tilting-pad bearing, calculation of rotor critical speeds and rotor stability is significantly affected. The net effect of the differing modal estimates on these rotordynamic analyses is an area for future study. More details on this topic are available in ROMAC report 537. Page 16

17 1. BEARINGS Modal Analysis and System Identification The identification problem for a TPJB is fundamentally one where there are more outputs than inputs. With the FFBTR design, the number of available independent inputs into the system is two, corresponding to horizontal and vertical excitation of the rotor. Deliberate misalignment of the shaft would introduce an additional freedom, but angular misalignment will affect the bearing coefficients due to fluid-structure interaction forces. The number of outputs is N p + 2, where N p is the number of bearing pads. Recent work in ROMAC, documented in reports 408 Rotor Bearing System Identification Using Time Domain Methods and 515 Stability of Rotors Supported by Tilting Pad Journal Bearings, discusses input-output and output only time-domain system identification techniques. Report 408 in particular documents methods that can be used to identify physical bearing parameters with limited information on the inputs. There has also been other research in signal processing and system identification that deals with this problem. These methods are under consideration for implementation on the FFBTR to identify the physical bearing parameters. 1.3 Dynamic Force Measurement Validator Students: Tanner Hall, Tim Dimond (ROMAC Test Lab Engineer), Simon Mushi Faculty: Paul Allaire Research Professionals: Wei Jiang, Robert Rockwell The development of a technique to accurately measure forces exerted by the magnetic bearings on the Fluid Film Bearing Test Rig (FFBTR) is necessary to determine the applied unit load, stiffness, and damping in the rig s test bearing. The motivation behind the Dynamic Force Measurement Validator (DFMV) is to provide data showing which force measurement method is more adequate for the FFBTR with the least amount of uncertainty. The DFMV will compare the force measuring capability of Hall sensors to fiber optic strain gages (FOSG). The FOSG method was originally developed by Texas A&M while the Hall sensor method has been used by Aenis & Nordmann and Knopf & Nordmann. The FOSG method has the benefit of being immune to electromagnetic interference, but could prove to be less accurate due to the expected low strain levels and the full scale of the sensors. The FOSG we will be using has a range of microstrain, but we expect actual strains to be only 50 microstrain. While Hall sensors are more effective over the full range of magnetic flux, they are susceptible to noise and will require an alteration to the e-core geometry, either by increasing the air gap or countersinking within the e-core material. The DFMV, shown in Figures 1 and 2, features a single-sided e-core linear magnetic actuator bearing suspended under a welded structure by an array of three dynamic load cells, which provide stability and a calibration standard for the rig. The e-core will levitate a target plate and experience strain and magnetic flux due the variable set of weights attached underneath the plate. FOSGs attached to the pole sides and Hall sensors embedded onto the faces of each pole will detect magnetostructural strain and magnetic flux produced Page 17

18 1. BEARINGS by the e-core as it levitates the load. The static forces caused by the hanging weights and the dynamic forces, introduced through perturbation frequencies, applied to the e-core will produce comparable strains and magnetic fluxes to those predicted in the radial magnetic bearings for the FFBTR. The major components of the DFMV have been assembled except for the installation of the FOSGs and Hall sensors on the e-core, and partial levitation has been achieved. Before the sensors can be installed and testing can begin, the DFMV s controller must be improved. The target plate, rod, and weight assembly, shown in Figure 3, was designed to only levitate vertically through a sleeve guide inside the support structure. However, the initial controller design exposed the rig s mechanical tolerances and it was found that the assembly is able to pitch inside this guide. Consequently, an analog PID controller was shown to be inadequate to maintain a constant air gap between the e-core s poles and the target plate, a critical requirement to accurately calculate force. A digital controller is now being implemented featuring independent pole control so that if the plate begins to pitch, the controller will compensate for this error by decreasing the current on one pole while increasing the current on the opposite pole to level the assembly. Test results are expected later this spring. Figure 1: Side View of Assembled Dynamic Force Measurement Validator on Concrete Base Figure 2: Cutaway View of Dynamic Force Measurement Validator A digital controller is now being implemented featuring independent pole control Figure 3: Schematic of E-Core, Target Plate, Weight Rod, and Weight Assembly Page 18

19 1. BEARINGS 1.4 Water Lubricated Rubber Bearing ROMAC Software Engineer: Amir Younan Faculty: Paul Allaire Project Start Date: August 2006 The project evaluates the elastohydrodynamic effects in compliant journal or thrust bearings such as water lubricated stern tube rubber bearings where the rotational speed is very low and the load is high. The analysis involves the deformation of the compliant materials in the presence of a liquid lubricant. These bearings are highly loaded, with a rigid shaft operating at low speed, and very small lubricant film thicknesses. The angular contact areas are extremely small. The classic EHD characteristics are a pressure profile with a peak value and a pressure spike near the trailing edge as well as a nearly uniform film thickness with a sudden film thickness decrease, also near the trailing edge. One of the major objectives of this project has been to develop a software program which solves the EHD compliant surface bearing problem automatically. Figure 1: The inner surface of the bearing (original and deformed) and the distribution of the pressure along the circumferential direction. There are two types of EHL: hard EHL which takes place with high elastic modulus such as in rolling element bearing and soft EHL which takes place with low elastic modulus as in rubber lined bearings. In the soft EHL, the effect of pressure on the viscosity is not pronounced due to the small pressure magnitude as opposed to the hard EHL. The average film thickness is typically 1-10 micrometers. The average pressure developed is 1 MPa. The elastic distortion due to the low elastic modulus is large even with light loads and these elastic deflections are the primary contributor to the lubricant film separating the contact surfaces. Figure 2: Typical dimensionless pressure profile (left) and film thickness profile (right) in a water lubricated rubber bearing Page 19

20 1. BEARINGS 1.5 Time Transient Squeeze Film Damper Analysis Romac Software Engineer: Amir Younan Faculty: Paul Allaire Project Start Date: January 2008 New squeeze film damper software has been developed to conduct nonlinear time transient analysis for different designs of dampers. The analysis can handle the regular configuration: with and without end seals, with and without oil supply grooves, with and without centering spring. The code accounts for air entrainment in the oil. Modified expressions for the viscosity and density of the oil are developed based on the volume fraction of the air with respect to the total volume. The expressions are linked to the Reynolds equation to update the oil properties based on the pressure distribution and air volume fraction. The pressure profile is calculated around the circumferential direction of the damper and it is integrated to provide the forces in the horizontal and vertical directions. Figure 1 shows the example pressure distribution for four air volume fractions. The upper figure shows the pressure distribution at quarter (left) and half (right) length of a squeeze film damper without a supply groove. The lower figure shows the pressure distribution at quarter (left) and half (right) length of a squeeze film damper with a supply pressure at 270 o. Figure 1: Pressure distribution for four air volume fractions at the quarter (left) and half (right) length of the damper for two designs without (upper) and with (lower) supply pressure Page 20

21 1. BEARINGS The input of the squeeze film damper solver is the current position and velocity of the shaft. The output is the pressure distribution, as shown in figure 2, as well as the damper forces. The solver can be linked to a rotordynamic code, as shown in Figure 3 (left), to calculate the transient response of a rotor system with a squeeze film damper. The Transient orbital response of a single mass inside the damper is illustrated in Fig. 3 (right). Figure 2: Pressure distribution at half length of the damper for two designs without (left) and with (right) supply pressure Figure 3: Flow Chart of the link between the squeeze film damper (SFD) solver and the rotordynamic solver (left).the orbital response of a mass inside the damper (right) Page 21

22 1. BEARINGS 1.6 Rolling Element Bearings Romac Software Engineer: Amir Younan Faculty: Paul Allaire The current rolling element code analyzes the elastohydrodynamic lubrication (EHL) in line contact (roller element bearings and spur gears) and point/elliptical contact (ball bearings and helical gears). The EHL problem is a multiphysics problem where the structure domain of the contact surfaces interacts with the fluid domain of the lubricant film. The finite element solver is based on solving the governing Reynolds equation and elastic equation simultaneously which proves to be the successful way to reach converged solution. The pressure distribution and film thickness profile are the main output of the code along with the minimum film thickness and pressure spike magnitude, as shown in fig. 1. A numerical perturbation is applied to the pressure and film thickness profile to calculate the contact stiffness taking into consideration the elastic contribution of the structure as well as the fluid contribution of the film lubricant. Point Contact Results The pressure distribution topology (upper left) and contour plot (lower left)) The film thickness profile topology (upper right) and contour plot (Lower right) Page 22

23 1. BEARINGS The rolling element bearing code is moving to a higher level of analysis where the three dimensional structure is modeled with a full 3-D finite element solver. The previous solver modeled the outer surface of the contact bodies with a half space assumption (Boussinesq solution). The Boussineq solution enforces the dependency of the elastic deflection at one point to the pressure distribution of all the contact point which results in full populated matrices. The three dimensional finite element solver removes the elastic dependency and improves the numerical solution of the problem by formulating the system matrices to be banded. Different mesh generation techniques have been applied to the contact bodies with a dense mesh in the contact area and coarse mesh in the rest of the body, as shown in fig. 2. The mesh generation depends on the isoparametric shape function of element (27-node isoparametric element) selected for the structural model. The solver will conduct the simultaneous solution between the Reynolds equation governing the fluid domain and the linear elasticity of the solid rolling element bearing structure. The Three dimensional mesh of a ball (coarse mesh on the upper and lower contact point/ coarse mesh for the rest of the central region of the ball) Left: The origin of the mesh generation is 6 cubes Right: The final mesh of the ball mapped using the shape function Page 23

24 1. BEARINGS 1.7 Three Mass Rotor Test Rig Student: Matthew Goodhart, Feng He Research Professional: Jim Durand Faculty: Paul Allaire The purpose of this project is to design and convert an existing fluid film bearing test rig as shown in fig. 1 below, into a small scale squeeze film damper (SQD) test rig to do initial testing on instrumentation requirements, and SQD geometry to improve the design and test results from the Engine Balancing Test Rig reported on elsewhere in this newsletter. Figure 1: Overview of original three mass rotor Figure 2: Overview drawing for SQD for Three Mass Rotor Test Rig The objectives for this work are: Measure displacement response during SQD lift off Measure unbalanced displacement response during run-up from 0 to 10,000 rpm Measure unbalanced displacement response at key operational speeds Measure the circumferential and axial pressure distribution in a squeeze film damper during run-up from 0 to 10,000 rpm for key SQD configurations in a bubbly mixture of oil and air/vapor. Calculate SQD damping forces during run-up and at key operating points Investigate air entrainment dynamics - under what conditions does it begin? How can it be controlled? How does it affect the stiffness and damping coefficients? Assess various SQD geometries, and determine which are best Compare to theory described in section 1.5 Figure 3: Section view of the SQD for the Three Mass Rotor Test Rig Page 24

25 1. BEARINGS Figure 2 is an overview drawing that shows the design of the SQD for the converted rig. Figure 3 provides a sectional view of the SQD. And finally, fig. 4 provides a look at how the pressure transducers are mounted in the journal housing and squeeze film. The SQD geometry is as follows: Rotor shaft diameter: 1 Rotor length: 2 feet Rotor weight: 40 lbs SQD journal length, L: 0.75 SQD film thickness C: 10 mils or SQD intermediate damper ring ID and thickness: 2.05 and SQD outer journal OD, D, and thickness: 3.57 and Journal orbit radius, e: Ratios: L/D = 0.21, e/c = 0.7, C/r = Type of Seal: O Ring Location & Type of Oil Distribution: central groove Figure 4: Pressure taps for the SQD on the Three Mass Rotor Test Rig The resulting rig is shown in Figure 5 below, as it is being assembled and readied for initial testing. Testing will begin soon on this rig. Initially, we will look at startup conditions, and compare the experimental results with computer code predictions. The results of this effort will make the work done on the larger Engine Balancing Test Rig much more directed and effective. Figure 5: Photo of assembled Three Mass Rotor Test Rig, ready for initial testing Page 25

26 2. SEALS 2.1 Hole Pattern Seal CFD Analysis Student: Patrick Migliorini Research Professional: Alex Untaroiu Faculty: Houston Wood, Paul Allaire, Chris Goyne Annular seals in compressors, turbines, and other rotating machines are commonly used to limit leakage between different pressure regions and are extremely important to rotordynamics. They are also some of the least well known components in any advanced rotordynamic analysis of high performance machines. ROMAC is developing detailed seal analysis for industrial annular seals using computational fluid dynamics (CFD) as the best method for obtaining the leakage rates, velocity and pressure distributions, stiffness, damping and mass coefficients. This analysis will be applied to hole-pattern seals (Figure 1). Figure 1: Hole-Pattern Seal Geometry and Computational Domain Currently, there are two methods used to determine the rotordynamic coefficients of gas annular seals, bulk flow analysis and computational fluid dynamics analysis. Traditionally, rotordynamic characteristics of gas annular seals have been analytically determined using bulk flow theory. A strong advantage of using bulk flow theory to determine rotordynamics characteristics for gas annular seals is that the analysis is computationally inexpensive and extremely fast. However, major disadvantages stem from the assumptions made in order to reduce the governing equations to an analytically solvable form. These solutions rely on empirical coefficients determined from experiment. It has been observed that for honeycomb gas annular seals, the friction factor is sensitive to changes in the Reynolds number, inlet pressure, clearance, and ratio of cell depth to cell height. This makes it necessary to adjust empirical coefficients on a case by case basis. Recently, computational fluid dynamics (CFD) has been used to model flow in gas annular seals to determine the forces generated by fluid motion and the rate of leakage out of the seal. The advantage of CFD is that the governing equations can be solved for complex geometries without loss of generality. CFD is better able to capture the flow physics than bulk flow models. The major disadvantage of CFD is that simulations can be computationally expensive, taking hours or weeks to run. With increases in computing power, CFD has become a more advantageous tool for determining rotordynamic characteristics. Page 26

27 2. SEALS A method to calculate leakage and rotor dynamic coefficients has been developed considering steady-state flow and known temperatures and pressure drops using 360-degree seal geometry. Even though the seal geometry is axisymmetric and only a small angular section can be modeled to accelerate convergence and reduce computational time, a full 360-degree analysis of the seal geometry is carried out to determine the stiffness and damping coefficients (Figure 2). A full model is required in order to capture the eccentricity generated by the off-centered position of the impeller, so that its subsequent impact on the magnitude of radial fluid forces is incorporated. Figure 2: Eccentric Rotor Model Figure 3: Seal Clearance Profiles CFD will be used to compare four types of clearance profiles: constant clearance, converging clearance, diverging clearance, and parabolic clearance. (Figure 3) These different seals will be compared under different pressure ratios and varying inlet preswirl ratios. The CFD prediction will be compared against bulk flow codes. 2.2 Seal Test Rig Full Scale Test Rig (ROSTR) Students: Marion Reid, Amir Younan, Tim Dimond, Tommy Meriwether Faculty: Chris Goyne, Paul Allaire, Houston Wood, Roland Krauss, George Gillies There is a need for accurate test data of annular labyrinth, hole pattern, and honeycomb seals in axial compressors and turbines. So far, many models have proven to yield very different results and must be justified with empirical data. To accommodate this need, a design for a seal test rig is being completed at ROMAC. Specifications were collected from industry, and after ten design iterations, a final conceptual design was settled upon that should accommodate most of their needs. For this final design, all mechanical parts have been outlined and justified. The drive system is in place after being refurbished and documentation is provided. Rotor dynamic analysis of rotating elements has been completed for critical elements, as well as analysis of many other components including the base which is needed to start assembly. The drive system is in place after being refurbished and documentation is provided Page 27

28 2. SEALS Seal Test Rig (Motor, Gearbox, Test rig and base) The rig will be placed in the south end of the Aerospace research lab (ARL) with the FFBTR. The oil supply for the gearbox, power supply and magnetic bearing amplifiers will all be shared with the FFBTR. Many of the modifications made to achieve this final design came from suggestions by industry members at last year s annual meeting. The new design is shown below. Test rig cross section. 1. Quill shaft flange for connection to high speed coupling. 2. Quill shaft. 3. Shear ring to lock all components in place. 4. Quill shaft angular contact bearings and bearing housing. 5. John Crane dry gas seal. 6. Quill shaft labyrinth seal. 7. Geared coupling. 8. Backup/Drop test bearings. 9. Test rig shaft bearing housing. 10. Magnetic bearing actuator. 11. Magnetic bearing target. 12. Test gas outlet. 13. Interchangeable teeth on stator (TOS) test seal section. 14. Inlet / Interchangeable swirl vane. 15. Tie bolt test shaft. 16. Interchangeable teeth on rotor (TOR) test seal section. 17. Pressure housing. Major changes include: A canned design to allow for better pressure containment, a quill shaft to minimize leakage, tie bolt built up test rotor for TOR interchangeability, shear rings for pressure containment, use of a John Crane Dry gas seal to minimize leakage, improved swirl ring to shorten test shaft. This design will allow for all components to be interchangeable by sliding out of the end of the pressure housing. Page 28

29 2. SEALS A full torsional analysis was completed for the final design and verified that the drive train would not experience torsional failure. Also, with the new shorter test shaft design, lateral critical speeds have been pushed above twice the highest operating speed. An analysis of the base design was completed using ANSYS and a contact model that included the major components of the test rig (motor, gearbox, test rig). The first problematic mode for the base and test rig is shown below at a frequency 20% above the maximum speed of the test rig. Modal analysis of test rig and base using ANSYS 2.3 Seal Test Rig Student: Max Depiro Faculty: Chris Goyne A small bench-mounted rig was constructed in 2007 to test the feasibility of Particle Image Velocimetry (PIV) in a labyrinth seal. The rig as it is today can be seen in figure 1. This rig was built to learn a great deal about this complex technique before it could be implemented on the ROSTR. Since its construction the rig has yielded many new discoveries, including the need to coat metal surfaces with a fluorescent paint, which shifts the wavelength of the incident laser light. This is extremely important because it is difficult to see particles close to solid boundaries due to reflections. Other important findings to date for the use of PIV include the choice of particles, the calibration technique, and other small factors encountered. Figure 1. The Seal Rig at the ARL PIV lab in a labyrinth seal. This is notable because PIV has never been applied to labyrinth seals with a compressible fluid. To get to this point we have simplified the seal insert so that all horizontal surfaces were flat, including both surfaces of the acrylic window shown in figure 2. We have also found a new acrylic base paint to further reduce laser reflections. Results will be obtained in the near future and be on our way getting a final design for a PIV system in the ROSTR. So far we have been able to get pressure and flow-rate data from the rig. However, a full three dimensional velocity field in the tooth cavity is a challenging problem. At this very moment however, we are extremely close to ROMAC s first successful measurements using PIV Figure 2. Seal rig close-up showing the new flat labyrinth seal insert with flat window. Page 29

30 2. SEALS 2.4 Seal Design Optimization Student: TBD Faculty: Costin Untaroiu, Houston Wood, Paul Allaire, Chris Goyne Research Professional: Alex Untaroiu An automatic approach to seal design employs advanced optimization algorithms to improve seal performance (Fig. 1). The generic geometry of seal is chosen (Fig.2) and its design process are cast as an optimization problem in the following standard form: minimize ƒ(x), with x={x 1,x 2...x n }, subjected to the constraints of the form a i x i b i ; i=1,n; where f(x) is the leakage rate selected as the objective function and x is a vector whose components are the main seal dimensions x i, selected as design variables (e.g. tooth width, cavity width, etc.) The ranges of design variables (a i,b i ) are chosen based on the manufacturing constraints. The optimum design (Fig. 2) is determined using advanced optimization algorithms, such as: Adaptive Response Surface Methodology, Genetic Algorithm, etc. To see the influence of each seal dimension on the seal leakage rate, a sensitivity study are performed using response surfaces which fit through the CFD predictions at the design points. Since current approach showed good results in the design of a labyrinth seal (Fig. 3, Romac Report no. 533, 2008), several applications in the design of seals and bearings are in progress. Mesh Generator Input files for CFD Analysis CFD Solver Proposed designs for evaluation Optimization algorithms Final Design Results of CFD Analysis Figure 1. Schematic geometry optimization of labyrinth seal design flow b t Initial design Optimized design α 1 α 2 Figure 2. Design parameters used in a design optimization study of a labyrinth seal (Romac Report no. 533) Figure 3. Optimization results (Romac Report no. 533, 2008 Page 30

31 3. FLUID FLOWS 3.1 Compressor Surge Test Rig Student: Kin Tien Lim, Se Young (Pablo) Yoon Faculty: Chris Goyne, Zongli Lin, Paul Allaire The safe operating range of pressure and mass flow for centrifugal compressors is limited by a dynamic flow instability known as surge. Surge occurs as the mass flow through the compressor is reduced to a critical point where the flow pattern becomes unstable. This critical point, called the surge point, separates the stable and unstable operating region of the compressor in the compressor characteristic curve. Surge causes largeamplitude low-frequency oscillations of pressure ratio and mass flow rate in compression systems and even flow reversal, putting a high load on the compressor impeller and casing. This causes undesired vibration, high temperature, compressor damage and lost productivity. For the experimental study of compressor surge, the Compressor Surge Test Rig was commissioned by RO- MAC as seen in Figure 1 and Figure 2. The test rig is a single-stage overhung centrifugal compressor, and the rotor is levitated on two radial active magnetic bearings (AMBs) that allow it to rotate freely without any mechanical contact. The rotor is axially supported by a thrust AMB, which also gives the flexibility to control the impeller tip clearance with high precision. The unshrouded impeller and volute of the compressor was manufactured and provided by Kobelco for this test rig. It operates with a vaneless diffuser. A modular ducting system at the compressor exhaust forms the variable plenum volume for the compression system, where the position of a throttle valve can be moved along the piping in order to change its volume. Table 1 summarizes the design and performance parameters of the centrifugal compressor. Figure 1: Compressor Surge Test Rig Page 31

32 3. FLOWS Parameter Value Maximum speed (RPM) 23,000 Design mass flow rate (kg/sec) Design pressure ratio 1.68 Inducer hub diameter (mm) 56.3 Inducer tip diameter (mm) Impeller tip diameter (mm) 250 Figure 2: Drawing of the compressor surge control test rig layout. Table 1: Compressor Design Parameters The objective of the Compressor Surge Test Rig is to develop an innovative active surge controller by taking advantage of the versatility of magnetic bearings. The general concept is to use the tip clearance variation to generate a pressure wave traveling towards the flow control valve to cancel the pressure wave generated by the initiation of surge. With measured flow information from pressure and flow rate sensors along the compressor, the proposed surge control algorithm calculates, at real time, the required impeller tip clearance variation to stabilize the compressor in the presence of surge, which is fed to the thrust AMB controller as the commanded impeller position. Assembly of the compressor was completed in 2008 and the test rig was commissioned. We are currently testing the compressor performance at different operating conditions in order to correlate the experimental setup to our mathematical model. One of the most important components of our model is the relationship between the compressor pressure ratio and impeller tip clearance. Figure 3 shows the results of the static clearance test, where we mapped the compressor characteristic curve at nominal and increased impeller tip clearance. It can be seen that as the clearance between the impeller and the shroud increases, the increased tip leakage lowers the compressor efficiency and pushes the compressor characteristic curve down. Figure 4 shows the waterfall plot of the frequency components for the compressor exhaust pressure ratio signal when the impeller was axially modulated at various input perturbation frequency. We can find in figure a predominant frequency component matching the frequency of the perturbation signal, which shows that the tip clearance has a dynamic effect on the compressor pressure ratio. Further work is needed to experimentally identify the transfer function of the system from impeller clearance to compressor pressure ratio, and fully validate its dynamic characteristics by comparing it to theoretical predictions. The dynamics of the compression system was tested in surge condition, where the mass flow through the compression system was kept below the surge point by closing the throttle valve. Figure 5 shows the compressor pressure ratio and mass flow rate measurement during surge, and a very dominant oscillation of about 7Hz characterizes both measurements. This frequency matches the frequency generally found in deep surge, and agrees with the frequency predicted by the mathematical model. The effect of surge on the compressor states can be seen clearly in the waterfall plot of the frequency components of compressor pressure ratio at different mass flow rates in Figure 6. We can see in the figure that, at flows rates below 0.5kg/sec, a dominant component appears suddenly at 7Hz, which matches the expected frequency of the surge oscillation. Page 32

33 3. FLOWS Figure 3: Compressor characteristic curve at various tip clearance. Figure 4: Frequency components of compressor pressure ratio at various impeller perturbation frequencies. After completing the tasks required to match the mathematical model of the compressor to our experimental setup, we plan to start the implementation of the surge controller. With the collected experimental data of the compressor characteristics, we will update the compressor model and surge control algorithm. We are continuing our effort to derive a robust surge controller that stabilizes the compressor system beyond the stable operating range. The derived controller will be implemented in our test rig, and its effectiveness to mitigate compressor surge will be tested. pressure Mass flow Surge initation Figure 5: Compressor exhaust pressure and mass flow rate oscillations during surge. Figure 6: Frequency component of compressor exhaust pressure at various mass flow rate. Page 33

34 4. ROTORDYNAMICS 4.1 Rotordynamics Students: Amir Younan, Jawad Chaudhry Faculty: Pradip Sheth, Paul Allaire The RotorDynamics activity in ROMAC includes a number of interrelated efforts. The following figure illustrates the capabilities of the available RotorDyanmics software tools from ROMAC these are the programs which have been developed, refined, and matured through many years of industrial and academic usage: Figure 1: Dynamic Software Research The ROMAC RotorDyanmics research and development activities can be described by the three pronged approach illustrated in the following diagram: Figure 2: RotorDynamics Software Tools Page 34

35 4. ROTORDYNAMICS The production use software is the software utilized by the ROMAC members, and includes such existing ROMAC software tools as: Lateral Stability FORSTAB, ROTSTB Lateral Forces Response FRESP2, RESP2V3 Lateral Critical Speeds MODFR2, CRTSP2 Torsional Transients TORTRAN3 Torsional Critical Speeds TWIST2 Lateral Transients COTRAN Balancing BALOPT Component Modeling such as THPAD, THBRG, MAXBRG, and SEAL3 for creating bearing and seal coefficients for rotordynamics These are just a few examples of the production software available to ROMAC members. Detailed User manuals and theoretical background are documented in the ROMAC library available on the web and on a CD. These production programs are the primary tools which are continuously updated and maintained by ROMAC. Elsewhere in this newsletter, Amir Younan and Jawad Chaudhry describe the continuing developments of these production tools. In addition, a fully integrated system called COMBOROTOR is under development, with a Beta version already released, which will combine the capabilities of all of the different codes indicated above. New and emerging issues in RotorDynamics have motivated new developments at ROMAC this year. These motivating issues include: Need to continuously correlate and calibrate the production tools Larger, more detailed models including foundation support structure Gear drives leading to coupled torsional/lateral vibrations Developments for Model reconciliation procedures Nonlinearities, including ability to analyze large orbits and limit cycles Model based diagnostic systems Modeling tools for repair/replacement of subsystems in existing machinery Need to investigate and implement Model reduction schemes for specific applications To address these issues, a research platform in MATLAB called MatlabRotor is continuing to be developed to experiment with new ideas that have not been investigated before. MATLAB is chosen for its extremely efficient and easy to use built in functions for numerical procedures for large and complex matrices, a very large selection of numerical integration algorithms for differential equations, and excellent visualization tools. The entire system is written in a compact and general form with explicit sub functions that are called during the modeling of the system. This platform is being utilized to develop more advanced and detailed models for the RotorDynamic systems and is described elsewhere in this newsletter. Page 35

36 4. ROTORDYNAMICS 4.2 Coupled Later-Torsional-Axial Rotordynamic Analysis (Combo rotor/matlab rotor) Students: Jawad Chaudhry, Amir Younan Faculty: Pradip Sheth, Paul Allaire Project Overview This project aims to advance the ROMAC rotordynamic software in several areas. First, it combines the capabilities of the most popular rotordynamic codes, including CRTSP2, TWIST2, ROTSTB and FORSTAB, into one code. Therefore, it will make it much easier for the user to conduct a comprehensive rotordynamic analysis. Second, it employs the finite element method that is more capable and reliable than the transfer matrix method used by some existing codes. Third, the tilting pad bearings and flexible support are modeled in the time domain, which eliminates a search that can potentially lead to missing modes. In addition, the project will provide the capability to analyze coupled lateral-torsional-axial vibration that exists in some geared systems. This new steady state computer program is continuing to be developed and will be the main Figure 1: Critical Speed Map solver behind the new ROMAC GUI. A Fortran version (Combo Rotor) and a Matlab version (Matlab Rotor) are being developed. Progress in the Past Year The beta versions of Combo Rotor and Matlab Rotor have been released and continue to be tested and confirmed with the production codes. Combo Rotor is currently incorporated in the old GUI (shell) and uses TECPLOT for its post-processing graphics. Work on coupled analysis in Combo Rotor is making good progress. The Combo Rotor model can be used by Matlab Rotor and uses Matlab visualization tools for post-processing graphics. Future Work The final code with full capability of lateral-torsional-axial analysis with fully tested and verified results will continue to be developed. Options to easily manipulate the model to compute coupled or independent analysis, such as lateral only or torsional only will be available. Page 36

37 4.3 MATLAB Rotor Student: Jawad Chaudhry Faculty: Pradip Sheth, Paul Allaire MATLAB ROTOR Code is a finite element based code that presently performs time-independent, forced response rotordynamic analysis. MATLAB ROTOR was created as a Research platform to develop, test and incorporate new tools for rotor dynamics within ROMAC research as dictated by the industrial members of ROMAC. The code is also released to ROMAC members who may be interested in utilizing the MATLAB environment for rotor bearing dynamics. New developments are prototyped, tested and validated by this code before migrating to the Production Codes. The present version of MATLAB ROTOR code combines most of the capabilities of current production ROMAC codes such as CRTSP2, ROTSTB, RESP2V5, TWIST2, and FORSTAB. Since the MATLAB ROTOR code is developed entirely within MATLAB, the numerical methods and visualization tools embedded in MATLAB functions are utilized. The embedded, built-in functions of MATLAB are kept up to date by MATHWORKS in their new version releases and one of the key objectives of MATLAB ROTOR is to rely on these externally developed tools for focusing on the rotor dynamics issues. MATLAB ROTOR code is written in an open source format such that the interested ROMAC members can utilize the current version and do customization, modifications, and additions that may be specific to their rotordynamic problems and requirements. The first version of MATLAB RO- TOR was released in June The code is compatible with Combo Rotor and utilizes the same model generated in the GUI. Additional exclusive features that are only available in MATLAB ROTOR (i.e. mode tracking) are included in the internal MATLAB ROTOR GUI. Current Capabilities of MATLAB Rotor Full 12 DOF beam element with axial, torsion and bending capabilities; Includes gyroscopic, rotary and shear deformation effects; each node thus has 6 DOF Reduced bearing stiffness and damping coefficients e.g., from THPAD or MAXBRG Steady State Synchronous Response Campbell diagrams Seals and Squeeze Film Damper Coefficients Coupled or Reduced Uncoupled analysis 2D/3D elliptical whirl mode shapes and Animations Multiple-level Rotors Mode Tracking Current Progress: More detailed and accurate models for RotorDyanmic systems have always been one of the goals of ROMAC and MATLAB Rotor is the best platform to test new methods and attempt to achieve these goals. The options to include more detailed models for the disks and blades will create models which will give the user a much better understanding of the dynamics of the system. Full 3-D models of the disks and blades using solid elements are being developed. 3D meshes to fully incorporate the shape of the disks and blades will seemingly merge with the beam models of the rotor. Page ROTORDYNAMICS Top: Figure1- Tracked modes for the 3-D Campbell Bottom: Figure2-3D Mesh of the Disk

38 4. ROTORDYNAMICS 4.4 Engine Balancing Test Rig Work is being done in conjunction with several ROMAC members, notably Pratt & Whitney and Boeing, to investigate one of the main sources of noise transmission the unbalance forces transmitted to the engine casing via the bearing system. The bearing system includes rolling element bearings combined with squeeze film dampers (SQD). To investigate this problem in detail, the bearing test rig shown below, will be converted to an engine balancing test rig. The new rig will be developed with sufficient versatility to explore squeeze film damper optimization, confirm engine balancing theory as developed, and validate predictive computer codes. The specific objectives for the work are: Design a rig that could experimentally validate the unified balancing theory that is being developed as a part of ongoing research at ROMAC. Figure 1: overview of the existing bearing test rig to be converted into the Engine Balancing Test Rig Design an SQD similar to those employed in modern aircraft engines to be used in this rig that will facilitate the measurement and optimization of reductions in noise transmission. Determine the necessary instrumentation to fully document system performance during startup and lift off, during steady state operation, and during off normal operation such as out of balance situations. At this point, a design has been completed to convert the existing rig into a test rig with either two or three bearings composed of rolling element bearings inside of squeeze film dampers, or conventional fluid film bearings. In the expected configuration, two of the bearings will be located near the existing locations, as shown in fig. 2 below. Provision for adding a third bearing, closer to the bearing furthest from the motor, will also be made to simulate the mid-shaft bearing that is present in some engines. The new balancing rig will be set up so that any of the three bearings can be configured as either a rolling element bearing with a squeeze film damper, or as a tilting pad fluid film bearing. The objective of having the capability of testing a set of tilting pad bearings is to compare the balancing of the rotor with the aircraft bearing configuration to the ground based system using fluid film bearings. An overview drawing of the entire test rig, configured with 2 SQDs is shown in fig. 2. Figure 2: Depiction of the Engine Balancing Test Rig assembly showing the fluid damper bearings, magnetic exciters, and unbalance disks in the center. Page 38

39 4. ROTORDYNAMICS The rotor shaft for the rig will be about 6 feet in length with a weight of approximately 200 pounds, with bearing spacing ratios that conform to typical engines of interest. The design, as shown in cross section in fig. 4, has a mounting system that is composed of a double vertical I beam. The I beam webbing will be designed to produce an appropriate level of stiffness that can be well characterized. The I beam will be 8 tall. This will allow the centerline of the rotor shaft to be far enough from the concrete base to permit a disc, of up to 12 in diameter, to be mounted on the end of the shaft and overhung over the last bearing as shown in figure 3. This disc can simulate the engine fan, allowing for investigation of out of balance problems at this location, as well as balancing that must be done at this location to correct imbalances located elsewhere in the rotor. Figure 3: Engine Balancing Test Rig shown with the unbalance disk on the side opposite from the motor. This disc will simulate a jet engine fan. Figure 4: Cross sectional view of the support spacers of varying levels of stiffness, may be fit between the double I-Beam support and the clamps to allow for variable ranges of support stiffness Stiff clamping rails, as shown in fig. 4, are located on either side of the I beam mounting system. There is normally a space between the mounting structure and these rails, so the I beam can vibrate in accordance with the response of a mounting system in an actual plane. If a stiff mounting system is desired, spacers can be bolted in between the stiff side rails and the I beam, making the entire mounting structure rigid. Replacing the stiff spacers with more flexible materials of known levels of stiffness provides the capability of simulating mounting systems of varying natural frequencies. The masses installed on the rotor will allow small weights to be attached to simulate a variety of imbalance conditions. In addition, three magnetic shakers will be located as shown in the diagrams to produce a variety of complex vibration patterns which can be programmed in to act on the rotor over time, including the ability to simulate aerodynamic and gyroscopic effects. As noted earlier, the SQD is directly in the noise transmission path. However, the pressures and vibration amplitudes of the shaft in the SQD are not well characterized in most engines. Thus, it is difficult to understand the noise transmission details through this component of the total path. Details of the SQD design are presented in fig. 5, showing a sectional view of the SQD and giving the locations of pressure and position transducer taps in the squeeze film. Page 39

40 4. ROTORDYNAMICS As shown, the squeeze film damper design includes a central oil groove with O ring end seals, but other configurations can be mounted in the bearing housing as well. The design will also allow for various centering springs, or no centering spring at all. The SQD will be 1.75 long (L = 1.75 ), with an OD of 6.5 (D = 6.5 and r = 3.25) and a clearance of (C = ). This will give an L/D of 0.27, and a C/r of One of the purposes of this rig will be to measure the noise or vibration attenuation accomplished by the squeeze film dampers and Figure 5: Section drawing of the SQD balancing. To measure the noise transmission, load cells will be mounted between the bearing housings and the I beam mounting system, and between the mounting system and the steel plate that makes up the base of the Engine Balancing Test Rig. The rig will allow for multiple bearing configurations to be tested on the rotor. One of the initial configurations will involve a squeeze film damper bearing with a squirrel cage centering spring attached. Ones without centering springs will be tested later. Each squeeze film damper bearing will have two position sensors mounted vertically and horizontally over the edge of the inner ring of the bearing. There will also be an arrangement of six pressure transducers, arranged in 2 groups of 3 sensors, at 2 axial positions as shown in fig. 5. This will provide a comprehensive look at the radial and axial pressure distribution that exists within the squeeze film. A full data acquisition system will be employed to capture, process, and store the information for the array of pressure, position, and load cell sensors. In summary, the engine balancing test rig will be a sophisticated platform to test a variety of phenomenon related to noise transmission in modern aircraft structures. The Engine Balancing Test Rig will be an invaluable resource for testing the unified balancing theory as it develops, for experimentally determining the design optimization process for squeeze film dampers, and for validating computer codes that predict squeeze film and rotor dynamic behavior. To our knowledge, there is nothing comparable to the proposed rig, and its capabilities will allow it to test a wide range of problems that are confronting engine designers and manufacturers today related to the reduction of noise transmission. 4.5 Rotordynamics Integration with NASTRAN Students: Jawad Chaudhry Faculty: Pradip Sheth, Paul Allaire This is a collaborative effort with MSC Software, Boeing, GE Aviation and other companies to use NASTRAN and other existing codes combined with the ROMAC codes to analyze very complex rotors such as gas turbines. Integration of ROMAC THPAD code has made major progress. ROMAC members will be able to use the THPAD to analyze the bearings in their NASTRAN model. The new squeeze film damper code will also be integrated in the same manner. Page 40

41 5. OPTIMIZATION 5.1 Optimization of Rotor-Bearing Systems Student: Amir Younan Faculty: Costin Untaroiu, Paul Allaire Research Professional: Alex Untaroiu Design of modern rotating machinery has become more challenging due to continuing and ever increasing demands of lighter weight, higher speeds, and greater reliability of the rotor-bearing systems. Many parameters which have a significant influence on the dynamics of the rotor-bearing system (e.g. bearing characteristics, inertial and stiffness distribution of the rotor) open the possibilities for application of design optimization methods. While the ROMAC codes are usually employed in the rotor-dynamic analyzes of the rotor-bearing systems, this study is attempting to extend their applicability to optimization studies of a rotor-bearing system design, which may complement/replace the traditional design approach based mainly on the experience and intuition of rotor-dynamic analyst. A good design of a rotor-bearing system should satisfy a significant number of requirements (stability, unbalance sensitivity, bearing performance factors, etc) which are usually defined as constraints or objectives in the optimization problem. Several geometric characteristics of the rotor together with the parameters defining the configuration of bearings are considered as design variables into the automatic optimization process. The analysis of the bearings and the whole rotor-bearing systems are employed using ROMAC codes (e.g. THPAD and Comborotor). Specific subroutines are developed to parse the data of interest from the output files of ROMAC codes, and then to transfer it to a commercial optimization software (Fig. 1). The complex design optimization problem is solved using heuristic optimization algorithms, such as genetic, particle-swarm optimization, or simulated annealing. In addition to finding the best design in the chosen design space, the results of iterative optimization process can be used in sensitivity and trade-off studies (Fig. 2) which are very useful in the design process. ROMAC Codes (e.g. THPAD, Comborotor) New designs for evaluation Optimization algorithms Results of Rotordynamic Analysis Lower limit Pareto front Final Design Figure 1. Automatic optimization of rotor-bearing design Figure 2. The trade-off of power loss in bearing versus the minimum damping ratio (ROMAC Report no. 538, 2009) As this research further develops, we anticipate developing an optimization code integrated with the other ROMAC codes (e.g. bearing, seal and finite element codes). Page 41

42 Y Y Y 6. GEARS 6.1 Geared Systems Students: Amir Younan, Jawad Chaudhry Research Scientist: Major Blake Stringer Faculty: Paul Allaire Expanding on the 12-dof gear-mesh stiffness model discussed in the last newsletter, US Army Major Blake Stringer continued his research in drive train technology by developing a rotor dynamic finite element model of a helicopter transmission. The transmission is a primary source of noise, vibration, and component fatigue in rotary-wing aircraft. This model determined the response to various forcing functions and could use both linear and nonlinear mesh stiffness approximations. Common gear and bearing transmission faults were seeded into the model to observe the response differences between normal and various abnormal operating conditions, such as gear contact surface wear, pitting, spalling, bearing failure, etc. It is hypothesized that such distinctions could be captured in a health conditioning database and compared to near real-time signals from aircraft data sensors to determine the amount of useful component life remaining, thereby maximizing the life of the component. output carrier planetary mesh ring gear spiral bevel mesh planet gear spiral bevel gear input sun gear spiral bevel pinion OH-58 Transmission Schematic (Left) and Disassembled (Right) X Shaft A Location (mm) 0 0 X Shaft B Location (mm) 0 0 X Shaft C Location (mm) normal conditions contact wear conditions normal conditions contact wear conditions Torsional Mode Shape Torsional Mode Shape Torsional Mode Shape Amplitude (Input Node) Amplitude (Carrier Node) Shaft A Location (mm) Shaft B Location (mm) Shaft C Location (mm) Frequency (ω) Frequency (ω) Transmission Shaft Mode Shapes (3417 Hz) Frequency Responses at Input Node (left) and Carrier Node (right) Page 42

43 7. MAGNETIC BEARINGS 7.1 AMB Test Rig for Aerodynamic Cross-Coupling Students: Simon Mushi (ECE) Research Professional: Wei Jiang (MAE) Faculty: Paul Allaire (MAE), Zongli Lin (ECE) A test rig with a 1.23 meter long flexible rotor completely supported on active magnetic bearings (AMBs) has been constructed to study aerodynamic cross-coupled stiffness forces and rotor drops on auxiliary bearings. Fabricated, assembled and balanced at Rotating Machine Technology Inc., this machine is supercritical as the maximum running speed of 14,200 RPM is above the first bending mode. In one configuration, the two outermost AMBs attempt to stabilize the rotor position while the mid-span or quarter-span AMBs generate cross-coupled stiffnesses and other destabilizing forces typical of a seal or impeller (see Figure 1). Presently we can generate up-to 15 MN/m of destabilizing cross-coupled stiffness at the mid-span AMB. The initial development of dynamic models of this plant and synthesis of various AMB control algorithms has been successful. Preliminary rotor levitation was performed in fall 2008 without the connection of the motor. Model refinement and uncertainty characterization through MIMO system identification is now underway and commissioning of the test rig will be complete by the end of spring Figure 1: Front view of the test rig showing the shaft and mounted discs: the drive-end and nondriver end support bearings and the mid-span and quarter-span disturbance bearings. The control PC and hardware rock is visible in the back. Project objectives: 1. Provide a platform to study some challenges faced in industry: (a) successful recovery of rotor position after auxiliary bearing contact; (b) reducing the effects of speed-dependent gyroscopic behavior and synchronous unbalance on performance; (c) reducing the impact of aerodynamic cross-coupled stiffness on machine stability; 2. Apply modern control synthesis techniques to stabilize a flexible rotor on AMBs subjected to disturbances; 3. Model different rotor-bearings configurations (e.g., between-bearing or over-hung configurations); 4. Serve as an educational tool for researchers and industrial members of the ROMAC Industrial Program. Page 43

44 7. MAGNETIC BEARINGS Test rig details As mentioned earlier, a 1.23 meter steel shaft weighing 440N is the main rotor in this test rig. A Rexnord Thomas miniature flexible disc coupling is used to attach this shaft to a shorter drive shaft holding the timing belt pulley. The coupling will reduce the effect of angular and offset misalignment on machine operation. The slave pulley runs on a pair of rolling element bearings (Timken S8KDD). The motor is a 5 HP Toshiba AC induction motor that is inverter driven to 7,200 RPM. In addition to the AMB lamination journals on the rotor, a large 9 diameter disk is also mounted to increase the gyroscopic character of the rotor. figure 2 and fig. 3 show the locations of the various parts of the test rig. Barely visible to the left of the coupling in fig. 3 is one of the backup rolling element bearings used to protect the AMBs. While the rotor- AMB air gap is (0.38 mm), the clearance between the rotor and the backup bearing is half of this i.e., 0.19mm. This will protect the AMBs from impact in the event of power failure or controller overload. Presently, NPB 6011 ball bearings are used, and the test rig design allows for their easy access and replacement. Figure 2 : Overhead view of the test rig clearly showing the AC motor Figure 3 : Close-up of the Motor and drive system Magnetic Bearings The AMBs used for this test rig have average outer and inner diameters of 7.7 and 3.6 respectively (see figure 4). The average axial length is 1.8 and the air gap is M-15 Silicon Steel is used for the AMB and rotor lamination stacks. The maximum predicted load capacity is 200lb-force per bearing. The bearings were inherited from a previous test rig in which their power loss under various conditions was investigated. Each AMB control quadrant is driven by a single power amplifier. Copley Controls Model 413 and 422 analog amplifiers are operated in torque mode (also known as current or trans-conductance mode) to supply the necessary current at either 75V or 160V (see figure 5) to meet our slew rate requirements. The bearing saturation current is 16.4 A which is well within the amplifier operating window. Figure 4: Three of the four AMBs used in the test rig Page 44

45 7. MAGNETIC BEARINGS Figure 5 : Power amplifiers to drive two AMBs Control and Instrumentation Hardware The test rig is fully instrumented with the following sensors: 1.Bently Nevada eddy current proximity probes (4 per bearing : total 16), 2.Optical keyphasor for speed and phase observations 3.LEM current transducers to monitor current in each support AMB coil (total 8). Custom electronics are used to perform signal conditioning i.e. gain/offset, galvanic isolation, anti-alias filtering, and ADC input buffering where necessary. High resolution 16-bit data converters are used to interface the analog sensors and actuators with the digital control system. A powerful DSP controller (Innovative Integration M6713) is used to execute controller code in real-time and provide user operation of the rotor-amb system. The custom graphical user interface allows the user to: 1. Observe rotor position and AMB coil currents at all four bearings in real-time 2. Adjust AMB bias currents 3. Select disturbance type, amplitude, frequency and phase for disturbance AMBs 4. Generate on-demand FFT to observe vibration spectra 5. Log the above data to a file for later processing 6. Control the AC motor Various modern state space control algorithms such as Linear Quadratic Gaussian, H-infinity and musynthesis are being proposed to stabilize the rotor over its entire speed range and with varying crosscoupled stiffness forces. The latter two approaches allow us to capture plant uncertainties such as speeddependent gyroscopics, uncertain mode shapes and unknown cross-coupled stiffness directly into our control problem and solve for less conservative, high performing controllers. Unbalance rejection algorithms that go beyond the ad-hoc inclusion of notch filters are also being tested. Heavy use is made of MAT- LAB and its Robust Control toolbox perform the above analysis and synthesis of robust controllers. An orbital simulation of the rotor position at 14,000 RPM with 5 MN/m of cross-coupled stiffness at the midspan AMB is shown in figure 6. The controller is able to keep the bearing centered within the 0.2mm backup bearing clearance (the red line). The unbalance rejection algorithm is not running hence the oscillating position of the rotor center. Page 45

46 7. MAGNETIC BEARINGS Figure 6 : Simulated rotor orbit at 14,000 RPM with a mu-controller and mid-span cross-coupled stiffness of 5 MN/m Continued work is being performed on various methods of experimental determination of the plant (rotor + AMB) model. We have performed standard rotor impact hammer tests to determine the bending natural frequencies, and attempt to reconcile the differences with the rotordynamic model (see fig. 7). Interference fits and simple modeling of the lamination stacks result in a rotordynamic model which under-predicts the bending frequencies. Single-input, single output (SISO) frequency response data was collected and attempts were made to extract modal data. While we could precisely determine the location of the first two bending frequencies, we were unable to sufficiently resolve their amplitudes and the mode shapes. Robust control design of a flexible lightly damped plant such as this rotor-amb system necessitates accurate knowledge of this modal data. Therefore, with the rotor levitated we are in the process of performing online multiinput, multi-output (MIMO) system identification. This approach uses the AMBs to excite the shaft while measuring the position and coil currents to accurately determine modal information, and AMB parameters. Our twin goals for spring 2009 are to complete the system identification of the rotor-amb plant and test controllers that have been developed at speed on this test rig. Further work is planned on non-linear control algorithms and the extension the digital controller development procedure to achieve full integration with Matlab. This will allow hardware-in-the-loop testing and provide a unique platform for continued AMB research and education in ROMAC. Page 46

47 7. MAGNETIC BEARINGS 7.2 Rotor Drop Analysis on Auxiliary Bearings ROMAC Engineers: Amir Younan, Tim Dimond Faculty: Paul Allaire The purpose of this research project is to design a suitable auxiliary bearing system for the fluid film test rig. The work includes a time transient analysis for the rotor response during the loss of the magnetic bearing support. The rotor along with the lamination was simply modeled as a three mass rotor. The analysis started from a steady state condition of the rotor under an unbalance force. At the drop time, the magnetic bearing coefficients were set to zero and the transient response of the rotor inside the bearings is monitored. The contact between the inner race of the auxiliary bearing and the shaft was modeled using the Hertzian contact Theory. The contact forces were calculated and monitored with time. Figure 1 shows the preliminary (left) and final (right) design of the auxiliary bearing. Figure 1: Preliminary and final design of the auxiliary bearing for the fluid film test rig 7.3 Matlab E-Core Magnetic Bearing Design Tool Students: Thomas Meriwether Lab Engineer: Tim Dimond Research Professional: Bob Rockwell Faculty: Paul Allaire A new design tool will soon be added to the list of ROMAC software tools. The primary purpose of the Magnetic Bearing Design Tool (MBDT) is to decrease preliminary design time. The program will be utilizing the fmincon function from the Optimization Toolbox in Matlab. The user will enter various input parameters such as a target force, material choice properties and the program will then optimize bearing size based on a simple circuit model and physical bearing constraints. Following completion, the code will then be compared against a more formal FEA analysis. Page 47

48 8. Graphical User Interface 8.1 RotorLab + Development Faculty: Bob Ribando Research Professional: Sean Travis RotorLab+, the working environment intended to replace the existing RotorGUI and RotorLab tools, has a new look. Expanding from its predecessor, the new interface allows analysis of any number of assemblies and components within the same project. The new interface also streamlines the workflow with features such as analyzing components and assemblies independently, showing component data in only one window per component, and automatic synchronizing of data in all windows. We have created a special collaboration web site, "RotorLab+ E-Adopters", for those ROMAC members interested in helping in the development of RotorLab+. As substantial changes and additions are made, early adopters get the opportunity to test out the latest version and provide feedback. If you want to join the early adopters group, please Bob or Sean. Existing features: Open and save a project. Project contains any number of assembly and component workspaces. Project tree view shows a hierarchical view of project containing the workspaces, components, and analyses, allowing for quick navigating and status viewing. "Factory" window allows creating and selecting workspaces, components, and analyses. Assembly workspace is a window containing: 3D view of assembly. Property panes for assembly components. Setup and results panes for assembly analyses. Workspace options and assembly summary. Component workspace is a window containing: Property panes for components. Setup and results panes for component analyses. Workspace options. One assembly analysis, Critical Speed Map (via CRTSP2), accessible; more will be added soon. Planned features: 3D view of component. Quick duplication of workspace, of component, and of analysis. All ROMAC analyses (codes) accessible. Import and export of data to and from Excel. Printing of analysis reports and graphs. Figure 1: The upper-left pane shows an outline of the various parts of the project (called Compressor Example in this screenshot). The right main area shows a tab for each workspace one assembly workspace and three component workspaces. Several screenshots of RotorLab+ as it exists in February 2009 are included in this article. All development is taking place in VisualBasic.NET with DirectX for high quality graphics rendering. Page 48

49 8. Graphical User Interface Figure 2: In the Assembly Workspace 1 tab, the top area shows a graphic view of the assembly. In the bottom section are tabs for the components, analyses, and various workspace options. Shown here is the Shaft 1 tab where the shaft data is entered. Figure 3: In the same Assembly Workspace as Figure 2, a Critical Speed map analysis has been added. Its Setup tab is shown here. After entering the data in the Setup, the analysis can be run from the Analysis menu or from a right-click context menu.

50 9. MAGNETIC BEARING SOFTWARE 9.1 Safety Critical and Fault Tolerant Operation of Magnetic Bearing Software. Student: Patrick Gradon Faculty: John Knight The digital control of magnetic bearings relies upon computers that perform the required control calculations in hard real time. These computers consist typically of complex software operating on an embedded, fault-tolerant processor. In many magnetic bearing applications, the failure of the bearing system can have serious consequences. The failure might lead to loss of control of a rotating shaft thereby necessitating emergency repair or replacement, or even destruction of the equipment within which the magnetic bearings operate. In such applications, it is necessary to design the bearing system so as to reduce the risk associated with failure to an acceptable level. Since correct operation of the system depends in part upon correct operation of the software, that software must be built in such a way that the system can justifiably depend upon it. We have developed a process for the simultaneous construction of computer software and a detailed, comprehensive argument that the software is sufficiently dependable for its intended applicaton Computer scientists have developed a wide variety of techniques for developing software and for ensuring its correct operation in various specific regards. It is possible, for example, to specify using a precise mathematical notation, what the output of a computer program must be and then to prove that the step-by-step algorithm implemented in a given program always produces this output. Researchers have also developed automated tools that can prove that a given program is free of certain classes of defects such as division by zero or arithmetic that might produce a value too big to be stored in the designated area of memory. Engineers building software for a magnetic bearing application must select a combination of these techniques that will produce software that is sufficiently dependable for that application. To do this, they require the guidance of an engineering process. As part of a computer science research program, we have developed a process for the simultaneous construction of computer software and a detailed, comprehensive argument that the software is sufficiently dependable for its intended application. We refer to this process as Assurance Based Development (ABD). We anticipate that ABD is an ideal approach for the development of magnetic bearing control software that is ultra dependable and therefore suitable for long-term use in critical magnetic bearing applications. Combined with the use of appropriate safety engineering techniques at the system level, such as system hazard analysis and fault tree modeling, we anticipate that the use of ABD will yield demonstrably adequate control software for magnetic bearing applications. In the long term, we plan to develop a pattern that can be followed to build ultra-dependable control software for general magnetic bearing applications. Page 50

51 10. HEARTPUMP 10.1 Redesign of Heart Pump Motor Research Professional: Wei Jiang Faculty: Paul Allaire The original design of heart pump motor uses the same stator core as magnetic bearing to reduce the cost, but actually windings are the major cost for the stator cores. The design generated a motor with rated power about 13 watts. The downside for the design is: large negative radial stiffness, stator core loss, high manufacture cost. The coreless design will generate less negative radial stiffness, easy to manufacture stator, smaller stator size; the downside is the rated power is lower, with about 6.28w at maximum efficiency. Power output can go higher with the cost of efficiency. The stator outer diameter for the initial coreless design is 30mm, which is down from 38mm of previous design and axial length is 14mm, down from 18mm of previous design e+000 : >1.323e e+000 : 1.257e e+000 : 1.191e e+000 : 1.125e e-001 : 1.059e e-001 : 9.924e e-001 : 9.262e e-001 : 8.601e e-001 : 7.939e e-001 : 7.277e e-001 : 6.616e e-001 : 5.954e e-001 : 5.293e e-001 : 4.631e e-001 : 3.969e e-001 : 3.308e e-001 : 2.646e e-001 : 1.985e e-002 : 1.323e-001 <9.524e-007 : 6.616e-002 Density Plot: B, Tesla The coreless motor is designed to be a sine wave PMSM machine. The torque/emf profile: The red line is ideal sine, the blue line is the FEM torque result, and the green line is error between the two. Page 51

52 10. HEARTPUMP The basic type of losses in coreless motor: 1. Stator metal shell loss caused by rotation rotor magnets. 2. Stator copper resistive loss. 3. Stator back iron hysteresis loss. 4. Rotor metal shell and PM eddy current loss. Item 3 and 4 are much smaller in coreless design compare with old design, due to the flux reduction generated by stator coils. Item 1: Stator Metal shell loss, speed, flux, axial length of rotor P Shell = 2 2 KB max ω FEM analysis shows the following result: (0.5 Tesla, 6000 rpm) Torque (Power Equation) τ = r J B dv 2D FEM model is used to optimize the motor design. The τfem torque is calculated for 1 A/mm 2 current density, and 10 mm axial length. The torque is proportional to current density, and axial length. So the output can be written as: Page 52

53 10. HEARTPUMP P out = 3 L( ) τ mm FEM J peak ω 2 10 The copper loss is related to the coils volume and current density. P copper 2 ρj peakvcoil = 3 pack _ factor ( x)/x 2 /y (x-3.18) y 2 V coil can be parameterized with axial length V coil = A ( 2L + 2d1 + π d ) coil Using the constraint of output power to be 6000 rpm-9000rpm With each coreless stator design, we can find the function of total power loss with respect to axial length and speed (6000rpm-9000rpm) y x The follow diagram is the power loss as parameter of speed (1-1.5 represent 6000rpm-9000rpm) and axial length (6mm-10mm). Best efficiency over the full speed range is found at axial length around 8 mm. Consider two end turns give total 14 mm motor axial length. Compare to the old design: The size reduces from 38mm diameter 18 mm axial length to 30 mm diameter and 14mm axial length. The negative stiffness cause by motor reduces from -7.86N/mm to -3.4N/mm. The efficiency of the motor is almost the same. New design power loss is about W from 6000rpm to 9000rpm. The old design copper plus stator eddy current loss is about w from 6000rpm to 9000rpm. Adding the unaccounted stator magnet core loss (which is much higher in old design), and eddy current loss of the rotor (which is also higher in old design) should make the two very close in efficiency. Compare of core loss: Old design stator core volume 3517 mm 3, max flux 1.2T New design stator core volume 1407 mm 3, max flux 1.1T Page 53

54 10. HEARTPUMP 10.2 Life Flow Ventricular Assist Device Students: Arunvel Kailasan, Brady Bolton, Simon Mushi, Patrick Graydon, Brad Nichols Research Professionals: Alex Untaroiu, Wei Jiang, Kim Wasson Faculty: Paul Allaire, Houston Wood, Zongli Lin, John Knight Collaborators: Curt Tribble, Kent Harmon, Charles Klodell, Charles Hobson, Stephen Evans (University of Florida-College of Medicine); Shashank Desai, Nelson Burton, Tonya Elliott, Lori Edwards (Inova Fairfax Hospital-Inova Heart and Vascular Institute) Recent work on the heart pump has been focused on optimizing overall performance by reevaluating the design of individual components. In particular, recent efforts have been put into verifying the magnetic performance of the heart pump using a finite element model. As a result of this analysis some important control parameters have been tweaked and the motor is being redesigned with a smaller footprint to achieve higher efficiency. In addition, as we move closer to a marketable product we are looking more closely at materials and the manufacturing techniques. Of particular importance to the function of the magnetic bearing controller are several physical parameters including the bias flux and magnetic stiffness. However, these parameters are very difficult to measure accurately on the physical rig. While two dimensional finite element models and basic linear magnetic circuit analysis are good at providing rough estimates of these parameters, they lack the accuracy desired at this stage in the design process. Thus, a full three dimensional nonlinear finite element model of the magnetic assembly was constructed in ANSYS Workbench to accurately predict these parameters and reevaluate our design. The model was first constructed without the magnetic coils in order to measure the bias flux and stiffness due to the permanent magnets alone. Later the coils were added to get active flux and stiffness measures. Figures 1-3 show the solid model, mesh (1/6 slice), and the resulting flux density respectively. Figure 1: Solid Model Figure 2: Mesh Figure 3: Flux Density The results showed some small discrepancies between the parameter values currently in use and those predicted in ANSYS. By implementing these new parameter values we hope to further optimize the controller performance. As a direct result of including a self sensing magnetic bearing design, the flow path of the pump has been Page 54

55 10. HEARTPUMP redesigned. The new design configuration of the pump is more compact and the overall length is reduced by 25% as compared to the previous design. Since this device is intended to be an implantable ventricular assist device, these two features create potential for a better anatomical fit while reducing the overall invasiveness of the implant. The LifeFlow s current design has the capability of providing flow rates from 2 to 8 lpm over physiologic pressures, for rotational speeds of the impeller varying from 5,000 to 9,000 rpm. A plastic pump prototype of the LifeFlow has been constructed (fig. 4) in order to assess the flow performance of the pump and to validate the CFD prediction of main flow parameters. The plastic pump prototype does not include magnetic system components, thus, the impeller is shaft-driven. The fluid test rig illustrated in fig.5 was constructed to facilitate flow measurements on the plastic prototype. Flow measurements demonstrated a very good agreement between the experimental data recorded for pressures and flow rates and the CFD estimations. Figure 6 is a plot comparing the pump-flow curves of the computational model to the pump-flow curves of the experimental test rig. The data taken from the experimental test rig corresponds closely to that of the computational results. The largest percent difference in the CFD and experimental data is approximately 10.4 percent and that is found on the plot where the pump speed is rotating at 7000 RPM. Figure 4: Plastic Pump Prototype Figure 5: Fluid Test Rig In parallel with the flow path re-design and testing, a magnetic suspension test rig was constructed in order to test and validate current suspension system design (fig.7,8). The magnetic suspension redesign and testing is currently underway. As soon as the experimental testing of both main components of the pump is concluded, the two will be combined and a magnetically levitated pump prototype will be build for in vivo testing. Figure 6: plot comparing the pump flow curves of the computational model to the pump flow curves of the experimental test rig. Full Suspension Figure 8: LifeFlow System - assembly and cut-away views of components Figure 7 Page 55

56 11. OTHER RESEARCH 11.1 OPTIMIZATION OF STRUCTURES AND ROTOR Student: Matthew Wagner Faculty: Paul Allaire and Randy Cogill (Systems and Information Engineering) The continuing and ever increasing demands of lighter weight, higher speeds, and greater reliability of rotating machinery and their structural systems demand a rigorous optimization process for their design. The basic analysis tools of rotor-bearing-seal dynamics, and modeling of structural systems are at a mature stage and the opportunity to further apply optimization strategies for the design of these systems is now practical and realistic. This research addresses the applications of model reduction methods, Finite Element Analysis and numerical optimization algorithms for the optimization of static and dynamic performance of structures to meet specific criteria such as vibration and stress levels, and transient dynamics under design constraints. It would then be possible to solve problems such as: given a rotor dynamic system, can we design a mounting foundation structure such that the entire system has a desired modal response, and physically realize such a structure assuming we cannot change the rotor bearing system. Traditionally in a multidisciplinary design optimization problem, material, structural, and dynamic properties of a system are simultaneously optimized, all while satisfying constraints on the integrity and economic feasibility of a design [1]. Multidisciplinary design optimization problems are typically a static optimization problem and do not explicitly account for the time evolution of the states of dynamic systems appearing in the model. Hence, these techniques are typically applied to static loading scenarios or to optimize frequency-domain characteristics of dynamic systems. However, multidisciplinary design optimization methods generally do not computationally exploit the structure present in many design problems resulting in the use of general gradient-based algorithms or general global optimization heuristics, such as genetic algorithms. As a result, convergence of these algorithms can be very slow, and the resulting solutions are only guaranteed to be local optima, there is not a guaranteed global optimal solution. A new method for structural optimization based on the theory of linear quadratic Gaussian (LQG) control is being developed. The most important objective of this work is the formulation of a guaranteed global optimization method that does not have the limitations of methods such as the gradient search method which may find only a local minimum. This method is applied to minimize an objective function such as displacement of a subsystem of a larger system composed of multiple subsystems. This makes it possible to isolate and optimize the effect that one subsystem has on the dynamic behavior of a larger system containing this subsystem. Model reduction methods are necessary to discretize large-scale systems and reduce the dimensionality of the problem, making analyses and optimization feasible. Component mode synthesis (CMS) and Modal Analysis (MA) methods are being explored and implemented to reduce complex systems. These methods allow for a large system to be discretized, reduced, and analyzed via Finite Element Methods to obtain state parameters that describe system responses to prescribed inputs. Thus, we have a dynamic system represented by its modes and have access to design variables for the system that include the geometric, material, and state variables of the components. The analyses of component systems by these methods means that we can formulate and reduce a structural problem into a feasible optimization problem that can provide meaningful and structurally realizable interpretation of design variables that describe the struc- Page 56

57 11. OTHER RESEARCH tures of each system. Consider figure 1 as a multi-stage compressor train (A, B and C) which can be discretized into system A (rotor dynamic system), system B (connecting system), and system C (supporting structure) and further reduced using CMS and MA methods. Having reduced the dynamic representation of the system it is then possible for instance to optimize the design variables of the connecting system (B) while minimizing the displacement of the support structure (C). Following the LQG methods it is then possible to formulate the inverse problem and determine physically realizable system characteristics and parameters to satisfy the optimization problem. The optimization and model reduction methods are being validated on finite element and experimental models, figure 2. The experimental work will provide experimental validation of the theoretical derivations of the methods used for structural optimization and the ability to physically realize these optimal systems. Figure 1: Linear representation of a rotor dynamic system with structural support Figure 2: FEA and experimental models for validation. [1] G.N. Vanderplaats. Numerical Optimization Techniques for Engineering Design With Applications Mcgraw Hill, Page 57

58 ROMAC EVENTS ROMAC Annual Conference The 2009 ROMAC Annual Conference will be held on June 15 - June 19, 2009 at the Bay Watch resort in Myrtle Beach, South Carolina. They have offered Queen and King Studios for $ inclusive. You can call the hotel at for reservations. Please let them know that you are with the University of Virginia/ROMAC conference to receive this discounted rate. You will also need to register with ROMAC, so please fill out the registration form that is on our website. You can , fax or mail the form back to Karen Marshall. Our fax number is The meeting will begin with a welcome reception on Monday at 6:30. The actual sessions will begin on Tuesday morning. Friday sessions will be all magnetic bearings and will end midday. For those of you not wishing to stay for the Friday sessions, Thursday sessions will be ending at approximately 5:00. As usual, we are looking forward to hearing talks from member companies. If you are willing to make a presentation, please contact me to be put on the list of speakers. We look forward to seeing you in Myrtle Beach!! For additional information please feel free to contact Mrs. Karen Marshall at (434) or by ing her at kfm2b@virginia.edu. ROTORDYNAMICS AND MAGNETIC BEARINGS SHORT COURSE The 2009 Rotordynamics and Magnetic Bearings Short Course is a course in rotordynamics, bearing dynamics, magnetic bearings and applied dynamics for industrial rotors. Other topics that will be covered include unbalance response and rotor balancing, stability of industrial compressor rotors, advanced fluid film bearing analysis, compressible flow seals, advanced analysis of pumps, motors, turbines and aircraft engines. Magnetic bearings design and control theory for magnetic bearings and specifications for industrial rotors will also be covered. The course will have presentations by University of Virginia faculty, staff and students, as well as many case histories given by speakers from ROMAC industrial members. The Short Course will be held on July 27-31, 2009 at the Mechanical Engineering Building in Charlottesville, VA. We have gotten a large number of high quality industrial speakers and we will have to cover travel costs for a number of them. A nominal fee of $500 per attendee for employees of ROMAC member companies will be charged. The fee for non ROMAC members is $1500 for the full course. If you have any questions, please contact Karen Marshall at (434) Page 58

59 ROMAC CONTACT INFORMATION ROMAC Faculty and Staff Faculty Phone Paul Allaire George Gillies Chris Goyne John Knight Zongli Lin Jim McDaniel Bob Ribando Houston Wood Staff Tim Dimond Jim Durand Bob Rockwell Sean Travis Alex Untaroiu Costin Untaroiu x151 Amir Younan Karen Marshall Current Companies 1. Aerojet 2. Bechtel Plant Machinery 3. Bechtel Marine Compulsion Corp. 4. Boeing Commercial Airplane Group 5. Cameron Compression 6. Curtiss-Wright 7. Siemen s Demag Delaval 8. Turbocare 9. Elliott 10. Dresser Rand 11. Electric Boat 12. EDAS 13. ExxonMobil 14. Flowserve 15. GE Aviation 16. GE Oil & Gas 17. Innovative Power Solutions 18. ITRI 19. Kingsbury 20. Kobe Steel 21. Lufkin Industries 22. Mechanical Solutions 23. MSC Software 24. ODS 25. Petrobras 26. Pratt & Whitney 27. Praxair 28. Renk 29. RMT 30. Rolls Royce Energy Systems 31. Rolls Royce Aviation 32. Shell 33. Solar Turbines 34. Statoil 35. Hamilton Sundstrand 36. TCE 37. Waukesha Bearings 38. Zollern Page 59

60 ROTATING MACHINERY & CONTROLS LABORATORY University of Virginia Rotating Machinery & Controls Laboratory (ROMAC) 122 Engineer s Way P.O. Box Charlottesville, VA Phone: Fax:: romac@virginia.edu We re on the web: