The 2 nd International Workshop Ostrava - Malenovice, 5.-7. September 21 COMUTER CONTROL OF AN ACCUMULATOR BASED FLUID OWER SYSTEM: LEARNING HYDRAULIC SYSTEMS Dr. W. OST Eindhoven University of Technology Department of Mechanical Engineering ower Transmissions and Tribology.O. Box 513, 5 MB Eindhoven, The Netherlands E-mail: w.j.a.e.m.post@tue.nl Abstract Load Sensing fluid power systems adapt to the time variations of the load and are capable to deliver the necessary energy to various loads. The design of a fluid power transmission for delivering such load in an energy efficient manner seems to be no easy task. With regard to repetitive or cyclic time varying loads accumulator based systems prove to be a fruitful concept for such systems. Combining the fixed displacement pump, delivering the mean flow, with a suitable accumulator, storing or delivering the flow difference leads to a smaller size of pump. Consequently the size of the energy source can be smaller and all components in the hydraulic power supply operate at sufficient high efficiencies. The usual design has some drawbacks: the system alters from a non-ideal flow source into a non-ideal pressure source and the system can not adapt efficiently to different load sequences. Additional throttle losses will occur and part or the complete increase of the energy efficiency is lost. Such systems can not easily compete with Load Sensing systems based on the much larger sized variable displacement pumps. Some kind of control in accumulator based systems is required to transform such systems in genuine Load Sensing systems for repetitive load sequences. This paper describes the essential steps to be taken to design such a controllable accumulator based system. Computer control is an essential part in the design of such system. INTRODUCTION Fluid power transmissions for use in industrial applications are often subjected to variable loads. Typically, such transmissions consist of a number of different motor parts, driving different loads in the system, and a central hydraulic power supply. Load Sensing systems adapt to the time variations of these loads and are capable to deliver the necessary energy to various load sequences. It may be noted that these time varying loads often are repetitive or cyclic due to the nature of the processes in production machines. The design of a fluid power transmission for delivering such loads in an energy efficient manner seems to be no easy task. A typical example of a repetitive load sequence, showing the pressure- and flow-sequence, is depicted in figure 1. This example is part of a class of load sequences; each based on the same pattern but having different values for the cycle time and different magnitudes for the pressure and flow. These patterns are based on measured data from the hydraulic system in a production machine. For the purpose of this research the measured data for pressure and flow is smoothed in order to obtain manageable data. It may be noted from this figure that the ratio of peak to mean value is quite large for both pressure- and flow-sequence (each about 2.5:1), resulting in an even higher ratio for peak to mean power (about 5:1). Different systems for delivering this load will be regarded and will be compared in this paper. Basis for this comparison is the energy, i.e. the mean power used in these systems to deliver the load sequence. However, different systems can not be easily compared on basis of mean hydraulic power, in particular accumulator based systems are problematic in this sense.
The 2 nd International Workshop Ostrava - Malenovice, 5.-7. September 21 p [Ma] 12 8 4 p L p L,m 3 3 q * 1 [m /s]..4.2 q L q L,m [kw] 4 2 L L,m 5 1 15 2. 5 1 15 2 5 1 15 2 Figure 1. An example of a typical repetitive load cycle: pressure-, flow- and resulting hydraulic power sequence. Therefore mean input power of the pump, i.e. the mean mechanical power of the shaft is used. The effect of pump efficiency is then accounted for too. Cycle-efficiency then can be defined as the ratio of mean power of the hydraulic load and the mean mechanical input power. Various concepts of Load Sensing systems will be compared on basis of cycle-efficiency. Variable displacement pump systems. Load Sensing systems based on variable displacement pumps enable pressure and flow to follow closely the desired pressure- and flow-sequence, for example the sequence of figure 1. In order to determine the necessary flow the pressure difference over a hydraulic resistance is measured and is used for flow control. This pressure difference may be variable or may be held constant, but is always leading to a rise of the pump pressure and hence to additional throttle-losses. In order to deliver the maximum flow the variable displacement pump must be sized large enough. Consequently, the pump and energy source (electric motor) are oversized for almost every working condition of the transmission. Total efficiency of the power transmission will decrease due to these part load conditions of the electric motor and pump. In particular efficiency decreases quite considerably for lower values of the displacement of the variable pump. To demonstrate this effect a Load Sensing system was designed for the load sequence of figure 1. The resulting output from simulations with this design can be found in figure 2. It may be noted that the pump pressure closely follows the pressure of the load, while delivering the desired flow exactly. Despite this excellent hydraulic behaviour pump losses are quite large, resulting in a cycle-efficiency of only 32%. The sequence of the mechanical input power shows large peak values (about 9 kw), indicating the necessity of a large electric motor. In this case the motor should be rated at 11 kw in order to deliver the load. The electric motor operates most of the time well below of 5% of its capacity. It may be concluded that such systems are not ideal for such applications with regard to energy efficiency. p [Ma] 15 1 5 p L p sys 5 1 15 2 [kw] 9 3 L LSS 3 V = - 32 [cm /rev] i 5 1 15 2 L,m E,nom = 87 [W] = 11 [kw] Cycle-efficiency: η cyc = 32% Figure 2. Variable displacement pump Load Sensing System: resulting pressure- and mechanical input power sequence (load sequence of figure 1) Conventional accumulator based systems. At first glance accumulator based systems seem to circumvent the drawbacks of a variable displacement system. A fixed displacement pump delivers the mean flow and is sized accordingly smaller. Thus the pump is loaded optimal and is operating at high efficiencies. Decrease of the size of
The 2 nd International Workshop Ostrava - Malenovice, 5.-7. September 21 the pump leads to lower values for the driving power and the electric motor can thus be sized accordingly. Both the pump and the electric motor are loaded optimal. The accumulator deals with the differences in flow delivered and the flow demanded. However, the usual design of accumulator based systems has some drawbacks to with regard to efficiency. The hydraulic power supply of the transmission alters from a non-ideal flow source into a non-ideal pressure source. Additional throttle losses will occur and part or the complete increase of efficiency will be lost. This is illustrated by the results of simulations for the design of such system, see figure 3. Again, the sequence of figure 1 used for this design. ump- and accumulator-pressure lie well above the maximum pressure of the sequence, indicating the additional throttle-losses. Due to these losses, the cycle-efficiency is low at a value of 29.2%. The sequence of the mechanical input power is nearly constant; indicating the optimal loaded electric motor. For this system, an electric motor of 4 kw could be used to drive this load sequence. Compared to the variable displacement system the accumulator-based system is sized smaller. Nevertheless, although being smaller and operating more optimal these systems can not really compete with the Load Sensing systems based on the much larger sized variable displacement pumps. p [Ma] 15 1 5 p L p sys 5 1 15 2 [kw] 9 3 L AS 3 V = 9.5 [cm /rev] i V = 2 [L] A = 87 [W] 5 1 15 2 L,m E,nom = 4 [kw] Cycle-efficiency: η cyc = 29% Figure 3. Conventional accumulator-based system: resulting pressure- and mechanical input power sequence (load sequence of figure 1) It may be interesting to study the possibility of combining the positive properties of both systems in an adapted accumulator-based system. The accumulator will be used for flattening the peaks, but should also follow the load pressure-sequence more closely. Some kind of control in the accumulator-based system is then required to transform such hydraulic power supply in a genuine Load Sensing system with high efficiency. CONCET FOR LEARNING HYDRAULIC SYSTEMS In accumulator based power supplies the existence of two different types of sources, i.e. the hydro-pump and the accumulator may be observed. These two sources can be used more beneficially in such system. A load sequence, for example the sequence of figure 1, can be divided into a number of particular subsections. For each of these subsections the best solution in terms of efficiency for the choice of type of source can be found. Either pump or accumulator can then drive each subsection in the sequence under the most suitable conditions. The complete load sequence can be composed from these part solutions. Unfortunately, this is usually not possible for adjacent subsections supplied by the accumulator. The actual state of the accumulator depends on the time history of the delivered flow. In order to deliver the difference in flow over a complete sequence the accumulator needs to be charged in time. The state of the accumulator at the start of a sequence has to be preserved at the end of that sequence. Otherwise, the supply for the next of the repetitive sequence is not allowed for. So independent and optimal part solutions can not be combined directly: this has to be done with regard to the behaviour of the accumulator. The control of this system should provide for this problem. Next, it will be necessary to control both the fixed displacement pump and the accumulator independently by hardware components. By means of valves, the pump can be by-passed to reservoir and the accumulator can be connected or disconnected to the system. Only on/off valves are suitable to ensure that the additional by-pass and throttle losses will be as small as possible, resulting in an on-off control for this system. The practical implementation of this kind of accumulator based system results in the hydraulic diagram of Learning Hydraulic Systems of figure 4.
The 2 nd International Workshop Ostrava - Malenovice, 5.-7. September 21 Control V a (t), Q a (t) u a (t) p a (t) M p p (t) u p (t) p s (t) Rest of hydraulic system p (t), L q (t) L Figure 4. Schematic of Learning Hydraulic Systems. DESIGN OF CONTROL FOR LEARNING HYDRAULIC SYSTEMS The combination of pump, accumulator and the two on/off valves result in four distinct states for the hydraulic power supply. These states are characterised by pump-on/accumulator-off, pump-on/ accumulator-on, pump-off/accumulator-on and pump-off/accumulator-off. Each of these four states can be applied on the distinct subsections of the load sequence. Control thus can be defined as finding the optimal time sequence of states resulting in the highest efficiency for the regarded load sequence. Optimal cycle- or sequence-efficiency is in fact the aim for this type of control. Hence, the load sequence has to be known in advance in order to reach this goal. In a real system disturbances may prevent preservation of the state of the accumulator for each cycle. Controlling such different type of accumulator based hydraulic power supply involves quite a few control actions. A supervisor control is recommended to take care of control of all these different activities. The main activities will be considered first. System optimisation. Main action of control can be defined as finding the optimal time sequence of states resulting in the highest efficiency for the regarded load sequence. Highest efficiency corresponds with minimising of the energy losses for that system and the particular load sequence. A mathematical model is necessary to describe the state of the accumulator-based system. reliminary investigations showed that a lowfrequency dynamic model is quite acceptable for this purpose. The information of the load sequence can be regarded accordingly, hence the smoothing of the actual measurements. The energy loss of the hydraulic power supply can be defined as the difference in hydraulic power delivered by the pump and the required hydraulic power at any time integrated over the complete cycle time. The hydraulic power delivered by the pump at any time depends on the actual state of the system (on/off-valve pump and accumulator) at that time and on the initial state of the accumulator. Minimising the energy losses hence can be regarded as an optimisation problem. In this case this is a non-linear constrained optimisation problem (BURGT 1994). It is a non-linear problem due to the mathematical model of the hydraulic power supply. In addition, it is linear and non-linear constrained, amongst others: the load has to be delivered (load pressure and flow) and the state of the accumulator has to be preserved over the sequence. The result of this optimisation is a set of optimal switch-times for the hydro-pump and the accumulator. A feasible practical solution can be archived by computer control. Both simulating the mathematical model of the system and the non-linear constrained optimisation can be implemented by means of software. Complete computer control of such accumulator-based system seems to be the most obvious solution, given this situation for an important part of the control. Sizing such system requires the solution of the same optimisation problem. Given an estimated size of the pump and of the accumulator and the number of switch-times, optimal switch-times are the result of such optimisation. The resulting efficiencies for each set of sizes of the pump and accumulator can be compared. Consequently, the best solution, i.e. the optimal size of the pump and
The 2 nd International Workshop Ostrava - Malenovice, 5.-7. September 21 accumulator and optimal switch-times, can be found. This applies for each distinct load sequence. Moreover, such sized system can adapt to different load sequences: it is a Load Sensing system. Each optimisation for such sequence will result in a set of optimal switch-times, resulting in the lowest losses. It is evident that that solving the optimisation problem requires some amount of time. Mostly, except calculations on recent available high-end desktop computers, computing time exceeds the cycle-time of the sequence considered. Hence, using optimisation in close-loop control is likely impossible or at least unfeasible. A secondary close-loop control seems a possible solution. Optimisation is then used for the generation of set points for the switch-times and the secondary control is used for eliminating the non-modelled effects and effects of small disturbances. Large disturbances and new load sequences are then provided for by the optimisation routine. Making intermediate, i.e. non-optimal results available enables the control to shorten the transition time from the non-optimal system into the optimal system. For the optimisation routine some start values for the switch-times (not necessary feasible) are needed. These values can be chosen randomly, but better start values may be estimated from the information of the load sequence and of the size of the pump and accumulator (OST, 1998). Secondary control. The optimisation routine only provides the set points for the optimal switch-times for new load sequences and in case of large disturbances. Therefor this routine is not feasible to fulfil the task of closed-loop control between adjacent cycles. A secondary control is necessary for eliminating the small disturbances and non-modelled effects. Deviation from the demanded load sequence is reflected in the state of the accumulator and is only noticed after each of the repetitive cycles. During a cycle no corrective action will take place. Closed-loop secondary control is applied on basis of the complete cycle. This type of control acts on one or more of the switch-times and shifts the chosen optimal set points in such way that the deviation will decrease. Typically, this type of control acts between some predetermined boundaries, closely following the demanded pressure-sequence, while delivering the flow accurately. Load characterisation. To perform the optimisation the load sequence has to be known to the control system. The load sequence is characterised by the time-sequence of the load pressure and flow. Measurements, identification or other methods, possibly a combination of methods can determine the load. ressures can be measured easily, but measuring the flow is generally complicated and costly. Flow can then be determined by reconstruction and identification techniques. From the machine control, information of the moments of changes for the actuators can be gained. The combination of these techniques introduces the raw data for the pressure- and flow -sequence. This information has to be processed to reduce the amount of data. Supervisor control. Some central or overall control is necessary to fulfil the complete task of control. The control of the complete hydraulic power system has to learn (to know) the load sequence and the optimal switchtimes (or sequence of states). Hence, the introduction of the name Learning Hydraulic Systems, or LHS for short. The supervisor controls the different routines in the control. Data exchange between these routines and the central control is an important element in supervisor control. The general layout of the complete control, including the supervisor, is shown in figure 5.
The 2 nd International Workshop Ostrava - Malenovice, 5.-7. September 21 Supervisor control Optimisation Load characterisation Secondary control Hydraulic power supply Figure 5. Computer control of Learning Hydraulic Systems (schematic). Decisions when to start or restart the optimisation routine and when to start the characterisation of the load are based on the state of the secondary control. The trend of the successive corrections, made by the secondary control, plays an elementary role in these decisions. Note that the supervisor always can reset the hydraulic system to an ordinary accumulator system in order to keep the system running, i.e. to enable the hydraulic system to deliver the demanded load. This is not an optimal situation in terms of efficiency, but enables the supervisor to regain control by starting the appropriate sequence of control actions (i.e. load characterisation, optimisation and finally secondary control). Consequently, the supervisor can manage all kind of disturbances. SOME RESULTS The results for Learning Hydraulics Systems can be categorised in several aspects, both theoretical and experimental. Feasibility of this system can be determined theoretical by means of simulations. A general sized design of LHS for a class of various load sequences is used as a testbench for the specific load sequence of figure 1. Hence, this system is not sized particularly for this load sequence. Nevertheless, LHS can adapt to this load by determining the optimal switch-times. The result of these simulations is depicted in figure. Comparing this result with the previous ones for the conventional accumulator system and the variable pump system shows a significant increase in cycleefficiency for LHS. It may be noted that pressure level more closely follows the desired sequence, but pressure peaks rise due to the smaller size of accumulator. For this load sequence the cycle efficiency is about 38% and an electric motor of about 7.5 kw is needed to drive this load. Comparison of this result with the previous results for other Load Sensing systems shows the improvement archived with LHS. This kind of improvement in efficiency can be noticed for other load sequences too, see for example (BURGT 1994), (OST 1994) and (OST 1995).
The 2 nd International Workshop Ostrava - Malenovice, 5.-7. September 21 p [Ma] 15 p L p sys 1 5 5 1 15 2 on off on off [kw] pump accu 9 3 L LHS V i = 1 [cm 3/rev] V A= 4 [L] = 87 [W] 5 1 15 2 L,m E,nom Figure. Learning Hydraulic Systems: resulting pressure- and mechanical input power sequence (load sequence of figure 1). = 7.5 [kw] Cycle-efficiency: η cyc = 38% It may be noted that the size of pump of LHS fits nicely between the sizes of the pumps of the conventional accumulator and the variable displacement pump system. The sizes of electric motor of these different systems compare accordingly. The conventional accumulator-based system, although being smallest and loaded equally, performs the least in efficiency of these systems. An experimental set-up was build to test the system of LHS under practical conditions. This design included the design of the computer control described previously. By means of a programmable hydraulic load simulator, the practical version of LHS can be tested for a wide range of load sequences. Not only the effects on efficiency can be tested this way, but also different aspects of the controller and the behaviour for disturbances. Figure 7 shows some results from this experimental setup. In this figure both measured pressure- and flow- sequence is depicted. Realised system pressure is shown in different intermediate stages of the optimisation, starting from conventional accumulator system (1) to final optimal result (3). p [N/m 2 ] x1 2 7 1.5 1.5 2 4 8 1 12 14 1 18 2 x1-4 p sys p L 1 2 3 p : sys 1. Start 2. Intermediate 3. Optimal (final) q [m 3 /s] 4 2 2 4 8 1 12 14 1 18 2 Figure 7. ractical results of LHS: measured pressure- and flow- sequence and intermediate results.
The 2 nd International Workshop Ostrava - Malenovice, 5.-7. September 21 The programmable hydraulic load simulator can programmed to introduce disturbances in the load. Quite large disturbances in pressure and flow in a particular cycle had to be introduced to force the supervisor to restart the optimisation, i.e. to run out of the bounds of control of the secondary control. In all occasions the supervisor succeeded in bringing the hydraulic system back to its original optimal state of efficiency. The supervisor or moreover the secondary control also succeeds in keeping the system running optimal during warming-up, i.e. starting from a cold hydraulic system. The measured pressure and flow sequence of figure 7 deviates from the input sequence for LHS of figure 1. Neglecting the higher order dynamics of the load sequence seems to be a quite acceptable approach for this kind of efficiency control. CONCLUSION Designing fluid power transmission for delivering time varying cyclic loads in an energy efficient manner seems to be no easy task. Resulting cycle-efficiency seems to be rather low for the conventional solutions. Learning Hydraulic Systems, based on an adapted accumulator-based system, seems to be capable to increase cycle-efficiency and can adapt to various load sequences. In this sense LHS is a genuine Load Sensing system. The combination of hardware and software enables the system to perform this task. Complete computer control plays an important role in this system. The layout of this control and its related modules allows the system to perform successfully under various circumstances. LITERATURE REFERENCES BURGT J. v.d., 1994, A new Approach to Saving Energy in Cyclic Loaded Hydraulic Systems: Learning Hydraulic Systems. h.d. Thesis, Eindhoven University of Technology, Eindhoven, The Netherlands, March 14, ISBN 9-38-13-8. OST W., BURGT J. v.d., 1994, Energy saving and system optimisation in cyclic loaded hydraulic power transmissions, Innovations in Fluid ower, Seventh Bath International Fluid ower Workshop, Bath, England, September, 21-23, ISBN -838-17-. OST W., 1995, Concepts for the design of circuits for changing cyclic loads, Design and erformance, Eighth Bath International Fluid ower Workshop, Bath, England, September, 2-22, ISBN -838-187-. OST W., 1998, Estimators for initial conditions for optimisation in Learning hydraulic Systems, Challenges and Solutions, Tenth Bath Fluid ower Workshop, Bath, England, September, 2-22, ISBN -838-227-3.