Alexander Horch and. Manfred Rode, ABB. ) UCL and ABB. Stevenage, Herts SG1 2AY, UK

Transcription

1 Organised by the Institution of Engineering and Technology Control and Automation Network The KnowAedge Network Control Loop Performance Assessment for Power Plants Nina Thornhill, UCL, Alexander Horch and Manfred Rode, ABB ) UCL and ABB Printed and published by the EET, Michael Faraday House, Six Hills Way, Stevenage, Herts SG1 2AY, UK 63 Authorized licensed use limited to: IEEE Xplore. Downloaded on January 6, 2009 at 06:20 from IEEE Xplore. Restrictions apply.

2 Abstract Control loop performance benchmarking is starting to cross over from the process industries into power generation. The talk will review tools and methods for measuring closeness of the control to the control objectives. An example in power generation will be presented showing the improvements achieved after the root causes of control loop problems were identified and rectified. About the Author Alexander Horch is a Senior Control Engineer in Plant Performance Improvement with ABB Corporate Research, Ladenburg, Germany. He completed his PhD in KTH, Stockholm before joining ABB in His research interests include control loop tuning and automatic analysis of plant data in the process and power generation industries. Nina Thornhill is a professor in the Department of E&E Engineering at University College London (UCL) and is a member of the Imperial College/UCL Centre for Process Systems Engineering. Her research interests are in industrial measurement and control, particularly in the diagnosis of control loop faults. She worked for ICI for 6 years before joining UCL in 1984 and since then has held a Royal Society Industry Fellowship with BP (1992-4) and a RAEng Global Research Award with the ABB Corporate Research Centre in Norway (2005). 64 Authorized licensed use limited to: IEEE Xplore. Downloaded on January 6, 2009 at 06:20 from IEEE Xplore. Restrictions apply.

3 1 Control Loop Performance Assessment for Power Plants Nina F. Thornhill E&E Engineering, University College London, UK Alexander Horch ABB Corporate Research, Ladenburg, Germany Manfred Rode ABB Corporate Research, Ladenburg, Germany Layout of the presentation 2 Control loop benchmarking Using measurements from routine operation Example from the process industries Plant-wide disturbances Detection and characterization Isolation and diagnosis of the root cause Application in a power plant

4 3 Control loop benchmarking Control loop benchmarking 4 Using measurements from routine operation Desborough and Harris (1992) approach: sp pv controller error, y C op y = yˆ + r ŷ - controller dependent term (predictable component) r - residual (independent of controller) Controller should have no d-step ahead predictable component ( d is a prediction horizon) Choose d to reflect engineering criteria, or use time delay. G pv

5 Control loop benchmarking 5 Examples from the process industries controller error (a) time/sample intervals CLPA index = η = predictable component (b) CLPA index = mean of (r 2) mean of (controller error 2) Control loop benchmarking 6 Commercial tools ABB: Optimize IT Loop Performance Manager and Plant Disturbance Analyser (PDA) (search for Loop Performance Manager) Aspentech: : Aspen Watch ControlArts: : CAT I, III, and V Entech/Emerson Process Management

6 Control loop benchmarking 7 Commercial tools,, continued ExperTune: Plant Triage Honeywell: : Loop Scout Matrikon: ProcessDoctor PAS: ControlWizard 8 Plant-wide disturbances

7 Plant-wide disturbances 9 Slowly developing Heat exchanger fouling, Abrupt Compressor trip, Persistent Valve limit cycle, hydrodynamic instability, tuning problems, controller interaction, recycles Signal analysis in time and frequency domain are used This talk focuses on persistent disturbances Plant-wide disturbances 10 A wish list from: Paulonis & Cox, 2003, A practical approach for large-scale controller performance assessment, diagnosis, and improvement. Journal of Process Control. 13, Desborough & Miller, 2002, Increasing customer value of industrial control performance monitoring Honeywell s experience. AIChE Symposium Series No , Facility-wide approaches, behaviour clustering; Detection of one or more oscillations; Detection of non-periodic disturbances; Determination of the locations and root causes;. Non-invasive stick-slip slip detection in control valves; Automated model-free causal analysis; Incorporation of process knowledge.

8 Plant-wide disturbances 11 oscillating measurement Plant-wide disturbances 12 PLANT-WIDE DISTURBANCE DETECTION & CHARACTERIZATION integrated error deviations oscillating spectral methods correlation spectral similarity methods zero damping Spectral envelope crossings spectral methods time-domain methods spectral similarity Spectral envelope dynamics of linear model zero crossings nonoscillating nonstationary wavelet For plant-wide analysis both detection and clustering are needed.

9 Plant-wide disturbances normalised time trend spectra The IET Seminar on time/sample Power Generation interval and Control frequency/sampling November frequency28 th Plant-wide disturbances Automated analysis Visualization of clusters with similar spectra. The vertical axis is a measure of how unalike they are; spectal dissimilarity tag number

10 Plant-wide disturbances 15 non-linear causes non-linear time valve diagnosis methods series analysis bicoherence; surrogate testing; Lyapunov exponent. limit cycle methods harmonics METHODS FOR ROOT CAUSE DIAGNOSIS no intervention Cross correlation; Qualitative trend analysis intervention Controller gain change linear causes tuning SISO methods interaction/ structural Causality methods Plant-wide diagnosis: (a) Find clusters; (b) Find the likely root cause. Plant-wide disturbances 16 Non-linearity testing Faulty valves, hydrodynamic instability, slugging flow; (power plant examples ) Non-linearity is most strong closest to the root cause; Two main branches: bicoherence analysis and surrogates analysis (U of Alberta) (UCL) There have been several successful studies.

11 Plant-wide disturbances 17 oscillating measurement 18 Application in a power plant

12 Application in a power plant 19 Process schematic 1 week load signal 33 observed loops Application in a power plant 20 Loop performance report week 1 week 2 week 3 week 4 whole month low load mid load high load load change Poor : T-CoalMillAir 4, T-CoalMillAir 3, F-Primary Air M4

13 Application in a power plant 21 Loop auditing report of one week (Diagnosis part) T-CoalMillAir 1,...,3 T-CoalMillAir 4 permanently in saturation, therefore out of valuation (poor) Application in a power plant 22 Loop detail plot for one data set SP & PV OP Raw data PV SP set point moving more than controlled var. power: actual (blue), optimal (red) tim e [s ] CE Power spectrum Oscillation detected frequency [Hz] ACF (blue) & CLimp (red) Load Norm alized quantities CE=SP-PV OP Au to-corre la tio n a n d closed loop im pulse response lags prediction horizon [samples] (T = 1 5 s ) sample controller output Bad control accuracy CLPA index Perform ance index as function of prediction horizon Phase plot control error

14 Application in a power plant 23 power plant data Application in a power plant 24 oscillating cluster root cause not strong 3 loops

15 Application in a power plant 25 Application in a power plant 26 Next steps Investigate these three loops. Discuss with personnel if this cluster is explained by the process schematic Automatic tools would be preferable! Outcomes Cluster identified Many signals due to dense instrumentation (many sensors measure similar effects) Combine PDA with LPM results and investigate most likely causes from both analyses.

16 27 Conclusions Conclusions 28 Conclusions Methods used for control loop diagnosis in the process industries have potential in power generation; Plant-wide disturbances affecting numerous measuring points can be identified and their root causes isolated;

17 29 END