Chiller Fault Detection and Diagnosis FDD Paul Riemer

Chiller Fault Detection and Diagnosis (FDD) Paul Riemer June 20, 2000 ECE/CS/ME 539 Semester Project Instructor: Prof. Y. H. Hu (a little piece of my MS research project)

What is FDD? A process of comparing quantities that characterize a system’s actual performance against their baseline values to determine deviation from accepted behavior and to identify which of the system’s components are responsible and how so.

Chiller Basics z Large Scale Air Conditioning Equipment z Cools Water To Be Piped Around Building z Vapor Compression Cycle z Uses Refrigerant such as CFC, HCFC, NH 3 z Large Energy Demands: Mostly Electrical y. Compressor (Centrifugal, Reciprocal, Screw, Scroll) y. Water Pumps

Chiller Schematic Cooling Tower Condenser 3 Expansion Device (Shell and Tube HX) 2 Centrifugal Compressor 4 Evaporator (Shell and Tube HX) Air Handlers 1

& GPMCW TCWS Condenser TCON D Expansion Device TCWR T 2 POWER Centrifugal Compressor Evaporator TCHWS TEVAP GPMCHW TCHWR

Independent Monitored Quantities z. GPMCHW - Chilled Water Flow Rate z. TCHWS - Chilled Water Supply Temp z. TCHWR - Chilled Water Return Temp z. GPMCW - Condenser Water Flow Rate z. TCWS - Condenser Water Supply Temp (a. k. a. Forcing Inputs)

Dependent Monitored Quantities z. TCWR - Condenser Water Return Temp z. TCOND - Condenser Saturation Temp z. TEVAP - Evaporator Saturation Temp z. T 2 - Compressor Exiting Temp z. Power - Electric Power Draw

Characteristic Quantities Evaporator z Heat Transfer z UA z Approach TCHWS-TEVAP z CHWDT TCHWR - TCHWS Condenser z Heat Transfer z UA z Approach TCOND-TCWR z CWDT TCWR - TCWS (CQs) Others z Isentropic Efficiency z Motor Efficiency z COP

FDD Process 1 Neural Network Predictor Forcing Inputs Data Reduction Code Dependent Quantities Physical Chiller Data Reduction Code Predicted CQs No Fault! Fault Classifier Fault X! Actual CQs Remedy?

FDD Process 2 Neural Network Predictor Forcing Inputs Predicted CQs Dependent Quantities Physical Chiller Data Reduction Code No Fault! Fault Classifier Fault X! Actual CQs Remedy?

Fault Classifier & End Goal z. Compare actual and predicted CQs z. Comparison criteria from a detail thermodynamic model of a chiller Actual CQ 1 2% Above Predicted CQ 1 Actual CQ 2 5% Below Predicted CQ 2 } =No Fault Actual CQ 1 15% Above Predicted CQ 1 Actual CQ 2 10% Below Predicted CQ 2 } =Fault X

Neural Network Predictor z. All Approaches y 5 Independent Quantities as Inputs y. Feed Forward Multi-layer Perceptron y. Created and Trained using Matlab Toolbox y. Linear Activation Function y. Fault Free Data Set - April

Neural Network Predictor z. FDD Process 1 y 5 Dependent Quantities as Outputs y. Approach 1 - 1 Network w/5 Outputs y. Approach 2 - 5 Networks each w/1 Output z. FDD Process 2 y. Approach 3 y 1 Network w/11 CQ’s as Outputs

Data, Valuable Data z. Available y 4 identical chillers for cooling season y 10 monitored quantities on 1 -minute interval z. Utilization y. Trimmed non-operating data y. Trimmed to expand interval between points y April as fault free training and testing data y. July as potential faulty data for FDD

Results z. Approach 1 = Run 34 z. Approach 2 = Runs 51 -55 z. Approach 3 = Run 74 z. Part 1 y. Approaches 1 & 2 and Actual Values y. Plots of 5 Dependent Quantities

Condenser Water Return Temp

Condenser Saturation Temp

Evaporator Saturation Temp

Compressor Exiting Temp

Electric Power Draw

Results Continued z. Approaches 1 & 2 not significantly different z. Approach 1 results converted to CQ’s by EES data reduction code z. Part 2 y. Approaches 1 & 3’s CQ’s vs actual values y. Plots of 11 CQ’s

Evaporator Heat Transfer Rate

Evap. Conductance Area Product

Evaporator Approach

Chilled Water Temp Difference

Condenser Heat Transfer Rate

Conductance Area Product

Condenser Approach

Condenser Water Temp Difference

Compressor Isentropic Efficiency

Motor Efficiency

Overall Coefficient of Performance

Conclusions z. Approaches 1 & 3 quite similar z. Training Set Predictions (April Data) y. Good Matches: QEVAP, DTCHW, QCOND, DTCW y. So-so Matches: APPREVAP, UAEVAP, NISEN, NMOTOR, COP y. Bad Matches: APPRCOND, UACOND

Conclusions Continued z. Actual FDD Predictions (July Data) y. Acceptable: QEVAP, DTCHW, QCOND, DTCW, UAEVAP, APPREVAP, y. Irrelevant: UACOND, APPREVAP x(recall training set prediction not acceptable) y. Interesting and worth further study: x. NISEN - decreased compressor efficiency? x. NMOTOR - increased motor efficiency? x COP - increased overall performance?

End Notes and Beyond z. Three approaches performed equally well on April Training Data z. Prediction worked on about half of CQ’s z. Future work as part of thesis project y. Modify Network Configurations y. Utilize More Data x. Training x. FDD Prediction