Predictive and Diagnostic Method for Vapor Compression Chiller

Abstract

The aim of thesis work is to develop a reliable and accurate Fault Detection and Diagnosis (FDD) tool for vapor compression chillers. Faults that normally occur in chillers can be categorized as:- (i) Abrupt and (ii) Degrading faults. Abrupt faults occur spontaneously and are easily detectable where obvious impacts on the chiller system operation can be observed. On the other hand, degrading faults are difficult to detect at the initial stages and develop gradually over time. Common degrading faults in chillers are refrigerant leakage and condenser fouling. In this study, we will only focus on the occurrence of degrading faults as they are more difficult to detect and affect the system efficiency over a period of time. Three degrading faults namely (i) Refrigerant leakage (ii) Condenser fouling and (iii) Refrigerant overcharge were simulated on a scroll chiller in this thesis work. The test runs were firstly conducted under fault-free conditions and was then followed by other faulty test runs. Fouling was simulated by blocking the condenser water tubes. Refrigerant leakage was simulated by undercharging the system while overcharge was performed by adding more refrigerant. Fault-free models were then developed for the selected fault detection parameters namely, (i) suction superheat temperature (ii) condenser approach temperature (iii) condenser sub-cooling (iv) overall condenser heat transfer coefficient and (v) condenser entropy generation. These models were developed using simple linear regression technique. Two statistical methods were employed in this study to detect and diagnose the three faults. The faults were detected using the Cumulative Summation (CUSUM) technique while the diagnosis was accomplished using the Bayes Theorem. Although five fault detection parameters were investigated, it was found that condenser approach and condenser sub-cooling temperatures were more than sufficient to evaluate the three degrading faults investigated in this study. These parameters are readily available in industrial chillers. A fault is considered detected if the non-reset drift (negative or positive) of any of the two fault detection parameters surpasses the second decision limit. A simple lag analysis is also proposed to enhance the diagnostic results. Once a fault is detected, the diagnosis routine is then activated to compute the fault probability. The proposed FDD scheme was able to diagnose the leakage fault at the lowest severity level of 10% with an acceptable probability of 86.8%. The occurrence of 10% fouling was also successfully diagnosed with a probability of 91.6%. Refrigerant overcharge of 10% and above was correctly diagnosed with a probability of 79.8%. The proposed CUSUM-Bayes approach was also validated using faulty data sets obtained from an external 90 Rton chiller. The capability of the developed scheme to diagnose faults on the existing scroll test rig and external centrifugal chiller was clearly demonstrated in the thesis work. The scheme was developed using only a handful number of measurements that is usually available in chillers. Apart from that, the proposed scheme can be easily programmed as a commercial software due to its simple and straight forward equations.