Investigate the Application of Statistical Technique in Condition-Based Maintenance

Fault detection is a vital process in Condition-based maintenance (CBM) as it is used to validate the health of a machine. One of the technique that is used as fault detection is by comparing the machine historical data such as the vibration data to a standard threshold limit value, such as in ISO 10816. This method has shown to be unreliable as the machine can deviates from its normal operating condition, showing signs of fault, before reaching the threshold limit value. Thus, this project focuses on to close the gap by applying statistical technique into the development of a fault detection model. As a case study, this project has used a crude heater pump vibration data has been used to construct the fault detection model. First, 8 vibration time-domain features such as root mean squared (RMS), kurtosis, skewness, mean, peak, crest factor, impulse factor, clearance factor were extracted from the filtered vibration signal and has been fitted into the 3 control charts namely, Individual, Exponential Weighted Moving Average (EWMA) and Cumulative Sum (CUSUM) control chart. Next, all of the 24 control charts were tested for false alarm error by using False Alarm Probability (FAP) test. The result shows that all of the EWMA control chart (λ=0.2), CUSUM and Individual control chart (integrated with Western Electric Rules) of RMS produces a lot of false alarm, indicating it’s too sensitive for the process, while the other control chart has passed the test. The best result came from Individual control chart that has not been integrated with all of the Western Electric rules, as it produces no false alarm error.