Determining Sensor Accuracy for Condition Monitoring Applications based on End of Life Metrics: Capacitor Case Study

Condition monitoring is playing a vital role in providing the information on health status of system. One of the approaches is to use sensors embedded in the system, and the measured data will be utilized for estimation of remaining life using end-of-life based diagnostic algorithms. Currently these sensors are selected based on the detectability of features for diagnostics algorithms, without attention to the diagnostic algorithms outcomes at the system level. Relating the diagnostic algorithm critical-to-quality (CTQ) metrics to system level outcomes for maintenance activities is addressed in this paper with a capacitor bank case study. Comparison of the diagnostic algorithm outcome with the actual scenario is carried out by using a confusion matrix and related with system measures such as no fault found and operational interruption rate. Normal probability density functions (pdf) are plotted for healthy capacitor bank and algorithm estimations for degraded capacitor bank at derived mean and standard deviation. Intersecting area of the actual capacitance pdf with the estimated pdf is calculated based on % degradation at various tolerance levels of capacitance estimation. Finding of this method is that the sensor must measure capacitance with 1 to 2% error tolerance for minimum impact on system level outcomes at all levels of capacitor degradation.