Towards Performance Prognostics of a Launch Valve

Abstract

Due to its criticality in aircraft carrier steam catapult operations, the performance of the Launch Valve is monitored using timer components to determine the elapsed time for the valve to achieve a set opening distance. Significant degradation in performance can lead to loss in end speed of the catapult and result in loss of aircraft/lives. This paper presents a method of using existing timing data for anomaly detection and predicting when maintenance is required (MIR) for a Launch Valve. Features such as mean and standard deviation of timing values are extracted from clock time data to detect anomalies. Neyman-Pearson Criterion and Sequential Probability Ratio Testing are used to formulate a decision on the degraded state. Once an anomaly is detected, an observation window of the previous N filtered samples are used in a risk sensitive particle filter framework. The resulting distribution is used in the prediction of shots until MIR. Performance degradation is extracted from training data and modeled as a third order polynomial. The algorithm was tested on two test sets and validated by Subject Matter Experts (SMEs) supplying the data. An Alpha-Lambda performance metric shows the time predictions until MIR fall inside an acceptable performance cone of 20% error.