Identifying key predictors of failure in electro-mechanical actuator

Linear electro-mechanical actuators (EMA) are increasingly being used in various industries due to their multiple advantages such as energy efficiency and superior controllability. Despite these benefits, EMAs remain susceptible to degradation and sudden failures, posing unacceptable challenges to reliability and safety.
This study identifies key failure predictors in EMAs using exploratory data analysis (EDA) and local outlier factor (LOF)-based anomaly detection, applied to run-to-failure experimental data from controlled tests. These statistical
methods are applied to multiple sets of data from initial and final phases of the experiments, representing new and degraded EMA conditions. In EDA, a sliding window extracts statistical features to compute correlation coefficients across
experimental parameters. Based on the findings of EDA, LOF-based anomaly detection is performed on the signals with consistent and most significant deviations in correlation coefficients. This confirms the anomalies in raw data of these signals.
Finally, four predictors are identified based on the relative change in the correlation coefficients of statistical features which shows a significant and consistent trend. The significance of these predictors is furthermore confirmed by LOF-based anomaly detection. The identified parameters are the standard deviations of the EMA’s load, the electric motor torque, and the EMA’s velocity, as well as the skewness and mean of the EMA’s load and velocity.