Fault Detection Method of Aircraft Sensor Based on Canonical Variate Analysis

Aircrafts heavily rely on sensors for effective control. However, sensor faults are bound to occur over time. It is imperative to devise a dependable sensor fault detection technique for aircrafts. This study presents a data-driven strategy for aircraft fault detection using canonical variate analysis (CVA), a multivariate statistical technique. Through the utilization of CVA, the study establishes optimal correlations in aircraft flight data, facilitating accurate and efficient diagnosis of faults. The T2 statistic and squared prediction error (SPE) are put forward as indicators for detecting faults in this method. Compared to the traditional model-based Kalman filter (KF) fault detection method, this approach involves learning from actual data and using the learned models for decision-making, without the need for prior knowledge. The effectiveness of CVA in accurately identifying abrupt and incipient faults in inertial measurement unit (IMU) sensors has been confirmed through comprehensive simulation and experimental verification. This method lays the foundation for active fault tolerance control of aircrafts and improves the overall reliability and safety of aircrafts.