MULTIVARIATE STATISTICS FOR ANOMALY DETECTION: APPLICATION IN A TURBOJET

Although the computational power of embedded systems has increased in recent years, these systems are increasingly being taxed with more tasks. This raises the interest for computationally lean algorithms which are able of rendering process operation more efficient and reliable. This is particularly relevant in the case of flight computers for autonomous aircraft. Fault detection, isolation and identification assist in management strategies to improve both predictive maintenance and operational safety. This article combines a principal component–based representation with multivariate statistics to detect and isolate anomalies in a process. The resulting algorithm is computationally lean and was validated with respect to experimental measurements in a turbojet before and after years of operation. The results show that the developed algorithm is capable of successfully determining the fouling components in the turbojet.