New Techniques for Fault Detection and Diagnosis for Safety-Critical Applications

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

Sensors are critical components of virtually all technological systems, especially those that depend on control for safe and reliable operation. Control systems play a key role in automotive, aerospace, robotic, building, and process control applications. Safety-critical implementations have increased dramatically and will continue to grow as systems become increasingly autonomous. In the context of autonomous systems, sensors malfunctioning or failure can lead to severe degradation of performance, can destabilize the control system and may produce destructive control inputs. All of these events can severely compromise the operation of the system under consideration and may lead to tragic consequences, such as loss of human life. For these reasons, there is a compelling need for techniques that assess the integrity of sensors that are critical to a system?s operation. The investigation of such techniques is the objective of this project, which is expected to contribute to improvements to the reliability of safety-critical autonomous systems. This project focuses on a technique for sensor fault detection called sensor-to-sensor identification (S2SID). S2SID takes advantage of freely available and unknown external or ambient excitation to identify a sensor-to-sensor model called a pseudo transfer function (PTF), which is independent of the excitation. A PTF is a time-domain generalization of a transmissibility used in structural dynamics. S2SID is performed under healthy sensor conditions in the presence of unknown excitation. Then using a validation data set, again induced by unknown external or ambient excitation, the identified PTF can be used to compute a sensor-to-sensor residual. This residual, which is the discrepancy between the predicted sensor output based on the identified PTF and the actual measurements, provides a metric for detecting and diagnosing sensor faults. Unlike frequency-domain techniques based on transmissibility operators, this project focuses on time-domain identification methods that can exploit transient signal characteristics, which can degrade frequency-domain estimates. S2SID will be demonstrated on flexible aircraft applications.