Fault Diagnosis of Time-Varying Parameter Systems With Application in MEMS LCRs

Abstract

Multiple-model adaptive estimation (MMAE) is a well-known technique used for model matching of deterministic parameter systems. This technique can be used in fault diagnosis by allocating a model to each type of fault. In each contingency, the model that represents the behavior of the actual system can indicate the type of fault occurrence. Kalman filters are generally used in modeling and residual-signal generation of time-invariant systems. Slowly time-varying parameter systems, however, require a system identification unit in addition to the model-matching core. This paper utilizes the least square forgetting-factor technique in parameter identification of slowly time-varying systems and combines it with MMAE for fault-diagnosis applications in microelectromechanical-systems (MEMS) lateral comb resonators (LCRs). Prescheduled faults were designed for simulations and experimentally examined in real-time implementations of estimation-based diagnosis technique for two fabricated MEMS LCRs. It is shown that the application of a system identification unit significantly increases the performance of the fault diagnosis in MEMS devices.