Fault Identification and Isolation in Dynamic Systems Using Multiple Models

Abstract

The demand for more reliability, safety and performance in industrial systems is rapidly increasing every day. The early detection of faults can avoid catastrophic events and the identification of the fault nature and severity can lead to the most appropriated and efficient maintenance task. Thus, an enhanced system diagnosis feature has the potential to increase safety and reduce the operational costs. In this context, fault detection and isolation techniques are used as the basis for building powerful decision making tools. This work's objective is to identify and isolate multiple faults in dynamic systems through signal processing. An approach based on a multiple-models architecture is considered whereas the plant output signals is compared with simulation data from a set of models representing the failure modes being analysed. The Autonomous Multiple Models (AMM) technique is chosen for further residue estimation and fault isolation. A case study using computational models representing an electro-mechanical system is carried out in order to validate the proposed method and evaluate its performance and limitations such as failure modes not mapped through the models and its capability to handle concurrent faults.