Real time algorithm based on time series data abstraction and hybrid bond graph model for diagnosis of switched system

Abstract

In this paper, we propose a real time algorithm to realize a diagnosis of switched systems for abrupt parametric faults. This algorithm is based on interaction between a Qualitative Diagnosis (QD) and a monitoring component that performs a Qualitative Trend Analysis (QTA) of residual signals generated from Bond Graph (BG) elements called residual sinks. The QTA is applied in order to on-line detect change in the mean of residual signal based on combination of Piecewise Aggregate Approximation (PAA) with Page-Hinkley Test (PHT). The QD procedure is performed in two stages. In the first off-line stage, Symbolic Fault Signature Matrix (SFSM) is generated from a Parameterized Temporal Causal Graph (PTCG). The PTCG is valid for all system modes and deduced from a unified Hybrid Bond Graph (HBG) model by converting its elements into node and labeled edge. Each entry in the SFSM matrix gives the residual symbolic signature which corresponds to the lower-order signature predicted using the PTCG model by propagating initial deviation from the fault parameter in the label of edge to the residual node. In the second on-line stage, trend extraction by linear regression is triggered after change detection in order to estimate the lower order time-derivative symbol for each residual sinks. Subsequently, we propose a stepwise similarity measure for fault isolation task. The functioning of this approach is illustrated in simulating with a switched quarter-car active suspension system.