Development of a Particle Filter-Based Prognostic Model to Predict Maintenance Dependent Processes

Components and systems in industrial processes undergo wear and degradation until they are either replaced or repaired. Maintenance actions, mostly imperfect repairs, may not entirely reduce the degradation and may expedite future degradation rates. Those repair and maintenance actions should be considered in prognostics models to accurately predict the components’ health conditions.

Different mechanical components have different fault modes in complex engineering systems, such as Nuclear Power Plants (NPPs). The change of a process variable may be caused by the degradation of a few different components. Effective ways to decouple the fault modes are required to predict each component's Remaining Useful Life (RUL).

This dissertation introduced a maintenance dependent monitoring and prognostics model (MDMPM) framework to detect anomalies, decouple faults, perform prognostic analysis on each fault, and reinitialize based on specific maintenance activities.

This dissertation also introduced an Auxiliary Particle Filter (APF) model, which has the advantages of dynamically updating the model parameters and optimizing computational speed for prognosis applications in real engineering problems.

The newly developed APF model is applied in the prognostics analysis and model reinitialization and further developed into the Auxiliary Particle Filter Prognostics Model (APFPM). The MDMPM framework and APFPM are demonstrated with a dataset from an electric motor accelerated aging experiment and also from a semi-simulated NPP Circulating Water System (CWS). The motor data results show that the APFPM can quickly and accurately predict the RUL and is robust to measurement noise. The semi-simulated data introduced pump degradations and condenser tube fouling into the CWS. These two faults are decoupled, and degradation indexes are generated to predict pump and condenser tubes' failure, respectively. The results show that the APFPM incorporated MDMPM provides a reliable way for prognostics of maintenance dependent processes.