PROGNOSTICS OF POLYMER POSITIVE TEMPERATURE COEFFICIENT RESETTABLE FUSES

Polymer positive-temperature-coefficient (PPTC) resettable fuse has been used to circuit-protection designs in computers, automotive circuits, telecommunication devices, and medical devices. PPTC resettable fuse can trip from low resistance to high resistance under over-current conditions. The increase in the resistance decreases the current and protects the circuit. After the abnormal current is removed, and/or power is switched off, the fuse resets to low resistance stage, and can be continuously operated in the circuit. The resettable fuse degrades with the operations resulting in loss or abnormal function of the protection of circuit. This thesis is focused on the prognostics methods for resettable fuses to provide an advance warning of failure and to predict the remaining useful life. The failure precursor parameters are determined first by systematic analysis using failure modes, mechanisms, and effects analysis (FMMEA) followed by a series of experiments to verify these parameters. Then the causes of the observed failures are determined by failure analyses, including the analyses of interconnections between different parts, the microstructures of the polymer composite, the properties (such as crystallinity) of the polymer composite, and the coefficient of thermal expansion (CTE) of different parts. The revealed failure causes include the cracks and gaps between different parts, the agglomerations of the carbon black particles, the change in crystallinity of the polymer composite, and the CTE-mismatches between different parts. Cross validation (CV) sequential probability ratio test (CVSPRT) is developed to detect anomalies. CV methods are introduced into SPRT to determine the model parameters without the need of experience and reduce the false and missed alarms. A moving window training updating based dynamic model parameter optimization (MW-DMPO) n-steps-ahead prognostics method is developed to predict the failure. MW methods update the training data for prediction models by a moving window to contain the latest degradation information/data and improve the prediction accuracy. For each updating of the training data, the model parameters for data-trending model are updated dynamically. Based on the developed MW-DMPO method, a MW cross validation support vector regression (MW-CVSVR) n-steps-ahead prediction is developed to predict failures of PPTC resettable fuses in this thesis. The cross validation method is used to determine the proper SVR model parameters. The CVSPRT anomaly detection method and MW-DMPO n-steps-ahead prognostics method developed in this thesis can be extended as general methods for anomaly detection and failure prediction