An intelligent approach to machine component health prognostics by utilizing only truncated histories

Numerous techniques and methods have been proposed to reduce the production downtime, spare-part inventory, maintenance cost, and safety hazards of machineries and equipment. Prognostics are regarded as a significant and promising tool for achieving these benefits for machine maintenance. However, prognostic models, particularly probabilistic-based methods, require a large number of failure instances. In practice, engineering assets are rarely being permitted to run to failure. Many studies have reported valuable models and methods that engage in maximizing both truncated and failure data. However, limited studies have focused on cases where only truncated data are available, which is common in machine condition monitoring. Therefore, this study develops an intelligent machine component prognostics system by utilizing only truncated histories. First, the truncated Minimum Quantization Error (MQE) histories were obtained by Self-organizing Map network after feature extraction. The chaos-based parallel multilayer perceptron network and polynomial fitting for residual errors were adopted to generate the predicted MQEs and failure times following the truncation times. The feed-forward neural network (FFNN) was trained with inputs both from the truncated MQE histories and from the predicted MQEs. The target vectors of survival probabilities were estimated by intelligent product limit estimator using the truncation times and generated failure times. After validation, the FFNN was applied to predict the machine component health of individual units. To validate the proposed method, two cases were considered by using the degradation data generated by bearing testing rig. Results demonstrate that the proposed method is a promising intelligent prognostics approach for machine component health.