With the popularization of big data, an increasing number of discrete event data have been collected and recorded during system operations. These events are usually stored in the form of event logs, which contain rich information of system operations and have potential applications in fault diagnosis and failure prediction. In manufacturing processes, various levels of correlations exist among the events, which can be used to predict the occurrence of failure events. However, two challenges remain to be solved for effective reliability analysis and failure prediction: (1) how to leverage various information from the event log to predict the occurrence of failure events and (2) how to model the effects of multiple correlations on the prediction. To address these issues, this paper proposes a novel reliability model, which integrates Cox proportional hazards (PHs) regression into survival analysis and association rule mining methodology. The model is used to evaluate the probability of failure event, which occurs within a certain period of time conditional on the occurrence history of correlated events. To estimate parameters and predict occurrence of failure events in the model, an effective algorithm is proposed based on piecewise-constant time axis division, Cox PHs model, and maximum likelihood estimation. Unlike the existing literature, our model focuses on the interactions among events. The applicability of the proposed model is illustrated through a case study of a manufacturing company. Sensitivity analysis is conducted to illustrate the effectiveness of the proposed model.
- discrete event data
- association rule mining
- reliability evaluation
- Cox proportional hazard model
- failure prediction