Incipient fault detection based on empirical likelihood

A fault detection method based on empirical likelihood is presented to deal with the incipient fault in process and equipment. The problem of incipient fault detection is studied in the view of distribution test by a moving window approach. The original fault detection problem is transformed into distribution test, and a set of empirical likelihood values is computed. Based on the likelihood values, the confidence limit is determined as the 95th or 99th percentile of the likelihood values. The main advantage is that it can improve the sensitivity and efficiency of incipient fault detection without making any assumptions on the distribution of variables, and the calculation is relatively simple. The simulation and application results show that the method is feasible and effective, and has higher detection rate and sensitivity than traditional principal component analysis.