Incipient Fault Detection And Application Research Based On Empirical Likelihood

Due to the corrosion and wearing of equipment, the deterioration of catalyst performance and so on, the incipient fault in production system and equipment will usually occurs before damage or danger. Incipient faults have characteristics such as slow changing, slight and some others, if it can be timely and accurately detected as early as the fault produced, it is possible to win more time to handle faults, and then avoid accidents. Therefore, incipient fault detection has more important significance compared to the conventional fault detection.Firstly, this paper researches the incipient fault detection method based on empirical likelihood in static process. Take the collected data of equipments or operating process as the monitor model, get fixed-length continuous subset of test data by a moving window approach, set the value of the likelihood function of subset,and then judge whether there is a fault based on whether the likelihood function value exceeds the confidence limits. This method transforms the one-sample test problem into distribution test, and is applicable to the static process while first-order statistics change. The application results of simulation model and gearbox show that,this method has higher detection rate and sensitivity of incipient fault compared to the traditional principal component analysis(PCA).To extend the application range of empirical likelihood method, an improved method of statistical pattern analysis based on empirical likelihood is presented. This method improves the statistical pattern analysis(SPA), take each order statistics of process data such as variance, kurtosis as the monitor model respectively by empirical likelihood method. In order to improve the comprehensiveness and accuracy of fault detection, respectively analyse and compare the statistics of every process variable to separate the fault variables after the fault is detected. The results of the application in Tennessee-Eastman(TE) process show that the improved method has better detection result compared to SPA methods, and it is suitable for monitoring complex multivariate static process.Finally, a dynamic process monitoring method based on empirical likelihood is presented in the process of dynamic characteristics. Separate the data by Kalman filter, and respectively monitor the obtained estimated value and residual error by empirical likelihood method. The results of the application in blast furnace process show that this method has better detection effect such as false positive rate and false negative rate compare to traditional dynamic PCA method, and it is suitable for monitoring complex multivariable dynamic process.The research results above show that, the incipient fault detection method based on empirical likelihood presented in this paper has better feasibility and application range, and it is more suitable for incipient fault detection compared to traditional methods.