Process Monitoring And Fault Diagnosis Using Recurrence Plot In Industrial Processes

With the rapid development of information technology, the productivity and the complexity of the industrial processes become larger. In order to ensure the product quality and process conditions, people have paid more attention on the process monitoring and fault diagnosis. If the process abnormal condition and process fault can be identified timely and precisely, people can predict the product quality and the facilities conditions to avoid process failure. Therefore, the researches for on-line process monitoring and fault diagnosis are very important in industrial manufacturing processes.The traditional process monitoring and fault diagnosis methods are mostly based on physical models or mathematical models. However, it is very difficult to obtain the appropriate physical model or mathematical model of real industrial processes. The reasons are that the real industrial manufacturing processes usually contain enormous process variables and multiple manufacturing stations with large correlations. The uncertainty of the processes is also very difficult to evaluate. Fortunately, with the development of new technologies, more sensors are used in manufacturing processes and massive data have been collected. These data contain rich process information and can record the information of real process conditions. It becomes new research topics that how to extract useful information from the massive process data to satisfy the requirements of process monitoring and fault diagnosis.The vibration, temperature, pressure, voltage and current signals are mostly collected in industrial processes which belong to nonlinear signals. The recurrence plot (RP) transforms the nonlinear signal into high dimensional phase space to obtain the recurrence matrix to represent the signal patterns. The advantages of using RP method are people don’t need to know the prior information of signal distribution, signal length and stationary when using the RP method. The RP method has been widely used in different areas for system performance evaluation in physics, astronomy, biology and other disciplines. However, the research of using the RP method for industrial manufacturing monitoring is rarely reported.Hence, we introduce the recurrence plot method in our research work to analyze the real nonlinear process signals. New process monitoring and fault diagnosis methods are proposed by integrating the recurrence plot method and statistical quality control methods. The main research works are showed below.(1) An automatic process monitoring scheme has been developed to analyze the progressive stamping processes based on recurrence quantification analysis (RQA). The relationship between normal, faulty tonnage signals and their RP plots are investigated. Several RQA features are extracted from the RP plots of the tonnage signals. The process faults, named the fault due to missing part, are successfully detected by the new method, especially for the process fault occurred at the weak operation. The detection ratio of the fault due to missing part occurred at’weak operation’by our proposed method is 91% while it is only 79% by using the PCA based method. In addition, a new RP parameter training algorithm, named self-learning parameter selection algorithm, is proposed to obtain a best setting of the RP parameter’threshold’. This parameter learning algorithm can be widely used in other areas when apply the RP method.(2) A new multivariate T2 bootstrap control chart is proposed for equipment condition monitoring by integrating the multivariate statistical process control technique. The T2 control chart monitoring statistic is constructed based on different RQA features. The control limit is estimated by bootstrap method due to the monitoring statistic does not follow normal distribution. The performance of the new T2 bootstrap control chart is demonstrated by the simulation study and the rolling element bearing testing data.(3) The new concepts of template RP and relative RP are proposed to represent the relationship between different RP plots. A new process monitoring and fault diagnosis method is proposed based on property of the relative RP which is that the magnitudes of the elements in relative RP directly reflect the process condition. This new method can detect the abnormal conditions and faults in manufacturing processes without any prior information and can diagnose where and when the process fault occurred. The performance of the new method is demonstrated by the simulation study and the real progressive stamping processes.