Process Monitoring Method For Complex Industrial Processes Based On Forecastable Partial Least Squares

With the development of the techniques of data collection, storage, transmission and processing, a lot of data which can reflect the productive process and device status can be obtained every day. How to make use of these offline and online data to extract the inner feature of the productive process, so we can monitor the process and improve the productivity and quality. The data driven based method came out and has been the hot spot of the fault diagnosis technique.Forecastable component analysis(ForeCA), a novel dimension reduction technique for temporally dependent signals. Based on a new forecastability measure, ForeCA finds an optimal transformation to separate a multivariate time series into a forecastable and an orthogonal white noise space.Based on the advantages mentioned above, this paper apply ForeCA into the fault detection and diagnosis field. I combine Fore CA with the partial least squares and present a fault detection method based on ForePLS algorithm. Combine active learning and least distance discriminant analysis with ForePLS for fault diagnosis. At last, a method based on ForePLS combined with vector autoregressive model to predict the slowly varying fault is proposed.Specifically, the main work of this article include the following aspects:Firstly, on the basis of ForeCA and PLS, forecastable partial least squares(ForePLS) method is proposed. This method can extract the forecatable feature related with the quality variable, which is very useful to reflect the system.Secondly, on the basis of ForePLS, a fault detection model based on ForePLS is proposed by presenting two monitoring statistics: CUSUM statistic and SPE statistic.Thirdly, a new method based on discriminant ForePLS is proposed for multiple fault diagnosis. In order to solve the imbalanced classification, active learning is introduced to select the most informative samples for training, which can avoid redundant samples effects to classifiers’ accuracy and improve the training efficiency.Finally, The ForePLS model with autoregressive model is proposed for slowly varying fault prediction, which can effectively prevent losses caused by such failure and avoid the frequent replacement of parts, improve production efficiency.