Research On The Improvement Of Fault Detection Method Of Industrial Batch Process Based On Mpca

The advance of science and technology level promote the growth of demand for products,that driving the development of modern industrial process toward the direction of integration and diversification. The effective monitor for the production of each link relation to product quality and production safety in the industrial process. Its maybe cause serious personal casualty or environment pollution or any other problem once a fault happened. Batch process is a kind of different process from continuous production process. As the intermittent batch characteristics and its dynamic performance of the process, that present the different characters to continuous production. So we need to dig deeper into the information of the batch process, exploiting and applying fault control system with automatic detection ability is practical significance.The meaning of the fault detection could inform operator the cause of the problem timely when a fault occurred. Data-driven fault diagnosis is an important research field in the fault detection. In the light of the characteristics of batch processes, the article illustrate the MPCA method as the main line to discuss the fault detection of industrial batch processes, and put forward some new improvements on traditional algorithm. The main content of this article:(1) Introduces the research status at home and abroad based on data-driven method of intermittent process fault diagnosis. Expounded the MPCA based on data-driven method of related technical principles, including the establishment of the relevant statistic and statistic control limit to MPCA. It also introduces the penicillin fermentation simulation platform Pensim2.0(including the default values and setting of the parameters,the simulation data path diagram).(2) Considering the SPE statistic T2 statistic different monitor performance,the same number of principal components(PCs) is difficult to meet the requirements of this two monitoring indexes at the same time. This paper research a SNR(signal noise ratio) method to select PCs. The method count the fault factors, separate to consider the performance of SPE and T2 to select the PCs. Simulation results show that the SNR of the fault has higher sensitivity, more superior to CPV method.(3) In view of the traditional time series auto-regressive model has big error to estimate model parameter by using the least squares method to establish model under the limited number of samples. The article put forward a kind of Bootstrap combine with least square method to set up statistic model.Primarily, set up a residual empirical distribution and fix the model parameters based on the model of repeated sampling residual. Then use the SPE model for online fault detection by Pensim2.0 platform. Compared the improved method with the least square method to determine the model parameters, the prediction error between real value became further smaller.