Research On Multi-block Fault Monitoring Method In Industrial Process Based On Data

The stable operation on the process and the good product quality are the key aspects for a modern industry to keep its competitiveness.The fault monitoring technology is one of the effective ways to improve the competitiveness.With the significant improvement of computer technology,and the rapid development of sensors and distributed control systems,massive amounts of industrial process data have been saved and recorded.Therefore,data-driven fault monitoring technology has emerged.As a new integrated monitoring framework,the multi-block model can perform timely and effective monitoring for complex process industries.Under the framework of the multi-block model,this paper conducts the following research on the data-based fault monitoring algorithm:(1)Considering the extraction of process local information and hidden information,a multi-block principal component analysis fault monitoring method based on hierarchical information extraction is proposed.The first layer considers the mutual information between process variables,and divides them into blocks according to the value of the mutual information to extract the local information of the process.The second layer extracts characteristic information such as cumulative error and second-order difference for each variable block,and combines the observation value to expand each variable block into three information sub-blocks.The monitoring results are fused based on the Bayesian method.Simulation experiments on Tennessee-Eastman process verifies that after hierarchical information extraction,the sub-blocks contain both process local information and characteristic information,the overall monitoring performance has been improved.(2)Considering that monitoring methods based on variable block have the problem of weakly related information loss,a fault monitoring method based on similar slow feature transform vector is proposed.Based on the slow feature analysis,the dynamic information of the process is effectively extracted,and the slow feature transformation matrix is obtained.The similarity is defined according to the distance,and the transform vectors with higher similarity are divided into the same sub-block.Dividing the transformation matrix based on the similarity between the transformation vectors can make slow features in the same sub-block have similar monitoring performance,and can enhance the expression of effective information.The statistics of each sub-block are integrated by support vector data description.The simulation experiments of the Tennessee-Eastman process and an actual blast furnace ironmaking process show that this method can effectively improve the fault detection rate and the detection delay.(3)In the fault monitoring method based on slow feature analysis,the selection of slow features only according to the degree of change,but the slowest-changing features are not necessarily more beneficial for monitoring.In order to further improve the selection of slow features in monitoring model based on slow feature analysis,a fault monitoring method based on fault-sensitive slow features is proposed.Through theoretical analysis of the statistics of the SFA-based fault monitoring model,the fault sensitivity coefficient is defined as a new slow feature ranking criterion to select the slow features which are most sensitive to fault in each variable direction.Taking into account the unknown characteristics of the fault in the real-time monitoring process,monitoring model is established for each variable direction based on the multi-block strategy.Finally,support vector data description is used to fuse the statistics of each sub-block to obtain intuitive detection results.This method can effectively select fault sensitive slow features without fault data and it can also both improve the monitoring performance and generalization ability of the model.