Multi-mode Process Monitoring Method Based On Sliding Window And Similarity Analysis

With the rapid development of modern industry and computer technology,process industry system structure tends to be more and more complex.Once the technological process breaks down,it may lead to huge economic losses and even casualties,Therefore,process monitoring and fault diagnosis technology has attracted people's more attention.In industrial process,there are many different working modes according to the different product specifications,and different modes present different feature information.In this thesis,the research background of the grinding classification process is studied,and the monitoring of the multi-mode industrial process is carried out.From actual conditions,some key issues are considered when multivariate statistical methods are used in solving multi-mode process,such as modal partitioning,similarity analysis,common space extraction,etc.A scheme for process monitoring of multi-mode complex industrial production is proposed.The main research work is as follows:(1)Multi-mode process partitioning based on sliding window.In process monitoring of multi-mode complex industrial flow,production data is divided corresponding to the work mode is the key problem in the whole multi-mode process monitoring.Extracting process features with sliding windows and data updating ideas,combining certain process knowledge,the partition of the stable mode and the transition mode is effectively realized.However,the transition mode is the conversion stage between two stable production modal,detrimental period to the production process,the stage which should be shortened.(2)Multi-mode fault detection based on similarity.On the basis of dividing the work modes,removal of the transition mode data,kernel principal component analysis modeling of each stable mode data is carried out and the similarity part of each load matrix is extracted.The similarity part of the weighting algorithm based on distance and angle is acted as the common part of the stable modes and the special features of the remaining information are respectively modeled as special part of each modal.The process parameters for the off-line modeling stage are identified and applied to the on-line monitoring.Firstly,use common space model to find the failure time point.Then,the special part of each mode is monitored to judge which stability mode is the system working on before the fault occurs.And then guide field staff in accordance with process experience to remove troubles.Finally,the method is applied to the fault detection in grinding classification process,and the practicability and accuracy of the method are verified.