Research On Fault Detection And Diagnosis Of Dynamic Chemical Process By Data-driven Method

Chemical process fault detection and diagnosis is an important branch of intelligent manufacturing,as well as a key factor in ensuring the safety of industrial operations,production stability,and environmental ecology.Under the background of large application of intelligent equipment,industrial data such as equipment,material and process state parameters,environment parameters can be collected.Information with strong relevance and high value can be mined around industrial big data to form a closed data loop,so as to realize real-time monitoring and adjustment of the production process,which is conducive to the optimization of the chemical process.However,modern chemical process system always has dynamic characteristics,which is the most essential characteristics in the process industry,and also determined by the inertia of the process equipment itself and the universal application of feedback control.It is of great practical significance to pay attention to dynamic characteristics of process data in fault detection and diagnosis for control optimization and safety production of chemical process.Therefore,in view of the actual industrial requirements of chemical enterprises,this paper relies on the data-driven method,focuses on deeply mining the dynamic features and temporal features of process data,and solves the problems such as nonlinearity and noise pollution in the data,to carry out research on fault detection and diagnosis.The main research content and innovation points of this paper are summarized as follows:1.Aiming at the problem that it is difficult to extract the dynamic features of data for fault detection under the background of strong noise in chemical process,a dynamic low-rank matrix and optimized long short-term memory(OPLSTM)robust fault detection method is proposed.Firstly,a new low-rank matrix decomposition method,dynamic principal component pursuit,is proposed for the dynamic and low-rank characteristics of process data,with the aim of separating critical detection information and useless noise,extracting dynamic low-rank matrices,and avoiding noisy information from masking dynamic features.Secondly,the optimized LSTM extracts the dynamic time series features from the dynamic low-rank matrix.In addition,the support vector data description method is used to describe the hypersphere with clear boundaries,and the distance-based statistics are constructed to achieve the fault detection results with high fault detection rate and low false alarm rate.Finally,the proposed method is tested on Tennessee-eastman process and real chemical electrolytic aluminum process,and compared with other methods.Experimental results show that,compared with other methods,the proposed method achieves the lowest fault alarm rate under the premise of the highest fault detection rate,showing superior robustness.2.To address the problem that traditional methods in deep learning are insufficient to characterize the dynamic faults of chemical process,a fusion model based on CNN,Squeezeand-Excitation(SE)attention mechanism and optimized LSTM is proposed for dynamic chemical process fault diagnosis.Firstly,the extended sliding window mechanism is used to transform the fault data into augmented dynamic data to enhance the dynamic features of the data.Secondly,SE attention mechanism is used to optimize the critical fault features of the augmented dynamic data extracted by CNN.Then,OPLSTM is used to balance the fault information and further extract the dynamic features of the time series data.Finally,the proposed method is validated in Tennessee-eastman process.The experimental results show that the proposed method has the highest fault diagnosis accuracy than the other seven compared methods due to its ability to represent dynamic information and capture the temporal and spatial structural features of the process data.The method proposed in this research makes up for the shortcomings of existing process fault detection and diagnosis methods,and can provide decision basis for risk assessment and abnormal situation management,reduce the risk of process,and contribute to the safe production of dynamic chemical process.