Research On Fault Diagnosis And Analysis Of Process Industry Based On Data-driven

In the development of process industry,with the increasing number of equipment,the possibility of equipment failure is also increasing.At the same time,the equipment correlation is constantly enhanced,so that the failure of a certain equipment will lead to the failure of other equipment,thus affecting the normal operation of the system.Therefore,more and more attention has been paid to how to effectively identify equipment faults.Considering the causal relationship between the faults caused by the correlation of equipment,this thesis studies the fault propagation network on the basis of fault diagnosis.For the problems that it is difficult to learn the characteristics of process industry data,locate the root cause when multiple faults occur,and determine the fault propagation path,so the data-driven modeling is used and the corresponding algorithm is established.The main work are listed as following:(1)A fault diagnosis model based on attention mechanism and 2DCNN-LSTM(Two Dimensional Convolutional Neural Network-Long Short-Term Memory)is proposed.The convolutional neural network is used to extract the static features of the data,while the sequential features between the data are extracted by the cyclic neural network,and the attention mechanism is used to improve the recognition ability of the model for minor fault features.Experiments show that this model can improve the performance of judging abnormal working conditions,and comparative experiments prove the effectiveness of this model.(2)A fault propagation network generation model based on transfer entropy(TE)and improved K2 algorithm is proposed.Through the judgment of high and low thresholds,the fault state is refined and the system is divided to reduce the complexity of modeling.For the network model obtained after the transfer entropy calculation,there is a ring and a relatively complicated phenomenon.Through the ring removal algorithm and the improved K2 algorithm is used to solve the problem,and the network structure algorithm can guarantee the fault propagation network with the highest score.Experiments show that this model can judge the root cause of the fault and generate a more correct fault propagation network.