Fault Diagnosis Of Chemical Process Based On Deep Learning And Signed Directed Graph

The chemical process has complex structures and many variables,where faults can cause severe economic losses and injuries.Effective fault diagnosis is an essential means of ensuring safe chemical production.Both historical data and process mechanisms can reflect the operating conditions of a process from different perspectives,leading to the development of two types of fault diagnosis methods,that is,data-based and modelbased methods.But both have their limitations.Data-based methods are simple to model but rely on data quality and are poorly interpretable.The model-based approach is highly interpretable,but the modelling costs are high.Therefore,how to combine the two types of methods to improve the fault diagnosis rate of the chemical process has become a problem demanding prompt solution.This paper proposes a chemical process fault diagnosis method based on deep learning and signed directed graph,which can detect,classify,reason and locate chemical process faults.The main contents of the proposed method are as follows.（1）A Bidirectional long and short-term memory（BiLSTM）fault identification method based on temporal feature extraction（TFE）is proposed.The TFE-BiLSTM model obtained an accuracy of 96.6% in the identification test of 18 types of conditions in the TennesseeEastman process.（2）A fault inference and location method based on time-delay-signed directed graph（TD-SDG）is proposed.First,dynamic simulation is used to analyze the dynamic mechanism of the process in order to determine the initial structure of the SDG.Then,a time lag analysis of the process data is performed using dynamic time warping to obtain the time lag of information propagation between variables,which is later added to the directed arc of the SDG model to form the TD-SDG model.Combining the time lag information in the TD-SDG model with bidirectional reasoning and conducting a ranking search for consistent paths based on the total time lag time can reduce the difficulty of locating fault sources and improve the efficiency of fault diagnosis.The fault diagnosis method proposed in this paper was verified in a TE process and subsequently applied to an actual mixed aromatics hydrogenation process.The results show that the TFE-BiLSTM model can accurately classify nine conditions of the mixed aromatics hydrogenation process with an accuracy of 95.5%.The constructed TD-SDG model has 141 nodes containing all key variables within the process and can effectively reason and locate the fault source.