Research On Process Control And Fault Diagnosis Method Of Reactor

As the scale of chemical production continues to expand,the chemical system has become more complicated,increasing the possibility of significant safety accidents.To ensure the stable operation of the chemical production process,it is necessary to solve the research difficulties in the field of the chemical industry,discover the failure in time and perform an accurate diagnosis.First,this paper begins by introducing the unique process characteristics of the reactor,which are deconstructed into several smaller procedural links for detailed analysis.A primary control circuit is designed to meet specific key control requirements,and a self-resistive controller is developed for temperature regulation within the reactor.Using PCS7 as the control system,we have successfully completed the configuration of the process control circuit.By gathering variables in Win CC,we’ve managed to compile a data set that encapsulates the reactor’s production process.Additionally,we have performed a mathematical modeling of the reactor’s temperature system and obtained a transfer function.The ADRC control method is applied to control the reactor temperature system,and the PID control method is used to conduct step response,disturbance tracking,and sine wave tracking experiments respectively.Our experimental results demonstrate that the ADRC control method performs superiorly within the reactor temperature system.Secondly,during the chemistry,the characterization or noise processing of the high-dimensional process signal can directly affect the fault detection and diagnosis performance.To study the characteristics of the high-dimensional process signal,this article proposes a new type of long-term memory network(Wavelet Transform MultiChannel Bidirectional LSTM and GRU(VOMLG)for process data set failure diagnosis,which is diagnosed by basic model BI-LSTM and GRU composition.First,use wavelet transformation to extract the features of high-dimensional process signals in each frequency band.Secondly,use the fast Fourier transformation to convert the time domain signal into a frequency domain signal to make the signal easier and intuitive.Finally,VOMLG can learn different time-frequency features from these multi-scale process signals.Finally,this article compares the existing fault classification methods in the process of Tennessee-Eastman in the reactor process.It reached 91.73%,and VOMLG achieved higher failure classification accuracy than other deep learning models.In the TE data concentration,the accuracy rate of VOMLG has reached 96.1%,especially in other papers,categories 3,9,and 15 of the faulty data in other papers;the model of this article has reached a higher level.The classification accuracy is 99.50%,98.63%,and 97.38%,respectively.