Methanol Reactor Fault Diagnosis

The primary objective of this study is to enhance fault identification capabilities in power to methanol systems by combining simulation based fault injection with machine learning techniques. In this study, a mathematical model of a methanol synthesis reactor was simulated in MATLAB and time series data was generated using input data from an Aspen plus simulation of the reactor. The developed simulation environment was exploited to simulate several fault scenarios and a comprehensive dataset was generated containing both fault free and faulty data. A one dimensional convolutional neural network(1D CNN) was employed to classify the injected faults in the dataset. The implemented 1D CNN structure, with simple and compact configurations, achieved high accuracy by learning optimal features directly from the provided data.