Machine Learning Based Fault Classification in Pilot Plant Batch Reactor: Using Support Vector Machine.

Area of Science:

• Chemical Engineering
• Process Control
• Machine Learning

Background:

• Effective fault identification is crucial for process industries to ensure plant safety and minimize downtime.
• Challenges in fault detection include complex data, non-linearity, and robust correlations.
• Real-time issue categorization is essential for process monitoring and operational efficiency.

Purpose of the Study:

• To introduce a novel fault identification technique for batch reactor experimental trials.
• To categorize internal and external faults, including reactor temperature, coolant temperature, and jacket temperature.
• To evaluate the performance of multikernel support vector machines (SVMs) for fault classification.

Main Methods:

• Utilized multikernel support vector machines (SVMs) for fault classification.
• Employed a dataset obtained from empirical research on batch reactor trials.
• Compared different classification methods, focusing on nonlinear classifiers with radial bias functions.

Main Results:

• The multikernel SVM approach demonstrated effective categorization of internal and external faults.
• The nonlinear classifier employing the radial bias function achieved at least 22.08% superior accuracy compared to other methods.
• Successfully identified faults related to reactor temperature, coolant temperature, and jacket temperature.

Conclusions:

• Multikernel SVMs offer a robust and accurate method for fault identification in batch reactors.
• The proposed technique enhances process monitoring and contributes to improved plant safety and reduced production costs.
• The superior performance of nonlinear classifiers highlights their potential for complex industrial fault detection.