Fault detection and isolation in the challenging Tennessee Eastman process by using image processing techniques

Area of Science:

• Industrial Systems Engineering
• Computer Vision
• Machine Learning

Background:

• Early fault detection and isolation are crucial for preventing equipment damage in industrial settings.
• Traditional methods often rely on time-domain sensor signals, which can be complex to analyze.
• A novel approach is needed to improve the accuracy and efficiency of fault diagnosis.

Purpose of the Study:

• To develop and evaluate a new fault detection and isolation method using image processing techniques.
• To assess the efficacy of using 2D image representations of sensor signals for fault diagnosis.
• To compare the proposed method's performance against existing techniques using the Tennessee Eastman benchmark process.

Main Methods:

• Sensor time signals were converted into 2D matrix images.
• Image processing techniques were applied for feature extraction, including texture, wavelet transform, mean, and standard deviation.
• Multilayer Perceptron (MLP) and Radial Basis Function (RBF) neural networks were employed as classifiers.
• The method was tested on the Tennessee Eastman benchmark process.

Main Results:

• The proposed image-based method demonstrated notable efficacy in detecting and isolating faults.
• Feature extraction from 2D images proved successful for fault diagnosis.
• The method showed superiority over previous fault detection and isolation techniques on the benchmark process.

Conclusions:

• The novel image processing approach offers a powerful tool for early fault detection and isolation in industrial systems.
• Representing sensor data as images simplifies analysis and enhances diagnostic capabilities.
• This technique provides a significant advancement over traditional methods for industrial process monitoring.