Dynamic Feature Extraction-Based Quadratic Discriminant Analysis for Industrial Process Fault Classification and...

Area of Science:

• Industrial Process Monitoring
• Fault Diagnosis
• Nonlinear Dynamics

Background:

• Dynamic nonlinear processes present challenges for fault classification due to high dimensionality and limited samples.
• Accurate and early fault diagnosis is crucial for reliable industrial operations and preventing failures.

Purpose of the Study:

• To introduce a novel method for enhancing fault classification and diagnosis in dynamic nonlinear processes.
• To address the limitations of high-dimensional and sample-limited fault classification problems.
• To enable early diagnosis of faults in online samples, even those with smaller amplitudes than training data.

Main Methods:

• Dynamic feature extraction from multivariate time series data.
• Augmenting feature sets using dynamic reconstruction errors.
• Employing weighted maximum scatter difference (WMSD) for dimensionality reduction.
• Utilizing quadratic discriminant analysis (QDA) for fault classification.

Main Results:

• The proposed method demonstrated superior performance in fault classification and diagnosis compared to traditional methods like LDA and KFD.
• Effective handling of high-dimensional, sample-limited fault classification scenarios.
• Successful early diagnosis of faults in a cold rolling mill simulation model, including low-amplitude online samples.

Conclusions:

• The novel method significantly enhances fault classification and diagnosis in dynamic nonlinear processes.
• The approach offers reliable industrial process monitoring and early fault detection capabilities.
• This technique provides a robust solution for complex industrial fault analysis.