A new elite opposite sparrow search algorithm-based optimized LightGBM approach for fault diagnosis

Area of Science:

• Industrial Engineering
• Machine Learning
• Data Science

Background:

• High-dimensional data in industrial processes poses challenges for traditional fault diagnosis.
• Existing dimension reduction techniques may fail with non-normally distributed or deformed data.
• Optimization algorithms like Sparrow Search Algorithm (SSA) require enhancements in convergence and optimization capabilities.

Purpose of the Study:

• To propose a novel fault diagnosis approach using an optimized LightGBM classifier.
• To introduce a new dimension reduction technique, cam locally linear discriminate embedding (CLLDE), for deformed high-dimensional data.
• To develop an improved optimization algorithm, elite opposite sparrow search algorithm (EOSSA), for hyper-parameter tuning.

Main Methods:

• Developed cam locally linear discriminate embedding (CLLDE) to improve feature extraction from deformed high-dimensional data.
• Proposed elite opposite sparrow search algorithm (EOSSA) by integrating elite opposite and orifice imaging opposite learning strategies into SSA.
• Utilized EOSSA to optimize the hyper-parameters of the LightGBM classifier for fault identification.

Main Results:

• CLLDE demonstrated superior performance over LLDE on the iris dataset for handling deformed data.
• EOSSA exhibited faster convergence and better optimization capabilities compared to five other algorithms on benchmark functions.
• The EOSSA-LightGBM approach achieved a superior fault recognition rate on the Tennessee Eastman (TE) process dataset.

Conclusions:

• The proposed CLLDE method effectively addresses data distribution changes in feature extraction.
• EOSSA significantly enhances the optimization performance of the Sparrow Search Algorithm.
• The EOSSA-LightGBM framework provides a highly effective and superior approach for industrial fault diagnosis.