A Machine Learning Approach for Gearbox System Fault Diagnosis.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Gearbox failures pose significant risks in industrial operations.
• Existing fault diagnosis methods often require detailed knowledge of gearbox design and operating conditions.
• There is a need for robust, automated fault detection systems.

Purpose of the Study:

• To develop a fully automated gearbox fault diagnosis approach.
• To create a method that does not require prior knowledge of gearbox construction or load.
• To improve the accuracy and reliability of gearbox condition monitoring.

Main Methods:

• Utilized an adaptive filter to evaluate prediction errors for gearbox monitoring.
• Applied a support-vector machine (SVM) classifier to the prediction error's standard deviation.
• Cross-validated the proposed method on a public dataset with 1760 test samples.

Main Results:

• The proposed automated method demonstrated superior accuracy compared to two reference methods.
• Achieved an average accuracy improvement of 9% over reference methods.
• The approach proved effective across various support vector machine settings.

Conclusions:

• The developed automated gearbox fault diagnosis system is effective and accurate.
• The method's independence from specific gearbox parameters enhances its applicability.
• This approach offers a promising advancement in predictive maintenance for gearboxes.