Gearbox Failure Diagnosis Using a Multisensor Data-Fusion Machine-Learning-Based Approach.

Area of Science:

• Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Gearbox failures are a primary cause of industrial equipment downtime.
• Condition-based maintenance and expert systems are vital for early failure diagnosis.
• Reliable failure detection and diagnosis ensure safe and continuous operation of industrial systems.

Purpose of the Study:

• To conduct a comparative analysis of two distinct machine-learning-based approaches for gearbox failure diagnosis.
• To propose an enhanced predictive method integrating multisource sensing data via early fusion.
• To evaluate the effectiveness of proposed methods using an experimental dataset and statistical metrics.

Main Methods:

• Method 1: Linear Predictive Coefficients (LPC) for signal processing and Long Short-Term Memory (LSTM) for learning.
• Method 2: Mel-Frequency Cepstral Coefficients (MFCC) for signal processing, Convolutional Neural Network (CNN) for feature extraction, and LSTM for classification.
• Proposed Improvement: Early fusion technique for combining multisource sensing data.

Main Results:

• Comparative evaluation of the two primary machine learning approaches for gearbox failure diagnosis.
• Demonstration of the effectiveness of the improved predictive method utilizing multisource data fusion.
• Statistical metrics used to quantify and validate the prediction accuracy of the tested methods.

Conclusions:

• The study highlights the efficacy of machine learning in gearbox failure diagnosis.
• The proposed early fusion technique offers an improved approach for predictive maintenance.
• Accurate failure diagnosis is critical for minimizing industrial downtime and ensuring operational safety.