An Upper-Probability-Based Softmax Ensemble Model for Multi-Sensor Bearing Fault Diagnosis.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rotating machinery bearing fault diagnosis increasingly uses multi-sensor data (acoustic, vibration).
• Existing methods often suffer from high computational costs and complex architectures, hindering real-time industrial deployment.

Purpose of the Study:

• To develop a lightweight and computationally efficient multi-sensor ensemble framework for high-accuracy bearing fault diagnosis.
• To address the limitations of current methods in real-time industrial environments.

Main Methods:

• Transforming vibration and acoustic signals into spectrograms.
• Processing spectrograms with modality-specific lightweight convolutional neural networks (CNNs).
• Integrating CNN outputs using an AdaBoost-based ensemble strategy for adaptive, high-confidence predictions.

Main Results:

• Achieved average classification accuracy exceeding 99.90% on benchmark and in-house datasets.
• Demonstrated significant reductions in FLOPs, inference latency, and model size compared to state-of-the-art methods.
• Exhibited robustness against false positives and missed detections in bearing fault diagnosis.

Conclusions:

• The proposed framework offers a practical and deployable solution for real-time bearing fault diagnosis.
• It effectively balances high classification performance with computational efficiency.
• The method avoids complex feature fusion, simplifying its implementation.