Non-Invasive Measurement and Multi-Channel Attention Network for Fault Diagnosis of Rotating Machinery Under Variable Operating Conditions

Rotating machinery is critical in industries such as wind power, aerospace, and manufacturing. These machines often operate under high loads and intense conditions, leading to frequent failures in critical components like bearings and gears. Such failures can result in significant economic losses and safety concerns. Fault diagnosis technology, which uses operational data to analyze and identify fault types, is essential for maintaining reliable operation. Traditional vibration signal analysis is effective but costly and invasive, while acoustic analysis is non-invasive and cost-effective but vulnerable to environmental noise. In contrast, three-phase current signals, measured using current transformers, offer a low-cost, minimally invasive alternative with excellent noise resistance. This paper proposes a method for fault diagnosis using three-phase current signals. The method employs Butterworth filtering to improve the signal-to-noise ratio, followed by advanced data analysis techniques such as multi-scale feature extraction and a multi-channel attention mechanism to realized fault diagnosis of rotating machinery. Experimental results demonstrate the effectiveness of this approach in achieving accurate fault detection, with a diagnostic accuracy of 96.11%. The proposed method shows robust diagnostic capabilities across various operating conditions.