A Low-Delay Lightweight Recurrent Neural Network (LLRNN) for Rotating Machinery Fault Diagnosis.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rotating machinery systems require robust fault diagnosis for safety and reliability.
• Existing Long Short-Term Memory (LSTM) network methods for fault diagnosis often suffer from high memory usage and computational delay due to numerous parameters.
• There is a need for more efficient and computationally lighter models for real-time mechanical fault diagnosis.

Purpose of the Study:

• To propose a novel low-delay lightweight recurrent neural network (LLRNN) model for mechanical fault diagnosis.
• To address the limitations of traditional LSTM models in terms of memory occupancy and computational delay.
• To develop an automated fault diagnosis system that requires minimal manual intervention.

Main Methods:

• A specialized LSTM cell structure incorporating a forget gate was designed for the LLRNN model.
• Input vibration signals were segmented into shorter sub-signals to reduce time sequence length.
• The segmented sub-signals were processed directly by the LLRNN for automated fault diagnosis.

Main Results:

• The proposed LLRNN model demonstrated significantly reduced memory space occupancy compared to existing methods.
• Experiments confirmed lower computational delay in the LLRNN model.
• The LLRNN model achieved diagnostic accuracy comparable to established fault diagnosis techniques.

Conclusions:

• The LLRNN model offers an effective solution for mechanical fault diagnosis with improved efficiency.
• The proposed method is suitable for applications requiring low latency and minimal computational resources.
• This research contributes to the advancement of intelligent fault diagnosis systems for rotating machinery.