MVDR-LSTM Distance Estimation Model Based on Diagonal Double Rectangular Array.

Area of Science:

• Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Belt conveyors are crucial in industrial settings, but roller faults can cause significant downtime.
• Accurate and real-time fault detection is essential for efficient industrial operations.
• Existing methods may struggle with complex structures and long conveying distances.

Purpose of the Study:

• To develop a novel and accurate method for estimating roller fault distance in belt conveyors.
• To address limitations of current fault detection techniques in challenging industrial environments.
• To leverage deep learning for enhanced predictive maintenance in conveyor systems.

Main Methods:

• Utilized a diagonal double rectangular microphone array for acoustic data acquisition.
• Employed the Minimum Variance Distortionless Response (MVDR) algorithm for signal processing.
• Applied a Long Short-Term Memory (LSTM) network for fault distance classification and estimation.

Main Results:

• The proposed MVDR-LSTM method achieved high-accuracy fault distance identification.
• Demonstrated superior performance compared to Conventional Beamforming Algorithm (CBF)-LSTM and Functional Beamforming Algorithm (FBF)-LSTM.
• Successfully estimated idler fault distance even in noisy environments.

Conclusions:

• The developed method provides a robust solution for roller fault distance estimation in belt conveyors.
• The approach shows significant potential for application in various industrial testing and monitoring fields.
• Deep learning, combined with advanced signal processing, offers a promising avenue for industrial fault diagnostics.