Listening Through Noise: Robust Ultrasonic Crack Detection in Coal Mine Drill Pipes Using Sliding-Window RMS and CNNs

Area of Science:

• Geophysics and Mechanical Engineering
• Non-destructive Testing
• Machine Learning Applications

Background:

• Coal mine drill pipes face severe operational stress, leading to micro-cracks and potential equipment failure.
• Existing crack detection methods struggle with the complex underground environment and noise interference.

Purpose of the Study:

• To develop an accurate and robust ultrasonic crack detection framework for coal mine drill pipes.
• To improve the safety and reliability of mining equipment through intelligent diagnostics.

Main Methods:

• Utilized a sliding-window root mean square (SWRMS) index for ultrasonic signal feature extraction, capturing crack position and size.
• Developed a convolutional neural network (CNN) model for intelligent crack classification in high-noise conditions.
• Employed data augmentation with alternating noise levels to simulate real-world interference from drill pipe threads.

Main Results:

• The SWRMS index effectively represents both the spatial location and size of cracks.
• The CNN model achieved a classification accuracy of 94.4% even at a 75% noise level.
• Demonstrated superior recognition performance and noise robustness compared to traditional methods.

Conclusions:

• The proposed ultrasonic crack detection method offers significant advantages in accuracy and robustness for coal mine drill pipes.
• This framework provides effective support for real-time monitoring and intelligent diagnosis in mining operations.