A Real-Time Deep Machine Learning Approach for Sudden Tool Failure Prediction and Prevention in Machining Processes

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence in Manufacturing
• Signal Processing

Background:

• Tool Condition Monitoring (TCM) is critical for industrial competitiveness, cost reduction, and quality improvement.
• Sudden tool failures are unpredictable in dynamic machining environments, leading to part damage and downtime.
• Existing prefailure detection methods struggle with threshold definition and sensitivity to material-specific phenomena.

Purpose of the Study:

• To develop and validate a real-time system for detecting and preventing sudden tool failures.
• To establish a robust prefailure indicator independent of cutting conditions.
• To overcome limitations of current approaches in hard-to-cut material machining.

Main Methods:

• Utilized Discrete Wavelet Transform (DWT) lifting scheme for time-frequency analysis of Acoustic Emission root mean square (AE_rms) signals.
• Developed a Long Short-Term Memory (LSTM) autoencoder for compressing and reconstructing DWT features.
• Defined a prefailure detection threshold based on LSTM autoencoder training statistics and reconstruction errors.

Main Results:

• The developed system accurately predicted sudden tool failures before they occurred.
• Sufficient time was provided for corrective actions, protecting machined parts.
• The approach demonstrated robustness across different cutting conditions and sensitivity to crack propagation.

Conclusions:

• The proposed DWT-LSTM autoencoder approach effectively predicts tool prefailure in real-time.
• This method offers a reliable and condition-independent threshold for detecting impending tool failure.
• The system successfully addresses limitations of prior methods, particularly for hard-to-cut materials.