Deep Anomaly Detection for CNC Machine Cutting Tool Using Spindle Current Signals.

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Sensor Technology

Background:

• Industrial automation relies heavily on machine cutting tools, impacting production efficiency and cost.
• Sudden tool breakage in manufacturing presents a significant challenge due to highly imbalanced datasets, hindering traditional supervised learning models.
• Accurate prediction of tool breakage is crucial for preventing costly downtime and ensuring product quality.

Purpose of the Study:

• To propose a novel and effective method for predicting machine cutting tool breakage, addressing the class-imbalance problem.
• To leverage high-precision Hall sensor data and combine anomaly detection with deep learning for improved prediction accuracy.

Main Methods:

• Utilized high-precision Hall sensors to collect spindle current data from computer numerical control (CNC) machines.
• Developed a novel method, CNN-AD, integrating anomaly detection and deep learning techniques to handle class-imbalanced data.
• Applied the CNN-AD method to predict tool breakage in CNC machining processes.

Main Results:

• The proposed CNN-AD method demonstrated superior performance in predicting tool breakage compared to traditional algorithms.
• The method effectively addressed the challenge of class imbalance inherent in tool breakage datasets.
• CNN-AD achieved faster convergence and higher prediction accuracy in experimental evaluations.

Conclusions:

• The CNN-AD method offers a promising solution for accurate and efficient tool breakage prediction in automated manufacturing.
• Integrating anomaly detection with deep learning is effective for handling imbalanced data in industrial predictive maintenance.
• This approach can significantly improve production efficiency and reduce costs associated with unexpected tool failures.