DARTS-CNN-BiLSTM: Intelligent Fault Diagnosis for Computer Numerical Control Machine Tool Feed System

Area of Science:

• Manufacturing Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Computer numerical control (CNC) machine tool feed systems are crucial for manufacturing quality and efficiency.
• Fault diagnosis in these systems is challenging due to variable speeds and strong noise.
• Existing methods often require manual tuning and feature engineering.

Purpose of the Study:

• To propose an automated deep learning model for robust fault diagnosis of CNC machine tool feed systems.
• To address challenges posed by variable-speed operations and high noise levels.
• To improve machining quality and efficiency through accurate fault detection.

Main Methods:

• A novel deep learning model, DARTS-CNN-BiLSTM, integrating differentiable architecture search (DARTS) with a CNN-BiLSTM framework.
• DARTS automatically optimizes convolutional neural network structures for spatial feature extraction.
• Bidirectional Long Short-Term Memory (BiLSTM) captures temporal dependencies, complemented by global average pooling and a softmax classifier.

Main Results:

• The DARTS-CNN-BiLSTM model achieved over 90% diagnostic accuracy under strong noise (SNR ≥ -6 dB).
• An average accuracy of 98.15% was recorded on a variable-speed dataset.
• The proposed method outperformed advanced models like Inception-BiLSTM and DenseNet.

Conclusions:

• The automated architecture design significantly enhances fault diagnosis performance compared to manual tuning.
• The DARTS-CNN-BiLSTM model demonstrates superior effectiveness and robustness for complex feed system fault diagnosis.
• This approach offers a reliable solution for maintaining high-end manufacturing equipment health.