A novel online anomaly detection model based on wavelet feature enhancement VAE for deep hole drilling

Abstract

In the task of real-time anomaly monitoring for deep hole drilling in CNC machine tools, the scarcity of anomalous samples and stringent real-time requirements pose significant challenges. Traditional anomaly detection methods often suffer from strong dependence on labeled data, sensitivity to noise interference, and low computational efficiency. To address these challenges, this paper proposes a wavelet feature-enhanced multi-scale variational autoencoder (VAE) unsupervised real-time anomaly detection network (WEVAE-Net) for deep hole drilling, using spindle current signal data collected from a normally worn situation in a boiler manufacturing plant. Specifically, a soft-thresholding strategy is designed to suppress noise in the high-frequency components of wavelet decomposition while preserving transient features. Then, a convolutional neural network-VAE module is constructed for each wavelet sub-band to extract low-dimensional latent features, which are further combined with wavelet energy features and statistical features reconstructed via inverse wavelet transform. Subsequently, these multi-dimensional latent and time–frequency features are fused and reduced via principal component analysis, followed by weighted aggregation to compute anomaly scores. Compared with traditional reconstruction error-based methods, the proposed approach demonstrates superior robustness. Finally, an online real-time anomaly monitoring model is developed by using a queue-based dynamic update mechanism, enabling processing of one current data point per second. The proposed unsupervised framework is deployed on deep hole drilling systems of Machine No. 1 and No. 3 in this boiler factory. Experimental results on 100 samples, including normal wear, tool breakage, and spiral groove defects in products, achieve an accuracy of 97%, providing an efficient, data-scarce, and unsupervised solution for tool condition monitoring in CNC machining, thereby improving industrial economic efficiency.