CdualTAL: multi-domain tool wear prediction using a dual-channel Transformer and cross-attention network

Abstract

Accurate tool wear prediction is crucial for manufacturing efficiency, yet effectively using multi-domain sensor features is difficult due to redundant noise. There is a critical need to strategically leverage highly predictive strong features and potentially informative weak features. To address this issue, we propose CdualTAL, an improved Transformer-based encoder–attention–decoder algorithm. Its name represents the model’s key components: a correlation-adaptive feature selection algorithm module, a dual-channel Transformer encoder, an attention mechanism, and a long short-term memory (LSTM) decoder. CdualTAL employs a dual-channel encoder to independently process the full set of multi-domain features, along with a subset of strong features selected using a designed correlation-adaptive feature selection algorithm. A custom cross-attention mechanism is then used to fuse these representations, sharpening focus on strong features while judiciously integrating information from weak ones. Finally, a hierarchical LSTM decoder captures deep temporal dependencies. Validated on tool wear datasets, CdualTAL outperforms 11 state-of-the-art methods, achieving superior prediction stability and accuracy with an average R2 of 0.983 and a root mean square error (RMSE) of 4.373.

1. Introduction

As information technology advances rapidly, traditional manufacturing industries are undergoing digital transformation toward greater intelligence. A critical component of intelligent manufacturing is tool wear monitoring, which can significantly enhance production efficiency and automation (Chehrehzad et al., 2024). During the operation of computer numerical control (CNC) machines, cutting tools inevitably experience friction with workpieces during machining. Tool wear worsens continuously with processing time until tool failure (Hou et al., 2025); therefore, real-time monitoring of tool status and accurate wear prediction are crucial. Such measures reduce cutting downtime, enable the more efficient allocation of resources and lower production costs, and improve product quality.

Tool prediction and replacement have traditionally relied heavily on operator experience. Premature replacement often leads to unused tool life, while delayed replacement results in tool failure and workpiece damage (Kumar et al., 2025). Subsequent to the widespread adoption of machine learning, researchers have made significant progress in tool wear assessment by integrating multi-source signals, such as cutting force, vibration, and acoustic emission, using various machine learning methodologies (Shi et al., 2020; Marani et al., 2021; Ou et al., 2021; Duan et al., 2022). Analyses conducted across time, frequency, and time-frequency domains have proven effective in characterizing tool wear status (Huang et al., 2020, 2024; Yan et al., 2021; Guo et al., 2022; He et al., 2022), enriching datasets and enhancing recognition accuracy and reliability. However, while multi-domain features offer comprehensive insights into the complex dynamics of tool wear, they inevitably introduce numerous irrelevant or noisy components. These so-called redundant features pose a significant challenge, as they dilute the model’s focus on the most predictive attributes (Gao et al., 2022; He et al., 2022).