Search

The tool is a component that is easily damaged and wasted in the CNC machining process. Accurate prediction of tool wear is conducive to reducing processing costs and improving processing efficiency. Most of the current research uses deep learning models to mine the degradation characteristics of tool wear. However, a single deep learning model and a simple sequential combination model can only learn some features, resulting in insufficient features extracted by the model, which seriously affects the accuracy of tool wear prediction. To solve the above problems, a tool wear prediction method based on parallel dual-channel adaptive features fusion is proposed. Firstly, the force, vibration, and acoustic emission signals collected by multi-sensors are preprocessed. Based on CNN-GRU and ConvGRU, a new dual-channel parallel structure is established to extract features from the input multi-sensor signal data. Secondly, the attention mechanism is used to fuse the features extracted from the parallel structure, and different weights are adaptively assigned to the tool wear features, thereby suppressing the influence of irrelevant or redundant features. Finally, the tool wear prediction values are output through the linear layer. The experimental results based on the PHM2010 data set show that the mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2) of the proposed method are 4.24, 6.41, and 0.966, respectively. The prediction performance of the model is better than other deep learning methods, which can accurately predict the wear state of the tool, provide information support for tool change decisions, and improve production efficiency. [ABSTRACT FROM AUTHOR]

Accurate tool wear prediction is crucial for preventive maintenance on time. Most of the existing data-driven prediction methods still need complex feature engineering, which reduces the prediction accuracy and efficiency. To address this problem, a tool wear prediction model based on Improved Particle Swarm Optimization (IPSO) Convolutional Neural Network (CNN) and Bidirectional long short-term memory (BiLSTM) network is proposed. Firstly, the cutting force, vibration, and acoustic emission signals are taken as the input features of the model. CNN is used for high-dimensional feature extraction in the raw signal. Then, the BiLSTM with long-term memory and time-series processing ability is used to model time-series data. Besides, IPSO is employed to enhance the prediction accuracy of the hybrid model. Finally, the fully connected layer obtains the tool wear prediction results. IEEE PHM 2010 challenge data was used to illustrate and validate this model. The experimental results show that the hybrid model has an average prediction error of 7.06% on this data set, with an average absolute error (MAE) and an average root mean square error (RMSE) of 1.49 and 1.81, respectively. The prediction performance of the proposed hybrid model outperforms other related deep learning models. [ABSTRACT FROM AUTHOR]