Research On Feature Extraction And Feature Transfer Method For Tool Wear Prediction

The state of tool wear directly affects the surface quality and machining accuracy of the part,so accurate prediction of the tool wear value is one of the effective measures to ensure the machining quality of the part.However,factors such as changes in machining parameters and cutting temperature,the manufacturing error and installation error of the cutting tool will lead to differences in the distribution of monitoring data,which weaken the generalization performance of the existing tool wear prediction methods using machining signal features.To address this problem,the feature extraction and the feature transfer methods are studied in this dissertation,which effectively improves the accuracy of the tool wear value prediction.The main contents of this dissertation are given as follows:For the feature transfer,two tool wear prediction methods are proposed.(1)Aiming at the complex parameter optimization process in the transfer component analysis method when applying to the tool wear prediction,a prediction method based on correlation alignment is studied,which do not require to specify hyperparameters and simplifies the parameter optimization process.Test results show that compared with other methods,the support vector regression using the correlation alignment has more stable feature transfer performance and better prediction effect.(2)Aiming at the problem of insufficient domain-invariant feature learning when the feature-based adversarial transfer learning method is applied to the tool wear prediction,a prediction method based on domain adversarial adaptation is studied,and a tool wear prediction method using the method and a multiscale time-distributed convolutional long shortterm memory network model is proposed,which improves the model’s ability to learn deep domain-invariant features.Test results show that this method has good prediction and generalization performance.For the feature extraction,compared with the cutting force signal,the torque or current signal of the feed motor has the advantages of low acquisition cost and no interference with machining,so the torque feature extraction is studied.To improve the poor prediction performance and weak stability of the existing motor torque or current features of the tool wear,a feature extraction method based on the machining process model is proposed.Taking advantage of the fact that the alternative component of the motor feed torque related to the cutting force is not affected by the factors such as friction in the feed system,the dynamic feature of cutting torque and cutting current are deduced by combining the cutting force model and the torque model of the feed system.The results show that the proposed method is simple and efficient,and the extracted features can effectively reflect the change of tool wear value with different machining parameters.Finally,combined with the above mentioned two methods,a tool wear prediction method for variable working conditions based on the dynamic feature of cutting torque and the correlation alignment is proposed.By employing the proposed dynamic feature of cutting torque with strong generalization performance in tool wear representation,and the correlation alignment method to realize feature transformation from the source domain to the target domain,the wear prediction accuracy is improved under variable working conditions.The results show that the proposed method can achieve better tool wear prediction performance under variable working conditions,compared with the traditional multi-sensor feature prediction method.The method was also verified in the current monitoring conditions,therefore has good engineering practical value.