Tool Wear Monitoring Software Based On Deep Learning

Tool wear is an important factor that influences machining precision and part quality in milling,and it is essential to seek an efficacious method to monitor tool conditions.An effective tool wear monitoring software can ensure good product quality,minimize tool failures and optimize production costs.The main goal of this paper is to develop effective tool wear monitoring software,which mainly solves the problem of how to monitor the wear of milling cutters through sensor signals during milling.This paper takes tool wear as the monitoring object,cutting force,acoustic emission and cutting vibration as the monitoring signals.Aiming at the large signal noise in the actual monitoring environment,the variational modal decomposition algorithm is used to reduce the noise of the monitoring signal,and extract Signal characteristics.It is mainly to reduce the impact of environmental noise on the deep learning model and reduce the complexity of the model,and to endow the monitoring method with good adaptability.And on this basis,the optimization of the deep learning network structure is discussed,and the influence of different signal feature inputs on the deep learning model has improved the software accuracy.The main research contents of the thesis are as follows:The tool wear information collection experiment was carried out using a vertical machining center.The vibration,force and acoustic emission signals of the workpiece in different directions during milling were collected through sensors,and the collected signal data was processed by a variational modal decomposition noise reduction algorithm.Extract the signal features from the time domain and frequency domain of the processed signal,and use the Pearson coefficient to analyze the correlation degree of the extracted features,and screen the features with strong correlation.Input the filtered features into the deep learning model to determine the maximum wear zone width of the flank of the tool.A tool monitoring method based on the multi-head self-attention mechanism and the bidirectional long short-term memory network(MS-Bi LSTM)is proposed.The sensor data after signal noise reduction and feature extraction is used as the input sequence,and the multi-head self-attention mechanism is used to give different moments.The signal characteristics are assigned different attention coefficients.Then,a two-way long and short-term memory network is introduced to encode time information.The fully connected layer is built on the bidirectional long and shortterm memory network and is used to calculate the knife value.Through the comparison of experimental data,it is found that compared with the currently commonly used monitoring models based on convolutional neural networks and bidirectional long-and short-term memory networks,the MS-Bi LSTM model adopted in this paper is more stable and more accurate.Finally,this thesis develops a software based on deep learning.The software uses a multi-threaded mechanism to realize real-time monitoring,signal processing,and wear value warning functions.Analyze the signal characteristics through the deep learning model to obtain the tool wear value and promptly alarm when the tool reaches the wear limit...