Numerical control machining tool wear status and product quality detection for digital twins

The tool wear status detection in numerical control machining and online detection of product surface quality are related to product quality, and enterprise survival and development. A tool wear state detection model is constructed based on attention mechanism—convolutional neural network—bidirectional gated recurrent units. This network model considers the correlation between timing signals in a single convolutional neural network, overcoming the gradient explosion and dispersion in recurrent neural networks. At the same time, the attention mechanism is introduced to improve the predictive accuracy of the model. For the product surface quality detection, a YOLO4 model is constructed. The original data images are clustered using the K-Means +  + method on the real bounding boxes. The model is composed of four parts, input, BackBone, Neck, and Predicto. The online tool wear detection network model could remove noise data from the original signal. The function loss value and accuracy were about 0.750 and 97.50%, both of which were superior to other network models. The overall function loss value of the YOLOV4 network model gradually decreased with the increase of iteration times, which was consistent with the trend in the other three types of object detection models. The mAP value was higher. The numerical control machining tool wear status and product quality inspection results can effectively detect quality issues in the production line, thereby improving production efficiency and promoting the intelligent manufacturing in enterprises.