Intelligent Manufacturing Product's Defect Recognition Based on Convolutional Neural Network, and its Feedback Algorithm's Learning Rate Optimization

Intelligent manufacturing is a kind of system composed of intelligent machine and human experts. In order to recognize the defects of products in intelligent manufacturing process, a new recognition method based on convolutional neural network algorithm is proposed. Using the method of transfer learning, the training neural network is realized by the Inception-v3 model in the deep learning framework TensorFlow. Based on the implementation of training network, this paper optimizes the learning rate of network feedback algorithm and improves the pass rate and system stability. At present, the method has been applied to the field of rubber pipe manufacturing, and the experimental results show that the method can recognize the defect of multiple rubber pipes without the complex pretreatment of the image. And it can recognize many kinds of defects of rubber pipes, with high precision and small complexity, which have strong robustness and adaptability.