Manufacturing Quality Prediction Based on Two-Step Feature Learning Approach

Manufacturing quality prediction performance is influenced by multi-parameter in manufacturing multi-stage processes. To solve this problem, a two-step feature learning approach (TFLR) is proposed in this paper. For the first step learning, the multi-parameter feature (high-dimensional information) is learned by a manifold learning algorithm (ML), which can enhance the representation of the multi-parameter inputs and reduce the calculation burdens. For the second step learning, the features of the low-dimensional information obtained by the ML are learned by a deep learning technique, which can learn sufficient features of the pattern between manufacturing quality and the low-dimensional information through layer-wise unsupervised training. Based on the two-step feature learning, the manufacturing quality predictions are achieved by a regression neural network. One type of manufacturing system with multi-parameter is investigated by the proposed TFLR model. The experiments show that the TFLR has good performances, and overwhelms the peer models. It is recommended from this study that the two-step feature learning, manifold learning for dimension reduction and deep learning for feature extraction, is much more promising in multi-parameter manufacturing quality prediction.