RETRACTED: Knowledge-based deep belief network for machining roughness prediction and knowledge discovery

Abstract

Abstract The surface roughness prediction and knowledge discovery in the machining process are very important to optimize those process variables (e.g., spin speed, feed rate) online and then obtain high machining quality of products. Deep neural networks (DNNs) consist of a complex structure and multiple nonlinear processing units to perform deep feature learning. It has achieved great success in computer vision, natural language processing, and speech recognition. It is very appropriate to apply DNNs for modeling complex non-linear relationship between the process variables (e.g., spin speed, vibration) and the surface roughness. Due to the “black box” and the huge data demand problem, there are still huge obstacles to the applications of DNNs in real-world cases. This paper proposes a new DNN model, knowledge-based deep belief network (KBDBN), which integrates the symbol rules and classification rules with the deep network. This not only enables the model to have good feature learning performance, but also can discover the knowledge (i.e., rules) from the deep network. A KBDBN-based prediction model for workpiece surface roughness prediction is further proposed, which not only effectively discovery the knowledge (i.e., symbolic rules and classification rules) for process control, but also shows better recognition performance than that of the typical machine learning models (e.g., support vector machine, artificial neural network, logistic regression) and DNNs (e.g., deep belief network, stacked auto-encoder). Moreover, the interpretable DNN model makes it be applied easily in real-world cases.