Cutting temperature field online reconstruction using temporal convolution and deep learning networks

Cutting temperature is one of essential signals to judge the status of cutting tool in the machining process, which is of great significance for improving the processing quality. For cutting tool temperature real-time monitoring, a method is proposed to sense the heat input of a nonlinear heat transfer system and reconstruct its temperature field. The study includes two problems: the forward and the reverse problem. For forward problems, a rapid computational model (RCM) is proposed to compute the node temperatures, which shows superiority in online applications due to the reusability of the system parameters. At each time step of the on-line process, the RCM program called the cooling and warming factor, to compute the node temperature. For inverse problems, a heat input sensing method (HISM), based on hybrid neural networks (HNNs), is developed to map temperature signals, nonlinearly, to heat inputs with an accuracy of 98%. The training data is obtained by offline finite element analysis. The coupled method, called HISM-RCM, is numerically tested in a system with nonlinear thermal properties and complex geometry with an accuracy of 99.76%. Compared with analyzed data and infrared thermography (IR), the HISM-RCM method has been shown to achieve an efficient temperature field reconstruction based on a limited number of temperature measurement points.