Robustness Improvement against Sensor Failure in Estimating Thermal Displacement of Machine Tools Based on Deep Learning

Abstract

Thermal deformation due to a machine tool's internal heat generation or heat exchange with the ambient environment deteriorates machining accuracy. To suppress the thermal displacement, this paper presents a scheme to estimate and then compensate for the thermal deformation on a turning center by using a deep learning neural network model. A critical issue with its practical implementation is its response to temperature sensor failures. If the compensation drives the machine in an abrupt, unpredictable manner when a sensor fails, it may damage the workpiece or the machine. In this paper, a scheme to train the deep learning model is presented such that it becomes more robust against temperature sensor failures. The deep learning model was trained considering the assumed profiles by the temperature sensors in failure. By using a commercial machine tool, the robustness of the thermal displacement prediction model against the sensor failures is experimentally verified.