Machine learning approaches for real-time monitoring and evaluation of surface roughness using a sensory milling tool

Abstract In the case of complex forming tools, the topographical properties of the component surfaces play a crucial role for the resulting quality of the formed metal sheets. To be able to process the free-form surfaces of these components, the used end-milling tools need to be long-cantilevered and slender. Today, the characterisation of the component surfaces is carried out afterwards and partly by extensive quality investigations. A 100 % inspection of the entire surface is usually not carried out. Detecting the surface properties during machining and thus making the process more efficient, a sensor integrated end-milling tool has been developed. The significant advantage of this tool is the high sensitivity due to the accelerometer installed close to the tool center point. In this paper, a machine learning approach is presented which can predict the surface roughness Ra of the component based on the acquired acceleration data from the sensory milling tool. To calculate the surface roughness Ra a convolutional neural network CNN is developed and trained using LabVIEW combined with a Python algorithm. Furthermore, results obtained from real-time investigations are presented using milling tests. In conclusion, the results are discussed, and an outlook is given showing how the CNN can be used for an active control of surface roughness in the milling process.