Surface Roughness Determination With the Help of Artificial Neural Networks as Enabler of Metal Machining Process Controlling System

Implementation of control systems for metal machining process is leading to better quality of products, increase of productivity and decrease of environmental impact. These control systems are analyzing various types of data, acquired by sensors and IoT devices, to predict, to make classifications and to generate decisions. These imply all machining process phases. By analyzing the surface roughness, the machining process can be controlled, the decision if the resulting product can be accepted or should be rejected can be taken, and decisions regarding maintenance tasks can be generated, tasks that may imply the replacement of cutting tools. This way, the control system integrates predictive maintenance features which in-crease its complexity and value. In our research we propose a new classification algorithm to determine the surface roughness, classification that is later used as input to predictive and decision algorithms of the control system of the machining process. Traditional methods used to determine the surface roughness require highly skilled specialists and in-vestigations with the help of high-quality measuring equipment, both of which are not in the grasp of every company. The method we propose is intended to become an affordable and reliable tool for everybody. Thus, we decided to use a low-cost microscope to acquire the images that will be analyzed to determine the surface roughness. For classification we used Feed-Forward and Autoencoder Artificial Neural Networks on samples, splitting material roughness into three categories. Using this approach we achieved over 88% recognition of surface roughness categories. Our best run gave us an average error of 0.79%. These results make this method a viable tool for control systems implementations for metal machining.