A learning fuzzy control approach to improve manufacturing quality

To guarantee a constant quality of manufactured products, it is necessary to optimize the process parameters immediately when a failure of the workpiece quality has been observed. Since this relationship of measurement and process parameters is complex and nonlinear in most cases, this feedback loop is closed manually by an experienced operator in general. In the paper the concept of a fuzzy model based quality control is introduced, which allows automated feedback. Based on a process model, the controller is able to interpret the measurement and to adjust the process parameters. To overcome the problem, that a complex process model has to be developed first, a learning approach is presented. As membership functions radial basis functions are used to approximate the control law, and the model parameters are recursively determined by Kalman filtering. The method is applied to control workpiece geometry and surface roughness in a turning process.