Monitoring manufacturing processes by utilizing empirical modeling

Application of acoustic emission analysis to the characterization of manufacturing processes and produces is demonstrated. The relations between characteristics of AE signals and process parameters are modeled empirically. The model is built nonparametrically by a self-organized information processing system which resembles a neural network. The network structure is derived based on the statistical description of natural phenomena. During learning the modeler creates a set of representative data which comprise acoustic emission and process characteristics. These data are utilized at the process monitoring for an associative estimation of process characteristics from the input acoustic signals. The performance of the complete sensory-neural network is demonstrated using examples of turning, grinding and friction processes. It is shown how the cutting tool wear, the roughness of the ground surface and the quality of the surface which is generating friction can be estimated on-line. (C) 1998 Elsevier Science B.V.