Automatic Input-Output Configuration and Generation of ANN-based Process Models and Their Application in Machining

Reliable process models are extremely important in different fields of computer integrated manufacturing. They are required e.g. for selecting optimal parameters during process planning, for designing and implementing adaptive control systems or model based monitoring algorithms. Because of their model free estimation, uncertainty handling and learning abilities, artificial neural networks (ANNs) are frequently used for modelling of machining processes. Outlying the multidimensional and non-linear nature of the problem and the fact that closely related assignments require different model settings, the paper addresses the problem of automatic inputoutput configuration and generation of ANN-based process models with special emphasis on modelling of production chains. Combined use of sequential forward search, ANN learning and simulated annealing is proposed for determination and application of general process models which are expected to comply with the accuracy requirements of different assignments. The applicability of the elaborated techniques is illustrated through results of experiments.