On hybrid learning and its application in intelligent manufacturing

For most real-world problems, the information concerning design, evaluation, realisation, control, monitoring, etc., can be classified into two groups, e.g. numerical information usually obtained by sensor measurements, and linguistic information obtained from human experts. Trainable systems must also rely on these kinds of information (sampled input-outputs pairs, and human experience). Artificial neural networks (ANNs) and symbolic (expert) systems can be mentioned as characteristic techniques, The paper demonstrates that neuro-fuzzy solutions can combine the above information sources, i.e. they have hybrid learning abilities. Combined use of the neural and fuzzy techniques in cutting tool monitoring is illustrated. The results are compared with ANN and previous neuro-fuzzy (NF) approaches. The paper shows that the NF technique can comply with the above fundamental requirements of intelligent manufacturing, i.e. real-time nature, uncertainty handling and learning abilities, with the additional benefits of managing both symbolic and numeric information, hybrid learning, and a kind of explanation facility. Finally, the integration of such a hybrid system in an intelligent manufacturing environment is investigated. (C) 1997 Elsevier Science B.V