Machine learning approaches to manufacturing

Continuous, steady improvement is a key requirement to manufacturing enterprises that necessitates flat and flexible organizations, life-long learning of employees on the one side, and information and material processing systems with adaptive, learning abilities on the other side. On the basis of two Workshops on Learning in Intelligent Manufacturing Systems, a thorough-going analysis of the literature, and with numerous contributions, the paper surveys machine learning techniques that seem to be applicable in realizing systems with intelligent behavior. Symbolic, subsymbolic approaches and their applications in manufacturing are equally treated, together with hybrid solutions which try to integrate the benefits of the individual techniques. In order to find appropriate techniques for given problems, the strengths, weaknesses and limitations of the methods are described on a wide range of manufacturing fields. Finally, future trends are enumerated.