Neural networks based learning and adaptive control for manufacturing systems

The problems associated with the quality control of spot welded joints for galvanised steel are discussed. The conventional spot weld-quality assessment techniques commonly available for industrial use and the probable reasons which make them difficult to employ in mass production industries are discussed. The paper then explores the capabilities of a multilayer neural network as a self-organisational structure for use in unknown pattern space. These explorations have resulted in a network that extracts features from the experience of the measurable data without the necessity of a teacher. This capability is exploited to devise a weld quality control monitor for zinc coated steel. It is essentially a nondestructive testing technique which can quite easily be incorporated into a welding robot.<<ETX>>