Modelling the enamelled wire manufacturing process to improve on-line quality control

This paper examines the potential for using empirical models derived from real plant data for online quality control for an enamelled wire production process. Existing procedures are based around a well-established offline technique known as Tangent Delta. Using data recorded from normal production operations, models representing parameter input/output relationships are fast developed using standard linear regression. The linear models do not capture the behaviour of the process sufficiently well so other methods based on nonlinear and fuzzy methods and artificial neural networks are developed. The ability of each of these methods to capture the process characteristics are compared using test set data. The results indicate that the nonlinear models derived using the group method of data handling (GMDH) approach offer considerable promise for online quality control in this industrial application