NADALINE connectionist learning vs. linear regression at a lamp manufacturing plant

The results of applying connectionist learning methods to find cause and effect relationships on a manufacturing line are described. The NADALINE learning algorithm is used to extract linear relationships between production variables and a quality measure. The result of NADALINE learning is compared with that of a conventional linear regression technique. These results show that a simple connectionist algorithm can operate using limited computing power, online, and give a meaningful interpretation of a manufacturing process. Possibilities of using these interpretations for control are explored. Filtering methods that were used to make the historical data more manageable are discussed