A Pattern Recognition System Based on Cluster and Discriminant Analysis for Fault Identification during Production

This paper focuses on one stage of a research project concerning online surveillance of the knitting process, which intends to detect faults as soon as possible. The objective of the paper is focused on the pattern recognition stage, i.e, distinguishing faults. For that purpose, discriminant analysis is proposed as the approach to be explored. The general problem is discussed, followed by the prototype developed up to this stage. The techniques used for detecting faults are also briefly presented in order to follow immediately into the main issue of the paper: pattern recognition using discriminant analysis. Results obtained from experiments on industrial weft knitting machines are presented and discussed and future improvements and approaches are also presented.