ESSC & DUPLEX 2019

Over the past decades, huge volumes of quality data collected through continuous and regular monitoring of individual processing lines from the rolling industry have been accumulated. Now is the time to reap returns on past investments made in expensive and precise quality measuring equipment. These data await being put to better use. Any anomalous or divergent quality data require an action to be taken by the monitoring personnel. This action may be repairing, rerouting or in the worst case scenario, scrapping a coil resulting in losses. A huge amount of time and effort is typically invested in pin-pointing the exact cause of the defect(s). This paper talks about finally pulling out these hidden or unused yet, available data for the purpose of continuous quality improvement. The data is scoured for previously unseen correlations using one or more classification algorithms. When relations are found, the data is arranged in relational trees. These trees can then be converted into new rules that are implemented immediately and further help in predicting and detecting errors and defects at an earlier stage on the processing line. The newly inducted rules thus help in generating better quality data. These newly generated and more precise data are again fed to the data correlator, which in turn comes up with newer and better correlations, which can be converted into even more intelligent rules based on a great deal of past experience as well as current, improved data. These rules continue to improve the quality of the output coils while at the same time allowing for an easier and earlier root cause analysis of a problem. The feedback based system contributes to bettering itself, thereby resulting in continuous quality improvement.