Associative Classification for Failure Prediction in Aluminum Wheel Molding: A Case Study

Bildiriler > Özet Bildiri · 2019

Yazarlar: İnanç KABASAKAL

International Conference on Data Science, Machine Learning and Statistics, Türkiye, Van, Uluslararası, Sözlü Sunum, Özet Bildiri, 2019-2019, ss 42-42

Özet

<p>In&nbsp;the&nbsp;verge&nbsp;of&nbsp;the&nbsp;Industry&nbsp;4.0&nbsp;transformation,&nbsp;the&nbsp;digitalization&nbsp;is&nbsp;a&nbsp;primary&nbsp;concern&nbsp;for&nbsp;manufacturers.&nbsp;As&nbsp;a&nbsp;popular&nbsp;concept,&nbsp;Smart&nbsp;Manufacturing&nbsp;emphasizes&nbsp;the&nbsp;collection&nbsp;of&nbsp;data&nbsp;captured&nbsp;through&nbsp;various&nbsp;processes&nbsp;for&nbsp;further&nbsp;use.&nbsp;Analysis&nbsp;of&nbsp;process&nbsp;data&nbsp;with&nbsp;machine&nbsp;learning&nbsp;algorithms&nbsp;might&nbsp;provide&nbsp;useful&nbsp;information&nbsp;for&nbsp;root&nbsp;cause&nbsp;analysis&nbsp;and&nbsp;failure&nbsp;prediction.&nbsp;As&nbsp;a&nbsp;supervised&nbsp;learning&nbsp;approach,&nbsp;Associative&nbsp;Classification&nbsp;(AC)&nbsp;is&nbsp;often&nbsp;used&nbsp;for&nbsp;prediction&nbsp;tasks.&nbsp;In&nbsp;our&nbsp;study,&nbsp;molding&nbsp;process&nbsp;data&nbsp;obtained&nbsp;from&nbsp;a&nbsp;global&nbsp;industry-leading&nbsp;wheel&nbsp;rim&nbsp;manufacturer&nbsp;was&nbsp;analyzed&nbsp;for&nbsp;fault&nbsp;prediction.&nbsp;The&nbsp;study&nbsp;might&nbsp;be&nbsp;described&nbsp;as&nbsp;a&nbsp;preliminary&nbsp;model&nbsp;developed&nbsp;for&nbsp;failure&nbsp;prediction&nbsp;on&nbsp;a&nbsp;real-time&nbsp;setting.&nbsp;The&nbsp;proposed&nbsp;approach&nbsp;for&nbsp;prediction&nbsp;consists&nbsp;of&nbsp;two&nbsp;steps.&nbsp;Firstly,&nbsp;the&nbsp;process&nbsp;data&nbsp;reported&nbsp;for&nbsp;subsequent&nbsp;steps&nbsp;were&nbsp;individually&nbsp;inspected&nbsp;with&nbsp;individuals&nbsp;control&nbsp;charts.&nbsp;The&nbsp;values&nbsp;out&nbsp;of&nbsp;control&nbsp;limits&nbsp;were&nbsp;marked&nbsp;as&nbsp;events&nbsp;that&nbsp;might&nbsp;correspond&nbsp;to&nbsp;potential&nbsp;causes&nbsp;of&nbsp;failures.&nbsp;With&nbsp;such&nbsp;assumption,&nbsp;an&nbsp;event&nbsp;dataset&nbsp;was&nbsp;organized&nbsp;where&nbsp;each&nbsp;event&nbsp;is&nbsp;linked&nbsp;to&nbsp;a&nbsp;part-code&nbsp;along&nbsp;with&nbsp;a&nbsp;class&nbsp;label&nbsp;that&nbsp;denotes&nbsp;the&nbsp;type&nbsp;of&nbsp;failure,&nbsp;if&nbsp;any.&nbsp;Among&nbsp;the&nbsp;classification&nbsp;techniques&nbsp;in&nbsp;the&nbsp;data&nbsp;mining&nbsp;context,&nbsp;Associative&nbsp;Classification&nbsp;was&nbsp;adopted&nbsp;due&nbsp;to&nbsp;the&nbsp;descriptive&nbsp;nature&nbsp;of&nbsp;rules.&nbsp;Accordingly,&nbsp;when&nbsp;the&nbsp;model&nbsp;rises&nbsp;a&nbsp;fault&nbsp;prediction,&nbsp;the&nbsp;classifier&nbsp;rule&nbsp;also&nbsp;involves&nbsp;a&nbsp;root&nbsp;cause&nbsp;for&nbsp;the&nbsp;problem.&nbsp;The&nbsp;event&nbsp;data&nbsp;constructed&nbsp;after&nbsp;the&nbsp;process&nbsp;control&nbsp;was&nbsp;analyzed&nbsp;to&nbsp;extract&nbsp;rules&nbsp;with&nbsp;Association&nbsp;Rule&nbsp;Mining&nbsp;(ARM)&nbsp;technique.&nbsp;RuleGenerator,&nbsp;an&nbsp;implementation&nbsp;of&nbsp;the&nbsp;Apriori&nbsp;Algorithm&nbsp;has&nbsp;been&nbsp;employed&nbsp;for&nbsp;the&nbsp;discovery&nbsp;of&nbsp;rules&nbsp;and&nbsp;classifier&nbsp;rules.&nbsp;Among&nbsp;the&nbsp;7295&nbsp;rules&nbsp;discovered,&nbsp;91&nbsp;classifier&nbsp;rules&nbsp;were&nbsp;presented&nbsp;in&nbsp;the&nbsp;results.&nbsp;With&nbsp;a&nbsp;minimum&nbsp;lift&nbsp;value&nbsp;of&nbsp;∼22.3&nbsp;among&nbsp;the&nbsp;classifiers,&nbsp;the&nbsp;model&nbsp;might&nbsp;be&nbsp;promising&nbsp;despite&nbsp;the&nbsp;limited&nbsp;size&nbsp;of&nbsp;the&nbsp;dataset&nbsp;analyzed.</p>

Detay görüntüle