Feature selection for fault detection systems: application to the Tennessee Eastman process

摘要: In fault detection systems, a massive amount of data gathered from the life-cycle equipment is often used to learn models or classifiers that aims at diagnosing different kinds errors failures. Among this huge quantity information, some features (or sets features) are more correlated with kind failure than another. The presence irrelevant might affect performance classifier. To improve system, feature selection hence key step. We propose in paper an algorithm named STRASS, which detecting relevant for classification purposes. certain cases, when there exists strong correlation between and associated class, conventional algorithms fail selecting most features. order cope problem, STRASS uses k-way class select assess we apply it on simulated collected Tennessee Eastman chemical plant simulator. process (TEP) has been many studies three specific faults not well discriminated algorithms. results obtained by compared those reference show selected always classifier whole set original better other