[2407.01644] Evaluating the Function of Information Enrichment Approaches In direction of Uncommon Occasion Evaluation in Manufacturing

[Submitted on 1 Jul 2024]

View a PDF of the paper titled Evaluating the Function of Information Enrichment Approaches In direction of Uncommon Occasion Evaluation in Manufacturing, by Chathurangi Shyalika and 4 different authors

View PDF
HTML (experimental)

Summary:Uncommon occasions are occurrences that happen with a considerably decrease frequency than extra frequent common occasions. In manufacturing, predicting such occasions is especially vital, as they result in unplanned downtime, shortening gear lifespan, and excessive power consumption. The incidence of occasions is taken into account frequently-rare if noticed in additional than 10% of all cases, very-rare whether it is 1-5%, moderately-rare whether it is 5-10%, and extremely-rare if lower than 1%. The rarity of occasions is inversely correlated with the maturity of a producing business. Usually, the rarity of occasions impacts the multivariate knowledge generated inside a producing course of to be extremely imbalanced, which results in bias in predictive fashions. This paper evaluates the function of information enrichment methods mixed with supervised machine-learning methods for uncommon occasion detection and prediction. To handle the information shortage, we use time collection knowledge augmentation and sampling strategies to amplify the dataset with extra multivariate options and knowledge factors whereas preserving the underlying time collection patterns within the mixed alterations. Imputation methods are utilized in dealing with null values in datasets. Contemplating 15 studying fashions starting from statistical studying to machine studying to deep studying strategies, the best-performing mannequin for the chosen datasets is obtained and the efficacy of information enrichment is evaluated. Based mostly on this analysis, our outcomes discover that the enrichment process enhances as much as 48% of F1 measure in uncommon failure occasion detection and prediction of supervised prediction fashions. We additionally conduct empirical and ablation experiments on the datasets to derive dataset-specific novel insights. Lastly, we examine the interpretability facet of fashions for uncommon occasion prediction, contemplating a number of strategies.