摘要：AbstractThis paper proposes an efficient multivariate time-series fault detection and classification approach aiming to detect faulty wafers (i.e. pieces of silicon) during semiconductor manufacturing process. This approach is based on using Independent Component Analysis (ICA) and several Machine Learning Ensemble Techniques. The main objective is to extract the most useful information from each time-series and combine them to build a set of fully concatenated features. Thereafter, Extra Trees, Random Forest, Gradient Boosting and Extreme Gradient Boosting, one of the prevalent evolutions of tree-based algorithms, are fitted to the extracted features subset to design and implement an efficient anomaly detection strategy. The obtained results show that the proposed technique is very efficient and very promising.
关键词：KeywordsAnomaly detectionData-driven diagnosis methodsMachine learningMultivariate time-series data
