A Self-learning Classification Framework for Industrial Time Series Streams. Issue 13 (2019)

• Record Type: Journal Article
• Authors: Ma, Zhenjie; Shi, Ke; Feng, Kang
• Is Part Of: IFAC-PapersOnLine. Volume 52:Issue 13(2019)
• Page Start: 1531
• Page End: 1536
• Publication Date: 2019
• Subjects: Self-learning -- Classification -- Incremental Clustering -- Time Series Streams; Automatic control -- Periodicals; 629.805
• DOI: 10.1016/j.ifacol.2019.11.417
• Languages: English
• ISSNs: 2405-8963

Abstract

Abstract: In Industry 4.0 era, a vast amount of time series streams is pumped by the sensors and other smart devices embedded in manufacturing process. Classification, a fundamental time series analytic task, plays a very important role to support process optimization. A lot of classification methods have been proposed in recent decades. Most of them need a training process to find the classes from training set which includes multiple labeled samples covering all the classes. This assumption does not hold in the processing of continuously coming and dynamically changing industrial time series streams where only partial classes' knowledge can be learned in advance. To address this issue, we propose a self-learning classification framework for industrial time series streams in which incremental clustering based classes learning is performed concurrently with the classification process. We demonstrate the utility of our ideas with experiments on real-world industrial time series streams.