 Time series fault prediction via dual enhancementYi Wang (),
Wencong Xu (),
Chaofei Wang (),
Yanbin Huang () and
Heming Zhang ()
Journal of Intelligent Manufacturing, 2025, vol. 36, issue 8, No 3, 5247-5262 Abstract: Abstract The industrial fault prediction based on time series data inference is a very meaningful work. Generally, this task has two key difficulties including the limited amount of fault data and learning of timeline-based knowledge. In this paper, we propose a novel dual enhancement method to conduct the time series fault prediction task. It consists of a time series sample augmentation (TSSA) strategy and a timing integration attention enhancement model, named time series enhanced transformer (TSET). Specifically, we decompose the original signal by variational mode decomposition and then add the noise signal for recombination, so as to augment the time series samples. Furthermore, to solve the problem that the traditional transformer is difficult to adequately extract the features from time series data, we propose a new timing association attention module to enhance the time series feature learning. It obtains the comprehensive temporal related features formed by a time integration function and then inputs them to the different channels of the attention mechanism in the transformer architecture. A TSET is constructed by the dual enhancement. We validate the proposed method on both public and industrial datasets. The experimental results show that our method can outperform the mainstream baseline methods. Note to practitioners—The model in this paper is applied to fault prediction in industrial automation processes. The log and fault record data generated in industrial processes are natural time series data. Time series prediction models generated in the field of Natural language processing need a lot of corpus data to generate reasonable inference results. However, the fault data of industrial processes usually cannot reach an order of magnitude sufficient for these models to operate well. On the other hand, these corpus based data models are not friendly to industrial fault data. The model in this paper is established from two aspects: data enhancement and temporal correlation enhancement, so that the industrial fault sample data can meet the needs of the prediction model as much as possible. By enhancing temporal correlation features, the prediction model becomes friendly to industrial temporal fault data. The method is validated through experiments on actual industrial process data and public datasets. In practical applications, inputting data into the model proposed in the paper generates prediction results based on time series, and distinguishes whether the predicted data is a fault, thereby providing fault warning, which is helpful for the automation of the entire industrial life cycle. Keywords: Data augmentation; Dual enhancement; Time series fault prediction; Timing integration attention; Transformer (search for similar items in EconPapers) Downloads: (external link) Related works: Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text Persistent link: https://EconPapers.repec.org/RePEc:spr:joinma:v:36:y:2025:i:8:d:10.1007_s10845-024-02515-y Ordering information: This journal article can be ordered from DOI: 10.1007/s10845-024-02515-y Access Statistics for this article Journal of Intelligent Manufacturing is currently edited by Andrew Kusiak More articles in Journal of Intelligent Manufacturing from Springer |
Is your work missing from RePEc? Here is how to contribute.
Questions or problems? Check the EconPapers FAQ or send mail to .
EconPapers is hosted by the
School of Business at Örebro University.