Evolutionary LSTM-FCN networks for pattern classification in industrial processes
| Autor: | Fernando Veiga, Basilio Sierra, Patxi Ortego, Alberto Diez-Olivan, Javier Del Ser, Mariluz Penalva |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Hyperparameter
General Computer Science Computer science Process (engineering) business.industry General Mathematics Deep learning 05 social sciences Evolutionary algorithm 050301 education 02 engineering and technology Machine learning computer.software_genre Nonlinear system 0202 electrical engineering electronic engineering information engineering Feature (machine learning) 020201 artificial intelligence & image processing Artificial intelligence Scenario testing Time series business 0503 education computer |
| Zdroj: | Swarm and Evolutionary Computation. 54:100650 |
| ISSN: | 2210-6502 |
| DOI: | 10.1016/j.swevo.2020.100650 |
| Popis: | The Industry 4.0 revolution allows gathering big amounts of data that are used to train and deploy Artificial Intelligence algorithms to solve complex industrial problems, optimally and automatically. From those, Long-Short Term Memory Fully Convolutional Network (LSTM-FCN) networks are gaining a lot of attention over the last decade due to their capability of successfully modeling nonlinear feature interactions. However, they have not been yet fully applied for pattern classification tasks in time series data within the digital industry. In this paper, a novel approach based on an evolutionary algorithm for optimizing the networks hyperparameters and on the resulting deep learning model for pattern classification is proposed. In order to demonstrate the applicability of this method, a test scenario that involves a process related to blind fastener installation in the aeronautical industry is provided. The results achieved with the proposed approach are compared with shallow models and it is demonstrated that the proposed method obtains better results with an accuracy value of 95%. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect