A Deep Learning Framework to Predict the Remaining Useful Life in Predictive Maintenance
Proper maintenance of equipment and machinery in the industry is essential to ensure operational efficiency and extend their lifespan. Thus, reducing replacement and repair costs, results in a long-term positive economic impact. The implementation of predictive maintenance strategies in the industry...
| Autores: | , , |
|---|---|
| Tipo de recurso: | artículo |
| Estado: | Versión enviada para evaluación y publicación |
| Fecha de publicación: | 2023 |
| País: | España |
| Institución: | Universidad de Sevilla (US) |
| Repositorio: | idUS. Depósito de Investigación de la Universidad de Sevilla |
| OAI Identifier: | oai:idus.us.es:11441/171988 |
| Acceso en línea: | https://hdl.handle.net/11441/171988 https://doi.org/10.2139/ssrn.4596645 |
| Access Level: | acceso abierto |
| Palabra clave: | Prognostics and Health Management Remaining Useful Life Deep Learning Convolutional Networks |
| Sumario: | Proper maintenance of equipment and machinery in the industry is essential to ensure operational efficiency and extend their lifespan. Thus, reducing replacement and repair costs, results in a long-term positive economic impact. The implementation of predictive maintenance strategies in the industry not only reduces costs associated with breakdowns and unexpected shutdowns but also improves resource planning, enhances productivity and product quality, and strengthens market competitiveness, generating a favorable economic impact. The general aim of predictive maintenance is to identify and address, through the use of data analysis, potential problems in the equipment before they become critical. One of the most important lines of work within this field is Remaining Useful Life (RUL) prediction, where the main goal is to estimate the amount of time a piece of equipment will function effectively before needing to be replaced or overhauled. RUL prediction can be achieved using several deep learning techniques, such as Deep Convolutional Neural Networks (DCNN) and Long Short-Term Memory (LSTM) networks. The aim of this article is twofold. On the one hand, we introduce a comprehensive framework for analyzing very long data sequences (time series in the case of RUL prediction) using stacked networks. The developed framework includes a well-designed cross-validation and hyperparameter optimization process to identify the best models for analyzing such sequences. Results show that the methodology can be applied automatically, without extensive knowledge on the context of the problem nor expertise with the machinery under study. On the other hand, we show the methodology applied on a case study (trying different techniques) where the aim is to predict the RUL of a fleet of aircraft engines, which may experience different failure types. Results showed that DCNN, instead of LSTM networks (which are designed to work with long sequences), is the most promising architecture. |
|---|