A predictive maintenance model for optimizing production schedule using deep neural networks. (January 2022)

• Record Type: Journal Article
• Authors: Zonta, Tiago; da Costa, Cristiano André; Zeiser, Felipe A.; de Oliveira Ramos, Gabriel; Kunst, Rafael; da Rosa Righi, Rodrigo
• Is Part Of: Journal of manufacturing systems. Volume 62(2022)
• Page Start: 450
• Page End: 462
• Publication Date: 2022-01
• Subjects: Predictive maintenance -- Scheduling problems -- Deep neural networks -- Process integration -- Industry 4.0; Manufacturing processes -- Periodicals; Production engineering -- Data processing -- Periodicals; Robots, Industrial -- Periodicals; Production, Technique de la -- Informatique -- Périodiques; Robots industriels -- Périodiques; Electronic journals; 670.42
• DOI: 10.1016/j.jmsy.2021.12.013
• Languages: English
• ISSNs: 0278-6125

Abstract

Graphical abstract: Abstract: Industry 4.0 (I4.0) provides connectivity, data volume, new devices, miniaturization, inventory reduction, personalization, and controlled production. In this new era, customization and data availability are essential to generate information that allows decision-making. The possibility of predicting the need for maintenance in the future and using this information for other processes is one of the challenges of the manufacturing process. In this context, this article presents a model to optimize maintenance and production schedules predictively and automatically, called Predictive Maintenance (PdM) & Schedule (PdMS). Usually, these solutions are addressed separately in literature. We aim to improve the machines' Remaining Useful Life (RUL) prognosis based on sensor telemetry and operating information. We also aim to predict if the machines will be part of the production schedule, considering the available resources. We propose a model that describes data engineering, validation, and normalization. It also describes our approach to create and combine degradation indices using similarity patterns that allow noisy data to identify time-based failures. This approach allows for the use of this type of prediction in scheduling problems. We compare several models based on deep neural networks (DNN) and recurrent neural networks (RNN), using criteria based on visual analysis, errors, regression coefficient R 2, and accuracy. The best result, RMSE = 8.789, MSE = 77.253, MAE = 2.262, R 2 = 0.848, Accuracy = 92.22, allowed for predicting failures with five-day anticipation, which avoids downtime and makes the intended integration possible.