Forecasting product quality using peephole long short term memory

Le, T., Vu, HC, Ponchet-Durupt, A., Boudaoud, N. et Cherfi-Boulanger, Z. (2023) Forecasting product quality using peephole long short term memory. Dans CIGI Qualita MOSIM 2023, Trois-Rivières, Québec, Canada DOI 10.60662/8gn5-5575.

Prévisualisation

Texte CIGI_QUALITA_MOSIM_2023_paper_7904.pdf Télécharger (835kB) | Prévisualisation

Résumé

Predicting the future health of production systems in manufacturing through forecasting is a reliable method. Manufacturers can detect potential issues early and take appropriate actions by using forecasting techniques. While many studies focus on predicting system health using process data such as sound, vibration, and magnetic readings of machines, there are limited studies that use product quality data to forecast the evolution of manufacturing systems. Besides, forecasting the short-term product quality time series in manufacturing systems, which are nonlinear, non- stationary, andnonseasonal, can be challenging. This study explores the use of peephole long short-term memory-based recurrent neural networks to forecast product quality data and address this challenge. Peephole LSTM-based RNN can effectively utilize the long-term dependencies in the product quality time series for more accurate forecasting. Experiments have been conducted to demonstrate that LSTM- based RNN can accurately forecast the complex quality of time series with a long forecasting horizon, and its performance is superior compared to many other forecasting methods.

Type de document:	Document issu d'une conférence ou d'un atelier (NON SPÉCIFIÉ)
Mots-clés libres:	Time series forecasting Deep learning and Smart manufacturing
Date de dépôt:	17 août 2023 13:23
Dernière modification:	25 août 2023 19:13
URI:	https://collection-numerique.uqtr.ca/id/eprint/2057

Actions (administrateurs uniquement)

Éditer la notice