Home // DATA ANALYTICS 2021, The Tenth International Conference on Data Analytics // View article
Authors:
Ariel Cedola
Rosaria Rossini
Ilaria Bosi
Davide Conzon
Keywords: predictive maintenance, machine learning, fea- ture engineering, manufacturing
Abstract:
Manufacturing systems suffer from progressive degradation due to wear, fatigue, cracking, corrosion, with respect to both age and usage. Reduced performance of system components and even catastrophic failure could be the main consequences of not being able to detect such faults at early times. Fault diagnosis and predictive maintenance aim at showing the machine working conditions, indicating current and possible future abnormal states, and allowing to take appropriate actions in advance in order to avoid damages, minimize downtime, improve the safety of the whole system and reduce manufacturing and repairing costs. In this paper, we successfully apply a data driven modelling approach designed for log data to a new scenario. The methodology proposed transforms the production and stops data of an industrial machine into a thoughtfully elaborated series of timestamped events, and applies a set of feature engineering techniques that enables to exploit the pipelines typically implemented in log-based predictive maintenance modelling. The transformed data is used to train a binary classificator that predicts with high accuracy (96.2%) the occurrence of a machine failure in the short-medium term.
Pages: 14 to 22
Copyright: Copyright (c) IARIA, 2021
Publication date: October 3, 2021
Published in: conference
ISSN: 2308-4464
ISBN: 978-1-61208-891-4
Location: Barcelona, Spain
Dates: from October 3, 2021 to October 7, 2021