Home // ALLDATA 2023, The Ninth International Conference on Big Data, Small Data, Linked Data and Open Data // View article
Authors:
Gerold Hoelzl
Bastian Fleischmann
Sebastian Soller
Jonas Zausinger
Matthias Kranz
Keywords: sensor based manufacturing dataset; industry; machine learning; anomaly detection; defect detection; industry 4.0; data sharing; toolset for result replication.
Abstract:
We aim at systems that make sense out of occurring anomalies to autonomously learn to predict and detect possible occurring machine drifts, failures and deviations, and the corresponding errors in the machines and products itself. To assess our prediction and classification methods, we collected data from a fully automated industrial machinery including 3 internal sensors in a large-scale dataset (> 87000 manufactured pieces with 39 different product types, in a timespan of nearly 7 months). We present the scenario and describe the collected data and the sensors. We describe the machine data and the corresponding errors, and present a generic tool that allows visualization, scripting, etc., especially when datasets have to be shared, as it gives an insight into the complexity of the data and the algorithms and make experiments as described in the paper reproducible. We argue to be currently in a replication crisis in data analysis that makes it close to impossible to replicate empirical findings due to the lack of the availability of the underlaying data and the implemented algorithms. We reached a point where we need to question if the results can be believed and how the datasets for evaluation are designed and recorded. To support an inevitable fundamental change towards the full openness of published results in collected data and the used algorithmic processing with minimum effort, we present and make publicly available (i) a large-scale dataset for IoT (Internet of Things) based predictive maintenance in an industrial setting combined with (ii) artificial intelligence algorithms used by our group, elaborated on the dataset embedded in (iii) a general tool to foster easily sharing of both for replicating results.
Pages: 11 to 18
Copyright: Copyright (c) IARIA, 2023
Publication date: April 24, 2023
Published in: conference
ISSN: 2519-8386
ISBN: 978-1-68558-041-4
Location: Venice, Italy
Dates: from April 24, 2023 to April 28, 2023