Skip to Main content Skip to Navigation
Theses

Contribution à l' analyse de causalité par apprentissage automatique pour l'aide à la décision, dans un contexte de supervision des systèmes pour l'industrie 4.0

Abstract : With the advent of Industry 4.0 and the accompanying acceleration of processes and data proliferation, the challenge for decision-making processes is to ensure rapid and reliable decision making. Key Performance Indicators (KPIs) are closely linked to decision making : they are both triggers and drivers. Thus, in order to improve decision-making processes, focus should be on KPIs. When a KPI reveals an abnormal situation, understanding the origin of the deviation is essential to look for solutions, and to select one among several. In this thesis, we are interested in this understanding, in particular in identifying the causal links between a KPI of interest and the manipulatable contextual variables, and in quantifying these causal links. To this end, we propose a causality-oriented decision support system that fulfils three functions : the identification of contextual variables causally linked to a KPI in the form of a causal structure ; the prioritisation of these variables according to their respective strengths of association with the KPI of interest ; and the possibility of predicting the KPI for proactive purposes. The first function aims to provide a better understanding of KPI deviations. It is implemented through a causal Bayesian network learning algorithm. The second function allows a better selection of best solutiond. It is implemented thanks to a computing that we propose to make on the final weights of a neural network having a good predictive power of the KPI. The third function, allows proactive decision making, it is made possible by the same neural network. The method was validated using two benchmarks and compared to other techniques with the same objectives.
Complete list of metadata

https://pastel.archives-ouvertes.fr/tel-03635890
Contributor : ABES STAR :  Contact
Submitted on : Friday, April 8, 2022 - 6:21:08 PM
Last modification on : Friday, August 5, 2022 - 2:54:01 PM
Long-term archiving on: : Saturday, July 9, 2022 - 7:18:28 PM

File

100801_AMZIL_2022_archivage.pd...
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-03635890, version 1

Citation

Kenza Amzil. Contribution à l' analyse de causalité par apprentissage automatique pour l'aide à la décision, dans un contexte de supervision des systèmes pour l'industrie 4.0. Génie des procédés. HESAM Université, 2022. Français. ⟨NNT : 2022HESAE003⟩. ⟨tel-03635890⟩

Share

Metrics

Record views

163

Files downloads

56