Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
Thesis offerMetaheuristics and user-driver learning algorithms for
accelerating the Simulation-Optimisation-Problem solving loop
Keywords : A.I, optimisation, industry 4.0, innovation, inventive design, simulationThesis supervisors : Roland DE GUIO (INSA-Strasbourg, ICube, Prof. HDR), Pierre PARREND
(ECAM Strasbourg-Europe, ICube, HDR)
Workplace : INSA Strasbourg
Laboratory : Laboratory ICube (UMR7357)
Contacts : [email protected]; [email protected] funding : VIRTFac project co-financed by “INTERREG V Upper Rhine”
Beginning date : October 1st 2019Term :36 months
Candidate profile:
Background in computer science required. Background in statistical optimisation, data science or AI is welcome
Ability to work in a multidisciplinary team
Context This thesis is part of the Interreg VIRTFac (Virtual Innovative Real Time Factory) project: Optimizing the path to Industry 4.0 by planning the right production system at the right time. VIRTFac and the thesis begin on October 1, 2019..
One of the objectives of the VIRTFac project is to support data sharing and analysis between simulation, optimization and invention means during the planning and the reconfiguration of production systems.
The PhD research benefits of the expertise of two research teams of Icube Laboratory: CSTB (Complex Systems and Translational Bio-Informatics) and CSIP (Conception, Système d’Information et Processus inventifs1). The VIRTFac project relies on academic and industrial partners, as well as on resources for data analysis and storage from the BICS Platform2 of ICube Laboratory among others.
1 Which can be translated by Design, Information System and innovative Processes2 http ://bics.unistra.fr/
Objectives The objective of the PhD is to propose a novel model for expert-based learning [1] using metaheuristics and approach to user-driver learning algorithms [2], [3].
This model will be applied to the VIRTFac use case dealing with optimization and innovations [4]–[15] for the Industry of the Future (IoF). It aims at improving the analysis, optimisation and inventive stages of the production system (re)configuration process. It will be implemented through a domain-independent library and a IoF-specific web-based application.
Following properties of expert-based learning are the focus of interest of this work: accelerated learning through expert-based, reinforcement, novelty learning through expert-aided classification, innovation support through expert investigation. According to the literature review and to the needs of the VIRTFac project, the thesis will focus on one of these three properties.
Organisation of work The work will be organised in three phases:
Phase 1. Positioning the PhD project in the VIRTFac project. This stage is overlapping in time with phase 2.
Phase 2. Proposal of a model addressing the problem chosen in phase 1 and first experimentations using synthetic data.
Phase 3. Evaluation of the model in the innovation process of VIRTFac, and suitable enrichment of the model.
Phase 4. Refinement of learning mechanisms and benchmark of the proposed model.
Phases (1&2), 3, 4 last approximately one year.
Phase 1 is short but very important as the PhD project takes place in the context of VIRTFac, which is a multi-disciplinary project. It will be a question of making, within the framework of the project team, a statement of the activities of the optimization, simulation and invention loop that can benefit from the metaheuristics and learning process. This step will help to select the most promising applications, ensure consistency between the various outputs of the VIRTFac project and guide the choice of the model proposed in Phase 2 and of one the three property mentioned above. This phase will study the relevance of the man-hill learning model in this context. This phase will conclude with the choice of a problem and a model to explore.
Phase 2 will start by the formalisation of model proposed in phase 1 and/or used in previous projects of CSTB team : HuMa [16], Juxta [17], Online learning [2], [3]. A running example including the representation of VIRTFac innovation data will be devised to ensure both a realistic evaluation of the proposed model and the compliance with the requirements of the VIRTFac project [9]. The model will support the three properties of man-hill systems: expert based reinforcement, expert-aided classification and expert investigation. It will focus on one single of these properties, to be identified according to the project requirement -expert-aided classification is a likely candidate to demonstrate the learning rate enhancement as well as enrichment of the model through expert feedback. The first implementation will be delivered as part of the v1 library.
In parallel, a literature review will be performed on 1) ant-hill approaches, in particular their application to learning like ant-miner [18], [19], 2) man-hill models and 3) expert-supported learning [20].
In Phase 3, a demonstrator will be developed to support gathering expert feedback through a project-dedicated application. This application will focus on the enhancement of CSIP data based innovation process. Required features as well as additional test cases will be included in the v2
document.docx 2
library. This library will be used in direct support of VIRTFac web software. Industry-level test cases, on the one hand, as well as representative synthetic data, on the other, will be defined and related data will be generated and made available through a platform. The library for generating synthetic data, DataCrafter will be made available through BICS platform too.
In Phase 3, the proposed model will be enhanced through one of following strategies: 1) support of additional properties of expert-based learning; or 2) deepening of the proposed model for a key characteristic to be identified during the first 2 years of the project. The model will undergo a systematic benchmark campaign.
6 months are dedicated to the redaction of the PhD thesis.
Collaborations This work is directed by Pierre Parrend (ECAM Strasbourg-Europe, UMR7357 ICube laboratory), and Roland De Guio (INSA-Strasbourg, UMR7357 ICube laboratory). The work will also be made in collaboration with Virtual Engineering Lab of University of Offenburg (Germany) which is partner of the VIRTFac project.
The e-laboratory 4P-Factory of the Unitwin3 UNESCO Complex System Digital Campus (CNSC) will host this project. In addition to VIRTFac plan for dissemination, the regional network of platforms REGESIF (Initiative Grand-Est des Services et Industries du Futur) also provides a suitable community for the dissemination of the result of the project.
Références [1] P. Parrend, F. Guigou, J. Navarro, A. Deruyver, et P. Collet, « Artificial Immune Ecosystems: the
role of expert-based learning in artificial cognition », in ICAROB 2018| The International Conference on Artificial Life and Robotics, 2018, p. 5.
[2] S. Gutiérrez, G. Valigiani, P. Collet, et C. D. Kloos, « Adaptation of the ACO heuristic for sequencing learning activities », in Proceedings of the EC-℡, 2007, p. 17–20.
[3] G. Valigiani, R. Biojout, Y. Jamont, E. Lutton, P. Collet, et others, « Experimenting with a real-size man-hill to optimize pedagogical paths », in Proceedings of the 2005 ACM symposium on Applied computing, 2005, p. 4–8.
[4] S. Bach, R. De Guio, et G. Nathalie, « Combining discrete event simulation, data analysis, and TRIZ for fleet optimization », J. Eur. TRIZ Assoc. Innov., vol. 04, no 02, p. 47–61, 2017.
[5] F. Ben Moussa, R. Benmoussa, R. de Guio, S. Dubois, et I. Rasovska, « An algorithm for inventive problem solving coupled with optimization for solving inventive problems encountered in supply chains », in SIL’16, 2016.
[6] F. Z. Ben Moussa, I. Rasovska, S. Dubois, R. D. Guio, et R. Benmoussa, « Reviewing the use of the theory of inventive problem solving (TRIZ) in green supply chain problems », J. Clean. Prod., vol. 142, p. 2677–2692, 2017.
[7] F. Z. BenMoussa, S. Dubois, R. De Guio, I. Rasovska, et R. Benmoussa, « Integrating the Theory of Inventive Problem Solving with Discrete Event Simulation in Supply Chain Management », in Automated Invention for Smart Industries, 2018, p. 330–347.
[8] L. Burgard, S. Dubois, R. de Guio, et I. Rasovska, « Sequential experimentation to perform the Analysis of Initial Situation », in TRIZ Future 2011, 2011.
[9] H. Chibane, S. Dubois, et R. D. Guio, « Automatic Extraction and Ranking of Systems of Contradictions Out of a Design of Experiments », in Automated Invention for Smart Industries, 2018, p. 276–289.
3 http ://unitwin-cs.org/
document.docx 3
[10] S. Dubois, R. de Guio, et I. Rasovska, « From simulation to invention, beyond the pareto-frontier », in 20th International Conference on Engineering Design (ICED), 2015.
[11] S. Dubois, T. Eltzer, et R. De Guio, « A dialectical based model coherent with inventive problems and optimization problems », Comput. Ind., vol. 60(8), p. 575–583, oct. 2009.
[12] S. Dubois, R. De Guio, A. Brouillon, et L. Angelo, « A Feedback on an Industrial Application of the FORMAT Methodology », in Automated Invention for Smart Industries, 2018, p. 290-301.
[13] I. Rasovska, R. de Guio, et S. Dubois, « Using dominance relation to identify relevant generalized technical contradictions in innovative design », in IESM 2017, 2017.
[14] L. Lin, S. Dubois, R. de Guio, et I. Rasovska, « An exact algorithm to extract the generalized physical contradiction », Int. J. Interact. Des. Manuf., vol. 9, no 3, p. 185–191, 2015.
[15] L. Lin, I. Rasovska, R. de Guio, et S. Dubois, « Optimization Methods for Inventive Design », in TRIZ – The Theory of Inventive Problem Solving, Springer., D. Cavallucci, Éd. Cavallucci, Denis, 2017, p. 151–185.
[16] J. Navarro, A. Deruyver, et P. Parrend, « Morwilog: an ACO-based System for Outlining Multi-Step Attacks », in IEEE Symposium Series on Computational Intelligence (IEEE SSCI 2016), 2016.
[17] P. Parrend, C. Maller, et E. Dietrich, « The Ant Reconciliation Algorithm (ARA): Ant-hill learning for label matching », in Artificial Evolution, 2017.
[18] F. E. Otero, A. A. Freitas, et C. G. Johnson, « cAnt-Miner: an ant colony classification algorithm to cope with continuous attributes », in International Conference on Ant Colony Optimization and Swarm Intelligence, 2008, p. 48–59.
[19] R. S. Parpinelli, H. S. Lopes, et A. A. Freitas, « Data mining with an ant colony optimization algorithm », IEEE Trans. Evol. Comput., vol. 6, no 4, p. 321–332, 2002.
[20] F. Guigou, P. Collet, et P. Parrend, « The Artificial Immune Ecosystem: a bio-inspired meta-algorithm for boosting time series anomaly detection with expert input », in EvoApplications, 20th European Conference on the Applications of Evolutionary Computation, 2017.
document.docx 4