What is CHAIKMAT?
CHAIKMAT is a research project funded by the French National Agency of research ANR that aims to
add flexibility and transparency to manufacturing through trustful automatic decision-making.
To do so, we propose a human-centric Artificial Intelligence (AI) approach that investigates whether
an available set of machines can perform a specific production processand then provides human experts
with meaningful explanations of how the decision process is conducted.
How?
CHAIKMAT propose a hybrid approach consisting of using semantic reasoning and machine learning process through a
manufacturing commonsense knowledge graph (MCSKG) that integrates the machines capabilities ontologies, the machines monitoring
data, and the manufacturing commonsense knowledge. Indeed, a knowledge graph is a new paradigm for data integration that inherently
represents a collection of interlinked descriptions of entities–objects, events, or concepts to put data in context and enrich
it semantically.
The Figure below presents the overall structure of the project. At the heart of the system, a hybrid predictive model
combining ML and semantic reasoning will be responsible for helping users (e.g., shopfloor manager, salesperson) to decide
whether a certain work order with constraints, such as due date, cost, and tolerances, can be produced using the manufacturing
resources in their disposal. However, this hybrid predictive model will also be able to explain its prediction so that the user
can fully comprehend the rationale behindsuch a decision. For example, how such a work order can be processed successfully if the answer is yes and for what reason such order cannot be processed, otherwise. The hybrid predictive model
will be trained with linked data coming from two different sources-technically, from two different triple stores.
The first triple store will contain the capability of various machines and equipment that the shopfloor possesses. This triple store will be populated by machine capability information by aggregating machine monitoring data, which will be collected by sensors capturing the real-time performance of the machines in the production line as well as from standard capability information provided by individual vendors of those machines. The second triple store will contain manufacturing commonsense data (e.g., know-how, background knowledge, and experience), which will be curated from domain experts. A semantic query (e.g., SPARQL, SHACL) interface will provide the predictive models access to the linked data stored in the triple stores. Additionally, an intuitive and user-friendly dashboard will be built to facilitate the visualization of the explanations generated by the hybrid predictive model.
MCSKG construction:
Acquire knowledge from domain experts and industrial stakeholders to develop the industrial
commonsense knowledge.
Define a domain ontology that formalizes machines’ capabilities. Then, the domain ontology will be instantiated
using real monitoring data generated by the flexible test plant to propose a knowledge graph.
Modeling of machine learning technique through MCSKG: Exploit the developed knowledge graph to model a graph-native machine learning technique.
Development of reasoning and explanation techniques: Propose or adapt/adopt an XAI technique that enables to explain the reason behind the results of the skill matching process and whether manufacturing a specific product is realizable or not within the factory.
Application and Validation: Design a prototype of a test plant; the assembly of specific products by various mini transportation systems andseveral mini robot stations with different capabilities is varied by additional product inlays as well as different shapes, materials, weights, and dimensions leading to a high variety in product customization.
Development of a visualization approach for explainable AI: Develop a human-centered approach to expose the whole reasoning and learning process througha user-friendly explanation. Such an approach will empower the trust of human actors in the proposed AI system.
 |
 |
 < |
 |
 |
|
● Prof. Hedi KARRAY is the scientific coordinator of the project. He is currently a Professor of Computer Engineering
at the National School of Engineering of Tarbes (ENIT). He received a Ph.D. degree in applied informatics from the University of
Franche-Comté, France, in 2012, an MBA in innovation from IAE Besancon in 2013, and the Habilitation to Lead Research (HDR) from
the National Polytechnic Institute of Toulouse in 2019. He joined ENIT as an Associate Professor in September 2013. He is at the
head of PICS Research Group at Systems Department of the Production Engineering Laboratory (LGP). Since 2016, he became a senior
scientist at the National Center for Ontological Research. Karray is also a technical committee member of several international
research groups such as IEEE SMC, IFAC 5.4, InteropVlab, and Industrial Ontologies Foundry. He is the president of the French
Chapter of IEEE SMC. He coordinated and participated in several collaborative research projects (H2020, Intereg, Region, PhC, etc.)
on the topics of ontology-based engineering, semantic interoperability, and decision support systems. He is the technical project
manager of the H2020 OntoCommons project. He is the head of the international master’s degree on Industry 4.0 at ENIT.
● Prof. Farhad Ameri is currently a Professor of Manufacturing Engineering and Technology in the Department of Engineering
Technology at Texas State University and the head of the Engineering Informatics (INFONEER) research group. His research focuses
on Data-driven Production Management, and Digital Manufacturing. Dr. Mayank Kejriwal, is a Research Assistant Professor in the Department
of Industrial and Systems Engineering at the University of Southern California (USC).
● Dr. Mayank Kejriwal 's research focuses on knowledge graphs and explainable AI. Currently, he is a Principal Investigator
on ISI's funded effort under the DARPA SAIL-ON (Science of Artificial Intelligence and Learning for Open-world Novelty) program
and a co-PI on ISI's funded effort under the DARPA Machine Common Sense program.
● Dr. Raymond Houe is Associate Professor in Computer Engineering at ENIT. His research interests are in the field of Decision-Making,
in relation to knowledge engineering and machine learning, combined with multi-criteria decision methods under uncertainty and incompleteness
in the field of systems monitoring.
● Prof. Rahul Rai is dean’s Distinguished Professor in the Clemson University International Center for Automotive Research (CU-ICAR).
He directs the Geometric Reasoning and Artificial Intelligence Lab. His research is focused on developing computational tools for manufacturing
by combining engineering innovations with methods from machine learning, AI, statistics and optimization, and geometric reasoning.
● Prof. Bernard Archimède is the Director of the LGP-ENIT laboratory. He received the PhD degree in Computer Sciences applied to Industry
from the University Bordeaux 1 in 1991. Since this date, he is researcher in the Production Engineering Laboratory of the National School of
Engineers at Tarbes where he is Full Professor. He has a broad interest in intelligent manufacturing-related issues such as distributed planning
architectures, distributed simulation, multi-agent systems.
● Dr. Linda Elmhadhbi is Associate Professor in Computer Engineering at the National Institute of Applied Sciences (INSA) of Lyon. She is researcher in Decision & Information Systems for Production systems Laboratory (DISP-lab). She
received her Ph.D. degree on applied informatics from the National Polytechnic Institute of Toulouse in 2020. Her Ph.D. dissertation concerned an ontological approach for services' interoperability in the context of crisis management.
Involved in different projects, she investigates decision support systems, knowledge management, domain and modular ontologies, upper-level ontologies, and the exploitation of ontologies to empower semantic interoperability.
● Dr. Arkopaul Sarkar is currently a senior researcher at ENIT, whose primary research area is in knowledge engineering
and its application in industries, especially manufacturing systems. He completed his doctoral study from Ohio University,
USA, with a dissertation titled “Semantic Agent-Based Process Planning for Distributed Cloud Manufacturing” which was
completed under the guidance of Dr Dusan Sormaz. Dr Sarkar has published several articles in peer-reviewed proceedings
and journals and continue his research in various aspects of modern industrial aspects, such as knowledge-driven
production planning, interoperability among product life cycle data, and capability of manufacturing resources and assets.
At present, he is a member of the European Commission-funded OntoCommons project, aiming to establish interoperability
among industrial data, and Industrial Ontology Foundry, which is a collaborative project to develop a core ontology
for industries.
● Syed Muhammad Raza Naqvi is currently a PhD student at ENIT. He received
his master’s degree in computer science from the Superior University Lahore,
Pakisatan and the Bachelors degree in Computer Science from the University of Lahore, Pakistan.
He has been working in the area of knowledge technologies and Human Computer
Interaction. He has participated in the development of advance drug-drug ontology
(ADDI), ontology driven health integration systems, and he investigates,
domain ontologies, knowledge graphs, and data integration.