Design and maintenance of control systems integrating AI techniques and Industry 4.0

Effective control loops are essential to manage and operate industrial processes such as pulp and paper mills. High performance of the control systems can minimize the undesirable process variations and then providing high-quality products with lower energy consumption and costs. However, designing and maintain control loops in complex and integrated industrial plants are very tedious tasks for operators and engineers. 
Recently and mainly during the last decade, Artificial intelligence (AI) techniques have shown their high capacity to model and understand a complex phenomenon, also, the more data is available, the more the performance of some AI techniques is high. The success of AI techniques has been achieved in part thanks to the development of advanced parameters optimization algorithms and the availability of power computing (GPU and distributed cloud computing). AI techniques are a powerful tool to extract patterns and recognize trends in structured and unstructured data. AI has had great success in the complete automation of driving, robot decision making, etc. Now, why not using AI to revolutionize the way industrial processes are controlled and to achieve fully autonomous plants. The use of AI techniques in the Industry 4.0 era to control industrial processes is still in its infancy. We believe there is a need for a very comprehensive study and analysis of how AI and data can be used to take current control system technologies to an entirely new level. The project aims to answer the following points with a view to defining new axes of research in this area:
1. How the technologies of industry 4.0 can facilitate the integration of AI techniques into the design and maintenance of control systems?
2. Is it feasible to auto-tune the parameters of the traditional PID controller using AI and data?
3. What are the existing restrictions and opportunities to apply Reinforcement Learning toward full automation of industrial plants?
4. Can the new development in optimization, deep learning, and fast computing make possible to use widely model predictive model (MPC)?
5. What is the role of digital twins (physical-based model or data-driven model) to design, commission, and maintain control-loops?
6. How interpretable AI techniques can help to find easily the manipulated variables and its dependencies with the controlled variables?                            

The study will target mainly the pulp and paper industry and can be extended to other sectors such as mining, refineries, etc.
 

Connaissances requises


Génie électrique, contrôle, AI 
 

Programme d'études visé

Maîtrise avec projet

Domaines de recherche

Énergie

Financement

Bourse de 12000$ pour une année 

Autres informations

Début  : 2020-09-01 
Partenaire impliqué : CanmetENERGY

Personne à contacter

Maarouf Saad | maarouf.saad@etsmtl.ca