AI to improve clutch management into mechanical CVT transmissions- QC-386

Project type: Research
Desired discipline(s): Engineering - computer / electrical, Engineering, Engineering - mechanical, Engineering - other, Computer science, Mathematical Sciences
Company: CVTCORP
Project Length: 4 to 6 months
Preferred start date: As soon as possible.
Language requirement: Flexible
Location(s): Sainte-Julie, QC, Canada
No. of positions: Flexible
Desired education level: CollegeUndergraduate/BachelorMaster'sPhDPostdoctoral fellowPreferred institutions: École de technologie supérieure, École Polytechnique de Montréal, Université de Sherbrooke, Université Laval

Search across Mitacs’ international networks - check this box if you’d also like to receive profiles of researchers based outside of Canada: 
No

About the company: 

CVTCORP is a company of approximately 60 employees that designs and manufactures transmissions for agricultural and construction vehicles of various capacities. We are therefore a multidisciplinary team made up of engineers responsible for the design of transmissions and various operating personnel dedicated to the production of the latter. We distinguish ourselves by using CVT (Continuously Variable Transmission) technology which maximizes engine energy and thus minimizes losses and fuel consumption.

Our transmissions are controlled by an on-board computer (TCU) which makes sure to select the optimal ratio at all times and to control the clutches engagement. The development of this TCU, as well as the algorithms it uses, is also done within the company by a dedicated electronics and embedded software team.

Keen to develop an ever more efficient and flexible product, we are now working to implement machine learning algorithms that will be used by this TCU to increase the reliability and control of our transmissions.

Describe the project.: 

The CVTCORP artificial intelligence project aims to improve the clutches control, which is a central part of our transmissions and our technology. These clutches are controlled by hydraulic pressure controlled by valves, itself controlled by our embedded computer (TCU). This control is divided into two main segments. The clutch filling and the transmission of torque through this clutch.

The proposed project addresses these two segments. First, we aim to build a reinforcement learning algorithm, dedicated to the embedded platform, to modulate and adapt the filling profile of the clutches according to a series of variables such as temperature, system wear, available oil flow, the control valve itself, and others.

In a second step, another AI algorithm will be necessary to evaluate the right command to be transmitted to the clutch valves to allow the engine to transmit the right torque without slipping the clutches but without overloading the engine or risking mechanical component damages.

At the end of this project, CVTCORP aims to have implemented and validated AI algorithms on the vehicle embedded system. In short, a vehicle will be able to continuously update its clutch calibration by itself.

The main tasks of the candidate will be to support the software development team to guide them through the AI ​​techniques best suited to address the problem mentioned. He will also have to explore and develop learning algorithms on PC platforms and help the embedded programming team to adapt these algorithms to the vehicle's computer. Assistance with additional data acquisition planning and their processing will also be part of the mandate.

CVTCORP already has a dataset containing more than 100 hours of continuous vehicle use recording to help the candidate get started. In addition, the project provides for the acquisition of additional data if necessary.

Required expertise/skills: 

CVTCORP is looking for a master's or doctoral candidate with the following expertise:

  • Analysis and processing of temporal signals.
  • Knowledge of different AI methodologies that can be applied to the control and modeling of physical systems such as clutches.
  • Experience in the development, and implementation of reinforcement learning algorithms and unsupervised learning.
  • General and varied knowledge of mechanical principles related to engines, transmissions, and their internal components such as clutches.
  • Knowledge of the various Python libraries intended for machine learning, Matlab, or other tools likely to facilitate the resolution of the proposed problem.