Vicomtech – Optimising the Design of Large CV Joints Using AI
Vicomtech
Sector: Industry
Business Case
The design and implementation of software based on machine learning involve the analysis of parameters and proposed designs to identify potential failure modes, eliminating the continuous use of test benches. This approach is expected to result in significant energy, resource, and production cost savings.
Objectives
Develop a new calculation tool for dimensioning and optimising CV joints, with the aim of predicting their damage and life under certain load conditions. This would serve as a basis for the development of software for real-time damage estimation and predictive maintenance of joints in vehicles, depending on the way they are driven.
Use case
The use of artificial intelligence algorithms trained on historical data from various sources, including test bed data or real-time data, will allow the prediction of both expected life and induced joint damage under a set of specified conditions, which will also allow automated preliminary joint sizing.
Infrastructure
Hybrid
Technology
Machine learning and deep learning
Data
Historical test bed data (tabular data), joint characterisations and damage assessment sheets. Real-time driving data: multisensor time series.
Resources
1 project manager, 2 data scientists. Support and guidance from product development engineers (boards). Technical development by Vicomtech, and technical validation and requirements by GKN.
Difficulties and learning
The main difficulty arises from the considerable variability in both the damage incurred and the service life under similar conditions and characteristics of both the test and the seals themselves. The challenge is to overcome the inherent bias of the models due to densely populated areas and noise by achieving sufficient predictive capability.
KPIs (business impact and metrics of the model)
RMSE: root mean square error MAPE: mean absolute percentage error Coefficient of determination (R-squared)
Funding
The project was funded by the Hazitek programme.
Collaborators, Partners
The project was developed for GKN.
