PILOT CASE
Aptiv
EV battery production
The APTIV pilot considers the production of battery trays for EVs, which includes the following steps. The welding process that is utilised in the production is a hybrid arc and laser welding process. The handling of the components is mainly automated with little intervention from the operators.
The main objective of APTIV is to minimise the number of defects that are generated during the production of the battery trays and enhance the reconfigurability of the process.
Pilot Description
Expected Goals
- Defects reduction by 50%
- Increased throughput by 5%
- Decreased production costs by 10%
Problem definition
An in-line quality assessment system for the welding process, and feeding control algorithms that adapt the process parameters in real-time are of utmost importance to ensure weld consistency. A closed-loop control system that will accurately measure the gap between the components before the weld and fix any misalignments is required.
Moreover, considering the optimisation of the production of the battery trays, as well as the need for digitalisation of the production, a solution is needed to collect production data in a non-destructive approach, incorporate data analytics to enhance the reconfigurability as a result of quality assessment of the process, aid decision-making and contribute to the goal of zero-defect.
Identified challenges
- A very significant challenge with battery tray welding is the length of the bead (~1m), making it very difficult to achieve a consistent, high-quality weld in the whole length of the bead.
- Another challenge is related to the gap control between the individual components.
Impact from openZDM technologies
The expected impact of openZDM project include reduction of defected welds on battery modules, early detection of defects and improvement the throughput. Through the reduction of defective parts, less time and resources will be consumed for reworking. Moreover, it will improve the sustainability of the production process.
The goal of the openZDM project is to work on the prediction of defected laser welding on battery modules images at before, during and after the process:
- Usage of images (2D and thermal) to predict possible defects, correlated with historical data.
- Help the operator to adjust in real-time machine parameters, in order to reduce the chances of a defected weld.
Currently, NDIs are both installed and online:
- 2D cameras at the entrance of the station, to take pictures before and after the welding process.
- IR cameras inside the station, to take pictures and make evaluation of the welding process.
Moreover, we are working on models fine tuning that are currently predicting 1 specific defect and aesthetical defects. The models that have been deployed are weld detection and segmentation with a dataset of more than 2000+ images. Two models to predict the defect have been approached (CNN and ViT), both have been trained with more than 65000 images of both good and defected welds.
Current preliminary results show good predictive capabilities for the short-term, but still presenting some bias in specific defects and process situations, but there’s a big window for improvements in reducing that bias.
Finally, different dashboards have been developed, showing in real time to the operator what is happening and proving useful information on the current status of the welding process.
Leading company
APTIV is a global technology leader, with more than 200,000 people across 131 manufacturing facilities and 11 major technical centers worldwide. With a presence in 48 countries, APTIV address mobility’s toughest challenges through deep software and systems integration expertise, delivering market relevant solutions for customers.
