CanadaManufacturing2023 – ongoingBallard Power Systems
Machine learning at test stations to shorten fuel-cell stack acceptance tests
Ballard Power Systems cut fuel-cell stack test time by 80% in a proof of concept with Acerta's machine learning
Problem
Factory acceptance tests on Ballard's PEM fuel cell stack assemblies took about 2.5 hours. The company also wanted to find the sources of test failures.
Approach
Acerta Analytics built custom models and ran them on a subset of Ballard's test data in a proof of concept. The two companies then started a two-year project, partly funded by NGen's AI for Manufacturing program. The models are to run on edge devices at the test stations, with Acerta's LinePulse software monitoring them.
Results
Takeaway
The 80% result came from a subset of test data and the vendor's own release. The full factory rollout was still ahead when it was published.
Categories
- Industry
- Manufacturing
- Business function
- Production and quality
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Partner co-development
- Evidence
- Vendor case study
Sources
Source summaries
Acerta announced a multi-year project with Ballard to shorten acceptance test times and find causes of test failures in fuel cell stacks. It states that a proof of concept cut test cycle times by 80%, from 2.5 hours to under 30 minutes. The figures are the vendor's own.
Full details are in the original sources.