Challenge
In structural and stress calculations, engineers must compare thousands of time series from non-regression tests after every change. That takes days - time that should go into interpreting the results, not comparing parameters.
Solution
- Decision assistant based on machine learning - it identifies which test is regressing and which parameters cause it.
- Focus on the engineer’s judgement - the machine compares, the human assesses.
- From idea challenge to production - the idea came from an internal SII innovation challenge, was developed in a four-year publicly funded research project and then industrialised.
Result
- About five times faster: a standard validation takes one day instead of five.
- Passed a blind test on unknown data - after which the client commissioned industrialisation.
- Rolled out across programmes from 2023 to 2025 and fully operational today.
- Fewer loops between design and calculation, better optimised model configurations.
Part of a bigger picture
The validation assistant is one of several AI research projects we drive along the entire V-model of product development:
Conception ↓
RequirementsRequirement conformity checking
Functional specificationCoverage analysis · traceability and impact analysis
System specificationGeneration of technical and functional specifications
DevelopmentAssistant for safety-critical software development
Integration ↑
SupportIntelligent ticket orchestration
ValidationAutomatic analysis of non-regression tests - the validation assistant
TestsGeneration and validation of test strategies · unit test generation
Cross-cuttingRAG chatbot for document analysis
* Project: we owned the delivery · ● in operation: we run the solution today

