AeronauticsARTIFICIAL INTELLIGENCE
From two days to two hours: generative AI takes on ECU test plan review
ALTEN in Morocco built a fine-tuned LLM assistant for a leading global automotive manufacturer, automating the review of ECU test plans against a catalogue of standard rules to cut review time from days to hours.
Electronic Control Unit (ECU) verification and validation relies on rigorous test plans, built from system requirements and checked against a large body of standard rules and recommendations before they can be signed off. Reviewing test plans against more than 100 such rules is essential to quality, but it is a slow, meticulous task that often generates client feedback even when the underlying tests are technically sound, making it a bottleneck in tightly timed delivery cycles. A leading global automotive manufacturer engaged ALTEN to help it respond to this challenge.
Challenge: Automate the review of ECU test plans against a large catalogue of standard rules, to reduce the time this step takes within a tightly timed delivery cycle.
Solutions: A fine-tuned LLM assistant that checks test plans against a 160-rule catalogue and generates a report indicating whether each rule is verified, including the location of any errors.
Benefits:
- Test plan review is expected to fall from two days to just two hours
- Faster turnaround within a demanding, seven-day delivery timeline
- Consistent, rule-by-rule verification that reduces reliance on manual checking
- Clear, actionable reporting that pinpoints the exact location of errors
A fine-tuned assistant for rule-based review
Test cases at the client are created from system requirements through three steps: requirement analysis, accounting for around 5% of effort, test plan creation, around 60%, and test plan review, around 35%. ALTEN focused its AI objectives on automating this critical third step. The fine-tuned LLM assistant takes the test plan, provided as an Excel file, together with a 160-rule catalogue, as its inputs, and produces a report indicating whether each rule is verified, along with the precise location of any errors. The solution is built using PyTorch, Microsoft Phi-3, Google T5, Python and NLP.
Faster reviews, steadier quality
Within a seven-day delivery timeline, test plan review alone currently consumes two full days, despite technical tests often being sound, simply because of the volume of rules to check manually. By automating this step, the solution is expected to bring review time down to around two hours, freeing teams to focus on resolving genuine issues rather than manually working through a 160-rule catalogue, and giving the client a faster, more predictable path from test plan creation to sign-off.
The power of generative AI is set to help this automotive manufacturer verify test plans faster and more consistently, without compromising the rigour ECU validation demands.
