Acerta Paper included in 47th International Vienna Motor Symposium

TORONTO, ON — April 28, 2026 — Acerta AI, a provider of operational AI solutions for discrete manufacturing, announced results from a production deployment of a machine learning and AI solution to reduce end-of-line testing time for hydrogen fuel cell stacks. When fully integrated at scale, the solution is expected to reduce testing time by up to 76%, improving production throughput while maintaining strict quality requirements.

End-of-line testing is one of the most expensive and throughput-limiting steps in scaling fuel cell production. Reducing test duration while maintaining quality directly increases available capacity and lowers cost per stack. This work is based on a two-year collaboration with a leading hydrogen fuel cell manufacturer, where the system was developed, tested, and deployed in a live production environment. By identifying early indicators of failure, the approach reduces test duration from over 2 hours to 15–30 minutes while preserving quality guarantees.

The differentiator is not model accuracy alone, but the ability to operationalize AI outputs into production decisions that increase throughput, reduce cost, and safeguard quality.

“In production environments, model performance alone isn’t enough,” said Greta Cutulenco, CEO of Acerta AI. “The challenge is turning predictions into trusted decisions that optimize throughput and cost without compromising quality.”

Acerta’s unique approach explicitly separates prediction from decision-making, converting model outputs into decision policies that define trade-offs between expected cost, test coverage, and resource usage. These policies can be tuned to reflect different operating modes, conservative, balanced, or aggressive, depending on production requirements.

“In manufacturing, there are no ‘perfect’ models,” said Sergey Strelnikov, VP of Engineering at Acerta AI. “That makes it critical to go beyond prediction and explicitly connect model outputs to production metrics such as throughput, cost, and resource usage. Our approach focuses on policy-based decisioning, where trade-offs between cost and risk are clearly defined.”

The system is trained centrally on large-scale datasets and deployed at the edge in production environments, where it operates under constraints on latency, reliability, and integration with physical systems. This cloud-to-edge deployment model ensures alignment between training pipelines and production behavior.

The results were shared with the global powertrain and propulsion community at the 2026 International Vienna Motor Symposium, underscoring industry relevance and operational readiness. The paper, “Accelerating Fuel Cell Stack End-of-Line Testing with Machine Learning: Early Failure Detection and Cost Savings in Production,” details the full system architecture, including data ingestion, model training, edge deployment, and monitoring.

The symposium, organized by the Austrian Society of Automotive Engineers (ÖVK) in collaboration with TU Wien, is widely regarded as a leading forum for powertrain and propulsion technologies, bringing together OEMs, suppliers, and researchers.