"An AI model trained in one setting could fail or degrade badly in another, including when factory benchmarks did not reflect real production conditions."
The researchers examine machine learning in real manufacturing operations, using data from a production facility supplying large metal components. Their study shows how changing operational distributions can undermine predictive performance, supporting the chapter's account of the gap between experimental models and production reliability.
View Source