Calibrated for Change, How Ford Is Integrating AI Into Propulsion Engineering

A modern powertrain controls software package spans millions of lines of code, engineered for flexibility across global markets and products through calibration parameters tuned to each application. A single product carries 50,000+ of these parameters, each bounded by thermodynamics, federal emissions regulations, and safety constraints — where a wrong AI answer can mean an unmet product capability, a recall, or customer risk. We don't know the internals of your models the way you don't know combustion chemistry — but we know what breaks when AI meets physics at production scale. This talk presents how Ford deploys AI/ML across this physics-constrained landscape: Bayesian optimization of base calibration, anomaly detection across 15 million connected vehicles, LLM-augmented engineering workflows that operationalize decades of fragmented information, and process efficiency analysis that was previously intractable due to complexity and sparse data. We cover architectural decisions forced by data-scarce environments where every training sample costs real test hours, explainability requirements, and embedded targets — then detail the governance framework for AI in our processes.