FEV Accelerates CFD Convergence with Reduced Order Modeling

The CFD Bottleneck in Hydrogen Powertrain Development

FEV Group is a global engineering services provider specialising in advanced powertrain and mobility systems, supporting OEMs and suppliers in automotive, transportation, and energy with expertise spanning combustion systems, electrification, and next-generation fuels including hydrogen, ammonia, and methanol. For engineering teams working on hydrogen combustion CFD simulations, the challenge is consistent: achieving convergence requires evaluating cold-flow scenarios across a wide range of operating conditions, and each run demands manually ramping inlet pressures - a process that is time-intensive, difficult to standardise, and repeated identically across every design study.

‍

There was no mechanism to capture the convergence logic once and reuse it. Every new combustion configuration started from scratch. The result was that full design of experiments (DoE) studies required far more engineering time and compute than the underlying physics justified. Every program began with the same manual preparation steps, contributing directly to long turnaround times.

‍

An AI-Driven Reduced Order Model Replaces Manual Simulation Initialisation

Using Key Ward, the FEV team implemented an AI-driven reduced order modeling (ROM) workflow that replaced manual simulation initialisation with a data-driven process. The team aggregated large-scale simulation data across operating conditions, structured those datasets for model training, and trained a ROM to predict steady-state flow behaviour directly. ROM predictions were applied to initialise CFD simulations automatically - eliminating the manual inlet pressure ramping that had previously consumed engineering time on every run.

‍

The workflow built inside Key Ward became a repeatable, governed engineering pipeline that captured convergence logic as reusable institutional knowledge. ROM predictions were integrated directly into CFD pipelines, standardising simulation initialisation across programs. The result was a reproducible workflow that eliminated manual iteration and improved consistency across the entire study portfolio.

‍

What the Full Case Study Documents

The full case study documents the structured dataset architecture used for ROM training, the validation methodology comparing ROM-predicted pressure fields directly against physics-based solver results, and the measured outcome: a 70% reduction in turnaround time for full DoE studies while maintaining the accuracy of physical testing. It also covers the reduction in manual setup effort, lower cost per design variant, and improved cross-program consistency. If your engineering team is constrained by CFD simulation turnaround time on combustion or fluid dynamics programs, the methodology documented here is directly applicable.