Context

SHW Automotive develops electric gerotor oil pumps for battery-electric drivetrains, where performance must hold across a wide operating range, including cold starts at temperatures down to −40 °C. That creates a difficult development problem: the same pump must satisfy flow, torque, efficiency, and power requirements under conditions where oil viscosity changes dramatically. In practice, this pushes learning late into development, because the most critical behaviors often only become visible in bench and climate-chamber tests.

In this case, late discovery was especially costly because cold-start validation is slow, expensive, and tightly coupled to release decisions. If the pump or motor sizing is wrong, the error propagates into multiple subsystems. This paper describes in detail how the team addressed this by building a simulation environment that could predict pump behavior across operating points before hardware existed, including the previously hard-to-model cold-start regime. This case study analyzes one aspect from Product Velocity point of view.

Shift Left

VrfVerify The central shift was to move learning about pump performance from physical validation into model-based development. StrStructure The team decomposed system-level requirements into subsystem behavior. They especially focussed on pump torque, flow, efficiency, and motor losses, so that the system could answer key design questions earlier. That made simulation part of requirements validation rather than engineering.

DsnDesign They combined a parameterized gerotor pump model with a more detailed treatment of leakage, friction, and thermal behavior, then added a novel viscous-friction-heating model to explain cold-start behavior that earlier models missed. This addressed the largest uncertainty, what happened under harsh startup conditions where fluid behavior changed rapidly.

Physical testing did not disappear, but its role changed. VrfVerify Bench and climate-chamber tests were used to validate and calibrate the model, rather than serving as the primary mechanism for discovering whether the design worked. SHIFTShift Left That is a strong Shift Left pattern: expensive downstream tests become targeted evidence loops instead of the place where learning takes place.

Accelerate

StrStructure The acceleration came from making simulation accurate enough to support real design decisions. Once the model could reliably predict behavior across operating points, the team no longer had to wait for hardware and climate-chamber availability to evaluate key trade-offs. That reduced iteration time and allowed more operating points to be assessed earlier in the cycle.

EvoEvolve Prediction accuracy was below 5% for volumetric flow and below 7% for torque and efficiency across relevant conditions. That level of reliability turns the model into a reusable development asset. Each validation cycle improves the underlying design capability, which means future pump variants can start from a stronger base.

The most important outcome is that testing moved from discovery to confirmation. VrfVerify Instead of finding fundamental issues only after prototype build, the team could predict performance earlier, reduce the number of costly physical iterations, and shorten the path to a releasable design. The paper explicitly states that this reduced development effort by an order of magnitude, from weeks to days.

Outcome

SHW Automotive replaced a slow, test-heavy validation pattern with a simulation-first development loop that exposed problems earlier and reduced dependence on late-stage hardware learning. The result was faster iteration, fewer surprises at cold start, and higher confidence in pump and motor design decisions before prototype build. It shows how validated simulation can reduce economic exposure by pulling critical learning forward and making each development cycle more informative.