SoftInWay's Foundation Models Target Faster Turbomachinery Preliminary Design

SoftInWay has unveiled what it describes as the first full-scale foundation model capability purpose-built for turbomachinery design, a development that could significantly reduce the time and data overhead required to produce a viable starting design for rotating equipment programs.

Available through the company's AxSTREAM AI platform, the pre-trained models integrate machine geometry, performance maps, and operating behavior across a requested speed range — connecting the multidimensional data that engineers typically spend weeks or months assembling before preliminary design can begin.¹

What Are Foundation Models, and Why Do They Matter for Turbomachinery?

Foundation models are large, pre-trained AI systems capable of generating predictions across a wide range of inputs without being retrained from scratch for each new application. In industries like natural language processing and image recognition, they have already reshaped workflows. Their application to turbomachinery, however, has been limited by the domain's physical complexity and the scarcity of large, structured design datasets.

SoftInWay's approach addresses both constraints. Rather than treating turbomachinery performance prediction as a single-output regression problem, the company's models connect flow-path geometry, performance maps, operating limits, and off-design behavior into a unified, engineering-ready output.¹

"Foundation models for turbomachinery cannot be built from generic data," said Maksym Burlaka, Associate Director of Technology and Project Management at SoftInWay. "They must account for the physics governing turbomachinery processes to accurately capture the relationships among geometry, performance, and operating behavior."¹

How the Technology Was Validated

The capability is backed by more than five years of development and was initially accelerated through NASA-funded research conducted in collaboration with Raytheon Technologies. During that program, the technology demonstrated the ability to reproduce real-world turbomachinery performance using experimental compressor data from Purdue University's Zucrow Laboratories.¹

In a subsequent internal validation study involving a single-stage centrifugal compressor, the model was trained on 34,305 designs encompassing 4.2 million performance points. It generated 51 flow-path parameters along with pressure-ratio and efficiency maps, surge limits, and choke limits across the full speed range. Geometry and performance predictions across the entire test dataset fell within a 3% error threshold relative to reference results — a level of accuracy that aligns with expectations for early-stage preliminary design, where rapid feasibility assessment is the primary objective.¹

For context, turbomachinery design processes traditionally begin with empirical correlations and meanline analysis before progressing through increasingly high-fidelity CFD and structural simulations — a sequence that can take months even before a credible flow path is established.² Foundation models capable of compressing that initial phase while maintaining physics-informed accuracy represent a meaningful shift in how design cycles could be structured.

Target Applications and Deployment

The capability is aimed specifically at organizations entering unfamiliar machine types, expanding product lines, or launching programs without an extensive proprietary design archive. Engineers input required operating conditions and design parameters; the model outputs a starting geometry and expected performance that can be transferred directly into AxSTREAM Flow Path for refinement and validation.¹

Organizations can also fine-tune models against their own engineering archives. SoftInWay supports deployment within customer-controlled infrastructure or private cloud environments to accommodate proprietary or export-controlled programs. Initial models cover select turbomachinery applications, with additional machine types and workflows described as in development.¹

Implications for Engineering Teams

For turbomachinery professionals, the practical significance lies in front-loading credibility into the design process. Rather than beginning with an empty database or a loosely applicable reference design, teams working with AxSTREAM AI would start from a physics-consistent machine concept that already reflects the target operating envelope — reducing iteration cycles and compressing schedule risk at the program's most uncertainty-heavy stage.

SoftInWay supports more than 750 companies globally, spanning OEMs, service providers, utilities, universities, and government organizations.¹

References
1. SoftInWay. "SoftInWay Unveils First-of-Its-Kind Full-Scale Foundation Models for Turbomachinery Design." Oct. 12, 2026.
2. Dixon, S.L., and Hall, C.A. Fluid Mechanics and Thermodynamics of Turbomachinery, 7th ed. Butterworth-Heinemann, 2014.