Transforming Engineering with AI – An Introduction to PhysicsX

Picture of an application of PhysicsX's technology

PhysicsX empowers engineers to build beyond human imagination using AI.

We build digital engineering solutions that leverage AI to accelerate CAE/numerical simulations to near real-time, enabling engineers to optimize the design, manufacturing, and operation of complex machines beyond anything possible today.

We do so in some of the most advanced and important industries of our time, including (aero)space, automotive, semiconductors, materials, and renewables. Our aim is to enable positive change at scale, be it by accelerating the energy transition or by minimising materials consumption.

We are a small but rapidly growing team, partnering with leading organisations across the globe, from North America over Europe to Australia.

The problem: Today’s simulations are too slow, processes are fragmented and learnings lost

Companies today are under pressure to:

  • Design better performing products faster, at a fraction of the cost
  • Manufacture with higher quality and throughput
  • Operate products, components, and systems more efficiently

However, today’s engineering workflows and tools can only marginally improve productivity and product performance with the same constraints due to the following challenges:

  • Slow and expensive simulations: numerical simulations (CFD, FEA, CEM, etc.) are slow and costly, making it impossible to optimally and exhaustively explore and leverage the search space in design optimization, configurations and parameters in discrete or process manufacturing, and control parameters and sensors to operate products and systems. As a consequence, experts rely on their experience and a limited amount of simulations to improve designs and processes, often resulting in marginal performance gains.
  • Fragmented workflows: workflows and tools for different stages of the engineering process are often siloed, requiring many handovers and limiting visibility of how engineering decisions affect performance, cost, and customer satisfaction over the entire lifetime of their projects and products.
  • Lost learnings: simulations rarely build on previous experiences. Engineers typically start from scratch for each new project, ignoring years of data and results from past projects, especially from other parts of the organisation.

The solution: Fast and easily accessible digital engineering solutions that connect the organisation, supported by deep domain expertise

A chart showing PhysicsX's approach

PhysicsX partners with organisations across advanced industries to solve their hardest engineering challenges, reducing time to market and enabling a step-change in performance. We empower engineers by providing easily accessible solutions to simulate physical systems in real-time, allowing faster iteration and automated optimization of designs and processes. These solutions are supported by a robust and secure enterprise platform, streamlining data and model management and connecting different parts of the organisation.

Our technology unlocks the following benefits:

  • Faster results and exhaustive optimisation: once set up, our solutions can simulate even the most complex systems in seconds (instead of days), automatically iterating through millions of designs to optimize performance and reach the global optimum of what is physically possible, while respecting manufacturing and other constraints.
  • Cross-functional workflows: any engineer can drive all parts of the workflow, including activities that were previously managed by dedicated experts, automating the most time consuming tasks while retaining the ability to inspect and manipulate each geometry, configuration or control parameter – thereby minimizing handovers and accelerating iterations.
  • Encoding of knowledge, data and models: our platform provides a unified environment for engineers to collaborate and build a common knowledge base, while deep learning models (so-called Large Physics Models, or LPMs) encode data and insights from previous work, as well as real-world data, resulting in 50-70% time and cost savings when starting new simulation projects.

We have also learned that the hardest and most impactful engineering problems tend to be incredibly complex by nature and can rarely be solved via out-of-the-box tools. Our field engineers and forward deployed teams therefore support our partners to set things up and help crack particularly tough challenges.

Our platform minimizes handovers and facilitates re-use and knowledge sharing across the enterprise. A single engineer can manage all parts of the workflow, automating the most time consuming tasks while retaining the ability to inspect and manipulate each geometry or configuration explored.

Example: Wind Turbine – An end-to-end optimization in design, manufacturing, and operations


As an example, consider optimizing a wind turbine. The optimization of such a large and complex machine will typically depend on hundreds of parameters across a large number of operating points, resulting in a search space that may contain trillions of configurations.

Since a single numerical simulation could run for days on a cluster, and because a systematic exploration of the design space would require thousands or even millions of simulations, engineers must rely on their intuition to determine which designs are feasible. They then perform a handful (~tens) of simulations in adjacent parts of the search space, typically achieving 20-30% of the potential performance improvements that could be achieved at the global optimum. This effect is further compounded if we move from the optimization of individual parts (e.g., a blade) to the optimization of the entire system (the whole turbine).

A second challenge is the number of handovers required. Designs are typically created by design engineers, but handed over to simulation engineers to model a system’s performance. Simulation engineers in turn request design engineers to make modifications based on their findings, and so on – a loop that may involve 10-50 handovers and typically takes 6-8 weeks in a real-world setting, with another round of iterations taking place once the “final” design is passed on to manufacturing. And since it is difficult to re-use traditional simulation workflows for a new design, not to speak of the fact that different teams often use different software and tooling, all of this starts from scratch for the next project, even if it is similar in scope.

Very similar challenges apply throughout the rest of the life cycle. For example, in manufacturing, the quality assurance process can be slow and inefficient: once the turbine is designed and is sent to manufacturing, 3D-scans of the blades are taken and sent to the Aero team for performance assessment and repair suggestions. This is a lengthy process that slows down time-to-market.

During operations, assessing the performance of wind turbines using different tools and data sources can create further inefficiencies and communication gaps. Operating the wind turbine is challenging as it requires assessing the performance of each wind turbine through a myriad of sensor data (e.g., SCADA) and CAE data. The tools assessing the performance of wind turbines during the operations phase are different to the ones used in design thereby creating gap between design and operations.

Crucially, engineering workflows in manufacturing and operations will rely on their own models and tool stacks which are created separately from the original design part, with all the issues that this entails. This separation can lead to inefficiencies, miscommunications, and delays, impacting the overall performance and time-to-market of wind turbines.

A symbolic illustration of a wind turbine simulation


In the new world, any engineer (i.e., not only simulation experts) can log into the PhysicsX platform and select a pre-configured workflow for turbine optimization. And if there isn’t one available yet, our engineering teams will work with them to build one. With a few clicks they can upload the relevant artefacts (e.g., simulation files, geometries, manufacturing constraints) and configure an optimization workflow. Once everything is set up, the solution will run automatically, training or fine-tuning the underpinning AI models. After the model has been trained, an individual simulation will run in less than a second on a single GPU (instead of hours or days on a cluster, as for traditional simulation software), enabling the exploration of vast parameter spaces across millions of designs and operating points. This allows engineers to optimize the performance of products and processes beyond anything possible today, pushing past the current limits of engineering.

Many parts of the existing workflow, such as 3D models of turbine blades, can be used to assess the performance of manufactured parts. These, when 3D scanned, can be used by the manufacturing team to directly assess their performance without the involvement of the Aero team – the same workflow can also highlight areas of the part with highest impact on performance, and can therefore enable the effective prioritisation of repairs. This information, among all the others, can be leveraged by both manufacturing and design teams at different stages of the product lifecycle.

The same model that is used in design optimization and manufacturing can also be used in operations – by enriching the model with sensor data (e.g., SCADA) – to assess the performance of the turbine, run diagnostics for predictive maintenance, and optimize its operating conditions. Using PhysicsX platform for design, manufacturing, and operations ensures all teams are working on the same tooling, thereby increasing efficiency and overall engineering and business performance.

Equations for a better world