@Tata-JLR: Why we invested in BeyondMath, leaders in advanced engineering simulation

InMotion Ventures has participated in an $8.5M Seed raise for BeyondMath, a UK-based startup revolutionising Computational Fluid Dynamics (CFD) through an AI powered simulation platform.  The round was led by by UP.Partners, with significant participation from Insight Partners and InMotion Ventures.

Sam Nasrolahi, Principal, and Oliver Fitz-Gibbon, Platform Lead, explore why we invested. 

The Market

Engineering design – using computers to create and test parts and systems – is an integral component of the engineering process, helping to optimise system performance, identify failure modes, speed up the development process and lower production costs.  

There are typically two key aspects to engineering design. Computer Aided Design (CAD), for creating 2D drawings and 3D models. Computer Aided Engineering (CAE), for simulating performance to identify inefficiencies and iterate on designs. 

Within the realm of CAE, Computational Fluid Dynamics (CFD) plays a crucial role in aerodynamics, thermodynamics, powertrain systems & climate control. CFD simulates and mathematically predicts the fluid flow around or through a part by solving complex governing equations of conservation of mass, momentum, and energy. However, the CFD market is grappling with significant challenges that are hindering its efficiency, reliability and ultimately widespread adoption. 

Traditionally, CFD simulations are run on time and cost-intensive legacy systems, often requiring supercomputers. It is not unusual for the process to take 100 hours to complete, leading to monumental compute costs. The process is inherently iterative, with any errors or design changes necessitating a complete re-run of the simulation. Moreover, the setup is labour-intensive, requiring engineers to meticulously refine surface meshes to ensure accurate results. Depending on the density and size of the mesh, the solution can take hours, days, or even weeks to solve. 

Engineers depend on simulations for the design of everything. From vehicles and aircraft, to lithium-ion batteries and the infrastructure of data centres. Today, teams across multiple sectors are seeking innovative approaches to streamlining the CFD process in the bid for faster, cost-effective solutions. 

What BeyondMath does:

ArtificialI Intelligence (AI) and Machine Learning (ML) are accelerating advancements in engineering simulation. There’s no shortage of companies experimenting to enhance the efficiency and accuracy of CFD simulations. The following list outlines some emerging approaches, each offering unique benefits and challenges.

  • AI accelerators for CFD: focussed on optimising the computational performance of existing numerical methods via specialised hardware or algorithms. AI accelerators improve computational speed, enable the handling of larger data sets and offer parallel processing abilities. However, high costs and compatibility issues make them an imperfect solution.  
  • Physics Informed Neural Networks (PINNs): integrating fluid dynamics equations into neural network architecture to enable accurate simulations. PINNs can handle complex fluid behaviour with sparse data, and offer faster computation times in comparison to the numerical method. However, the upfront computational cost of training the neural network is vast, and a reliance on the quality and quantity of data leads to scalability challenges 
  • Surrogate Modeling: training a ML model on simplified mathematical models to approximate complex CFD simulations. This approach enables faster simulations and lowers compute requirements and cost. Limitations include the ability to generalise beyond the training data, resulting in the need for repetitive training and a careful validation of results.  
  • Neural Operators: using neural networks to directly learn and approximate the underlying operators of partial differential equations. Neural Operators offer more efficient solutions to complex conditions, and significantly reduce time and computational cost. However, much of the work in this area is still in the research phase  

Against this backdrop, BeyondMath has positioned itself as a frontrunner in tackling the CFD challenge. The company employs a deep learning approach using neural operators, with their platform employing cutting-edge AI, trained to decipher the calculations required for CFD. This method avoids the need for extensive training data to approximate fluid behaviour and properties.  

BeyondMath’s technology could offer several advantages over traditional numerical and iterative CFD methods, and even other AI-based approaches. First and foremost, their simulations run at least 1,000 times faster. This dramatic increase in speed not only saves time, but also significantly reduces costs associated with engineering resources, computational processing, and model training efforts. 

The efficiency gains provided by BeyondMath’s technology could have far-reaching implications for the engineering workflow in future. Engineers could work iteratively and explore more designs for a specific problem, vastly improving their ability to optimise and innovate. This capability is particularly valuable in the automotive and aerospace sectors. In industries such as these, the smallest aerodynamic improvements can lead to significant performance gains and savings.  

Reducing simulation time to seconds significantly increases the opportunity for Design Space exploration, leading to improved product performance without straying from the vision. The expedited nature and ability to combine simulations improves x-attribute optimisation and drives efficiencies throughout the process. Ultimately, BeyondMath offers the potential for simulation democratisation. Their approach allows engineers to assess the implications of their design iterations directly: reducing feedback loops and significantly speeding up the development process.

Moreover, BeyondMath’s model requires little pre-training, which acts as a precursor to significant scale. A lack of pre-training allows the technology to be easily applied to different industries and CFD challenges cases, without the need for extensive customisation. Crucially, it also reduces the need for enterprise engineers – who typically lack ML skills – to pre-train the models themselves.  

Why we invested:

BeyondMath’s innovative approach to engineering simulation aligns perfectly with our commitment to investing in early-stage startups tackling the critical enterprise challenges of our era.  

The pain points in engineering design are well-documented and felt across multiple industries. By offering a solution that is orders of magnitude faster and more cost-effective than current methods, BeyondMath is positioned to capture a significant share of this growing market. The market potential for this AI simulation platform is undeniable.  

In our assessment of this rapidly growing sector, BeyondMath’s unique approach stood out. We believe their focus on neural operators, which directly learn and approximate the underlying operators of partial differential equations, offers significant advantages over other AI-based CFD solutions. As previously stated much of the work in this area is still in the research phase, however BeyondMath has made significant strides in bringing this technology to practical application.  

The commercial viability of this approach is strengthened further when considering that BeyondMath will be among the first to adopt an NVIDIA DGX H200 system, which will significantly enhance the capabilities of its platform. DGX H200 systems provide advanced AI supercomputing, allowing BeyondMath to train its physics solver at industrial scales and helping it deliver even more groundbreaking solutions to its customers.  

From an automotive industry perspective, BeyondMath’s technology has the potential to significantly accelerate JLR’s design and engineering processes. As vehicles become increasingly complex and aerodynamics play a crucial role in performance and efficiency, the ability to run faster CFD simulations will provide a significant competitive advantage.  

The strength of the founding team was one of the primary drivers for our decision to invest. Alan Patterson and Darren Garvey bring a wealth of experience and expertise to the business. Alan is a leading figure in AI with over 30 years of experience in applied ML and software engineering. His track record of holding senior leadership positions at the time of successful exits, including acquisitions by Google, Amazon, and Rolls Royce, speaks to his ability to build and scale novel technologies. Darren complements this with his 15+ years of experience in AI and software development across organisations of all sizes. He too has successfully led technical teams to exit, previously holding the position of Director, Principal Research Engineer at HomeX prior to its acquisition by Rolls Royce. Alan and Darren’s combined experience, along with their history of working together provides a solid foundation for BeyondMath’s success. 

Congratulations to the entire BeyondMath team on this funding round. We believe their unique approach, domian expertise, strategic timing and technical firepower will perfectly combine to represent a fundamental leap forward in simulation technology. We’re excited to work closely with the team as they continue to set new standards in engineering design. 

We’re always interested in speaking with exceptional founders setting new benchmarks in quality, technology, and sustainability. If you are a founder or know a company in the space, get in touch with our team. Contact us via LinkedIn or through our investment form.

Back

Go to Source