Since its release two months ago, Neural Concept Shape 4 (NCS 4) has already been adopted by many companies such as Subaru, Mahle, Valeo, Airbus, Safran and 25% of the top 100 tier-1 suppliers.
Developed over a period of 6 years, this new version marks the fourth major release of our platform. During this time, we have been able to measure a shift in the market and demonstrate an appetite for combining deep learning with world-class engineering know-how.
By harnessing NCS, designers and engineers across aerospace and automotive industries are reducing product development times by up to 75%, accelerating product virtual simulations by 10x, and improving product characteristics including efficiency, safety, speed, and aerodynamics by up to 30%.
In the face of pressing global issues like climate change and sustainability, we have entered a new era of engineering. The need for radical solutions has never been more critical, yet the time to bring new innovations to market is slow. Engineers are now tasked with reaching new levels of development efficiency and product performance at unprecedented speeds.
Importantly, new computational methods are now empowering engineers with immense possibilities. Deep-learning, software integration, LLM-based user interfaces, and scalable computing are the pillars of a new paradigm of engineering software. Engineering AI that relies on integrated platforms and connected software components are bridging the gap between compute capabilities, domain knowledge, and people.
These technologies are powering new engineering processes which remove current bottlenecks, whereby computers can translate high-level functional requirements into actual optimized designs at unparalleled speeds and scale.
People’s skills are also evolving alongside technology. The world where coding was necessarily complex and time consuming, while UI-based was simple and efficient, has changed massively since python has become so widespread and interactive development environments and copilots have emerged. The world where software was centered around a single user, is now outdated.
In its place, we are witnessing the emergence of a new type of engineer, CAE data scientists, who are CAE practitioners trained in data science through university studies or online courses. They understand the latest techniques of AI, are familiar with core data science tools such as Python, notebooks, and deep-learning concepts. These people are at the forefront of an unprecedented transformation of the engineering world, holding a unique mix of knowledge with the potential to change everything. Even though they currently represent 2% of engineers, they will be the ones powering the emergence of tomorrow’s design workflows, massively driven by AI. Those visionary engineers need a new category of cutting-edge tools. It's not enough to have the knowledge base and vision for the future - we must uplift these individuals, create a 10x acceleration to their work and provide them with the tools they need to usher in the next era of innovation.
Enters NCS - the interactive and scalable development platform for end-to-end Engineering Intelligence application development.
The first version of NCS, in 2019, was the first product, globally, that made Deep-Learning based surrogates commercially available to engineers. It was based on a full fledged web interface that allowed users to configure models, train them and then run inference through an API.
We progressively realized how these models were difficult to train inside this interface, how every slight deviation to the nominal case was making models fail, and how every realistic case was a deviation to the nominal situation. This is even more true whenever the teams want to build complex workflows combining optimization, generative model, hybridization, generative design. And we also realized how this is where the real value of Engineering AI lies.
At first, we came to the conclusion that for the real application cases that we wanted to solve, we had to take the product out of the hands of our customers, go back to coding and edit the whole data pre/post processing pipeline, the model architectures. We took one of our most difficult, but best decisions: drop the fancy UI, go back to working with scripts and focus on the technology. We developed our 3D generative models, our morphers, our optimization routines and connected all of it together. This phase allowed us to collaborate very efficiently with many customers and partners, to prove the first examples of practical value. As a result, we built what was probably the first very large scale optimizations combining generative design models (Neural Design Modules) and 3D DL based surrogates to design cutting edge products which have been purchaed by millions of people across the globe. We deployed large scale APIs for CFD predictions that are more accurate than RANS calculations, used by hundreds of users.
However, at this stage, the product was essentially an sdk, very difficult to actually customize and adapt outside our team. As we were starting to work more and more with external service partners and found skilled users at our customer’s, willing to handle customization on their own, we introduced the concept of plug-ins, which are very nicely built interfaces to modify some of the native components of the software. As more and more people wanted easy access to our product, and as we needed to support a growing base of application engineers, we consolidated our cloud architecture, and learnt how to build a unified product allowing deployments across cloud, private cloud and on-premise HPC, essentially offering the advantages of a SaaS product, on every environment. Our team reached more than 15 application engineers, and external expert users grew to more than 30, developing on advanced works mixing data-science and plug-in development. These expert users, in turn, were building applications used for day to day design by four or five times more engineers.
Working with so many different customers pushed our software to the limits and made it become a very versatile and cross application toolbox. We learnt how to handle files of 100 million points, predict on meshes with 5 million vertices, how to dynamically create slices within our predictions, how to handle detailed features and large objects. And this in a fully vendor-neutral way, building compatibility with a wide range of software. We introduced processing of signals, scalars data and optimized our architecture for each of these scenarios.
At this stage it appeared clear to us that the time-consuming work of experts, and their availability was a bottleneck to scaling more. Instead of reducing the range of possibilities and limiting our experts, we decided to work more to make their work more efficient as we believed they could be massively sped up if we could create a new user experience for them.
As we are going to market with NCS 4, the feedback from users and companies goes beyond our expectations. The tool is the closest offering on the market to satisfying the needs of the new generation of engineers and users, the CAE data-scientist.
To summarize, we are tremendously excited to see our field evolving in so many directions. Neural Concept, its team, and all of its customers have contributed more than any other players to develop it.
NCS 4 is enabling a new era of Engineering Intelligence – one in which engineers are provided with a superior user experience for experts and junior engineers alike - but which allows solving real engineering challenges with ever more ambitious workflows.
But we’re not finished yet. We have an incomparably wider range of AI capabilities in our software that we continue to develop, refine, and evolve. These new features incorporate 3D surrogates, of course, but also hybrid physics modeling, end-to-end differentiable optimization, reinforcement learning, and a powerful library of design modules that includes 3D advanced generative AI.
Beyond this unique technological core, we thrive to make NCS’ transformative impact efficient, and scalable. Due to the constant and rich feedback of our incredible user community of engineering experts, coupled with fast development iterations enabled by a highly modular software architecture made of different lightweight layers on top of our cutting-edge algorithms, together, we are redefining the contours of Engineering Software UX. We are also harnessing to the maximum the power of modern technologies - from LLMs to massively distributed compute architecture on the cloud to fast 3D web visualization - to bridging the gap between deep, scientific AI tech, and scalable industrial impact.
Whether you are a customer or a service and implementation partner, your contribution to building the future of engineering with NCS is changing the industry. You are getting humankind closer to the day where a computer will be able to design or re-optimize a whole car or aircraft engine end-to-end.
Neural Concept CEO