0:15 So yeah, my name is Shahroz. I’m a co-founder of Compute Maritime. So I think it’s been quite exciting time that the future of engineering AI or physics AI is quite bright specifically its implication into very critical industries and I’m quite excited for that but I have been thinking a very of the strange question rather than why we are pouring millions and if not billions into optimizing vehicles that only transport humans or the bodies why not looking at something that is more critical in a sense that that carries or transport the world around us I’m saying this because so far over this yesterday and today I’ve seen a lot of use cases and solutions circulating around automotive and aviation industry and I haven’t seen a solution of a problem of I would say rather the biggest logistical problem on planet that is just floating past us and that is ships.

1:24 Arguably shipping is one of the most critical driver of the global industry. But a question for the audience now how many of you have a solution or a use case something related to shipping ship design ship building or maritime apart from revenue? Wow, there are a few people. Okay, I’m not as pissed as I was yesterday. Okay, that’s good. So yeah, sorry, don’t be careful. So, it’s very important for us to understand why shipping is so important.

2:06 Let us look ourself around. The more than I would say 90% of the goods in this room have been arrived by ships. It’s your phones, your laptops, your bags when all transported to a certain location. If not by a single ship, then definitely by multiple ships. But if you look at this map around there, at any given moment, there can be more than 100,000 ships strong quoting goods around the globe.

2:34 So that means shipping has a very huge carbon footprint and that is contributing around 3% of global greenhouse gas emission. An interesting fact that I was just reading that if shipping were a country it would rank against alongside Germany. Any Germans here? Sorry about that. Yeah, that means you’re really emitting as much as mission as shipping. So so yeah so international maritime organization which is a body of UN and EU has started to take strict measures in term of decarbonizing shipping and they set like very ambitious I would say rather targets to make this industry net zero by 2050 which I believe the traction or the pace that we have it will never achieve that and then IM also knows and I think that was the reason that Trump just didn’t want to participate into the recent policies that we were contributing.

3:39 So just give you an example if imagine a ship that is transporting goods from Singapore to Rotterdam and it will take the shortest route and that was emit around 3,000 ton of CO2 and that would equate to around 700 plus cars. And if if that single vessel wasn’t I would say fulfilling the requirements set by EU and IMO they can end up paying around 1 to2 million penalties as carbon taxes for that single voyage.

4:11 And now imagine this going towards the entire fleet of a certain stakeholder. I think if I was a shipping owner I would definitely not like that. And if somebody say like that’s okay I think then then they are crazy. So let me step back and ask you a very I would say silly question. What’s that? It’s it’s a keyboard right? And and the layout is is this quadp keyboard layout that everyone I think of us have been using for past a decade now.

4:42 Have you ever thought like why this layout why why this layout came from? And interestingly this layout is is not something that is optimized. Actually, this layout was invented by Christopher Alahim just to slow down the the types so that mechanical typewriters they’re not jammed. So why this is relating to shipping? It’s it’s the legacy. We can see here a ship that is built today is looks exactly the same as a ship that was built 50 years ago.

5:15 And it’s not a coincidence. It’s just a legacy. And we have just seen it from the from the example of this this keyboard that legacy doesn’t always mean an efficiency. It’s just a legacy. And this legacy has shaped I would say not only the maritime but also like the various other industries how we design engineer things and how we take engineering and design decisions and and these inefficiencies in designs they scale throughout not only in term of I would say like certain aspect throughout the life cycle either it’s fuel consumption either it’s the regulatory analysis either its maintenance overhead or its recyclability and one example again consider like this this amor is a is a ship which has this number plate this IMO number is it’s literally like a number plate of a certain ship if it emits around the cost of its fuel is around 23 million and and if we provide this ship the single ship at 2% efficiency that means it will be saving in term of cost around half a million a year so great savings and and this the inefficiencies also scale not only related to the I would say cargo industry but also towards your passenger vessels your leisure industry your work boats and also on the on the naval vessels in naval it’s not always they don’t we don’t want to to see the efficiency from the fuel consumption point of view but we want to see the efficiency from being in still and that’s exactly the reason we at Compute Maritime decided to drive AI enable maritime sustainability and we are working towards bringing at the core of this $150 billion ship building industry generative AI and high performance computing and one thing I would mention here before I proceed that we have seen a lot of applications so far on building foundation models for physics that is great that is amazing because this is in our simulation design pipeline this is one of the bottleneck Because a single simulation takes so much time and imagine running hundreds of simulations.

7:31 But in our case, we see that a foundation model for generating design itself is as important as the foundation model for physics if not more. Because if you don’t have the design, even if you have the most fastest simulation tool on the planet, even if you don’t have a design generation capabilities, you’re still these tools are still more or less useless. Right? So, give you an example about like how a typical design process looks in a maritime industry, but I would say like again, I’m a mechanical engineer.

8:02 Although I have a PhD in marine engineer, I will still not call myself a marine engineer. So you a customer will come up with the design requirements. Designers then going to going to go and explore existing design basis to find a a baseline design that closely resembles these new requirements. They going to start creating some variations either manually or parametrically. They going to evaluate those variations see like if the new requirements are met.

8:28 If they are not met go back and change the design and and evaluate it again. This process, this process continues until we reach all we satisfy all the conditions and then we go towards the detail design. Again as I said like the biggest bottleneck is the performance because each simulation can take days even on high performance computing. We can solve it. We have seen all the great solutions coming up from from various of colleagues.

8:56 But again the problem related to the design generation stays there, right? You have to still manually create those variations. You even if you can parameterize those variation, you still have to handcode those rules that will make sure that that each design that has been generated or each variation that been generated is is a valid surface is a valid geometry that will go to your your solvers. Right?

9:20 So another point is that all these design tools these CAD design tools right they are really built I would say you have your your your concept your your whatever concept you have in mind to just create a 3D model of that concept. This is this is where these these CAD pool or any design pools I would say stay today. Although you can use agents to expedite the workflow but still they remain unintuitive.

9:48 Still they remain they don’t carry any intuition that could work interactively with the designers and help them actually innovate something instead of just working as a platform for your 3D design development. So to overcome this we build this foundation model called Neuroshipper. So, new shipper is the world’s first generative tool powered by a suit of foundation models. It is capable of not only designing but also optimizing and simulating helping naval architects design offices to build batter ships, batters, batter vessels and significant at significantly reduced time and a cost.

10:35 So I intentionally made the presentation not so technical because I knew I was the last one to present. So the neuros shipper is said like the it’s based on this foundation model that is both geometry and the physics in form. A lot will people ask like what do you mean by geometry in form? If you point me during the break come to me and I will explain more in the technical depth what does I mean from the geometric form and it’s trained on on 3D geometries and one thing that is differentiates us from from other foundational geometry-driven foundation model is that we are able to train our foundation model directly into spline spaces the spline spaces I mean directly in the surface representation not just the meshes meshes are the great but in term of meshes are great in term of predictive models but for generative models where both the input and the output is a design they they’re pretty terrible because you have to keep the surface resolution high.

11:28 You but if you want to do that you have to increase your mesh density. If you increase the mesh density the complexity of your model will will increase exponentially. And once you have we have this this foundation model built the only thing you need is to use that model to just define your design specifications. What type of shape you need and what constraint it has and the system going to start coming up with with thousands of thousand design possibilities which otherwise for a typical design or a naval architect would be difficult if not impossible to create solely relying on their intuition and as I said like because our model is directly creating pad representations CAD in that sense is spline spaces directly surface representations and in our case as I said like the workflow is not that that you have an LLM that is creating a CAD code and the CAD code is rendered by a CAD kernel but in our case it’s end to end 3D training within the 3D space and also if you have an existing design so you can simply upload the design and the model going to give you in matter of minutes a completely fully parametric model that will show 100% design validity and I will show you a case study on that yes as I was saying like yes in our case It’s it’s directly surface representation and we all know who are working in algebraic geometry.

12:51 Anyone here working in algebraic geometries or spline basic spline representation? No one. Okay, that’s great. So yeah, so you get like a very smooth surface representation and then this smoothness that means like the surface will be C2 and G2 continuous and in ter of meshes you will see have this like abrupt changes like value. You can increase the the resolution of the surface but you will have to increase the mesh complexity on that.

13:12 And if you have the surface that means it is intrinsically parametric. It is ready for manufacturing is mathematically precise. It preserve the the design in intent and it also doesn’t require any rework like the meshes it requires. So we are a very young company I would say like just a year old and early on we were very lucky to have a partnership with SEMA digital industries where Sim Industries came to us and they said like if you’re claiming such I would say technological breakthroughs and why don’t you give us your tool and then we’re going to connect it with our solvers like star CCM and see like how efficient efficient your your foundation models are in terms of design generation capabilities and so see independently tested did where they created where they took a bul carrier which cost grains and they created a parametric model in NX which took around a month and they created a similar parametric model in neural shipper and that literally took 3.47 for 7 minutes and the optimized design saved around 2,000 of fuel savings that translates to approximately these are approximate numbers in term of cost because the fuel prices very and if I recalculate this fuel price definitely they’re going to be a lot higher because of the current geopolitical situation.

So yeah so what we’re aiming for here is we’re not aiming for this large foundation models but we’re actually aiming for is small generic foundation model or a specific industry to solve specific aspect of the problem. So so far we have tackled the vessel design some energy saving devices like sails foils propellers and now we’re going towards handling some other aspects coming from the manufacturing and also from the operation point of view and coming towards the the manufacturing side we recently had a project funded by UK department of transport via innovate UK it was in partnership with five companies Seaman’s rapid fusion which are building industrial scale 3D printers HP provided us the computational power we needed.

15:25 University of Southampton and BYD. So this BYD is not a car company. It’s it’s a BYD naval architects. They are building boats and especially offshore roads. So in this project the aim was that well with our foundation models we are creating we have I would say intrinsically embedded this designdriven physics informed capabilities but can we also make this model manufacturing aware because these model can give you very nice complex surfaces which are very efficient can reduce a lot of fuel consumption like they can reduce drag and the direct drag translate into fuel reduction but most of the time if these designs are not manufacturable or built table then there’s no use of having such a I would say like optimize or efficient design right so we wanted to see how we can embed right from the primary phase the manufacturing constraint and here we look at the applications from editing manufacturing and that results to a ship we just released yesterday so from I would say like what we know this will be the very first offshore vessels that is built that was designed simul ated and optimized and we built using AI end to end.

16:38 It’s not AI assisted. It’s not AI as a suggestion. The geometric generation handled by our large geometric foundation models and also physics prediction handled by our foundation models and the geometric I would say like manufacturing constraints were also been embedded into those foundation models. So that the design was manufacturing ready or manufacturing aware. So a lot of people ask us why maritime industry. So that’s the reason that I want to always set up the story the theme about the maritime industry and give an introduction.

17:10 Yes, we have an interest that we will go towards automotive and aerospace. We recently have set up a partnership with a one of the biggest I would say aerospace company in Europe. We were seeing like how are they going to use our models to to design spacecraft. The biggest challenge is not about the scalability of the model because the model can be scalable for every every anything that is in motion but is the data set and for our case creating that data set took us more than 5 years literally.

17:43 So with this I would like to thank you and let you go for your coffee break. Sorry I hope I wasn’t boring everybody and if you have any questions happy to answer during during the break. Cheers. To learn more about the CDF fan computational design symposium, access the archive of previous presentations, interviews with speakers, and information about future events around the world, visit CDFAM.com. Come.