Generative Design
Assembly Line
Want to design the car of the future? Here are 8,000 designs to get you started.
Car design is an iterative and proprietary process. Carmakers can spend several years on the design phase for a car, tweaking 3D forms in simulations before building out the most promising designs for physical testing. The details and specs of these tests, including the aerodynamics of a given car design, are typically not made public. Significant advances in performance, such as in fuel efficiency or electric vehicle range, can therefore be slow and siloed from company to company.
But now, the engineers have made just such a dataset available to the public for the first time. Dubbed DrivAerNet++, the dataset encompasses more than 8,000 car designs, which the engineers generated based on the most common types of cars in the world today. Each design is represented in 3D form and includes information on the car’s aerodynamics — the way air would flow around a given design, based on simulations of fluid dynamics that the group carried out for each design.
Each of the dataset’s 8,000 designs is available in several representations, such as mesh, point cloud, or a simple list of the design’s parameters and dimensions. As such, the dataset can be used by different AI models that are tuned to process data in a particular modality.
DrivAerNet++ is the largest open-source dataset for car aerodynamics that has been developed to date. The engineers envision it being used as an extensive library of realistic car designs, with detailed aerodynamics data that can be used to quickly train any AI model. These models can then just as quickly generate novel designs that could potentially lead to more fuel-efficient cars and electric vehicles with longer range, in a fraction of the time that it takes the automotive industry today.
Here come manufacturable generative designs
Big CAD vendors may scoff, claiming that their generative design software has options to create manufacturable parts, and indeed, they do. However, whether they produce machinable part designs is a matter of debate. Such software has yet to spur widespread industry acceptance of generative design. The machinable shapes offered as proof are made by carving away non-machinable details and smoothing out otherwise gnarly, organic shapes resulting from a bone-growth algorithm on which they are based. It’s a roundabout approach at best and, at worst, incomplete — leaving non-machinable details behind.
Bogomolny, once the creator of shape optimization tools (ParaMatters CogniCAD) now has a better way. His InfinitForm software, still to be released, will optimize with cutting tools very much in mind. The size of the cutting tool is seen at the onset of optimization. This sets InfinitForm apart from conventional shape optimizers, which work by adding elements along stress paths or removing elements where no stress is present, all without consideration of the CNC machines that make 80% of aerospace and automotive parts.
Explainable generative design in manufacturing for reinforcement learning based factory layout planning
Generative design can be an effective approach to generate optimized factory layouts. One evolving topic in this field is the use of reinforcement learning (RL)-based approaches. Existing research has focused on the utilization of the approach without providing additional insights into the learned metrics and the derived policy. This information, however, is valuable from a layout planning perspective since the planner needs to ensure the trustworthiness and comprehensibility of the results. Furthermore, a deeper understanding of the learned policy and its influencing factors can help improve the manual planning process that follows as well as the acceptance of the results. These gaps in the existing approaches can be addressed by methods categorized as explainable artificial intelligence methods which have to be aligned with the properties of the problem and its audience. Consequently, this paper presents a method that will increase the trust in layouts generated by the RL approach. The method uses policy summarization and perturbation together with the state value evaluation. The method also uses explainable generative design for analyzing interrelationships between state values and actions at a feature level. The result is that the method identifies whether the RL approach learns the problem characteristics or if the solution is a result of random behavior. Furthermore, the method can be used to ensure that the reward function is aligned with the overall optimization goal and supports the planner in further detailed planning tasks by providing insights about the problem-defining interdependencies. The applicability of the proposed method is validated based on an industrial application scenario considering a layout planning case of 43 functional units. The results show that the method allows evaluation of the trustworthiness of the generated results by preventing randomly generated solutions from being considered in a detailed manual planning step. The paper concludes with a discussion of the results and a presentation of future research directions which also includes the transfer potentials of the proposed method to other application fields in RL-based generative design.
To excel at engineering design, generative AI must learn to innovate, study finds
“Deep generative models (DGMs) are very promising, but also inherently flawed,” says study author Lyle Regenwetter, a mechanical engineering graduate student at MIT. “The objective of these models is to mimic a dataset. But as engineers and designers, we often don’t want to create a design that’s already out there.” He and his colleagues make the case that if mechanical engineers want help from AI to generate novel ideas and designs, they will have to first refocus those models beyond “statistical similarity.”
For instance, if DGMs can be built with other priorities, such as performance, design constraints, and novelty, Ahmed foresees “numerous engineering fields, such as molecular design and civil infrastructure, would greatly benefit. By shedding light on the potential pitfalls of relying solely on statistical similarity, we hope to inspire new pathways and strategies in generative AI applications outside multimedia.”
🧠 AI PCB Design: How Generative AI Takes Us From Constraints To Possibilities
Cadence customers are already reaping the benefits of generative AI within our Joint Enterprise Data and AI (JedAI) Platform. Chip designers are realizing Cadence Cerebrus AI to design chips that are faster, cheaper, and more energy efficient. Now, we’re bringing this generative AI approach to an area of EDA that has traditionally been highly manual—PCB placement and routing.
Allegro X AI flips the PCB design process on its head. Rather than present the operator with a blank canvas, it will take a list of components and constraints that need to be satisfied in the end result and sift through a plethora of design possibilities, encompassing varied placement and routing options. This is hugely powerful for hardware engineers focused on design space exploration (DSE). This has long been par for the course in IC design yet it has more recently become critical to PCB due to the fact that today’s IC complexity doesn’t reduce when it gets onto the PCB—it increases.
However, it’s important to understand that this isn’t Cadence replacing traditional compute algorithms and automation approaches with AI. We remain as committed to accuracy and “correct by construction” as we’ve ever been, and while Allegro X AI is trained on extensive real-world datasets of successful and failed designs, we don’t use that data to determine correctness.
Physics-Driven Generative Design: The Future of Engineering
The Impact Of Machine Learning On Chip Design
The Making of the Impossible Statue
Chemix Brings Transformative AI Technologies to EV Battery Industry, Launching the First AI-Designed Battery in 2023
Chemix, the startup leveraging AI to rapidly build high-performance and environmentally sustainable EV batteries, unveiled MIX™, its AI-powered design platform specifically developed to accelerate the commercialization of next-generation EV batteries. By leveraging MIX, Chemix is poised to revolutionize the decades-old EV battery industry, similar to how AI has transformed drug discovery by accelerating pharmaceutical research and development. This will enable the pace of battery innovation to finally catch up with the ever-growing demand for better-performing, safer, and more sustainable EV batteries.
A battery is to an electric vehicle what a processor is to a computer – a critical technical component determining the entire system’s performance. Despite this central importance, the approach researchers have used to develop new battery materials and systems has remained largely unchanged for decades – until now. As opposed to the conventional method that relies on time- and labor-intensive processes for battery development and testing, Chemix adopts a revolutionary AI-based approach. This accelerates the discovery of the best battery materials, formulations, and recipes by leveraging large proprietary experimental datasets and cutting-edge proprietary algorithms. As a result, the company has created a vertically-integrated end-to-end battery development approach from scratch.
Chemix’s innovation combines battery-specific AI algorithms and their vast collected data to accelerate and optimize battery design – seamlessly integrated with the MIX platform. Chemix has used MIX to experimentally test over 2,000 unique battery material designs across more than 40 variations of commercially-relevant battery formats, accumulating nearly three million test cycles.
3D Printed Generative Design Drone Chassis
Training ChatGPT on Omniverse Visual Scripting Using Prompt Engineering
TOSHULIN is using Generative Design to Transform CNC Machine Design
Digitise and dematerialise: Divergent CEO Kevin Czinger on supplying automotive structures to the world's biggest brands
The manufacture of lithium-ion phosphate battery cells at Coda’s facility in China relies heavily on coal-fired power. And because of that, ‘well over’ 200 kilogrammes (kg) of Co2 per kilowatt hour (kWh) is being produced in battery manufacture. At this time, kg of Co2 per kWh is the most important metric on Czinger’s mind and the cogs whirring in his head only intensify as he does the workings out to reveal that these batteries and EVs aren’t having enough impact.
Post Coda, Czinger educated himself on lifecycle assessments, figuring only a holistic approach would return the energy emission reduction that is required in an era of climate emergency. He also came to realise that the way automotive structures are manufactured, and the costs required to do so, need optimising – particularly as EVs, hybrid cars and internal combustion engine vehicles (and all the tooling and fixturing to come with them) continue to emerge. “The amortisation period, the competition, the driving down of values, you’re looking and saying, ‘this is environmentally and economically broken,’” Czinger says.
Czinger and his team developed the Divergent Adaptive Production System (DAPS) to ‘digitise and dematerialise’ automotive production and provide the technical competency for the company, in time, to become a Tier One supplier to the automotive industry. What Divergent is willing to talk about, however, is how its DAPS workflow works. Its engineers start by understanding the static stiffness targets of a structure, then the typical load cases it will be exposed to, then what its boundary conditions are, then its crash requirements, durability requirements and dynamic stiffness response requirements. This information is the input for the Divergent design algorithm, which is where the company enters the concept phase. Here, Divergent gives the OEM ‘optionality’ to, for example, reduce stiffness in a certain area of the structure to reduce mass. After the concept phase comes the detailed design phase, and after that, it’s time to print the part.
Deep learning in product design
Digitization has also allowed engineers to give computers a more active role in the engineering process. Generative design and related optimization approaches work by programming a computer to run hundreds or thousands of simulations, tweaking the design between each run until it finds the best solution it can. The resulting geometries can outperform the work of the most experienced human designers.
At its outset, a deep learning surrogates (DLS) process looks a lot like other digital design optimization approaches. The engineering team defines the constraints and desired performance characteristics of the product, and the computer runs multiple conventional simulations on different design options. That’s where the approaches diverge, however.
As those initial simulations are run, they are used to train a neural network, which is set up to take the same inputs and attempts to replicate the outputs of the simulation system. When training is complete, this deep learning model will work just like the conventional simulation, but much, much faster. In real-world projects, deep learning simulation models can run orders of magnitude more quickly than their conventional counterparts.
Designing in the Age of AI
Design of a Ni-based superalloy for laser repair applications using probabilistic neural network identification
A neural network framework is used to design a new Ni-based superalloy that surpasses the performance of IN718 for laser-blown-powder directed-energy-deposition repair applications. Current high-performance engineering alloys commonly suffer from issues when processed using additive manufacturing methods. These include cracking, porosity, elemental segregation, and anisotropy. The computational method reported here enables the identification of new alloy compositions that have the highest likelihood of simultaneously satisfying a range of target properties, including criteria specific to additive manufacturing. The efficacy of this method is demonstrated with the design of a new alloy more amenable to laser-blown-powder direct-energy-deposition. The method may be readily extended to the optimization of other alloy types and process methods.
How Volkswagen and Google Cloud are using machine learning to design more energy-efficient cars
Volkswagen strives to design beautiful, performant, and energy efficient vehicles. This entails an iterative process where designers go through many design drafts, evaluating each, integrating the feedback, and refining. For example, a vehicle’s drag coefficient—its resistance to air—is one of the most important factors of energy efficiency. Thus, getting estimates of the drag coefficient for several designs helps the designers experiment and converge toward more energy-efficient solutions. The cheaper and faster this feedback loop is, the more it enables the designers.
This joint research effort between Volkswagen and Google has produced promising results with the help of the Vertex AI platform. In this first milestone, the team was able to successfully bring recent AI research results a step closer to practical application for car design. This first iteration of the algorithm can produce a drag coefficient estimate with an average error of just 4%, within a second. An average error of 4%, while not quite as accurate as a physical wind tunnel test, can be used to narrow a large selection of design candidates to a small shortlist. And given how quickly the estimates appear, we have made a substantial improvement on the existing methods that take days or weeks. With the algorithm that we have developed, designers can run more efficiency tests, submit more candidates, and iterate towards richer, more effective designs in just a small fraction of the time previously required.
JITX Launches General Availability And Announces $12M Series A From Sequoia Capital
Today we’re announcing the general availability of JITX and that we raised a $12M Series A round, led by Sequoia Capital, with participation from Y Combinator, Funders Club and Liquid 2.
Hardware engineers need a credible way out of the trap they find themselves in. JITX helps by letting them write code that automates their engineering process. To get ahead they can’t just do one design after another – they need reusable code that designs hardware for them. To illustrate this point: a software engineer can upload code to GitHub and thousands of people can reuse that code in their own projects. Using a traditional hardware design flow, each one of those thousands of engineers would have to re-design and re-analyze the same circuit to make sure the design will behave correctly in their product. JITX brings the productivity of software to hardware.
At the same time we were working with enterprise design teams like Northrop Grumman. It turns out that they also needed JITX to address some specific problems. Like everyone else, their biggest challenge is finding and retaining skilled engineers. There just aren’t enough experts to go around, and even entry level positions are getting harder to fill (turns out new EE graduates are more interested in AI than drafting circuit boards). So they use JITX as a way to make their existing experts more productive. They find a lot of value out of checking designs automatically – a manual derating analysis on a complex FPGA board can take months but JITX automates the whole procedure. They are also excited about using code as a more efficient way to coordinate across different teams in the organization. At the end of our iteration process we were quickly designing boards that were at the limit of what traditional factories could build (our thanks to Gerry Partida for an 8/4 stacked microvia with sub 70um trace and space!). For example we built this silicon validation board that included 2500 pins in a complex 300um grid.
Sim2Real AI Helps Robots Think Outside The Box
At Ambi Robotics, our robotic systems learn how to handle diverse items using data generated by advanced simulation. We fine-tune our simulations to the performance of our sensors, our robots, and variations on the items our robots will handle. Our simulations run extremely fast, hundreds of times faster than robots training in the physical world, so we can train our robots overnight. This is what enables our solutions to work reliably from day one.
How IGESTEK Produces 40% Lighter Automotive Parts
Autonomous Design Automation: How Far Are We?
As an industry, we will refine the different levels of Autonomous Design Automation further over the years to come. Eventually, the combination of the different steps of the flow with AI/ML will unlock even further productivity improvements. How long will it be until designers define a function in a higher-level language like SysML and, based on the designer’s requirements, autonomously implement it as a hardware/software system after AI/ML-controlled design-space exploration?
Improving PPA In Complex Designs With AI
The goal of chip design always has been to optimize power, performance, and area (PPA), but results can vary greatly even with the best tools and highly experienced engineering teams. AI works best in design when the problem is clearly defined in a way that AI can understand. So an IC designer must first see if there is a problem that can be tied to a system’s ability to adapt to, learn, and generalize knowledge/rules, and then apply these knowledge/rules to an unfamiliar scenario.
Calculating the best shapes for things to come
Maximizing the performance and efficiency of structures—everything from bridges to computer components—can be achieved by design with a new algorithm developed by researchers at the University of Michigan and Northeastern University. It’s an advancement likely to benefit a host of industries where costly and time-consuming trial-and-error testing is necessary to determine the optimal design. As an example, look at the current U.S. infrastructure challenge—a looming $2.5 trillion backlog that will need to be addressed with taxpayer dollars.
Generative Design for Milling Lightweights EV Motorbike Part
Generative design software uses a set of user-input parameters and constraints to develop efficient part designs. These shapes are often organic forms no human would design on their own, and in its earliest years generative design was locked to additive manufacturing and production methods facilitated by additive manufacturing. Not long after Lightning and Autodesk developed their first iteration of the generatively designed motorcycle swing arm, Autodesk updated its solver to support milling and other conventional manufacturing methods. Design candidates generated for milling generally cannot reach the same level of optimization as their AM siblings, but they are much easier to manufacture while still reducing the weight of the part.
What Is Generative Design, and How Can It Be Used in Manufacturing?
The primary use case of generative design in manufacturing is to automatically trigger design options that are pre-validated to meet the requirements you’ve established. That can be especially important for efficient manufacturing. Sometimes a part or tool must fit into an entrenched workflow or pipeline—methodologically or physically—as part of a larger device or process.
Chip floorplanning with deep reinforcement learning
Rolls-Royce Finds New-Engine Benefits in Old Test Data
The goal, according to Peter Wehle, head of innovation, research and testing at RRD, is to use this information to reduce new-engine weight and mass, while maintaining structural integrity.
Both parties are hopeful that using ML and AI will significantly reduce the number of sensors needed to obtain present and future data, thereby saving RRD millions of euros annually. According to Mahalingam, the software lets engineers choose the data they want from a data silo, select the algorithms they want to employ and decide whether or not they want to use a neural network to train an ML model.
Wehle notes that the disruptive tool is based on the interaction between a communication endpoint of the engine simulation and neighboring points. It carefully analyzes the effects of loads on physical structures.
Accelerating the Design of Automotive Catalyst Products Using Machine Learning
The design of catalyst products to reduce harmful emissions is currently an intensive process of expert-driven discovery, taking several years to develop a product. Machine learning can accelerate this timescale, leveraging historic experimental data from related products to guide which new formulations and experiments will enable a project to most directly reach its targets. We used machine learning to accurately model 16 key performance targets for catalyst products, enabling detailed understanding of the factors governing catalyst performance and realistic suggestions of future experiments to rapidly develop more effective products. The proposed formulations are currently undergoing experimental validation.
How Machine Learning Techniques Can Help Engineers Design Better Products
By leveraging field predictive ML models engineers can explore more options without the use of a solver when designing different components and parts, saving time and resources. This ultimately produces higher quality results that can then be used to make more informed decisions throughout the design process.
Evolutionary Algorithms: How Natural Selection Beats Human Design
An evolutionary algorithm, which is a subset of evolutionary computation, can be defined as a “population-based metaheuristic optimization algorithm.” These nature-inspired algorithms evolve populations of experimental solutions through numerous generations by using the basic principles of evolutionary biology such as reproduction, mutation, recombination, and selection.