Applications of AI in Design
BY:
ROHAN KULKARNI(11),VINEET KUMBHAR(12),KHSITIJA KUMBHARKAR(13),MADHURA KUNACHI(14),GANESH LOHIYA(15)
INTRODUCTION OF AI IN DESIGN
Engineering was a job carried out with pencils and paper not all that long ago. Calculations were made by hand and designs on large sheets were sketched out. Physical models will be created from actual blueprints to figure out how the final product should look and be made.
Of course, engineering today is a discipline that is deeply concerned with software and machine tools. Some of the basic techniques that engineers deploy when designing new product designs are computer-assisted design, computational fluid dynamics, and finite-element analysis applications. Prototypes may be printed directly from machine files when physical models need to be checked.
Although these instruments have strengthened the capabilities of engineers, the engineer is still clearly in charge of the design process. That power, however, is now in doubt. Growing interest is being expressed in using emerging artificial intelligence and other innovations to achieve higher levels of product automation and drive new product innovation. Advances in AI, coupled synergistically with other innovations such as cognitive computing, the Internet of Things, 3-D (or even 4-D) printing, advanced robotics, virtual and mixed reality, and interfaces with human machines, are changing what, where and how goods are built, created, manufactured, delivered, serviced, and updated.
ROLE OF AI IN DESIGN
This revolution will allow for a new kind of design process, one where, with little human interference, AI-enabled programs iterate and optimize. The resulting designs seem extremely complex, but are no more difficult to print than traditional designs, thanks to advanced printing technology. In commercial aircraft and other vital structures, parts that are the result of this generative design process are already being prepared for use.
The shift from drafting boards to CAD to engineering was disruptive. It is expected that the next transition to generative design will be more disruptive.
Artificial intelligence is a notion that includes a wide variety of technology, and for some time, some kinds of AI have been applied to engineering systems. In the 1980s, many of the mundane activities for engineers were first used to automate knowledge-based systems and AI rule-based expert systems. In the 1990s, the intelligent agent model was introduced and a shared language was given to identify issues and share their solutions.
These apps are considered to be “weak” AI.
“Strong” AI, on the other hand, will function more like general intelligence and be able to sense, interpret, learn from and react to the environment and users. Strong AI, also referred to as Artificial General Intelligence (AGI), refers to deep learning and machine intelligence, systems that demonstrate complex behavior similar to living systems such as swarms, colonies of ants, and neural systems. The ability to adjust to most circumstances will be provided to these systems.
In leaps and bounds, artificial intelligence is moving forward (indeed some researchers are now talking about the development of artificial superintelligence-ASI) and much of the AI enthusiasm is targeted at applications where computer systems work with great autonomy. The self-driving car is the poster child for AI, however there are a range of interesting applications, from robotic doctors that can more reliably diagnose diseases than any human doctor to AI-directed businesses that can orchestrate business operations without flesh and blood management.
Existing artificial intelligence has already impacted the product-design process, and AI will change the way we embed connected sensors and use mixed or augmented reality headsets in the future. Based on the current trend, in the next decade, we are likely to see AI influence product design and the development of engineering systems in three distinct phases.
Second, the laborious tasks faced by designers, such as having to continuously search for suitable content, correct mistakes, find optimal solutions, communicate changes, and check for design failure, will be eased by artificially intelligent systems. It would be possible for machine learning to take on certain jobs and do them much quicker.
Next, in generating sophisticated designs, AI would be able to help. At the designer’s elbow, intelligent systems will work, propose alternatives, incorporate sensor-based data, produce design precursors, optimize supply chain processes, and then deliver the designs to smart manufacturing facilities.
ACTING ON INTENTION
The final step will have more significant effects. During the design and development process, engineering systems that integrate stronger AI would be able to act more like human assistants. True human designers will only be able to design” by communicating intent and curating outcomes, while in order to produce new design iterations for analysis, intelligent systems and machines will act on these intentions.
However the AI wouldn’t approach the project the way a human designer might. Instead the computational power will be used to replicate the evolutionary method of Nature, taking the best current solution to a problem and iterating in each setting to maximize efficiency. In this way, beyond what is actually possible using the conventional design process, the AI will explore the variants of a design. This approach is called Generative architecture.
The engineer or industrial designer, along with design criteria and constraints, including material type, manufacturing capacity, and price points, sets high-level design objectives.
The AI generative design framework, such as Autodesk’s Dreamcatcher, explores the permutation of a design solution with the limits of the design problem identified, cycling rapidly through thousands or even millions of design choices and running performance analyses for each design. The device will tap available cloud computing processing resources for the most intensive calculations.
Its machine-learning algorithm is a core component of a generative design method. Without human guidance or interference, the algorithm detects patterns inherent in millions of 3-D models and generates taxonomies. Generative design software may use that skill to learn what all the components of a complex system are, define how they relate to each other and decide what they do.For a particular dimension of a piece, it can then serve up hundreds of different design options and provide them as parts for the next design.
PROJECT DREAMCATCHER
Dreamcatcher is Autodesk’s experimental platform to explore the potential of AI techniques and generative design methods in product development, from conceptual design to manufacturing.
The Autodesk dreamcatcher includes:
Designers’ methods for explaining design concerns. Solutions become scalable and accretive by pattern-based definition, thereby extending the quality and number of alternatives searched for in each design session.
Form synthesis tools, including several purpose-built methods that from a wide collection of input parameters, algorithmically generate designs of different types.
Exploration tools, offering a range of potential solutions to designers and their related solution strategies. These tools help designers create a mental model with high performance alternatives compared to all others in the package.
The designer can export the design to manufacturing tools or export the resulting geometry for use in other software tools until the design space has been explored to satisfaction.
The human re-enters the process once new designs have been created by the AI system. Based on the various choices of designs offered by the generative design method, he will review various options and then change the design goals and constraints to narrow down the options and optimize the available ones. Using the data, another collection of designs will then be iterated by the generative design method.
The most appropriate solution will be chosen by a combination of artificial intelligence and human intuition over the course of many of these periods.
Generative design techniques are not particularly fresh, but new excitement has been generated by integrating these deep reinforcement machine learning algorithms with cloud computing.
EVOLVING AN ANSWER
The generative design method may sound like something for the far future, but it has recently been applied to a real-world challenge involving a part of the Airbus A320 aircraft, one of the world’s most high-profile and costly items.
The portion was a partition separating the passenger cabin from the plane’s galley and supporting a flip-down seat during take-off and landing for flight attendants. Airbus engineers were searching for ways to reduce the weight and volume of the partition while maintaining enough power to bear flight attendants’ loads. It also had to hold up in case of a crash landing under the force of 16 g.
With a mix of generative design, biomimicry principles for material and structural design, and additive manufacturing, a group of Airbus designers turned to Autodesk and other collaborators to see if they could come up with an improved partition.
The generative modelling processes used by the team employed two biological model-derived algorithms. The first was focused on the slime mould adaptive networks: a single-celled organism that can expand, spread and aggregate with the minimum number of lines to form multicellular structures. In case a line fails, these systems have a built-in redundancy to maintain communication within the network. The construction of the bracing of the overall partition was informed by this algorithm.
The lattice that makes up each member was developed using a second algorithm derived from the microscale structure of mammal bones. Several different load cases were considered, some of which allowed the partition to contain more than 66,000 micro-lattice bars.
The generative design program (in this case, Autodesk Within) cycled through thousands of design variants after the design parameters were set. To determine the prototype, the human design team digitally mapped the various created options against weight, stress, and strength parameters.
A latticed structure that looks random but is based on the growth of mammal bones is the resulting design. The partition is thick at points of tension, like natural bone, but lightweight everywhere else. The feature called the bionic partition by Airbus and Autodesk, is 45 percent lighter than the conventionally built compartment divider used on current aircraft. The finished product, produced using additive manufacturing, needs only one-twentieth of the raw material compared to a partition constructed using conventional design processes.
THE AI CONTINUUM
In items ranging from smart gadgets and smartphones to drones, robots and autonomous vehicles, artificial intelligence is integrated. The AI spectrum can currently be divided according to complexity into three general areas:
ASSISTED INTELLIGENCE: AI automates repetitive and structured simple activities, operating from rules that are clearly specified. The main decisions are still being taken by humans. Examples include automated assembly line robots as well as software-based agents simulating humans’ online activities.
AUGMENTED INTELLIGENCE: AI increases humans’ ability to perform tasks, and humans and machines learn from each other. Examples include intelligent virtual assistants, some systems of generative design, and systems that can bring human attention to rare or noteworthy events.
AUTONOMOUS INTELLIGENCE: Some decision-making is taken over by AI, but only after a person completely trusts the computer or becomes responsible for the prompt performance of a task. Self-driving cars are only one example of autonomous intelligence, currently in production by over 30 companies.
More than 100 separate parts were 3-D printed and then assembled, made from a high-strength metal alloy developed by Airbus. The resulting partition is the largest 3-D printed aircraft cabin feature in the world, and it more than satisfies the demands of the Airbus team. It is thinner and stronger than the portion it will replace and each bionic partition will save approximately 3,180 kg of fuel per plane per year because it is 30 kg lighter.
Final testing and approval are ongoing for the partition. When finished, the final design will be used for next year’s A320 aircraft.
The lessons learned by Airbus in the design of the bionic partition pave the way to transform how an entire aircraft is built and constructed. The next generation of Airbus aircraft will be composed of components based on generative design, constructed using advanced materials by 3-D printing. For example, the cockpit wall, which is twice the size of the bionic partition and needs to be bulletproof to protect the pilots, or the structure that houses the galley for food and beverage service. Airbus aims to improve its methods for manufacturing larger structures within an aircraft.
In other high-profile projects, Autodesk has also implemented its generative design and AI software. To design the chassis of a race car, Autodesk partnered with the digital industrial engineering company Hack Rod. The company is also interested in the MX3D bridge project in Amsterdam. Not only will the bridge be generatively constructed, but multi-axis industrial robots will also print it on-site.
The generative design instruments created by Autodesk and other companies are just the beginning of an engineering revolution, one that will change the way we think about it.
Given the power of generative design instruments and the near-omniscient artificial general intelligence that will soon weaken it, some will wonder whether it will make the human designer and engineer redundant. We think it is the wrong way to look at artificial intelligence’s contribution. Instead the ingenuity of designers will be enhanced by these innovations and their productivity will significantly improve.
For example, a designer would be able to start a conversation with a cognitive design assistant using a system connected to a powerful voice recognition system. “He begins, “Let’s build an autonomous vehicle. Several distinct vehicle types appear. The designer chooses a category, defines the priorities and constraints of high-level design, and the design tasks begin with the cognitive assistant doing much of the work and creating many alternatives beyond what can be achieved by existing generative design systems. In assessing the various choices and making the final decision, the cognitive assistant will also assist the designer. It will also use the new explainable AI (XAI) tools to explain the logic of the chosen options to the designer. The cognitive assistant would automatically submit the files to an automated additive manufacturing facility to have a prototype made for the device after completing the design of a complex system.
It will not only be projects that undergo iterative processes that are biologically influenced. Self-evolving, restoring or even redesigning themselves as needed would become future complex systems.
A recent study by a team in the Department of Informatics at the University of Oslo in Norway provides an example. The study team, led by Kyrre Glette, found a way for robots to use generative design tools to design, grow, and produce themselves. The robots are designed, and their movements replicated in a virtual world. The scientists then add those criteria, such as getting from one side of the room to the other and each generatively generated design is checked against certain objectives by the AI. Fresh prototypes spawn the best results, and then the final design is printed automatically.
The goal is to build robots which are self-evolving. In the assembly of the printed pieces, the only human interference is
Ultimately, this research and related activities will contribute to the creation of self-repairing, self-healing, self-adaptive, self-reconfiguring systems and “operationally” improving products. Such products could grow over time instead of depreciating in value and ability.
WILL AI REPLACE ENGINEERS?
The response is NO. The position of the human engineer will in time, be that of a director rather than a producer. Humans may not be the ones carrying out the tasks, but we will select the path we want the system to follow and provide the most important feedback: if we are pleased with the performance.
Most of the technical aspect of engineering will be shifted to the machine-based design method, just as a good engineer today does not need to be able to operate a slide rule or complete an isometric drawing. To some degree, in a working partnership with an artificial intelligence that can find the solution as long as it knows what the problem is the programmer will become someone adept at interpreting the inchoate human desires for goods with a more elegant shape or using less resources or operating more efficiently. Engineering will be altered until computers know how to build, even how to design themselves, but engineers will still be highly trained. AI technologies can augment them cognitively, mentally, and perceptually. And thus, with a different set of skills, they will simply have to develop their abilities, including teaching the AI systems how to innovate and become successful collaborators in potential human-AI organizations.
