Artificial intelligence plays an important role in many industries and technologies. In simulation, too, AI can be used to speed up basic processes and administrative tasks in order to save time or simplify procedures. Setting parameters is one such area: a machine learning engine can observe experienced engineers as they use simulation tools and set parameters. ML can then replicate this process to a certain degree of accuracy to allow less experienced engineers to use the tool more efficiently.
Another area that AI and ML can help with simulation is using data-driven or physically informed neural network solvers to speed up simulation by several orders of magnitude. Instead of solving second-order partial differential equations (PDEs) using traditional numerical methods such as finite element or finite volume methods, these newer AI and ML methods use neural networks to solve PDEs. It has been proven that these methods work with simple geometries and boundary conditions. The challenge now is to apply these methods to complex problems in practice.

Multiphysical interaction

The concept of multiphysics has been around for 50 years. In the course of its development, it has been confronted with many challenges. The challenge today is the interaction between the different physics tools. In the past, an engineer would use different physics simulation tools to solve a variety of design problems in a single product. Take a computer chip, for example: You would first simulate the heat given off by the chip, then analyze how it affects the circuit board it sits on, and finally come up with a solution on how to cool the chip to protect the circuit board from cracking. While the aforementioned step-by-step approach was the best option for many years, engineers today are demanding a way to solve these problems simultaneously.
This multidisciplinary optimization reduces the time it takes to analyze the product, whether it's a chip or another product, and find the solution to each problem engineers face. This leads to optimized products and lower costs.

For example, Ansys has acquired Dynardo, a provider in the field of simulation process integration and design optimization. This has brought the company one step closer to multiphysics interactions and enables its users to identify optimal product designs faster and more cost-effectively. Additional efforts will be made this year to further advance the technology.

Microservices and hyper-scaling

Simulation is used in many areas. In wind turbines, for example, the wind flow on the rotor blades can be simulated.

© Ansys

We will also see progress in the area of 'microservices for simulation'. Here, the main parts of the simulation, for example geometry, then meshing, followed by solver and finally post-processing - are transformed from a monolithic process to dedicated separate parts. The steps required for the simulation are independent of each other. Instead of a single process, many services are provided, such as for geometry, meshing, solver and post-processing; these can then be used by different products that communicate with each other via APIs (Application Programming Interfaces) and can be executed in a scalable manner on cloud computing platforms such as Microsoft Azure or AWS. The result is better accessibility, more flexibility and better reusability to solve many different tasks. With the help of APIs, users of simulations can, for example, connect tools from one company with the systems of other companies to create an open platform.
One of the biggest challenges for users of many types of software is the runtime. They are increasingly demanding faster runtimes. Simulation is no exception here, but the focus on it will increase significantly.

One way to improve runtime is through parallel computing. Over time, parallel computing has taken many different forms. It has evolved from Shared Memory Processing (SMP) to Message Passing Interface (MPI) to fine-grained GPU-based parallelism and task-based parallelism. The idea for Hyper Scale is that we utilize all forms of algorithms on supercomputers. If users can run hyper-scale simulations, they will likely be able to run simulations in a few minutes or hours that might have previously taken 10,000 hours.
There is still a lot of work to be done in this area. 2020 will be too soon to see hyper-scaling on the scale described. But we can definitely expect it within the next decade.

Forward-looking and robust design

In an effort to achieve efficiency and cost savings, many manufacturers and service providers have eliminated over-engineering and instead focused on minimalist design. Where 10 cm of asphalt is needed for a highway, exactly 10 cm is used, not 20 cm as was common in the past 'just in case'. The problem is that there is variation in all materials. This means that the calculation of the required asphalt volume can vary from project to project: 10 cm of asphalt would be sufficient in one case, but not in another.
Robust design through simulation addresses such uncertainties and will be increasingly used this year. Using simulation to evaluate materials and calculate uncertainty prevents both over- and under-development of products and services. While a safety factor of 500% would be too high and therefore inefficient, 100% leaves no room for material variations. Based on different information, the safety factor for robust design would be 110%, for example.

Digitization of the physical world

Air flows and heat dissipation are simulated during the development of electronic devices. This prevents possible damage caused by overheating of individual components in the finished device.

© Ansys

Simulation is already digital, right? Yes, but simulation also increasingly encompasses the physical world. Thanks to the Internet of Things (IoT), the use of digital twins has increased recently. Engineers digitize information from a physical part so that they can analyze its performance and monitor systems. This allows them to avoid problems with the real component or machine before they even occur. Now that the technology has become established and proven to minimize downtime and the associated costs, digital twins will increasingly find their way into companies.
Augmented (AR) and virtual reality (VR) also need to be considered. Engineers currently visualize their simulated designs on 2D screens. However, as VR and AR technologies become faster and more accessible, designs will soon be able to be visualized in a 3D environment on an AR or VR headset. The data can then be evaluated more easily and the designs are easier to understand, edit and test. This leads to a leaner and more effective process.

Design, testing, maintenance - simulation plays a crucial role in these processes. But what happens when a product or part fails? How can simulation help? By digitally mapping their processes, manufacturers can coordinate all their activities from initial design through production to sales and track every part. So if a problem occurs with the brakes on a particular vehicle model, this can be traced back to the original simulation design. This design is then reviewed to quickly identify the fault. If the model is recalled, the problem can be fixed faster and more cost-effectively than through a manual testing process. Digital transformation is a big task for most companies, but many are already tackling it. It can be assumed that the digital transformation will accelerate this year and that large manufacturers will be able to complete the change before the end of the year.

New areas of application through simulation

Prith Banerjee is Chief Technology Officer at Ansys.

© Ansys

Simulation is well established in a number of fields with real multi-physical conditions. This year, simulation software providers will push the boundaries of the technology to solve problems in other areas of physics. Healthcare chemistry is currently an area that is not yet sufficiently covered by multiphysics simulation, but would benefit significantly from it.
would benefit from it.

What could this look like in practice? Clinical trials for new drugs require testing on humans, but one day trials could be replaced by simulations: The need to test the drug on thousands of subjects would then be eliminated, as would the enormous cost of the stu-dies. And where studies cannot be carried out, for example on children, the potential uses for simulation are enormous. Even in healthcare, where a blood clot can lead to a heart attack, simulation could be used to identify the right drugs to dilute the blood and dissolve the clot.

Today's technology makes it possible for products and tools that previously took years to develop to be manufactured overnight. For example, a start-up could develop a completely new way to accomplish a task or solve a problem, disrupting existing processes. Many forgotten companies have failed to prepare for such a situation. But instead of ignoring the remote possibility of this happening, companies need to address the risks directly by innovating.