On April 20 – 21, the TechNet Alliance Spring Meeting 2018 was held in Berlin, Germany. On this occasion, Prof. Wolfgang A. Wall, shareholder and co-founder of AdCo EngineeringGW GmbH, gave a presentation entitled “Predictive simulation of complex coupled problems – and the rising importance of uncertainty quantification and machine learning”. After an introduction into computational methods for multi-field problems, uncertainty quantification (UQ) and machine learning, innovative predictive physics-based simulations models for three highly interesting fields of application were presented. The fields which are attracting particular attention are 3-D printing in the form of selective laser melting (SLM), battery simulation, and biomedical applications.
The technique of 3-D printing in the form of SLM, where pre-defined contours in successive layers of powder are selectively melted by using a laser beam, offers several benefits, such as a near net-shape production rate and a Novel advanced simulation technology can provide crucial support for the design process of, for instance, complex load bearing components. The second presented field of application is predictive battery simulation, particularly with an eye to electric and hybrid cars. As already outlined in a previous blog post, AdCo EngineeringGW acts at the forefront of this research and development, providing its customers with advanced simulation technology that is key to a better understanding of new types of batteries such as solid-state batteries. In this context, we are currently collaborating very closely with BMW AG.
Finally, as the third field, predictive computational methods for challenging biomedical applications such as abdominal aortic aneurysms (AAA) and the human lung were addressed. All three fields of application have in common that uncertainties have to be taken into account when doing predictive simulations. With our UQ methods, confidence intervals for simulation results as well as worst-case estimates can be provided. Furthermore, global sensitivity analyses are enabled. For all these goals, among others, machine learning such as Gaussian processes can be utilized. Furthermore, deep neural networks can be exploited as surrogate models for UQ (“deep UQ”), where data-efficient deep neural networks are generated with built-in symmetries respecting the individual physics in the form of physics-based deep learning.
More information on TechNet Alliance:
The TechNet Alliance, founded in 1998, is a global network of computer-aided engineering (CAE) companies and, as such, constitutes one of the longest standing CAE Alliances in the world. In addition to members offering solutions for the CAE market (Principal Members), the network also includes companies supporting business activities (Business Support Members). CAE Experts from customers (Corporate Members) as well as worldwide acknowledged professors and retired CAE managers (Honorary Members). Today, the TechNet Alliance consists of over 70 members in Europe, Africa, Brazil, Asian countries such as China, India, Japan, and South Korea, United Arab States, and the United States of America. Since 2000, the TechNet Alliance meets twice a year to share experience and knowledge, with the Spring Meeting 2018 in Berlin being the most recent one.