The spin-off Renumics GmbH uses machine learning methods to make Computer Aided Engineering more efficient and take the strain off computational engineers.
Crash tests are an expensive affair. In early development stages, collision experiments are therefore often replaced by computer simulations that can be performed thousands of times taking various influential factors into account. These simulations are usually based on computer-supported processes, on so-called Computer Aided Engineering (CAE). This concept centres on computational engineers who compile numerical models, thus assisting constructors in the analysis and optimisation of their designs. The crucial time and cost factors here are the many manual work steps involved. For example, computational engineers invest a considerable amount of time in routine activities such as pre-processing geometries and integrating data instead of being able to concentrate on modelling and analytical work, which is precisely where Renumics comes in. This KIT spin-off has developed a software with which CAE can be automated. In this context, machine learning methods help make simulations workflows considerably more efficient.
Contrary to what might be expected, however, the technology has its origins in medical engineering, and not in mechanical engineering. Two of today’s Renumics founders, Stefan Suwelack and Markus Stoll, worked as scientific assistants in the “Cognition-Guided Surgery” research project that was conducted by KIT in co-operation with the University and the University Clinic of Heidelberg. The idea pursued was an intelligent, cognitive assistance system for surgeons that would “think for itself”, similar to a human being. Here too, the combination of machine learning methods with simulation automation formed a crucial field of research. A mechanical engineering conference became a key moment for the duo. Their ideas caught on, and specialisation subsequently focused on mechanical engineering. “We had already had the notion of founding our own company for some time,” Stoll explains. “As encouragement grew, this endeavour suddenly assumed a concrete shape.” Steffen Slavetinsky, who had developed deep neuronal networks for the classification of 3D models in the course of his Master thesis, became the enterprise’s third founding member.
Since November 2016, the three young entrepreneurs have been supported by the EXIST grant for founders, and in February 2017, they officially started their business. Stoll and Slavetinsky are responsible for advancing the technology and customising it according to clients’ needs, while Suwelack sees to distribution. Although all three have studied informatics, the diversity of tasks involved in setting up a company are no problem for them. Suwelack explains: “It was important to us that all team members hold similar values and are excellent in their fields of expertise. Covering as many different skills as possible tended to be of more secondary importance. Mostly, we acquired the knowhow we were lacking through learning by doing. But we were also able to draw on an excellent network of mentors who we had got to know in several Accelerator Programmes.” For example, Renumics took part in CAT – a KIT promotion programme supporting founders in the early stages when they are fleshing out their business model.
“When we started our venture, many critics thought our idea was too unspecific, since we did not want to focus our company on any particular branch. But this did not deter us from carrying on. It was above all the support of Professor Rüdiger Dillmann, Director of the KIT Institute for Anthropomatics and Robotics that encouraged us to start our own business,” says Stoll. “And then there were the important contacts we were able to develop while participating in the STARTUP AUTOBAHN, an initiative that Daimler AG took part in launching. This platform for established enterprises and start-ups offered us the opportunity to talk directly to engineers, who explained to us where their problems occurred in the workflow and what was particularly time-consuming.”
Today, Renumics co-operates in pilot projects with many companies in the automobile, chemistry and mechanical engineering branches. “In addition to solutions tailored individually to large-scale firms, we are planning a sort of assembly kit for 2018,” says Stefan Suwelack, looking forward optimistically. “It will enable computational engineers to adapt software according to the requirements of the respective company. Our goal is an open, highly scalable web platform with which we will also be able to establish our company as a leading provider of CAE automation.”
Dr Stefan Suwelack