AI / MACHINE LEARNING

26 | Engineering Reality Magazine

Enabling Accurate DesignDecisions While Compressing

Engineering Timelineswith CADLM Technology

By Kambiz Kayvantash, CEO, CADLM Hemanth Kolera-Gokula ,Fabio Scannavino, Manuel Chene, Raoul Spote, MSC Software

Volume X - Winter 2019 | mscsoftware.com | 27

28 | Engineering Reality Magazine

In a recent proof-of-concept, Adams, the Multi-Body Dynamics (MBD), and Lunar, the supervised Machine Learning solution from CADLM, were used to create Reduced Order Models (ROMs) of vehicle behavior. The simple scenario involved a vehicle accelerating on a straight road encountering an obstacle in its path. The goal was to predict any interference with the obstacle using real-time reduced-order models. The process occurred in two distinct phases. A design space for exploration that spanned two dimensions, vehicle velocity, and mass was defined in Adams. The process then involves creating a dataset in Adams that spans the design space. Lunar then consumed this dataset and created a purely data-based model that emulated the high-fidelity representation in Adams.

The transient predictions from the ROM are in line with the Adams results both in terms of the trend and magnitude with the added advantage of real-time predictions. The caveat being that the ROM is valid only within the bounds of the design space and

Real-Time Reduced Order Models

less computational overhead. Reduced Order Models (ROMs) provides an opportunity to create a virtuous Real-Time loop between Design and Operations with real time information sharing. It also provides an opportunity for simulation software providers like MSC to truly democratize engineering simulation across the product life cycle in a scalable manner without compromising on model fidelity.

In the past decade, much effort has been made to develop various methods of model reduction. MSC and CADLM have partnered to develop model reduction approaches for a variety of engineering problems while remaining agnostic to the underlying physics type. Using this approach, the engineering simulation community can tailor the level of model fidelity to the underlying simulation intent. For example, reduced-order surrogates of high-fidelity models can be used to explore the design space and execute computationally intensive, vehicle reliability, and optimization tasks.

Dynamic processes can be described using differential equations, and the solution to these equations can provide insight into the

nature of these processes. However, the simulation of such equations utilizing computational techniques, such as finite element or finite volume methods can become computationally very expensive or for some industrial problems unfeasible. The modeling of optimization problems and Multiphysics phenomena in practical engineering applications is often very challenging as repeated numerical simulations are required. A remedy is a simplification of the physics-based model but that relies on the experience and intuition of the engineers. Another avenue is reduced-order modeling, a mathematical approach serving to overcome high computational costs of the simulations. The primary goal is to approximate the large-scale problem by a much smaller one, which yields somewhat less accurate results but can be solved with considerably

28 | Engineering Reality Magazine

Volume X - Winter 2019 | mscsoftware.com | 29

Training the ROM Lunar Model to create a ROM Validation between ROM and Adams results Prediction based on validated ROM

any explorations beyond these bounds would require additional ROM training.

CADLM capabilities can be leveraged across the MSC portfolio and are agnostic to the physics type. Reduced order representations of both systems and component can be created. MSC is also looking to leverage ROMs to extend its industry leading capabilities in the area of multi-physics. As an example, ROMs simulating highly non-linear Fluid-Structure interactions have been created from time intensive Marc-Cradle simulations. The application goal was to optimize the displacement sensitivity of a flexible membrane with respect to various inflow velocities and membrane mechanical properties. This involved the costly computations and co-simulation between Marc and scFlow. A Reduced Order Model was constructed based on only eight completed runs and employed within an optimization loop. The solution was found by LUNAR within seconds and a final co-simulation was conducted in order to validate the optimized solution.

CADLM technology allows simulation users to make effective and precise engineering decisions while compressing design timelines. The ability to create traditional high-fidelity models and extract reduced order surrogates from them provides a simulation user the opportunity to tailor the model fidelity to the simulation intent and conduct real time optimization or reliability studies.

For More Information: www.cadlm.com

Figure 1: ROM Lunar analysis of an Adams Car Model

Figure 2: Marc-Cradle analysis of an Diaphragm Valve Model