This presentation was made at the NAFEMS Americas Seminar "Model-Based Engineering: What is it & How Will It Impact Engineering Simulation" held on the 1st of October 2019 in Columbus Ohio

Resource Abstract

As the electrification of systems continues, it is becoming increasingly important to be able to evaluate tradeoffs and alternatives as soon as possible, so that engineers can determine which components would work best and how they should be connected within an overall system. At the same time, organizations are looking to artificial intelligence in order to control these systems as they get more and more complex.



In this talk, we will review two application areas that can be accelerated through the use of Model-Based Design: full vehicle simulation and reinforcement learning. Full vehicle simulation models are used to assess alternatives according to specific objectives, such as fuel economy or performance. At times, this requires the integration of models from different engineering teams, who use different modeling and simulation tools, into a single system level simulation. This can be difficult to do in a traditional testing environment where different coded algorithms need to be stitched together. In the first part of this talk, we will review a case study to see how to evaluate different architectural candidates for an electrified powertrain and how using a simulation integration platform can help with the selection of the best candidate, given our objectives.



In the second part of the talk, we will review how Model-Based Design can be applied toward reinforcement learning. Reinforcement learning is an exciting new area due to its potential to solve complex problems in areas such as robotics and automated driving, where traditional control methods can be challenging to use. In addition to deep neural nets to represent the policy and algorithms to train them, reinforcement learning requires a lot of trial and error. Although this can be done with hardware, it can not only take a lot of time to collect the required amount of data, but for some applications it could also be both expensive and dangerous for systems, even if they are prototypes, to repeatedly fail during training. This is where using a simulation platform is advantageous – you can run thousands or millions of simulations to train your system to complete its objectives in an optimal manner. Using a walking robot case study, we will go through the steps needed to set up and solve a reinforcement learning problem, and how to take advantage of parallel simulations and automatic code generation to quickly get to an implementation that works.