Evolving Legacy Modeling Tools with the VERDE Toolchain at GM

At General Motors (GM), the Energy Model and Toolchain Development team creates and manages tools that GM engineers use to predict, analyze, develop, and validate core vehicle performance attributes—including driving range, fuel efficiency, drive quality, and acceleration. As part of this team, we are always looking for ways to advance energy analysis at GM through vehicle energy modeling and simulation.

GM engineers have been using energy modeling tools for more than 20 years. We developed these tools in-house for the greater control, speed, flexibility, and confidentiality that this approach provides instead of relying exclusively on commercial-off-the-shelf (COTS) packages. Though the tools we developed continued to increase in capability and efficiency, the number of users of these in-house solutions had begun to decline. This decline was attributed to several factors, including outdated user interfaces, difficulties integrating the tools with third-party software for cosimulation, and limited collaboration with engineers working in other domains. 

To address many of the longstanding challenges with our legacy modeling and simulation tools, we have created the Vehicle Energy and Range Development Environment, or VERDE. Built using MATLAB^® and Simulink^®, VERDE incorporates a modern user interface that makes it easier to learn and use for building and executing models (Figure 1). VERDE supports cosimulation with third-party tools for modeling thermal characteristics, high-voltage batteries, and noise, vibration, and harshness (NHV), among others. It is also both accurate and scalable: Our engineers use it to perform large-scale optimization studies and design of experiments (DOE) on high-performance computing (HPC) clusters. With the flexibility to support conventional, hybrid, electric, and fuel cell powertrains and the ability to streamline a wide variety of use cases from basic simulation, data visualization, and reporting to more advanced virtual hardware-in-the-loop simulation and real-time driver-in-the-loop testing, VERDE has attracted interest from diverse engineering groups, resulting in a 40% increase in users compared to our previous modeling environment. This article provides an overview of how VERDE was built and how it is used to advance energy analysis across GM.

The VERDE User Interface and Model

Among our top priorities in developing VERDE was improving the user experience. The precursor to VERDE was a custom MATLAB toolbox that had been developed in-house at GM over many years. Its user interface was complex and challenging to learn, which made it difficult to both onboard new engineers and add new features.

We built the VERDE user interface using MATLAB App Designer. It includes numerous features—including multilayer tabs, dynamic tree options, and interactive graphics—that make it intuitive and easy to navigate. As part of the redesign, we made the decision to structure VERDE so that the user interface and underlying full-vehicle model are semi-coupled, enabling VERDE to support multiple vehicle architectures.

With this semi-coupled structure, the VERDE user interface interacts with the full-vehicle models we’ve built to set up and run simulations. As we developed the models in Simulink, Simscape™, and Simscape Driveline™, we took advantage of several key features to manage the scale and complexity of modeling a system with more than 2,000 parameters and 200,000 blocks (Figure 2). For example, we used a bus architecture to lessen line complexity and clutter, model references to create and maintain an ordered model hierarchy, and model variants to vary the fidelity levels of individual components within the model as projects move from program framing to development and refinement, and finally to production.

VERDE Preprocessing and Postprocessing Tools

While a simple usage scenario for VERDE involves configuring the model, initiating a single simulation run, and checking the results, our engineering teams often have requirements that go far beyond this basic use case. To support these requirements, we built extensive preprocessing and postprocessing capabilities into VERDE.

On the preprocessing side, we added features that enable engineers to compare models against one another, set up and run DOE using a seed model, manage input data from a database, and submit jobs to an HPC cluster (Figure 3). The data management feature represents a recent enhancement to VERDE that is part of our ongoing efforts to improve the tool. In the past, VERDE used individual data files that were managed and stored locally by the user. Today, VERDE uses a PostgreSQL^®-based application named InSPIRE (Innovative Simulation Parameter Inputs and Results Environment). Integrated into VERDE with Database Toolbox™, InSPIRE stores model inputs and simulation results in a relational database, enabling enhanced automation, reporting, analytics, and traceability. 

We have also developed a set of bespoke tools for postprocessing data from simulation runs. Our plotter tool, for example, is used to plot time-series simulation results; it can also import data from other sources to be used in plots and export data native to MATLAB into other formats (Figure 4). Other postprocessing tools include an energy calculation tool for energy balance analysis and a model correlation tool for automating comparisons between simulation and test results.

Cosimulation with COTS Tools

From the outset, it was not our goal to recreate all simulation capabilities used at GM by implementing them directly in VERDE. Rather, the idea was to integrate tools already used by our engineering teams within the VERDE framework via cosimulation. To this end, we built in support for direct cosimulation with a variety of COTS software, including tools for modeling EV thermal systems, internal combustion engine thermal performance and emissions, and vehicle and driveline dynamics (Figure 5). VERDE supports Functional Mock-up Interface version 2.0 (FMI 2.0), making it possible to integrate it with COTS software that also complies with the standard.

Due to VERDE’s modular design, individual subsystems can be extracted as standalone modules for integration with lap-time simulators and other tools. VERDE is also capable of running in real time to enable driver-in-the-loop simulations (Figure 6).

Flexibility, Fidelity, Scalability, and Accuracy

In the time that our GM engineering teams have been using VERDE, it has delivered on the key goals that we had for it when we began its development.

From a flexibility standpoint, VERDE offers teams a comprehensive set of options for energy modeling and simulation. Engineers can set up simulations using a combination of vehicle architectures, drive scenarios (focusing on fuel economy or performance, for example), control and calibration settings, and cosimulation tools (Figure 7).

As a project progresses from the early exploratory stages toward production, engineers can use VERDE to increase the fidelity of their models. For example, initial simulations might use simple models for high-voltage battery temperature, such as a constant value or a temperature profile used in past simulations. Later, when greater fidelity is needed, the simulations can incorporate cosimulation with a variant based on a more realistic thermal modeling tool.

VERDE has also proven to be scalable. Its global user community has grown to almost 400 individuals across diverse engineering domains. Users can, when needed, submit large-scale simulations (for optimizations or DOE) to an HPC cluster directly from the VERDE user interface.

To validate the accuracy of VERDE simulations, we compared the results of real-world tests with simulation results produced by VERDE. For example, in one test, we simulated EV range in miles for a specific vehicle configuration and then compared the results with measured results from an identically configured electric vehicle. The comparison yielded an R-squared (R²) value of 0.9998, indicating an excellent fit between the model and the measured data, with the model being on average about 0.4% more pessimistic (Figure 8).

We are continuing to enhance and extend VERDE with an ongoing schedule of well-defined and consistent releases. We have several improvements planned that will further improve cross-domain collaboration and alignment, including support for new cosimulation tools, support for durability modeling, and speed improvements to run large studies faster. 

Conclusion

As the engineering requirements for GM have evolved, so have the simulation tools we use to analyze our vehicles. Due to the flexibility of the MATLAB and Simulink platforms, we’ve been able to improve our simulation tools to meet those needs. Modernizing the user interface is a key aspect of supporting our engineering teams, but equally important are enhancements to the underlying vehicle models, the architecture for cosimulation, the database for managing results, and the pre- and postprocessing tools. These improvements have required significant investment, but they have provided an even stronger foundation for the next generation of GM engineers to design the next generation of GM vehicles.

Acknowledgments

Special thanks to Nate Wilmot for guidance through this work and significant contribution to this article and presentation.