Copyright 2025 The MathWorks, Inc.
This example shows the workflow to create a surrogate AI model using training data from a multibody model of a vehicle. The resulting AI model can be used for design space exploration and for finding the optimal design parameters.
- Early-stage physical physical design is supported by creating a reduced order model to rapidly evaluate hardpoint locations.
- Sensitivity analysis is supported by running many simulations in parallel and analyzing the influence of design parameters on performance metrics
- Training data for the AI model is produced using Design of Experiments to ensure the entire design spaces is covered.
- Machine Learning and Deep Learning are both used to create surrogate models that are automatically validated against the generated data.
- Optimization algorithms are used to identify the set of design parameters that balance the tradeoff between multiple performance metrics.
- A MATLAB App enables exploration of the design space using responses surfaces.
Open the project file SSVT_Susp_Opt.prj to get started.
View on File Exchange:
You can also open in MATLAB Online:
Simscape Multibody is used to model the vehicle. The multibody model has 94 parameters defining the front and rear suspensions which can be tuned. This includes hardpoint locations, spring stiffnesses, and damping coefficients. The parameter values can be varied without recompiling the model so that parameter sweeps can be run as efficiently as possible.
The virtual test used to evaluate the performance of the design has three stages to measure three performance metrics.
- Ride Metric: Drive over a bump and measure pitch, roll, and vertical acceleration
- Roll Metric: Complete a double-lane change maneuver (ISO 3888) and measure roll angle
- Braking Metric: Brake to a stop and measure braking distance
The simulation results below show the measurements taken to determine the performance metrics.
Design of Experiments is used to generate parameter sets that cover the entire design space. The model is simulated with those parameters and performance metrics are calculated. Statistical methods are used to identify the relative sensitivities of each performance metric to each parameter.
The plot below shows the 10 parameters identified as most influential. Reducing the design space from 94 to 10 parameters accelerates the process of generating the surrogate AI model.
Design of Experiments is used to generate parameter sets that distribute the parameters evenly throughout the reduced design space covered by the parameters with the largest influence on our performance metrics. The scatter plot below shows the parameter sets generated for our Design of Experiments and the distribution of performance metrics.
Training data is used to train surrogate AI models. After the models are trained, we assess the accuracy of each model using the test data. We compare the performance metric predicted by the trained model against the true value from the original Simscape simulation. The plot below shows how the predictions compare with the ground truth simluation results as well as the error. The most accurate models will be used for the following steps of the workflow.
To accelerate the optimization, AI surrogate models will be used. The AI surrogate models are loaded from a file. The models can predict a performance metric based on the design parameters. The function call to obtain the prediction from the AI surrogate model is added to the optimization variable which already has the design parameters and constraints.
The optimization is performed using five different objectives that use the roll metric, ride metric, or both. The resulting parameters are tested in the full simulation model to see the improvement in performance. The plot below shows the results from the full simulation model for each of the five optimizations and the default parameters.
A MATLAB App has been created using App Designer. When the AI Surrogate tab is selected, a response surface is plotted. Sliders allow the user to adjust the values of certain parameters. The response surface shows how sensitive a performance metric is to the selected parameters. Two of the parameters are swept over the design space while the other parameters are held at user-selected fixed values.
When the Multibody tab is selected, the App lets you run the simulation model using a single set of parameter values defined in the App. The plots show the values for the performance metrics as well as the transient simulation results.
To learn more about vehicle modeling and simulation with Simscape, please visit Simscape Vehicle Templates.