Cummins Uses AI-Based Reduced Order Modeling to Predict Engine Performance and Emissions
Approach Enhances the Speed and Precision of Engine Performance Models
“There are a lot of benefits associated with using MATLAB, such as the low to no coding experience required, so a beginner user can also develop these models…. We can get more out of the platform without having to spend a lot of time on the code development work.”
Key Outcomes
- Using MATLAB enabled Cummins to reduce engine cycle simulation run times to one-eighth of real time
- Low-code tools allowed technical experts to focus on analysis over coding
- Accelerated end-to-end AI model development workflow, cutting down cost, effort, and memory footprint
To accurately predict engine efficiency and emission levels, it is essential to build models that simulate engine cycles. However, developing these models involves various 3D-to-1D simulations—often using third-party tools—which can take more than 20 times longer to complete when compared to real time.
To improve the speed and accuracy of these models, Cummins, a global leader in engine development, used MATLAB® to build LSTM-based neural networks. The team modeled 26 different engine responses—including pressure, temperature, and engine brake torque—using Deep Learning Toolbox™ and Statistics and Machine Learning Toolbox™.
Using MATLAB required little to no coding experience from the Cummins team and helped increase model speed to eight times faster than real time. In the future, the team plans to integrate their models with real hardware and control components.
