Sanden Improves Modeling Efficiency for Automotive Air Conditioning with MATLAB
Accuracy and Computational Speeds Improved with New Models
The background of this project was to see if it was possible to use a neural network to ensure accuracy while reducing the sampling time in motor model simulation…. MathWorks proposed using a deep learning method called LSTM and this initial introduction was necessary to accelerate future work.
Key Outcomes
- Created a high-precision, surrogate model of the automotive air conditioning system
- Computation speed for models was improved by up to 100 times
- Successful modeling set up a path to incorporate real-world parameters such as heat in the future.
Using AI, Sanden engineers automated part of their component modeling process.
Sanden is a top manufacturer of automotive air conditioning systems in both Europe and China. The company has a long history of adapting its technology to meet changing trends in the automotive industry, such as developing heat pumps that enable efficient heating in electric cars that cannot use engine waste heat. However, as the automotive industry undergoes substantial changes, leading to a rise in the man-hours needed for developing new systems, model-based development is becoming more important for parts manufacturers.
When creating models of electric compressors for air conditioning, Sanden must balance requests from clients for both faster model computation time and improved accuracy. To help meet these contradictory goals, Sanden used MATLAB® and Simulink® tools to incorporate a deep learning neural network into its models.
The team used Deep Learning Toolbox™ to train a long short-term memory neural network that was integrated into the motor system model with Simscape Electrical™. This network was used as a surrogate model for air conditioning motors and controllers to help streamline signal processing and remove unnecessary data. These tools resulted in highly accurate models and improvements in computation speed.
