Poclain Hydraulics Develops Soft Sensors to Measure Motor Temperature in Real Time Using Deep Learning and Kalman Filters
“We identified two neural networks that were already implemented in MATLAB and these neural networks helped us embed the codes into hardware to do real-time predictions of temperature.”
Key Outcomes
- MATLAB accelerated testing with pretrained neural networks
- Simulink enabled testing of a simplified, extended Kalman filter
- MATLAB enabled code generation in several languages, including C and C++
Poclain Hydraulics is a world leader in developing hydrostatic transmissions and motors that power machinery in industries such as construction, agriculture, and mining. These motors generate power by converting hydraulic energy into mechanical energy, which can raise the temperature of the motor and potentially lead to failures.
Poclain Hydraulics used MATLAB® and Simulink® to create a soft sensor that uses either a deep learning or Kalman filter approach to monitor motor temperature in real time. To be successful, a deep learning or extended Kalman filter model must take into consideration the motor’s load history and environmental components, such as external temperature. The main drawback of a neural network approach as compared to the Kalman filter was the lack of explainability, which was not seen as an issue in this case.
The team implemented a complete AI industrialization process, starting from data extraction and randomization, followed by training, testing, and validation of neural networks, and finally deployment to their hardware. MATLAB and Simulink facilitated the industrialization process by enabling C or C++ code generation, testing before deployment, and managing large data sets. The team also took advantage of pretrained neural networks available in MATLAB to speed up the process.
As part of the industrialization process, Poclain Hydraulics used MATLAB and Simulink to generate data by building and simulating a physics-based model of the motor. They were able to design experiments on data generation; test various motor parameters, such as pressure, speed, time, and risk factors; and manage the experiment results.
