7.7.21. MOSFET Junction Temperature Prediction

Predict MOSFET junction temperature for thermal management in power conversion systems.

7.7.21.1. Overview

High voltage switches are used widely in power conversion and motor drive applications. Accurate temperature prediction of these switches is critical for protecting the switching devices while delivering the best system efficiency. However, direct measurement of the switch temperature is often difficult and expensive.

This example provides a generic approach for predicting switch temperature based on the power loss of a switch and indirect temperature measurement from a temperature sensor in the vicinity of the switch.

The approach uses a linear + AI modeling method:

  • Linear model: A linear autoregressive-moving-average (ARMA) model describes the linear thermal behavior from power loss to the difference between case temperature and NTC sensor temperature.

  • AI model: An MLP model captures the residual nonlinear error between the linear model prediction and true measurement.

This combination provides more accurate models, more constrained AI behavior, and a more robust overall prediction.

Application: Thermal management in power converters, motor drives, EV power electronics

Task Type: Time Series Regression

Data Type: Multivariate (43 input variables)

7.7.21.2. Configuration

common:
  task_type: 'generic_timeseries_regression'
  target_device: 'F29H85'

dataset:
  dataset_name: 'mosfet_temp_prediction'
  input_data_path: 'https://software-dl.ti.com/C2000/esd/mcu_ai/01_03_00/datasets/mosfet_temp_prediction.zip'

data_processing_feature_extraction:
  data_proc_transforms: ['SimpleWindow']
  variables: [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43]
  stride_size: 1
  frame_size: 1

training:
  model_name: 'REGR_3k'
  batch_size: 128
  training_epochs: 100
  learning_rate: 0.01
  quantization: 0
  output_int: false

testing: {}
compilation: {}

7.7.21.3. Running the Example

cd tinyml-modelzoo
./run_tinyml_modelzoo.sh examples/mosfet_temp_prediction/config.yaml

7.7.21.4. Dataset Details

Input Variables (43 total):

  • Past NTC temperature data (20 time points)

  • Past power loss data (20 time points)

  • Meta information (ambient temperature, coolant temperature, coolant flow rate)

The dataset contains temperature data for various ambient temperatures, coolant temperatures, and coolant flow rate configurations.

Output:

  • T_case - T_linear (difference between expected case temperature and linear model output)

Dataset Download:

The dataset is automatically downloaded from:

dataset:
  input_data_path: 'https://software-dl.ti.com/C2000/esd/mcu_ai/01_03_00/datasets/mosfet_temp_prediction.zip'

7.7.21.5. On-Device Deployment

This example targets the F29H85x MCU. After running ModelMaker, copy the compiled model files to the CCS example project:

  • mod.a - The compiled model library

  • tvmgen_default.h - Header file for model inference APIs

The CCS example generic_timeseries_regression for F29H85x provides the application framework.

7.7.21.6. See Also