Deep RACE is a real-time reliability modeling and assessment of power semiconductor devices embedded into a wide range of smart power electronics systems. Deep RACE departures from classical learning and statistical modeling to deep learning based data analytics, combined with full system integration for scalable real-time reliability modeling and assessment. In this regard, it leverages the Long Short-Term Memory (LSTM) networks as a branch of Recurrent Neural Networks (RNN) to aggregate reliability across many power converters with similar underlying physic. Also, It offers real-time online assessment by selectively combining the aggregated training model with device-specific behaviors in the field.

First make sure you have already installed pip3, Tkinter, and git tools:

sudo apt install git python3-pip python3-tk

You should also install follwoing python packages:

sudo -H pip3 install tensorflow scipy matplotlib seaborn

You only need to clone the Deep RACE repository:

git clone https://github.com/TeCSAR-UNCC/Deep_RACE

Change the path to the Deep_RACE directory and run the train.py:

cd Deep_RACE
python3 ./train.py

All the training models will be saved automatically in ./inference_models/ folder. You can load them by running inference.py file.

The ./train.py will generate and save the predition out put in a text file. The file name is based on the selected MOSFET device number. As an instance, a text file with the name of ./prediction_output/res_dev2.txt will be generated for dev#2.

You can test different devices from RoIFor5Devs.mat by altering this line in the ./train.py.

Please cite the Deep RACE if it helps your research work.

@ARTICLE{8629973,
author={M. {Baharani} and M. {Biglarbegian} and B. {Parkhideh} and H. {Tabkhi}},
journal={IEEE Internet of Things Journal},
title={Real-Time Deep Learning at the Edge for Scalable Reliability Modeling of Si-MOSFET Power Electronics Converters},
year={2019},
volume={6},
number={5},
pages={7375-7385},
keywords={electronic engineering computing;Internet of Things;Kalman filters;MOSFET;neural nets;power aware computing;power convertors;power electronics;real-time systems;reliability;scalable decentralized devices-specific reliability monitoring;MOSFET convertors;Internet-of-Things devices;real-time deep learning processing capabilities;Deep RACE solution;MOSFET data;1.87-W computing power;edge IoT device;scalable reliability modeling;advanced high-frequency power converters;active reliability assessment;power electronic devices;real-time reliability modeling;high-frequency MOSFET power electronic converters;edge node;real-time reliability awareness;deep learning algorithmic solution;collective reliability training;collective MOSFET converters;device resistance changes;edge-to-cloud solution;Si-MOSFET power electronics converters;deep learning reliability awareness;Kalman filter;particle filter;Reliability;Real-time systems;Deep learning;MOSFET;Power electronics;Predictive models;Degradation;Deep learning;high-frequency power converter;long short-term memory (LSTM);MOSFET;reliability modeling},
doi={10.1109/JIOT.2019.2896174},
ISSN={2372-2541},
month={Oct},}

• Reza Baharani: Python code - My personal webpage

Copyright (c) 2018, University of North Carolina at Charlotte All rights reserved. - see the LICENSE file for details.

The five Si-MOSFET ΔR_ds(on) are extracted from NASA MOSFET Thermal Overstress Aging Data Set which is available here. Please cite their paper if you are going to use their data samples. Here is its BibTeX:

@article{celaya2011prognostics,
title={{Prognostics of power {MOSFET}s under thermal stress accelerated aging using data-driven and model-based methodologies}},
author={Celaya, Jose and Saxena, Abhinav and Saha, Sankalita and Goebel, Kai F},
year={2011}
}