Verification of Embedded AI (Deep Learning Results Implemented in MCUs)

"e-AI Solutions" provided by Renesas Electronics：By placing AI in the endpoint area, separate from cloud dependency, we aim to strengthen application systems and products that can make decisions and respond in real time.
Introducing an example of verification of “e-AI solution for motor failure detection” using the RX66T MCU with 32-bit CPU core “RXv3”.

Renesas Electronics' e-AI Solution is being developed to enhance and enable systems and products that can make decisions and respond in real time by deploying AI in the endpoint domain, separated from cloud dependencies.The “e-AI translator” provided by Renesas Electronics can convert the open-source deep learning library “TensorFlow” developed by google, the “Caffe” being developed at the University of California, Berkeley, trained Neural network information, etc., can be converted into code that can be implemented on the MCU.
With the cooperation of Renesas Electronics, we exhibited an e-AI verification demo at the Spring 2019 exhibition using our product RAMScope (a multi-measurement and calibration tool for verifying control algorithms) and an "e-AI solution demo kit for motor fault detection (MCU: RX66T)." The e-AI learned inference model implemented in the RX66T extracts characteristic points of characteristic data (current values) indicating motor operating conditions in real time and inputs them to the inference model to determine faults.The feature points for detecting faults are extracted from specific band samples of the frequency spectrum obtained by FFT processing of the AD converted data of the phase currents (torque currents).
RAMScope can verify inference models by measuring and logging in real time the internal variables of the feature points of the frequency spectrum input to the AI model and the internal variables of the judgment results output by the AI model.The trained e-AI model implemented in the Fault Detection e-AI Solution Demo can be verified together with the vector control algorithm.

For more information on Renesas e-AI.