This project is based on digikey's TinyML course using STM32 CUBE AI.
(https://www.youtube.com/watch?v=crJcDqIUbP4)
development target is stm32F769i disco board and built based on STM32CubeMX and STM32CubeIDE.
--Version--
STM32CUBE F7 v1.17.1
STM32CUBE IDE 1.15.1
X-CUBE-AI 9.0.0
STM32_AI_Sinewave
├─ .ai
├─ .cproject
├─ .gitignore
├─ .mxproject
├─ .project
├─ .settings
├─ Core --------------------------- Main application
│ ├─ Inc
│ ├─ Src
│ └─ Startup
├─ Drivers ------------------------ Drivers
│ ├─ CMSIS
│ └─ STM32F7xx_HAL_Driver
├─ Middlewares
│ └─ ST
│ └─ AI ----------------------- X-CUBE-AI Middlewares
├─ README.md
├─ STM32F769NIHX_FLASH.ld
├─ STM32F769NIHX_RAM.ld
├─ Sine_wave.ioc ----------------- CUBE MX .ioc file
├─ Sine_wave.launch
├─ X-CUBE-AI
│ ├─ App ------------------------ AI data file created with CUBE MX
│ └─ LICENSE.txt
├─ Sin_Wave.ipynb ---------------- Jupyter notebook AI model file
└─ sine_model.tflite ------------- tflite file
(It is my first deep learning model. 😅)
1. Training Model
2. Model Layer
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
dense_16 (Dense) (None, 128) 256
dense_17 (Dense) (None, 128) 16512
dense_18 (Dense) (None, 1) 129
=================================================================
Total params: 16,897
Trainable params: 16,897
Non-trainable params: 0
_________________________________________________________________
3. Loss Graph
- Execute .project file for add project to CubeIDE
- Build project. (Target: stm32F769i disco board)
- Connect target board and run it.
- When program starts, input value (x_val) is increased by 0.1 from 0 and predicted output value Sin(y_val) is output to VCP.
- AI modeling input ranges from 0 to 2pi, so any value higher than that will result in an error.