Cargo drop detection with tinyML and edge computing

This study focuses on developing a system to detect and monitor potential damage to cargo during transportation. The system involves the use of sensors, microcontrollers, machine learning models and communication modules to collect, analyse and transmit data. The thesis summarizes the use of various electronic modules and communication protocols such as DS1307 RTC module, I2C communication protocol, micro SD card module, SPI communication protocol, ESP8266 WiFi module and UART communication protocol. In addition, information about development tools such as STM32F4 Discovery development board, STM32CubeIDE, STM32CubeMX, X-CUBE-AI package is given. In the study, a device using accelerometers was developed to collect data during the movement of the cargo and to detect fall situations. A Kalman filter was used to prevent deviations in the collected data and the model was trained with artificial neural networks. The model trained with TinyML was loaded onto the embedded system. The measured values are classified locally thanks to the model trained on the device and the fall is detected. The trained model correctly predicted the fall with 97% success. By making this determination on the device, edge information processing is provided and cloud load is reduced and real-time detection is provided.