Characteristic mechanical vibration recognition using neural network

• Published in: 2021 5th International Conference on Electrical, Electronics, Communication, Computer Technologies and Optimization Techniques (ICEECCOT) pp. 637 - 642
• Main Authors: Dudak, Juraj, Brida, Peter, Gaspar, Gabriel, Sedivy, Stefan, Bednarcikova, Katarina
• Format: Conference Proceeding
• Language: English
• Published: IEEE 10-12-2021
• Subjects: Accelerometers; embedded system; firmware; Light emitting diodes; Microcontrollers; Neural networks; SPI; STM32; Time measurement; Vibration measurement; Vibrations
• DOI: 10.1109/ICEECCOT52851.2021.9708021

This paper deals with the development of an independent hardware module applicable to the recognition and categorization of different types of vibrations. The introduction of the paper is devoted to the description of the technologies used and the theoretical background needed in solving the task. The application part describes the development of the firmware for the microcontroller that provides communication between the devices and processing of the measured data. We then describe the creation and training of a convolution neural network and its implementation in the firmware of the microcontroller. Finally, we evaluate the results of manual testing of the implemented neural network and measure the time required to receive and process the measured data. The result of this work is a fully independent device capable of recognizing and categorizing five categories of vibration based on a one-second sample created from the data obtained from the accelerometer.