A Tiny Machine Learning implementation with low-power devices in Structural Health Monitoring Applications

The thesis work focuses on the practical implementation of Machine Learning models on Embedded Systems and the selected target for the tests is the Arduino Nano 33 BLE Sense board. The workflow starts with study of TinyML concepts, encompassing model conversion to TFLite and finally to a hex model ready for deployment on the microcontroller board. Examples from the literature will be discussed and experimentally implemented, such as, “Hello World”, “Magic Wand” and “Micro Speech-Recognition” tasks as per the book “TinyML - Machine Learning with TensorFlow Lite on Arduino and Ultra-Low-Power Microcontrollers” by Pete Warden, Daniel Situnayake. The final aim of this manuscript, which constitutes the core part of the work, is to implement novel TinyML models in SHM applications: specifically, two types of Neural Networks (NNs) namely the Associative Neural Network (ANN) and the One Class Classifier Neural Network (OCCNN) on Arduino Nano 33 BLE Sense board. These NNs are meant for damage detection and binary classification problems, whose output consists of a structural bulletin specifying whether the monitored is healthy or damaged.