Machine Learning-based Approaches for Advanced Monitoring of Smart Glasses

With today’s growing demand on productivity, product quality and effectiveness, the importance of Machine Learning-based functionalities and services has dramatically increased. Such paradigm shift can be mainly associated with the increasing availability of Internet of Things (IoT) sensors and devices, the increase of data collected in the IoT scenario and the increasing popularity and availability of machine learning approaches. One of the most appealing applications of ML-based solutions is for sure Predictive Maintenance, which aims at improving maintenance management by exploiting the estimation of the health status of a piece of equipment. One of the main formalizations of the PdM problem is the prediction of the Remaining Useful Life (RUL), that is defined as the time/process iterations remaining for a device component to perform its task before it loses functionality. This work investigates a possible application of predictive maintenance techniques for the monitoring of the battery of Smart Glasses. The work starts with the description of the considered devices, the modalities of data collection and the Exploratory Data Analysis for better understanding the task. The first experimental part consists in the application of an unsupervised anomaly detection technique, useful to initially deal with the partial and unlabeled data. The last part of the work contains the results of the application of both classical machine learning and deep learning approaches for the estimation of the RUL of the devices battery. A section for the interpretation of the machine-learning models is included for both the anomaly detection and RUL estimation approaches.