Industry 4.0 Efficiency: Predictive Maintenance with TinyML and an Incremental Model

Predictive maintenance (PdM) represents a strategic approach to the management of industrial assets that has gained significance in the era of the Fourth Industrial Revolution. In a context where operational efficiency and downtime minimization are becoming crucial imperatives, predictive maintenance stands out as an innovative solution. Instead of relying on scheduled maintenance methods based on fixed calendars, predictive maintenance leverages technological advancements, including sensors, artificial intelligence, and machine learning algorithms, to anticipate potential failures and breakdowns of industrial equipments. However, in actual circumstances, resistance to adopting predictive maintenance persists due to a myriad of variables, as well as the high customization it requires. In this paper, we propose a new framework for PdM, which is based on TinyML technology and incremental machine learning models capable of recognizing new problems reported by those operating the machinery. Our framework, deployed on the ESP32 card and tested on a rotating machine shows a continuous improvement in system performance, demonstrating its effectiveness and adaptability.