Edge computing enables innovative control and diagnostics of advanced mechatronic systems

Inspired by the vision and the activities carried out proposed by the ACTEMA research group at the University of Bologna, this doctoral thesis introduces innovative methodologies as crucial elements to redefine the paradigms for designing and developing key components of modern industrial manufacturing systems, with servomechanisms at their core, toward task-specific customization, sustainability, and adaptability. In this respect, high-performance embedded platforms, advanced control algorithms and innovative diagnostic methods are fundamental. Therefore, the thesis is structured into three parts: The first part establishes the technological foundation required to achieve the computational power necessary for high-demand real-time tasks, supporting the advanced diagnostic and control methodologies envisioned in this framework. This is accomplished on multicore heterogeneous platforms combining microcontrollers and microprocessors and oriented to edge computing. A novel software architecture is proposed to fully exploit these platforms in real-time applications. The potential of this approach is evaluated using a demanding control algorithm as a benchmark, with the STM32MP1 selected as the reference platform. The second part focuses on advanced control strategies. It begins with the introduction of a novel observer design for Permanent Magnet Synchronous Machines, combining recursive least squares estimation with adaptive sampling to ensure robust performance under challenging conditions. Additionally, this section presents the modeling of an advanced servomechanism prototype and demonstrates its control using an innovative repetitive controller. Moreover, energy-efficient trajectory optimization techniques for the mechanism are explored. The final part addresses advanced diagnostic methods. A mixed model-based and data-driven approach is developed, leveraging the properties of a Lagrangian kinematic system with 1-DoF, under periodic motion, to generate residuals for identifying machine conditions. Furthermore, an innovative Bayesian-like classifier is introduced and validated using the aforementioned residuals.