Abstract
This article advances innovative approaches to the design and implementation of an embedded intelligent system for predictive maintenance (PdM) in industrial applications. It is based on the integration of advanced artificial intelligence (AI) techniques into micro-edge Industrial Internet of Things (IIoT) devices running on Armr Cortexr microcontrollers (MCUs) and addresses the impact of a) adapting to the constraints of MCUs, b) analysing sensor patterns in the time and frequency domain and c) optimising the AI model architecture and hyperparameter tuning, stressing that hardware–software co-exploration is the key ingredient to converting micro-edge IIoT devices into intelligent PdM systems. Moreover, this article highlights the importance of end-to-end AI development solutions by employing existing frameworks and inference engines that permit the integration of complex AI mechanisms within MCUs, such as NanoEdgeTM AI Studio, Edge Impulse and STM32 Cube.AI. Both quantitative and qualitative insights are presented in complementary workflows with different design and learning components, as well as in the backend flow for deployment onto IIoT devices with a common inference platform based on Armr Cortexr-M-based MCUs. The use case is an n-class classification based on the vibration of generic motor rotating equipment. The results have been used to lay down the foundation of the PdM strategy, which will be included in future work insights derived from anomaly detection, regression and forecasting applications.