Objectives
The project aims to develop and validate a low-power multimodal Edge AI platform capable of processing data from heterogeneous sensors locally, improving accuracy, efficiency and sustainability compared with cloud-based solutions.
A secondary objective is to integrate self-supervised learning techniques so that the system can adapt autonomously to changing operating conditions and learn directly from unlabelled data.
Initial Challenge
Traditional industrial monitoring systems often rely on single-purpose sensors and supervised AI models that require large labelled datasets and continuous cloud connectivity. As a result, these systems can be expensive, slow and poorly suited to environments in which operating conditions vary over time or network availability is limited.
Solution
The proposed solution is a new generation of industrial monitoring systems based on multimodal artificial intelligence running directly on low-power sensor nodes. The system integrates self-supervised learning techniques that allow devices to learn autonomously from unlabelled data and adapt to changing operating conditions over time, without requiring centralised retraining. By processing data directly at the edge, the platform reduces its dependence on centralised computing infrastructure while enabling faster and more resilient monitoring.
Benefits
The benefits of the Edge AI approach range from improved operational efficiency to reuse across multiple sectors:
- scalable and sustainable monitoring capable of operating autonomously even in environments with limited computing and network resources;
- lower deployment costs and energy consumption compared with cloud-based solutions;
- autonomous adaptation to new operating scenarios through continuous self-supervised learning;
- a domain-agnostic platform that can be reused in agritech, food processing, smart manufacturing and energy;
- reduced need for manual interventions and unscheduled maintenance in industrial applications.
IFAB’s Role
IFAB acted as both funder and technical partner, supporting the project through its research and innovation investment programmes while also contributing directly to the development of the platform. Its technical work focused on the design and development of a multimodal Edge AI platform capable of processing data from heterogeneous sensors locally, directly on low-power sensor nodes. This approach reduces reliance on centralised computing infrastructure.
A distinctive feature of the solution is the implementation of on-device self-supervised learning mechanisms, enabling the system to learn from unlabelled data and adapt over time without the need for centralised retraining. The models were optimised for execution on resource-constrained devices in accordance with TinyML principles and integrated with IoT sensors to provide energy-efficient anomaly detection.
The approach was designed to be domain-agnostic and was validated in a specific use case: monitoring biomass levels in anaerobic digesters used for biomethane production. The Adapt-AI project enabled Fondazione IFAB to consolidate advanced expertise at the intersection of Edge AI, self-supervised learning and TinyML applied to industrial contexts. In this field, the ability to bring intelligence directly to the sensor, without relying on network connectivity, represents an increasingly significant operational and energy-efficiency advantage.
For further information, please contact: projects@ifabfoundation.org
