Test-before-invest Experiments

The experiment will make the lab software architecture more decentralized and resilient thanks to the installation of new hardware directly on the machine. This will substitute the existing one, limited to the gathering, filtering and formatting of raw data, to make the system more resilient towards unexpected events and more sustainable in environmental terms, as the data processing will be moved on the more efficient edge layer. 

The main goal of the experiment is to develop an edge solution, from hardware and software point of view, with algorithm for predictive maintenance. The edge solution is a key point to divide the overall computational algorithms in two main level: in edge computer solution, and a remote solution on or on premises. The key part of experiment is to design a build on a new edge computing, new features from algorithms. 

The experiment “Self-evolving monitoring systems for assembly production lines” consists of an industrial set-up representing a part of the DC drive assembly process. The experiment will set up a testbed for self-evolving solutions for monitoring production asset performances, targeting repetitive processes, such as assembly production lines. Demonstrated diagnostic systems will self-learn by analyzing the patterns from process time-series measurement, while using machine-learning methods.

In the first iteration, the experiment focused on generating synthetic datasets to improve recognition of 3D printed parts in high tech (additive) manufacturing using large language models, which has been increased to 98%.

During the second iteration, the focus shifted to developing a plug & play web application to automatically calculate carbon footprint emission in high tech manufacturing, extended with AI-generated carbon footprint reduction advice in the organization as well as the supply chain: CarbonInsight.

The experiment SmartBox: Scalable IIoT Monitoring for SMEs demonstrates a plug-and-play IIoT system—SmartBox—designed for easy deployment in SME environments. It enables real-time monitoring of machines through multimodal sensors (vibration, sound, temperature, etc.) without interfering with operations. Using edge computing and unsupervised learning, SmartBox autonomously detects machine states and performance anomalies. The solution is validated in real SME settings, offering an affordable and scalable path to digital transformation.

The experiment will deal with specific topics of work based on Digitalisation/AI used to reduce costs/increasing revenue according to a circular economy approach, such as quality monitoring in the whole production process, predictive maintenance, product traceability, management of industrial by-products. 

In the first iteration, AI-based processing of data from the Smart Maintenance setup was moved from the central computer/cloud to the individual drivetrains, under the form of a portable measurement box. This measurement platform was then used to demonstrate AI-on-the-edge for vibrational condition monitoring and predictive maintenance to SMEs.

In the 2nd iteration, we demonstrated the possibilities of acoustic condition monitoring, using the same AI-on-the-edge box, but relying on microphones rather than vibration sensors to measure anomalies and predict failures in components with rotating elements.

The purpose of this experiment is to take a typical production scenario, such as an assembly process, and use the already available data to create a stream of information which the IWOK system can use to suggest a solution to a problem to the Operator, who can in turn use their ‘human’ I/O to verify the solution and ‘teach’ the machine how to fix the problem itself. This then creates a living Knowledge base between the Operator and the Machine that can be used to avoid problems before they occur. 

BEhAI is a DF driven experiment in Lombardy focused on implementing AI inside manufacturing. What the experiment wants to demonstrate is how the application of the AI linked to sensors’ data can enhance human-machine interactions by identifying the state of the operator and adjust its patterns and functions to it, accordingly.

This will be done to address two main challenges:

- Understand and adapt to human states;

- Understand and adapt to context.

QUAD-AI@E experiment will develop a product defect detection system able to acquire product images, to manipulate images by an embedded AI tool, to extract info about the nature and severity of the product defect and to send alert to the personnel. The experiment builds on integrating AI at the edge technology into a product defect detection system prototype suitable for quality control automation in textile industry. 

The experiment aims to demonstrate an AI-driven real-time monitoring of a robotic assembly line equipped with a delta robot assembling a small 4-wheel vehicle. The core problem of the monitoring is the detection of non-OK state from force and torque time series acquired by sensors positioned on the robot end effector. The main goal is to detect a problem quickly after it appears so that a corrective action will come early, which can significantly improve the assembly process. The problem is formulated as a real-time anomaly detection from multidimensional time-series, which can be efficiently solved by machine learning techniques.  

The experiment will update the AAU IoT suitcase to include more advanced data analytics and machine learning capabilities. One machine type from the host SME will be selected and used as basis for experimenting and validating data-driven technologies.  

The SUNSYNC experiment aims to optimise recycling processes in industrial environments. The first iteration of the experiment focused on automating and streamlining recycling methods, utilising solar power as the primary energy source, extending battery life, and making data-driven decisions throughout the project.

 The implemented AI-driven system integrates data from light detectors, weather forecasts, and accumulator monitoring systems. The process includes data integration, AI-based data processing, and real-time decision-making to optimise energy use and align recycling operations with available clean energy.

In the second iteration of the experiment, we extended the usable machines by generalising the interfaces. We also replaced the data collector and database with a widely used solution. We created a new AI solution that predicts the expected solar energy. Due to the generalisation, the main focus is now on production planning.

Gradiant is developing an experiment in this DF consisting of providing real-time monitoring through AI at the Edge in an additive manufacturing (AM) cell for early defect detection. The idea is to showcase how these cutting-edge technologies can help us improve the quality control of a complex manufacturing process, moving from “afterwards” quality control, discarding the piece if defects are found, to an early defect detection, allowing operators to fix the process before defects appear.