Publications

The Industrial Internet of Things (IIoT) has created unprecedented opportunities for data-driven insights in areas such as predictive maintenance and process optimization. However, industrial devices sometimes produce 'broken data', stemming from sensor malfunctions, communication errors, or other system faults. Such corrupted information severely undermines the reliability of downstream analytical models. Realtime data validation is needed, but cloud-based solutions introduce latency, unpredictable costs, and privacy concerns. To overcome these challenges, we introduce CoreTrust, a novel industrial data quality architecture for edge node deployment. CoreTrust contributes (i) a closed-loop corrective architecture that detects anomalies, communicates corrections back to PLCs, and sanitizes persisted data in realtime; (ii) a comparative evaluation of rule-based versus AI-based anomaly detection under explicit industrial production requirements, demonstrating that rule-based methods achieve competitive detection accuracy with significantly lower resource consumption on constrained edge hardware while guaranteeing the approach explainability; and (iii) an automated GitOps-driven deployment pipeline enabling continuous delivery to heterogeneous edge nodes. We evaluate CoreTrust on an industrial case study, demonstrating zero false negatives across millions of data points, a negligible false positive rate, and a resource footprint suitable for constrained edge hardware.

JLiSA is the extension to the Java programming language of LiSA, an analysis engine that works on a generic and extensible control flow graph representation of the program to analyze. LiSA implements several standard abstract domains and analyses aimed at approximating numerical values, strings, and heap structures. At the end of the analysis, it produces an abstract state for each program point. Then, checkers produce warnings indicating whether a property of interest is respected. JLiSA provides a front-end to translate Java programs into the internal LiSA control flow graph representation, the semantics of various parts of the Java standard library, and checkers to verify assertions and detect whether exceptions might be thrown and not caught. This paper presents our first participation in SV-COMP in the Java category, where we achieved 3rd place.

The Internet of Things (IoT) revolution is reshaping how physical devices embedded with software connect to the Internet, facilitating seamless data exchange and driving automation. Industrial IoT (IIoT) extends these capabilities to industrial devices and Cyber-Physical Systems (CPS), driving Intelligent Manufacturing. This integration supports advanced applications like remote monitoring, predictive maintenance, machine learning (ML), and artificial intelligence (AI) optimization, enhancing production efficiency, adaptability, and decision-making. However, managing the vast amounts of data generated requires scalable, automated software architectures. Many small and medium-sized enterprises (SME) face challenges in building such systems due to limited resources and expertise, often starting with manual data collection and basic automation. This paper presents a solution: a fully automated, configurable, and scalable software architecture for Intelligent Manufacturing. Our system has been operational for almost one and a half years on 21 plants, processing about 17K tasks, amounting to more than three months of computations. The experimental results show that automation and elasticity have been needed by such systems since the beginning, while scalability is not required during an initial experimentation phase.