FBK conducted an in-depth analysis of the algorithms and decision intelligence solutions developed by the company to ensure their consistency, reliability and performance in real-world settings.
As part of the AI-MATTERS project, FBK’s Digital Industry Center (DI Center) collaborated with Rulex, an Italian company active in the development of end-to-end solutions for data management, decision intelligence and decision-making automation in various fields, including manufacturing, by providing services of “Application of formal methods for reliable industrial systems” and a “Feasibility study and evaluation of AI technology.”
The objective of the work was to support Rulex in validating and benchmarking its artificial intelligence algorithms, with a focus on transparency, robustness and traceability. Rulex’s software uses proprietary algorithms to extract decision rules from data, automate business processes, and optimize industrial activity planning. As with any AI-based system, there was a need to verify that the automated decisions were consistent, reliable, and understandable; hence the request to FBK for services that were realized by combining two different actions.
In the first strand of activities, formal verification methods were applied to validate the rules generated by Rulex’s software, defining the conditions under which they were most robust and generalizable. This made it possible to identify any anomalies or borderline cases that could have compromised the reliability of the system.
In parallel, in the second strand of activities, a quantitative benchmarking analysis was conducted on the optimization algorithms developed by Rulex. The objective was to objectively evaluate the effectiveness and efficiency of the
proprietary solutions, comparing them with a selection of open-source and academic reference solvers for discrete and continuous optimization problems.
The comparison looked at measurable performance indicators, such as the quality of the solutions obtained with respect to the known optimum, speed of convergence, and computational scalability with respect to problem size.
This approach made it possible to build a detailed comparative profile of the capabilities of the Rulex algorithms, highlighting strengths and areas for possible improvement in relation to real-world application contexts, such as maintenance planning or industrial scheduling.