Aramix (3rdPlace srl) is leading the CAusal Modeling and Bayesian Optimization for predictive DIAgnostics (CAMBODIA) project, which stems from a collaboration with TekRevolution and the University of Messina to develop innovative solutions based on Artificial Intelligence to support industrial safety, risk management, and the sustainability of production processes.

The solution: AI, data science and multidisciplinary know-how

This project was established to develop advanced and distinctive methodologies in two key areas of Industry 5.0:

• Predictive Diagnostics (FP2)
• Computational Fluid Dynamics (CFD) (FP3)

Our strategy is based on the use of an innovative causal modeling framework to identify cause-and-effect relationships, and on the application of optimal design of experiments techniques to minimize the number of required laboratory and cfd tests to achieve the desired level of knowledge.

As a deep-tech company, Aramix will apply these solutions in support to the development of advanced engineering systems, enhancing their safety, efficiency, and reliability.

Partners and expertise

3rdPlace: A leader in the application of Artificial Intelligence to industrial and managerial processes, with consolidated experience in data governance, machine learning, and predictive analytics.

TekRevolution: A technology company specializing in high-tech digital solutions. With extensive experience in Computational Fluid Dynamics, they bring key expertise in developing and optimizing components using innovative materials and manufacturing processes for advanced industrial applications.

University of Messina: A center of research excellence with an expert team in industrial chemical safety, risk management, catalysis, and innovative processes, which enriches the project with scientific knowledge and advanced analytical capabilities.

Achievements and Completed Activities

To date, our activities have validated the project’s methodological pillars through concrete applications, demonstrating the feasibility and effectiveness of the proposed approach.

Causal Modeling for Root Cause Analysis

Causal modeling has been successfully employed to analyze operational data from an FPSO (Floating Production Storage and Offloading) system. The analysis allowed us to identify and quantify the production and environmental factors with the greatest causal impact on pipe corrosion processes, facing the issues of large measurement uncertainty, limited data availability and large correlation between environmental factors. This use case validates the effectiveness of our approach in the field of diagnostics, enabling a shift from predictive analysis (estimating when a failure will occur) to prescriptive analysis (understanding why it occurs and how to address the root causes).

Bayesian Design of Experiments (BOED) for Experimental and Computational Efficiency 

We have successfully applied Bayesian Design of Experiments (BOED) to drastically reduce the cost of information acquisition in both physical experimentation and computational simulation contexts. We used two distinct technical approaches depending on the domain:

• In the Reliability domain: To optimize an accelerated reliability test, we used a direct failure model using the well known peck-arrhenius model. The BOED framework selected the optimal experimental settings by maximizing the Expected Information Gain (EIG), which was calculated directly on the statistical model describing the failure process. This approach is ideal when an explicit and reliable model of the phenomenon is available.
• In the CFD domain: To map the impact of varying operational conditions, we built a surrogate model based on Deep Gaussian Processes to model the epistemic uncertainty in simplifying a complex CFD simulation. The DoE was then applied to this computationally less expensive metamodel to choose the subsequent simulation points yielding the largest expected increase in information, thereby minimizing the total number of required CFD runs, with large savings in human and computational time.

Want to learn more?

Write to info@aramix.ai for more information.

Follow the updates on the Aramix LinkedIn page.