Artificial Intelligence to help increase industrial safety

Fast and precise detection and localizing of the damage facilitates increasing the safety in industrial processes and can help eliminate putting human lives and health at risk, contamination of the environment, or the risk of fire. However, knowledge about the existing damage is indispensable to taking proper security action.

“Failures may be a result of damage, human errors, and, recently, cyber-attacks,” says Professor Jan Maciej Kościelny, PhD, DSc from the Warsaw University of Technology. “To prevent failures, we need to recognize them very early to make the right decisions. Resilient and precise diagnostics is simply a must,” he adds.

Commonly used alarm systems have many defects and can generate even a few hundred alarms in a few minutes in an emergency  (the so-called alarm flood). Within the implementation of a research grant from Research Centre POB Artificial Intelligence and Robotics, a team of researchers from the Faculty of Mechatronics at WUT has devised a new method of resilient localizing of damage designated for complex objects of diagnostics. The algorithm resilience was gained due to the application of reasoning based on a hybrid Bayesian-fuzzy approach. The precision of diagnostics was achieved by applying a three-value assessment of residues and using the knowledge (often incomplete) about the order of symptom occurrence. The devised methods take into account the possibility of multiple damages and apply the dynamic decomposition of the diagnosed object.

“Our method is based on reasoning based on the observed symptoms, which occur in states with defects,” explains Professor Kościelny. “It is about a system capable of making true and precise diagnoses, in spite of numerous diagnostic uncertainties,” he adds.

The results of the research conducted within the grant have been published in, among others:

• Engineering Applications of Artificial Intelligence – „Diagnosing with a hybrid fuzzy-Bayesian inference approach”;
• Energies – „Diagnostic Row Reasoning Method Based on Multiple‐Valued Evaluation of Residuals and Elementary Symptoms Sequence”;
• Energies  – „Diagnostic Column Reasoning Based on Multiple-Valued Evaluation of Residuals and Elementary Symptoms Sequence”.

-

Project “Development of resilent inference methods for diagnostic advisory systems” is funded within the research grant of Research Centre POB Artificial Intelligence and Robotics under the “Excellence Initiative – Research University” programme implemented at the Warsaw University of Technology.

Research team:
Professor Jan Maciej Kościelny, PhD, DSc; Michał Bartyś, PhD, DSc, Associate Professor; Michał Syfert, PhD, DSc