Data-driven prognosis: a multi-physics approach verified via balloon burst experiment.

Area of Science:

• Multi-physics modeling
• Data-driven prognosis
• Computational mechanics

Background:

• Traditional predictive methods require extensive offline measurements and parameter training.
• Implementing predictive models for systems with difficult or impossible a priori testing is challenging.

Purpose of the Study:

• To develop a novel multi-physics formulation for data-driven prognosis (DDP).
• To enable accurate failure prediction using solely in situ measurements.
• To overcome limitations of traditional predictive strategies in complex systems.

Main Methods:

• A deterministic mechanics framework is employed for the DDP algorithm.
• Stochasticity arises naturally from assumptions on conservation potential order and dimensions.
• The method relies exclusively on in situ measurements, obviating offline testing.

Main Results:

• The DDP scheme successfully predicted the onset of instabilities in a balloon burst experiment.
• Prognosis was achieved using only online visual observations.
• The DDP method demonstrated high accuracy with no false positives.

Conclusions:

• The developed DDP formulation provides a robust approach for real-time system prognosis.
• DDP is applicable to diverse datasets and can be implemented where offline testing is infeasible.
• Prediction time horizons are adjustable by modifying memory time windows.