A BAYESIAN FRAMEWORK FOR STRUCTURAL HEALTH MANAGEMENT USING ACOUSTIC EMISSION MONITORING AND PERIODIC INSPECTIONS

Abstract

Many aerospace and civil infrastructures currently in service are at or beyond their design service-life limit. The ability to assess and predict their state of damage is critical in ensuring the structural integrity of such aging structures. The empirical models used for crack growth prediction suffer from various uncertainties; these models are often based on idealized theories and simplistic assumptions and may fail to capture the underlying physics of the complex failure mechanisms. The other source of uncertainty is the scarcity of relevant material-level test data required to estimate the parameters of empirical models.

To avoid in-service failure, the structures must be inspected routinely to ensure no damage of significant size is present in the structure. Currently, the structure has to be taken off line and partly disassembled to expose the critical areas for nondestructive inspection (NDI). This is an expensive and time-consuming process.

Structural health monitoring (SHM) is an emerging research area for online assessment of structural integrity using appropriate NDI technology. SHM could have a major contribution to the structural diagnosis and prognosis.

Empirical models, offline periodic inspections and online SHM systems can each provide an independent assessment of the structural integrity; in this research, a novel structural health management framework is proposed in which the Bayesian knowledge fusion technique is used to combine the information from all sources mentioned above in a systematic manner.

This work focuses on monitoring fatigue crack growth in metallic structures using acoustic emission (AE) technology. Fatigue crack growth tests with real-time acoustic emissions monitoring are conducted on CT specimens made of 7075 aluminum. Proper filtration of the resulting AE signals reveals a log-linear relationship between fracture parameters (da/dN and ΔK ) and select AE features; a flexible statistical model is developed to describe the relationship between these parameters.

Bayesian regression technique is used to estimate the model parameters using experimental data. The model is then used to calculate two important quantities that can be used for structural health management: (a) an AE-based instantaneous damage severity index, and (b) an AE-based estimate of the crack size distribution at a given point in time, assuming a known initial crack size distribution.

Finally, recursive Bayesian estimation is used for online integration of the structural health assessment information obtained from various sources mentioned above. The evidence used in Bayesian updating can be observed crack sizes and/or crack growth rate observations. The outcome of this approach is updated crack size distribution as well as updated model parameters. The model with updated parameters is then used for prognosis given an assumed future usage profile.