Identification of structure-specific damage growth properties and its impact on improved prognosis.

摘要

Structural health monitoring (SHM) employs sensor data to monitor fatigue-induced damage growth in service. Damage growth information can be used to improve the characterization of the material properties that govern damage propagation for the structure being monitored, turning aircrafts into flying fatigue laboratories. Initially, these properties are often widely distributed between nominally identical structures because of differences in manufacturing processes and aging effects. SHM data, in particular, measured crack growth are used to narrow the distribution of damage growth parameters using the Bayesian inference technique. The improved accuracy in damage growth parameters allows a more accurate prediction of the remaining useful life (RUL) of the monitored structural component. It can also help in predicting damage growth of other similar components.;In the absence of actual SHM data, we had to simulate measured data by applying error models to damage sizes obtained using Paris' law as a damage growth model. We consider that there are two kinds of errors, random noises resulting from the measurement environment and a deterministic bias resulting from the sensor's calibration or modeling error.;The Bayesian inference method is used for progressively reducing the uncertainty in structure-specific damage growth parameters in spite of noise and bias in sensor measurements. However, the Bayesian inference method is computationally intensive due to uncertainty propagation in likelihood calculation, which results in a difficulty in handling multi-parameter identification, and may not be feasible to use with an extremely large number of measurement data. On the other hand, least-square fitting of damage growth parameters is efficient, but it does not provide good statistical information on the uncertainty in their estimates and in RUL estimates. It is in particular efficient to identify deterministic variables such as bias. In the proposed research, we combined the two approaches by using the least-square approach to filter the data and then performing Bayesian inference. The proposed approach is applied to crack growth in fuselage panels due to cycles of pressurization. It is shown that the proposed method rapidly converges to accurate damage parameters with much smaller uncertainties. Fairly accurate damage growth parameters were also obtained with measurement errors of 5mm. Using the identified damage parameters, it is shown that the 95% conservative RUL converges to the true RUL from the conservative side.