Numerical and experimental study on improving diagnosis in structural health monitoring

Structural Health Monitoring (SHM) is a procedure of assessing structural integrity during its service period to improve maintenance in terms of cost and reliability. In SHM for aircraft applications, crack identification using reflected ultrasonic Lamb waves from a crack is one of the most active research areas. In addition to detecting a crack, estimating its size is important for judging the severity of the damage, and this can be done by various techniques. Specifically, we focus on the relationship between the sensor signal amplitude and crack size through experiments and simulation for help in size estimation. The maximum received signal amplitude is found to vary linearly from simulation and this agrees with measurements with crack size up to 30 mm.;However, SHM using embedded sensors have limitation in terms of accuracy of detection. Our approach to overcome this is as follows. If measurements are frequently performed using the above mentioned techniques while the crack grows, then a better estimation of crack size may be possible by analyzing sensor signals for the same crack location at different sizes.;The main objective of this research is to improve the accuracy of current diagnosis by using the prediction from previous inspection results. Unlike manual inspection, SHM can take frequent measurements and trace crack growth. By taking advantage of this aspect, higher accuracy about current crack size can be achieved. First, using the previous SHM measurements and the crack propagation model, we predict the statistical distribution of crack sizes at the next SHM inspection cycle. Then, this predicted distribution is combined with the SHM measurement at the next cycle by using the Bayesian approach for more precise estimate. The propagated distribution from the previous inspection is used as a prior and the variability at the current inspection is used to build the likelihood function. The uncertainty in measurements is modeled by a lognormal distribution. Results show substantial improvements in accuracy.