Probabilistic Damage Size Estimation for Structural Health Management

Airframe structures encounter severe environments, fatigue cyclic loadings, and aging. Micro-defects in structures will nucleate macro-cracks and grow into catastrophic damages after a period of service. Structural health monitoring (SHM) and damage prognosis (DP) have attracted much interest for increasing the reliability and reducing the operative risks of aircraft structures. The SHM techniques are booming for years, while models from fractural mechanics have been applied for damage prognosis in many studies. Damage estimation from online SHM imaging results is mainly studied in this work and tries to fill the gap between the online damage detection and damage prognosis.;A damage image segmentation technique based on Bayesian updating framework is proposed for diagnosis and prognosis in the context of structural health monitoring (SHM). This framework takes into account the prior information of the conditions, e.g. spatial constraints and image smoothness. Markov random field is employed to model the prior due to the spatial constraint in the neighboring grids in SHM images. Results show that the Bayesian segmentation method has the potential to segment damage areas reasonably in SHM images and achieve more accurate damage sizes automatically than the conventional K-means clustering method through eliminating random noises and inhibiting the fuzzy edges. To enhance the image segmentation by the online SHM detection, the previous detected images and multiple frequency excitations can be employed to achieve more reliable segmentation.;A morphological gradient prior enhanced three-step procedure for extracting a probability density function (PDF) of damage size from a damage image by an structural health monitoring (SHM) system is proposed for providing reliable and comprehensive information about the damage using probabilistic concept. This procedure can fill the gap between damage imaging and damage prognosis. The results can also implicate the accuracy and precision of various damage imaging techniques. Compared to conventional deterministic damage quantification methods, such as threshold crossing, the advantage of the proposed method is that it provides the probability of all possible damage sizes. Since damage size is given as a probability distribution, the failure is defined as a probability of damage being larger that the pre-defined critical damage size. Several examples are presented and compared to illustrate the validity of the proposed PDF extraction method. Results show that the PDF generated reflects not only the most likely damage size but also the confidence of the estimation. The damage prognosis is carried out to obtain the remaining useful life (RUL) to verify the damage estimation procedure, and an improved diagnosis by prognosis model is proposed to reduce the uncertainty of detection based on the previous detections.