On-line structural health monitoring and prognosis of a biaxial cruciform specimen

The current research of on-line damage state estimation techniques offers adaptive damage state prediction and residual useful life assessment. The real-time damage state information from an on-line state estimation model can be regularly fed to a predictive model to update the residual useful life estimation in the event of a changing situation. The present paper discusses the use of an integrated prognosis model, which combines an on-line state estimation model with an off-line predictive model to adaptively estimate the residual useful life of an Al-6061 cruciform specimen under biaxial loading. The overall fatigue loading history is assumed to be a slow time scale process compared to the time scale at which, the sensor signals are acquired for on-line state estimation. The fast scale on-line model is based on a non-parametric system identification approach such as correlation analysis. A new damage index equivalent to quantitative damage state information at any particular fatigue cycle, is proposed. The on-line model regularly estimates the current damage state of the structure based on passive strain gauge signals. These damage states information is regularly fed to the slow scale off-line predictive model as it becomes available. The off-line predictive model is a probabilistic nonlinear regression model, which is based on Bayesian statistics based Gaussian process approach. The off-line module adaptively updates the model parameters and recursively predicts the future states to provide residual useful life estimate.