Prediction of Progressive Damage State at the Hot Spots using Statistical Estimation

The conventional prediction of fatigue crack growth relies on a priori data set and hence, a calibration procedure is required. However, the occurrence of damage and the prediction of damage growth in real time are very unpredictable. Several factors inherently influence the wave signal and make it difficult for the user to interpret the data through direct comparison. To minimize this error in damage estimation, a statistical approach would be the best available solution. A model which can explicitly represent the a priori data set and can also estimate the output parameter (such as damage quantity) from the current data set is extremely important. Simultaneously, there should be a minimum power consumption for the proposed algorithm. Hence, in this effort a two-step analysis is adopted. First, the signal features are extracted using newly proposed discrete time energy model and then Gaussian mixture model (GMM) is used as a classification modeling technique to estimate and quantify the progressive damage. In this article, the effort is not limited to estimating the damage growth but extended to generate an ‘early alarm system’. The early alarm system is achieved by the damage occurrence identification and damage extent which can be calculated using the Mahalanobis distance. GMM is frequently used in the field of computer science for pattern recognition and classification. This would be the first attempt towards implementing such models for hot spot monitoring problems.