A B-Spline Based Gaussian Process Regression Approach for Fatigue Crack Length Estimation Using Ultrasonic Wave Data

The diagnosis and prognosis of fatigue cracks, which greatly influence the long-term durability of structures, is an important issue for structural health monitoring (SHM). This paper presents a study on the estimation of fatigue crack length using ultrasonic wave data. The measured signal is first denoised and truncated to extract the informative period of the signal. If a crack is detected, features are extracted to represent the distortion of the signals while reducing the influence of noise with a B-spline based method. Gaussian process regression obtained from an integration of mean and covariance functions is used for the estimation of the crack length. Real-world experiments validates the effectiveness of the proposed method.