Analysis of Feature Performance for Damage Detection Using Data Analysis

This paper emphasizes the use of data analysis to predict the pipeline behavior, focusing on vibration signals as a primary method for detecting the extent of corrosion or damage. Such methods provide good insights by enabling the early detection of damage before it leads to failure. This study employs ANSYS for building numerical simulations for modeling pipeline conditions and extracting desirable features, and the accuracy of these models is validated through experimental work. Key features are extracted from sensor data using a surrogate model and are subsequently analyzed through Principal Component Analysis (PCA). PCA helps to reduce the data dimension and keeps the variations of the data through the first principal components. Following this, Independent Component Analysis is used to further refine the data by extracting and comparing statistically independent components for different models. Data separation enhances the detection of anomalies such as corrosion or structural damage by "whitening" the data.