Combinação de ensaios ultrassônicos e aprendizado de máquina na detecção de defeitos em revestimento de tubulações industriais

Abstract

The inspection of industrial pipelines with internal mortar linings poses a challenge, especially for pipelines that do not allow internal inspection with automated equipment such as in-line inspection (ILI) tools. This study evaluates the capability of a Linear SVM classifier, trained with different ultrasonic signal processing techniques, to distinguish between intact and defective regions in these linings. Pulse-echo ultrasonic testing was conducted on carbon steel pipe specimens, and the acquired signal s were analyzed in both time and frequency domains. From the time-domain A-scan signals, features such as signal energy, attenuation coefficient, and statistical descriptors were extracted. In the frequency domain, the processing involved the application of the Discrete Fourier Transform (DFT) and the Discrete Cosine Transform (DCT), both individually and in combination. Dimensionality reduction techniques, including Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and their combination (PCA-LDA), were assessed. Experiments were conducted in MATLAB®, using one specimen for training with five-fold cross-validation and an independent test set comprising 20% of the samples. The trained model was subsequently tested on a second specimen to evaluate its generalization capability on previously unseen data. The study confirmed the effectiveness of the proposed methodology in classifying the integrity of internal coatings in pipelines, with the combination of DFT and DCT with LDA emerging as the most efficient approach.