Analysis of Acoustic Emission Signal for Prediction of Corrosion on Carbon Steel Pipelines

• Main Authors: Kafi, N.A., May, Z.B.
• Format: Conference or Workshop Item
• Published: Institute of Electrical and Electronics Engineers Inc. 2021
• Online Access: http://scholars.utp.edu.my/id/eprint/33434/; https://www.scopus.com/inward/record.uri?eid=2-s2.0-85130069154&doi=10.1109%2fAPACE53143.2021.9760548&partnerID=40&md5=cd27d0dd759b87fc8825a080da383501

Naturally, all materials deteriorate over time and this is an obvious in the case of carbon steel pipelines. The harsh environment it is placed in and the corrosion assisting fluids that flow through it increases the rate of structural deterioration even further. Ineffective and inadequate corrosion monitoring often lead to pipeline explosions that can damage the surrounding living things and environment. This project trained and tested two prediction algorithms, the quadratic Support Vector Machine (SVM) and ensemble RUSBoost trees, which classified Acoustic Emission (AE) data into three regions. Region 1 represents AE activity experiencing decreasing corrosion rate. Region 2 represents AE activity experiencing stagnant corrosion rate whereas Region 3 represents AE activity experiencing increasing corrosion rate. The AE data consists of 17 AE features extracted by an AEWIN software as well as the date and time for each hit. These features were analyzed in the time domain and frequency domain using Fast Fourier Transform (FFT) in MATLAB. Then, the Kruskall-Wallis test ANOVA test was conducted using SPSS software to check if the median between two or more groups are significantly different from each other for all possible AE feature pairs. Features that show most significant differences between regions were found to improve classification accuracy. When the significant features are not considered in the training and testing of classification models, it showed a decrease in classification accuracy. Six features were selected to be used as input into the two algorithms. A maximum accuracy of 86 was reached using ensemble RUSBoost trees algorithm. © 2021 IEEE.