Acoustic Emission Signal Localization in Steel Specimen Using Neural Networks with Multi-Sensor Feature Concatenation

The conventional acoustic emission (AE) signal localization algorithms in two-dimensions depend on velocity and time-of-arrival (TOA). Therefore, accurate calculation of such parameters is crucial. The difficulty is in identifying the velocity since they travel with separate modes depending on the element. Simultaneously, the complexity and computational time of the algorithms increases with the number of sensors. A novel localization algorithm is suggested to defeat these issues using a multi-sensor feature concatenation incorporated with a neural network (NN). Three sensors are used to capture the AE signals generated from a single pencil lead break-up test performed on a steel plate. After capturing, signal processing tools are employed to gather the required features. The fast Fourier transform is utilized for plucking out the frequency features, and the wavelet transform is applied to concurrently examine the signals in time-frequency domains. Then, the features are concatenated to construct the data set of the proposed localization algorithm. The input is the wave’s dominant frequency, TOA, amplitude, and duration, and the crack location coordinates are the NN model’s output. Results show that the proposed model achieves a mean squared and mean absolute error of 0.2313 cm<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> and 0.3463 cm, which is 81.77% more accurate compared to the conventional triangulation method, demonstrating robust localization performance without requiring explicit wave velocity estimation.