Wavelets Application in Prediction of Tunnel Defects in Friction Stir Welding of Alloy Joints from Vibroacoustic ANN-Based Model

This paper examines a connection of acoustic emission signal to the core parameters of the friction stir welding process, based on the artificial neural networks (ANNs). AE Instrument NI USB-9234 has obtained acoustic Z and Y emission signals. Wavelet Transform was used as the ANN's output through numerical and time parameters for discomposed acoustic pollution signals. The ANN inputs include rotation speed and frequency of the device, the machine profile and the tensile strength parameters. A multi-layer neural feed-forward network was selected and trained using the Levenberg Marquardt algorithm for different network architectures. Ultimately, an overview will be provided of the correlation between the estimated and analyzed results. The prototype obtained can be implemented with the aid of acoustical emission signals to design and improve automated system parameters and mechanical properties of the joint.