Multi-sensor fusion-based time-frequency imaging and transfer learning for spherical tank crack diagnosis under variable pressure conditions

In this paper, a crack diagnosis framework is proposed that combines new signal-to-imaging technique and transfer learning-aided deep learning to automate the diagnostic process. The objective of convert one-dimensional (1D) acoustic emission (AE) signals from multiple sensors into two-dimensional (2D) image capture information under variable operating conditions. process, short-time Fourier transform (STFT) first applied AE signal each sensor, STFT results different are then fused obtain condition-invariant 2D cracks; scheme denoted as Multi-Sensors Fusion-based Time-Frequency Imaging (MSFTFI). MSFTFI images subsequently fed fine-tuned (FTL) model built on convolutional neural network (CNN) for diagnosing types. (MSFTFI + FTL) tested with standard dataset collected self-designed spherical tank validate performance pressure suggest strategy significantly outperformed classical methods average improvements 2.36–20.26%.