<p>In the field of traditional hidden crack detection in metal structures, detection technologies are often limited by low sensitivity and imprecise data analysis. To overcome these challenges, the advanced Lamb wave ultrasonic testing technology is applied in this article. This technology combines signal preprocessing techniques such as wavelet transform denoising and high-pass filtering to improve signal quality. Through short-time Fourier transform and continuous wavelet transform for time–frequency analysis, combined with convolutional neural network and long short-term memory network to build an automatic crack detection and expansion prediction model, the accurate diagnosis and prediction of hidden cracks are achieved. Taking steel as the experimental object, the results show that the fusion of Lamb wave ultrasonic detection technology and advanced data processing methods significantly improves the sensitivity and resolution of crack detection. The system has a detection success rate of more than 88% for cracks of different sizes and types, and the error is controlled within 0.1&#xa0;mm; the crack expansion prediction accuracy rate is 87.5%, which effectively predicts the development trend of cracks. The study verifies the reliability of Lamb wave ultrasonic technology in the diagnosis and prediction of hidden cracks in steel, and provides a scientific basis for structural maintenance and repair. </p>