AI-Based Bolt Loosening Diagnosis with Deep Learning from Laser Ultrasonic Wave Data

Large-scale production of steel requires an exceptionally strong bolt connection. Throughout functioning, loosening of bolts and preload losses can lower a structure’s capability to transport loads while maintaining security and longevity. Among the most intriguing areas in this subject is applying Artificial intelligence algorithms like deep learning (DL) to non-destructive assessment. This investigation intends to contribute to this work by examining the adoption of DL methods for a bolt-loosening diagnosis from laser ultrasonic wave data. Ultrasonic Wave Propagation Imager (UWPI) was used to construct a 3D ultrasonic wave dataset. Various methods for signal processing were then used to produce the processing statistics. The determined variance was computed for assessing the efficiency of a convolutional neural network (CNN) on various processing data sets employing a regression framework depending on a VGG-like framework. To further illustrate the durability of the suggested method, comparisons were made with different AI algorithms for regression. It was discovered as a result that the suggested strategy has possibilities for integrating DL methods and laser-generated ultrasound. Furthermore, it has been demonstrated that DL efficiency for the automated looseness estimates is significantly influenced by the signal analysis method.