Detecting Loosening of Bolt-Flanged Pipe Using Smoothed Pseudo Wigner–Ville Distribution: Digital Twin-Based Approach Using Convolution Neural Networks

Abstract

In this paper, an integrated digital twin-based fault-diagnosis framework is proposed to detect bolt loosening in flanged pipe connections. The approach combines finite element-based numerical modeling, experimental modal analysis, and advanced signal processing-assisted deep learning to achieve reliable loosening detection under operational excitations. First, a finite element model of the bolted flange pipe is developed and updated using experimentally identified natural frequencies to establish a validated digital twin. The dynamic responses of healthy and loosened bolt conditions are then simulated and experimentally measured. To extract fault-sensitive features, the vibration signals are decomposed using a Haar filter bank, followed by transformation into the time–frequency domain using the Smoothed Pseudo Wigner– Ville distribution (SPWVD). The resulting time–frequency representations are subsequently used to train a convolutional neural network (CNN). A key contribution of this study lies in the hybrid digital twin learning strategy, in which CNN training is performed using numerical data and healthy experimental data. At the same time, fault classification is conducted under experimental loosening conditions. The results demonstrate that the proposed framework effectively captures bolt loosening signatures and provides a practical, physics-consistent solution for condition monitoring of bolted flange connections.