Integrating Virtual Sensor Data Augmentation Into Machine Learning for Damage Quantification of Bolted Structures Under Assembly Uncertainty

Machine learning algorithms have significantly advanced structural monitoring by achieving accuracy levels outperforming traditional methods. These approaches facilitate uncertainty modeling and statistical pattern recognition analysis, supporting decision-making and manipulating broader data fusion. Efficient condition assessment of bolted structures, widely used in engineering systems and structural steel members, is crucial for maintaining stability, preventing unwanted loosening, and enabling scheduled maintenance. A critical issue in bolted systems, torque loosening, is often caused or aggravated by excessive vibrations, shocks, temperature variations, and improper usage, increasing the risk of structural faults. Predicting and monitoring bolt loosening remain a significant challenge, as it typically requires expensive inspections and operational controls. This work proposes an enhanced machine learning–based condition assessment model for estimating bolt torque loosening using the spectrum of raw vibration signals and data-driven augmentation strategies. The condition monitoring accounts for intrinsic variability introduced during the assembly process, with damage indexes derived from dynamic responses serving as feature extractors. The machine learning model utilizes data augmentation and fusion to enhance the dataset, relying solely on experimental data, thereby eliminating the need for numerical models. The results demonstrate significant enhancement in the model performance by adopting the integrated dataset, yielding improved torque estimation accuracy with lower error rates. In addition, the monitoring process incorporates uncertainty quantification associated with torque estimation, providing a more reliable assessment of the system’s condition. Furthermore, this study highlights the potential of data-driven machine learning damage assessment techniques in bolted joint monitoring, providing an effective and efficient method for detecting bolt torque loosening using raw vibration spectra. The proposed approach accelerates inspection and establishes a robust technique for monitoring bolted systems.