Damage detection in beam structures using bayesian deep learning and balancing composite motion optimization

Abstract

This research proposes a new indicator based on change in a vibration signal’s probability spectrum centre to assess structural change. It relies on the change of a central position of the probability spectrum (C-PSD). For better performance than in previous research, this study covers three key issues. Firstly, the input data set using the balancing composite motion optimization (BCMO) method has been enriched and optimized. Accordingly, any missing part is compensated for and optimized before analysis. This is a major difference between this study and previous ones in which most pre-processed data are often filtered. As a result, the reliability of the indicator is significantly increased. Secondly, the optimized results derived from the BCMO method will be processed and trained using a deep learning platform with a probability distribution transfer function to develop a new set of indicators. The result shows that a deep learning-probability model can easily assess and detect a damage change since it is highly signal change-sensitive. The training process using deep learning probability can layer and categorize the same or different damages, which makes it simpler and more systematic for assessing damage levels for various types of structures. Thirdly, the article proposes using the central coordinates of a probability spectrum instead of the structure’s actual power spectrum, which is a new assessment method. A vibration signal’s probability spectrum is a novel spectrum never previously mentioned in any paper. It enables the model to have greater sensitivity to a structure’s change. In addition, instead of the conventional use of natural frequency, this model only applies the natural frequency centre for assessment of the damaged structure. This more sensitive indicator will thus work better and more effectively than other indicators in assessing and detecting damages.