A multi-task learning-based automatic blind identification procedure for operational modal analysis

Ai-Driven Blind Source Separation for Fast Operational Modal Analysis of Structures

“…3 is a benchmark case study used by several researchers in the literature (e.g. [41,42]). The structural matrices used in the model are:…”

“…This intuition was pursued by Lin et al [41], who encapsulated the principles of BSS into the loss function of a self-coding NN, yielding promising results in a real-world steel boxgirder cable-stayed bridge. Similarly, Shu and co-authors [42] recently developed a multi-task DNN for the automated identification of independent modes extracted from SCA. In this approach, mode shapes are directly extracted as the weights between the last two layers of neurons in the NN, and resonant frequencies and damping ratios are estimated from the independent components using the random decrement technique (RDT).…”

“…Unlike the previous similar AI approaches reported in the literature (e.g. [41,42]), the present approach employs a multi-task learning deep NN capable of extracting both the real and imaginary components of the independent modal sources, offering comprehensive modal identification capabilities. The developed AI-based OMA approach takes ambient acceleration data as input and provides quasi real-time identification with minimal computational demands.…”

“…Likewise Ref. [42], the mode shapes (here in complex form) are extracted from the weights between the last two layers of the NN. Subsequently, the resonant frequencies and damping ratios are estimated from the (complex) independent modal components through the robust Ibrahim time-domain (ITD) identification method.…”

Vibration Signal Evaluation Based on K-Means Clustering as a Pre-Stage of Operational Modal Analysis for Structural Health Monitoring of Rotating Machines

“…In this context, machine learning techniques were used to solve challenges of automation, with emphasis on the works of [8,9], who used unsupervised learning to automatically interpret the results of the Stochastic Subspace Identification (SSI) technique. More recently, the matter has been addressed by [10][11][12][13][14][15][16][17][18], and existing approaches were already in use in [19,20] for condition monitoring of wind turbines, highlighting the relevance of the subject to current research.…”