A stacked ensemble learning model is developed to predict the modal parameters of space grid steel structures under environmental effects. Potential damage is detected via statistical analysis of the prediction residuals. For this purpose, five standalone heterogeneous machine learning models were trained for predicting natural frequencies; each model used the principal components of the environmental data as input parameters. Next, a stacked ensemble learner was built using the outputs of the five standalone models as its inputs. Finally, a damage indicator combining the predicted residuals of multiple orders of natural frequencies is proposed and statistically analyzed for accurate damage detection. To verify the effectiveness of the proposed method, a space grid model was created in the field environment and measured for a period. Dynamic and environmental data were collected, such as ambient temperature, humidity, wind speed and direction, and structural surface temperature. An automated procedure of the covariance-driven stochastic subspace identification method was conducted to identify bulk mode. The environmental dependence of the natural frequencies, damping ratios, and vibration modes was analyzed. Then, the method was validated based on short-term monitoring data from the baseline health state and unknown future states. The results show that the natural frequencies and damping ratios of space grid structures fluctuate significantly on a daily basis due to environmental influences. Stacked ensemble learning utilizes predictions from multiple heterogeneous models to produce a better predictive model. The statistical analysis of the prediction residuals by ensemble learning effectively removes the environmental influences, allowing for timely structural damage detection.
Structural health monitoring (SHM) is vital to the maintenance of civil infrastructures. For rail transit systems, early defect detection of rail tracks can effectively prevent the occurrence of severe accidents like derailment. Non-destructive testing (NDT) has been implemented in railway online and offline monitoring systems using state-of-the-art sensing technologies. Data-driven methodologies, especially machine learning, have contributed significantly to modern NDT approaches. In this paper, an efficient and robust image classification model is proposed to achieve railway status identification using ultrasonic guided waves (UGWs). Experimental studies are conducted using a hybrid sensing system consisting of a lead–zirconate–titanate (PZT) actuator and fiber Bragg grating (FBG) sensors. Comparative studies have been firstly carried out to evaluate the performance of the UGW signals obtained by FBG sensors and high-resolution acoustic emission (AE) sensors. Three different rail web conditions are considered in this research, where the rail is: (1) intact without any defect; (2) damaged with an artificial crack; and (3) damaged with a bump on the surface made of blu-tack adhesives. The signals acquired by FBG sensors and AE sensors are compared in time and frequency domains. Then the research focuses on damage detection using a convolutional neural network (CNN) with the input of RGB spectrum images of the UGW signals acquired by FBG sensors, which are calculated using Short-time Fourier Transform (STFT). The proposed image classifier achieves high accuracy in predicting each railway condition. The visualization of the classifier indicates the high efficiency of the proposed paradigm, revealing the potential of the method to be applied to mass railway monitoring systems in the future.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.