Statistical methodologies for removing the operational effects from the dynamic responses of a high‐rise telecommunications tower

Abstract

Summary This paper describes statistical methodologies for removing the influence of operational effects from the dynamic responses of a telecommunications tower. The characterization of the dynamic responses of the structure, over a period of 3 months, was based on a continuous monitoring system that included accelerometers, anemometers and a meteorological station. The analysis of the results allowed identifying a significant number of critical events, for which the dynamic response under wind action is significantly amplified, as well as sporadic events, associated with high peak acceleration values, due to the influence of operational effects related to the movement of the lift, technical staff, and equipment. The automatic identification of critical events, based on extreme acceleration values, required the prior removal of operational effects from the records using two distinct methodologies, one based on the principal component analysis (PCA) and the other based on the crest factor (CF) and on autoregressive models (AR). Both methodologies showed efficiency and robustness in eliminating acceleration peaks due to operational effects; however, the methodology based on the CF and AR models proved to be computationally more efficient and resulted on a smaller number of false‐positive occurrences in the identification of critical events. The developed methodologies showed potential to be integrated in Structural Health Monitoring (SHM) systems to assess the structural safety and serviceability of telecommunications towers.

Detection of Wheel Polygonization Based on Wayside Monitoring and Artificial Intelligence

“…After the implementation of a CWT on the evaluated signals, a PCA is performed followed by the extraction of four statistical parameters, resulting in significant data compression [ 17 , 40 ]. Based on the p -by- q matrix X containing the extracted CWT features from the original signal evaluated by each sensor, where p is the measurement points number and q is the wavelets coefficients number, the principal components can be determined using the following equation: where represents the scores matrix, and T is a q -by- q orthonormal linear transformation matrix.…”

Damage Identification of Railway Bridges through Temporal Autoregressive Modeling

“…Conversely, an excessively large number of retained PCs can lead to over-fitting and the subsequent loss of generality in the model. A common rule of thumb in the SHM literature is to choose l PCs explaining more than 80% of the cumulative variance [42,56].…”

Online unsupervised detection of structural changes using train–induced dynamic responses

“…To overcome this problem, principal component analysis (PCA) [8] is used here for feature modelling. PCA is a multivariate statistical method that produces a set of linearly uncorrelated vectors called principal components (PCs) or scores, from a multivariate set of vector data [34]. PCA can be used to remove the linear effects of EOVs on the responses of the structure during a training period, from which scores are obtained.…”