Dynamic strain signal monitoring and calibration with neural network based on hierarchical orthogonal artificial bee colony

Abstract

The dynamic strain monitoring system is an important means to reflect the deformation and evolution of the structure in the field of traffic and transportation visually. The exact measurement of dynamic strain measurement system is inseparable from the traceability system of high-precision strain gauge. In this paper, the resistance strain monitoring system installed on the web surface of the main girder box of Jiujiang bridge is calibrated online by using the parallel method at a similar location. By means of dynamic strain measurement system calibration sample signal feature extraction, put forward online calibration method based on neural network algorithm for dynamic strain signaler, move the noise of passive nonlinear signal under the circumstance of incentives, set up and designed a calibration method of HOABC-NN for dynamic signal analysis of on-line monitoring system. After verification, it is found that the algorithm is superior to the traditional neural network training method, the calibration precision is improved obviously, and can support the dynamic strain of bridge monitoring system online calibration

Introduction

Under the action of long-term traffic load and environmental dynamic load, the fatigue problem of the structure is prominent especially [1]. In order to grasp the mechanical performance and fatigue characteristics of the bridge better, it is necessary to monitor the stress condition of the critical stress section.

In the field of engineering, structural stress is the direct reflection of bridge stress state and an important index to measure the safety performance of component materials [2]. Accurate and reliable measurement of dependent variables is crucial to engineering safety and product quality improvement. There are many ways to measure strain, such as electric method, piezoelectric method, magnetic method, vibrating string method, speckle correlation method and so on. Bridge strain monitoring system is an important part of safety performance evaluation of transportation infrastructure.

Dynamic strain monitoring system is practical and has higher precision, so it is used in bridge monitoring widely.

In the direction of bridge structure monitoring methods and monitoring signal processing, there are many scholars at home and abroad have taken deep research. Many scholars also carried out a lot of exploratory theoretical surveys on the processing and analysis of structural monitoring system data. Based on the deterministic analysis, someone as Seo has focused on how to modify the theoretical finite element model based on the load and structural response information obtained from the structural monitoring system data, so as to evaluate the bridge’s carrying capacity more accurately and effectively. The selected bridge type is steel–concrete composite system bridge. The parameters considered for correction in the finite element model include the moment of inertia of the key construction, the elastic modulus of the material, and the rotational constraint at the support, etc. Through multiple iterations and least square regression analysis, the finite element model of the structure was modified to reduce the difference between the measured response and the theoretical response to a reasonable range [3]. The above research provides a validation for the engineering application that the health monitoring system can truly reflect the stress and deformation response of the structure itself. Based on the probabilistic analysis, Kwon used the monitoring data(steel box girder strain) to analyze the reliability of steel bridge’s yield strength and fatigue strength. The research object is the reliability of a single component, not the whole structure. The limit state equation of the reliability of a single component is given, and the structural response is obtained through measured strain data [4], [5], [6], [7], [8], [9], [10]. The research on probability analysis improves the practicability of monitoring data. With the development of artificial neural network and the development of intelligent precision instruments, higher requirements are also put forward for the measurement accuracy [11]. In 2016, based on the cubic spline interpolation principle.Tian and others increased the data density through interpolation and solved the engineering accuracy problem caused by sparse measured data. Combined with the method of least square method, the F–f curve of vibrating string sensor was fitted and modified by Matlab software, obtained more accurate smooth curve of the accord with engineering practice and the reasonable conclusion. This method achieves accurate fitting of the F–f​ curve of the vibrating string sensor with limited measured point data, providing a feasible and fast measurement method for the long-buried vibrating string sensor F–f function, and obtaining the actual working curve [12]. In 2017, N Chen and others, based on fast Fourier transform method which uses digital Fourier filter automatically filter out noise, Quinn algorithm is then used for high precision frequency calculation. Based on the STM32 processor platform of vibrating wire sensor system frequency measurement in the method of experimental test, the test results show that under the condition of no noise, the frequency of the system measuring relative error less than 0.01%. In the case of severe white noise interference (signal-to-noise ratio is −20 db), the system’s relative error of frequency measurement is less than 0.3%. Compared with other frequency measurement methods, this method has better ability of moving noise and frequency measurement accuracy [13]. Yue and others proposed an improved artificial bee colony algorithm to reduce data redundancy and network traffic in the whole life cycle of the network by comparing the random distribution of nodes, thus achieving improved efficiency [14]. Herrmann and others applied the model of interlayer propagation network to the damage diagnosis of plane truss structure, and found that the stiffness value of each member could be expressed as a nonlinear mapping between node displacement and input load, which was realized by the interlayer propagation neural network. For the damage patterns in the training samples, the network diagnosis error is less than 13.51%, but for the damage patterns in the training samples, the network diagnosis error is 24.19% [15]. Kaminski uses the neural network to diagnose the approximate location of the damage by analyzing the frequency change, and obtains that the frequency drop after the speech is the most appropriate input parameter [16]. In 2018, Mao by monitoring dynamic characteristics of Sutong Cable-Stayed Bridge (SCB), including acceleration and strain responses as well as modal frequencies, are investigated through one-year continuous monitoring data under operating conditions by the structural health monitoring system. One-year continuous modal frequencies of SCB are identified using Hilbert–Huang transform method. Variability analysis of the structural modal frequencies due to environmental temperature and operational traffics is then conducted. Results show that temperature is the most important environmental factor for vertical and torsional modal frequencies. The traffic load is the second critical factor especially for the fundamental vertical frequency of SCB [17]. Chance and others conducted damage detection research on cantilever beam and cantilever beam plate through finite element data training neural network, and vibration mode and mode curvature derived from vibration mode were used for training data [18]. Muhammad proposes novel reference-free bridge displacement estimation by the fusion of single acceleration with pseudo-static displacement derived from co-located strain measurements. First, this team propose a conversion of the strain at the center of a beam into displacement based on the geometric relationship between strain and deflection curves with reference-free calibration. Second, an adaptive Kalman filter is proposed to fuse the displacement generated by strain with acceleration by recursively estimating the noise covariance of displacement from strain measurements which is vulnerable to measurement condition. Both numerical and experimental validations are presented to demonstrate the efficiency and robustness of the proposed approach [19]. Studies have shown that wavelet neural network can successfully make intelligent diagnosis of damage types, and can still make diagnosis correctly when the input mode is interfered by a certain degree, and the system has good robustness [20].

Strain sensing system will be affected by its nonlinear error with long-term work, the influence of traffic and environmental load, in the process of monitoring. Even resistive strain sensing system of high precision sensitivity decline, error value increase, it is usually by regular maintenance calibration of strain sensor performance tested during the period of use. However, as the monitoring value of strain monitoring sensor is cumulative and dimensionless, it is extremely difficult to measure and calibrate in the working environment. In the process of on-line calibration of strain sensing system, since off-line disassembly and reinstallation of sensing elements pose risks to the continuity and consistency of monitoring data, load test method is mainly used at present, but load test cannot eliminate the instability of parameters such as stiffness and strength of the structure itself. Therefore, at the present stage, the traceable high-frequency dynamic monitoring sensor is usually installed in the close position of the sensor to be corrected for parallel measurement. Based on time series, the noise is eliminated by combining with the artificial neural network algorithm optimized with a novel artificial bee colony [21], [22], [23], and the maximum signal is fitted. It has been proved that the evolutionary computation can obvious improve the accuracy of neural network [24], [25]. Thus, the online metering service of long-term monitoring sensor can be realized.