Dynamic strain signal monitoring and calibration with neural network based on hierarchical orthogonal artificial bee colony
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.
Section snippets
Dynamic strain signal calibration method
In this section, how to sample dynamic strain signal is introduced first. Then, the characteristic of the calibrating sample signal is analyzed. The calibration method of Dynamic Strain signal based on neural network is proposed.
Orthogonal artificial bee colony
In this section, the classical artificial bee colony algorithm is introduced first. Then, the modified artificial bee colony algorithm with the help of orthogonal method and hierarchical method is proposed.
Settings
In order to confirm the performance of HOABC, four other artificial bee colony optimization algorithms have been selected for experiments and comparisons. They are: canonical artificial bee colony algorithm (ABC), cooperative artificial bee colony algorithm (CABC), hybrid artificial bee colony (HABC), gbest-guided artificial bee colony (GABC). The population size of all test algorithms are set as 50 and the maximum iteration number is 1000. The parameter Limit is set as . For HABC,
Calibration results
From above analysis, it is obvious that the proposed HOABC can solve complex optimization problems with superior and stable performance. It is potential that HOABC may own the ability of optimizing the weights of neural network with higher precision for calibrating the dynamic strain signal.
Conclusion
There can be found through the above study, only by the combination of high-precision measuring equipment and multi-measurement methods, we can accurately evaluate the relationship between deformation and locality, accumulate a lot of local and overall data, monitor the micro-variables, judge the trend of micro-deformation and prevent micro-duration.by dynamic strain monitoring system is relatively stable environment load, the influence of traffic load, in use process can produce signal drift
CRediT authorship contribution statement
Lu Peng: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, Writing - review & editing. Genqiang Jing: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing - original draft, Writing - review &
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The research was supported by the Project of National Key R&D Program of China (No. 2018YFB1600300, 2018YFB1600302, 2017YFF0206305), Key Projects of Public Scientific Research Institutes, China (No. 2018-9031) and Beijing Natural Science Foundation, China (No. 8192046).
References (25)
- et al.
Trust-based secure clustering in WSN-based intelligent transportation systems
Comput. Netw.
(2018) - et al.
Bridge rating protocol using ambient trucks through structural health monitoring system
Eng. Struct.
(2013) - et al.
Time-dependent reliability assessment of ship structures under progressive and shock deteriorations
Reliab. Eng. Syst. Saf.
(2018) - et al.
Fatigue reliability assessment of retrofitted steel bridges integrating monitored data
Struct. Saf.
(2010) - et al.
Application of neural network algorithm in fault diagnosis of mechanical intelligence
Mech. Syst. Signal Process.
(2020) - et al.
Cooperative two-engine multi-objective bee foraging algorithm with reinforcement learning
Knowl.-Based Syst.
(2017) - et al.
Forecasting stock markets using wavelet transforms and recurrent neural networks: An integrated system based on artificial bee colony algorithm
Appl. Soft Comput.
(2011) - et al.
Back propagation neural network with adaptive differential evolution algorithm for time series forecasting
Expert Syst. Appl.
(2015) - et al.
Effects of twin tunnels construction beneath existing shield-driven twin tunnels
Tunn. Undergr. Space Technol.
(2015) - et al.
Maintenance, monitoring, risk and life-cycle performance of bridges
Struct. Infrastruct. Eng.
(2019)
Expansion prediction of alkali aggregate reactivity-affected concrete structures using a hybrid soft computing method
Neural Comput. Appl.
A hybrid GABC-GA algorithm for mechanical design optimization problems
Intell. Autom. Soft Comput.
Cited by (3)
Editorial special issue on AI-Driven sensing and computing for cyber-physical systems
2024, Computer CommunicationsOnline evaluation method for the performance of in-service fiber optic strain monitoring systems
2022, Hongwai yu Jiguang Gongcheng/Infrared and Laser EngineeringIn-situ calibration of in-service strain sensing system under passive excitation: Methodology and impacts assessment
2021, Measurement Science and Technology