A hybrid support vector regression with multi-domain features for low-velocity impact localization on composite plate structure
Introduction
Low-velocity impacts on the composite plate structure of the ship have attracted more attention since composite materials with high weight-to-strength ratio have been increasingly applied. These impacts caused by collisions with floating ice, stranding, and explosion can lead to barely visible impact damage (BVID). It is difficult to detect BVID on the composite plate structure by the visual inspection, but BVID can reduce the strength of the structure [1]. Therefore, to detect the locations of low-velocity impacts, it is crucial to develop structural health monitoring (SHM) for the composite plate structure of the ship. In the SHM system, low-velocity impact signals can be received by different types of sensors, such as piezoelectric (PZT) sensors [2], [3] and fiber optic sensors [4]. Among various available sensors, fiber Bragg grating (FBG) sensors are more suitable for SHM applications in large-scale structures. This is because FBG sensors have the advantages of flexibility, multiplexing capability, corrosion resistance, small sizes, and high sensitivity, and they can easily form a sensor network without causing damage to the composite plate structure. Additionally, FBG sensors are immune to the electromagnetic interference, which means that signals acquired by FBG sensors are not affected by the electromagnetic interference [5], [6], [7].
An increasing number of impact localization methods have been developed to locate the low-velocity impacts on the composite plate structure, which can be divided into active methods and passive methods. Different from active methods such as the time reversal method [8], passive methods are simpler and more effective when only the locations of low-velocity impacts need be determined. Passive methods mainly include geometric methods [9], machine learning methods [10], and reference database methods [11].
Since the low-velocity impact localization problem can be considered a classification problem or regression problem, machine learning methods such as the support vector machine (SVM) [12] and the back propagation neural network (BPNN) [13] have been widely utilized. Based on the frequency characteristics of impact signals which were achieved by the Fourier transform method, Jiang et al. [14] applied the extreme learning machine (ELM) to the identification of impact areas on a carbon fiber reinforced plastic (CFRP) plate. Yu et al. [15] achieved satisfactory impact localization results on a CFRP plate by using the short-time energy characterization extraction method and the SVM model whose parameters were optimized. For the impact localization on a CFRP plate, Lu et al. [16] combined the wavelet packet energy spectra of the low-velocity impact signals with the support vector regression (SVR) model to determine the locations of low-velocity impacts. Datta et al. [17] extracted the features, including the peak value, the mean value, the standard deviation, and the energy index, from time signals and strain signals and chose them as the input of the least squares support vector regression (LS-SVR) model. Moreover, Sai et al. [18] established the ELM model with the energy of frequency band signals which were obtained by the wavelet packet decomposition method. For the low-velocity impact localization system using FBG sensors, most of the current research focuses on using the single domain features to implement the localization or identification of low-velocity impacts on composite plate structures, which may lead to low localization accuracy.
Additionally, the selection and application of machine learning methods also have a significant influence on the localization accuracy for low-velocity impacts. Many studies [19], [20], [21] have demonstrated the superior generalization performance of SVR compared with other nonlinear methods, whereas an inappropriate determination of hyperparameters in the SVR model can dramatically decrease the method’s performance on regression problems. In addition to the grid search (GS) method [22], metaheuristic algorithms are common optimization methods for strengthening the performance of SVR. For example, the particle swarm optimization (PSO) algorithm [23], the genetic algorithm (GA) [24], and the differential evolution (DE) [25] algorithm have been successfully applied to the parameter selection in the SVR model.
In this paper, we propose a hybrid support vector regression with multi-domain features for the low-velocity impact localization problem on the composite plate structure. To extract more effective features, the trend component in the low-velocity impact signals is eliminated by utilizing the empirical mode decomposition (EMD) method. Then, three feature extraction methods, which are the time-domain feature method, the frequency-domain feature method, and the short time Fourier transform method, are employed to obtain multi-domain features as the input of the SVR model. Moreover, the optimized v-SVR model (BA-SVR) is designed to enhance its regression capability, in which three parameters involving the kernel parameter γ, the penalty parameter C, and the parameter v are tuned by using the bat algorithm (BA). The low-velocity impact localization system using four FBG sensors is adopted to analyze and evaluate the performance of BA-SVR with multi-domain features.
The organization of this paper is given as follows. In Section 2, the relative theories of the FBG sensor and the proposed method are presented. The optimized v-SVR model is given in Section 3. Section 4 provides the details of the hybrid support vector regression with multi-domain features. Section 5 discusses the results obtained by the proposed method in five sets of experiments. Finally, the conclusion and the direction of future work are given in Section 6.
Section snippets
Principle of FBG sensor
The FBG sensor can reflect part of the spectrum of incident light, and its refractive index changes periodically along with the Bragg grating length [26]. Based on the effective refractive index nneff and the grating period Λ, the Bragg wavelength λ is calculated by Eq. (1).
The shift in the Bragg wavelength of the back-reflective light depends on the change of the refractive index and the grating period with the temperature and the strain. In the laboratory environment, the impact
Optimized v-SVR
The v-SVR model has three parameters which are the kernel parameter γ, the penalty parameter C, and the parameter v. The determination of these parameters commonly depends on the trial and error search method based on manual operation, which is time consuming and may have a negative influence on the regression performance of the v-SVR model. Generally, a larger penalty parameter C and a smaller kernel parameter γ indicate that the v-SVR model has a greater nonlinear fitting capability and
Hybrid support vector regression with multi-domain features
To increase the localization accuracy for low-velocity impacts on the composite plate structure, the hybrid support vector regression with multi-domain features is proposed. After acquiring low-velocity impact signals by the FBG sensing system, three steps, which are the signal preprocessing, the multi-domain feature extraction, and the impact localization, are executed. The procedures of the proposed method are illustrated in Fig. 1 and given in detail below.
Step 1: The trend component induced
Experiments and discussions
In this paper, a low-velocity impact localization system is utilized to monitor the low-velocity impacts on the composite plate structure. The contributions of the multi-domain features and influence of the number of FBG sensors and the number of optimal parameters are analyzed, and then the performance of the hybrid support vector regression with multi-domain features in solving the low-velocity impact localization problem is evaluated.
Conclusions
In this paper, a hybrid support vector regression with multi-domain features is proposed for the low-velocity impact localization problem on the composite plate structure. The proposed method has three steps including the signal preprocessing, the multi-domain feature extraction, and the impact localization. First, the low-velocity impact signals acquired by FBG sensors are preprocessed by the EMD method to remove the trend component. Second, the time domain features, the frequency domain
CRediT authorship contribution statement
Qi Liu: Conceptualization, Methodology, Validation, Investigation, Writing - original draft, Writing - review & editing. Fengde Wang: Software, Validation, Investigation, Writing - review & editing. Jindong Li: Validation, Visualization, Writing - review & editing. Wensheng Xiao: Validation, Resources, Data curation, Supervision, Project administration, Funding acquisition.
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.
Acknowledgement
The authors would like to acknowledge the High-tech Shipping Research Project from Ministry of Industry and Information Technology of China for supporting this study through the project “Research of full life-cycle reliability guarantee technology system for subsea oil and gas production system” with the grant number of 2018GXB01-02-003.
References (52)
- et al.
Uncertainty quantification for acoustic nonlinearity parameter in Lamb wave-based prediction of barely visible impact damage in composites
Mech. Syst. Signal Pr.
(2017) - et al.
Low velocity impact damage evaluation in fiber glass composite plates using PZT sensors
Compos. Part B-Eng.
(2013) - et al.
In situ barely visible impact damage detection and localization for composite structures using surface mounted and embedded PZT transducers: A comparative study
Mech. Syst. Signal Pr.
(2016) - et al.
Impact localization on composite structure using FBG sensors and novel impact localization technique based on error outliers
Compos. Struct.
(2016) - et al.
Composite plate low energy impact localization system based on FBG sensing network and hybrid algorithm
Opt. Fiber Technol.
(2015) - et al.
Localization of impact on composite plates based on integrated wavelet transform and hybrid minimization algorithm
Compos. Struct.
(2017) - et al.
Impact localization on a composite stiffened panel using reference signals with efficient training process
Compos. Part B-Eng.
(2016) - et al.
Support vector regression based prediction of global solar radiation on a horizontal surface
Energ. Convers. Manage.
(2015) - et al.
Support vector regression with chaos-based firefly algorithm for stock market price forecasting
Appl. Soft Comput.
(2013) - et al.
Hierarchical time series forecasting via support vector regression in the European travel retail industry
Expert Syst. Appl.
(2019)
System reliability prediction by support vector regression with analytic selection and genetic algorithm parameters selection
Appl. Soft Comput.
Implementing support vector regression with differential evolution to forecast motherboard shipments
Expert Syst. Appl.
Low velocity impact localization on composite wing structure using error outlier based algorithm and FBG sensors
Compos. Part B-Eng.
Structural response reconstruction based on empirical mode decomposition in time domain
Mech. Syst. Signal Pr.
Structural health monitoring using empirical mode decomposition and the Hilbert phase
J. Sound Vib.
A novel intelligent method for mechanical fault diagnosis based on dual-tree complex wavelet packet transform and multiple classifier fusion
Neurocomputing
Multi-fault classification based on wavelet SVM with PSO algorithm to analyze vibration signals from rolling element bearings
Neurocomputing
Research on the unified mathematical model for FT, STFT and WT and its applications
Mech. Syst. Signal Pr.
Distributed collaborative probabilistic design for turbine blade-tip radial running clearance using support vector machine of regression
Mech. Syst. Signal Pr.
Mixed kernel function support vector regression for global sensitivity analysis
Mech. Syst. Signal Pr.
An improved support vector regression model for estimation of saturation pressure of crude oils
Fluid Phase Equilibr.
Gear fault detection using artificial neural networks and support vector machines with genetic algorithms
Mech. Syst. Signal Pr.
A new algorithm for acoustic emission localization and flexural group velocity determination in anisotropic structures
Compos. Part A-Appl. S.
Impact localization of composite stiffened panel with triangulation method using normalized magnitudes of fiber optic sensor signals
Compos. Struct.
Impact localization on composite wing using 1D array FBG sensor and RMS/correlation based reference database algorithm
Compos. Struct.
Strain monitoring and damage detection of a filament wound composite pressure tank using embedded fiber Bragg grating sensors
Key Eng. Mater.
Cited by (20)
Data-physics hybrid-driven deep learning method for impact force identification
2024, Mechanical Systems and Signal ProcessingPrediction of high-embankment settlement combining joint denoising technique and enhanced GWO-ν-SVR method
2024, Journal of Rock Mechanics and Geotechnical EngineeringAdvances in dynamic load identification based on data-driven techniques
2023, Engineering Applications of Artificial IntelligenceBinary dynamic stochastic search algorithm with support vector regression for feature selection in low-velocity impact localization problem
2023, Engineering Applications of Artificial IntelligenceA two-step localization method using wavelet packet energy characteristics for low-velocity impacts on composite plate structures
2023, Mechanical Systems and Signal ProcessingCitation Excerpt :It indicates that the proposed impact localization method can provide greater localization performance than most of the comparative methods whether considering the number of FBG sensors or not. Additionally, the sampling frequency employed in the present work is lower than the sampling frequency employed in the previously published studies except [24,26], and [41]. However, the proposed impact localization method collects a larger number of impact samples than the comparative methods.