Recent advances in soft computing tools of structural ...

2 downloads 0 Views 436KB Size Report
Structural health monitoring of civil structure is a complex engineering problem ... In this paper, application of soft computing tools such as Neural Networks, Fuzzy .... Random damage states are generated through finite element simulations, and are essentially ... Based on this observation, Loh and Huang model the seismic.
Recent advances in soft computing tools of structural health monitoring for civil infrastructure systems Harish Kumar Mulchandania*, Manoj Kumarb b

a Graduate Research Student, Civil Engineering Department, BITS Pilani, Pilani campus, 333031, India Associate Professor and Head of Department, Civil Engineering Department, BITS Pilani, Pilani campus, 333031, India

Abstract Structural health monitoring of civil structure is a complex engineering problem which require the use of artificial intelligence. Artificial Intelligence such as machine learning has been used from long time in field of computer science and have given promising results in monitoring systems when applied along with engineering technology. In this paper, application of soft computing tools such as Neural Networks, Fuzzy logic, Support Vector Machines, Genetic Algorithm and Hybrid systems in the field Structural Health Monitoring has been presented. Since damage detection is an inverse problem so soft computing tools act as robust, redundant and highly adaptive, so they are fit for structural heath monitoring.

Keywords : Structural Health Monitoring, Machine Learning, Damage detection

1. Introduction Structural health monitoring (SHM) usually refers to the process of implementing a damage detection strategy for aerospace, civil or mechanical engineering infrastructure. SHM is applicable to the wide range from sky scrapers to aircrafts to turbines. At present structural health monitoring in civil infrastructure system is in initial stages, whereas in the field of aircrafts systems it is much developed but it is still a primarily a research topic since every structural health monitoring system is an individual system. This process involves the observation of a structure or mechanical system over time using periodically spaced dynamic response measurements, the extraction of damage-sensitive features from these measurements and the statistical analysis of these features to determine the current state of system health. Statistical analysis requires the use of pattern recognition and machine learning algorithms to determine the damage and current state of system. Physical modelling of real-world systems generally is a challenging task since it has wide variety of damages associated with different length and different time scales, further geometric complexity and changing environmental and operational conditions of structures, so the use of the accurate pattern recognition approach is needed.

* Corresponding author. Tel.: +91 850 3999 321 E-mail address: [email protected], [email protected]

In order to detect the damage of the building structure, the following four steps are needed: 1. Operational evaluation: It describes the main purposes of SHM. The structure has relations with economic, life-safety, environmental, operational conditions and limitations. 2. Data acquisition: It describes how to obtain and store data, pre-processing, lter for sensors, location and number of the sensors, data sampling rate, etc. 3. Feature extraction: It determines the properties of the measured vibration signals. It decided which signals are useful for damage detection. The features normally are: stiffness, mass and energy dissipation of the system 4. Statistical model for feature classification: This area receives less attention. It focuses on implementation of classification algorithms. Algorithms used here can be classified in three main categories: regression, outlier detection and group classification. Soft computing tools are related to statistical modelling for feature classification. Neural networks (NN), Genetic Algorithms (GA) and Support Vector Machines (SVM) are major soft computing tools and are prominently used in the inverse problem of error detection. This study presents various models of Neural Network, Genetic Algorithms and SVM have been proposed by authors, their accuracy and it application in Structural health monitoring of civil structures. The neural network is a powerful pattern recognition tool; it forms a nonlinear mapping between input and output data sets using training data. A network is developed using interconnected neurons. The inputs of each neuron consist of the weighted outputs of other neurons. A commonly used neural network is the feedforward neural network. There is one input layer with n inputs, a hidden layer with m neurons, and one output layer with two outputs. For SHM problems, the two outputs typically represent damage size and location. The inputs are the damage indicators for the problem. The GA is a global searching process based on the Darwin's principle of natural selection and evolution. A simple GA consists in three main operations: selection, genetic operations and replacement. The GA starts with an initial population; the individuals of this population are subjected to the three operators and evolve. The result is a population with a higher fitness than the initial one as in natural selection. This process is iterated for a number of generations until a convergence criterion is achieved. There are three selection processes: roulette wheel, normalized geometric and tournament selection. To increase the speed of convergence an elitist strategy can be adopted. Support vector machines may be used to create a nonlinear time-series model that provides an alternative to these linear AR models. The support vector machine autoregressive method is superior to traditional linear AR in both its ability to handle nonlinear dynamics as well as the structure of the model. Specifically, the support vector approach compares each new testing point to the entire training set whereas the traditional AR model finds a simple linear relationship to best describe the entire training set, which is then used on the testing data.

2. Neural Networks Many damage detection schemes utilize neural networks to detect, localize, and quantify damage in structures. Rytter and Kirgegaard (1997) evaluate two neural networks for damage assessment, namely the multilayer perceptron (MLP) network with back propagation and the radial basis

function (RBF) network. Both the MLP and RBF networks consist of one hidden layer in addition to the input and output layers. A finite element model of a four-story building is used for this work. Random damage states are generated through finite element simulations, and are essentially stiffness reductions in beams and columns. It is concluded that the MLP network demonstrates the possibility of being used in connection with vibration-based inspection, whereas the RBF network completely fails. However, the authors caution that the performance of the RBF network is highly dependent on an appropriate selection of damage cases used in the training. Loh and Huang (1999) observe that a nonlinear neural network can be regarded as a general type of nonlinear auto-regressive moving-average (NARMA) model, which is a representation of a nonlinear discrete time series. Based on this observation, Loh and Huang model the seismic response of a half-scale five-story steel frame structure with three different networks. The structure is subjected to different levels of peak accelerations. These excitations range from 20% of the El Centro peak ground motion, to 20%, 40%, and 60% of the peak ground motion recorded in the Kobe earthquake. The second-floor response of the steel frame structure is predicted using the responses of the first and third floors. The first model is a linear model where the current response of the second floor depends on the past responses at the first and third floors with a time lag of 4. In the second network, the time lag is increased from 4 to 8. The third network differs from the second one in that the current second floor acceleration nonlinearly. Chan et al. (1999) construct an auto associative neural network to detect changes of cable tension on the Tsing Ma suspension bridge in Hong Kong. The Tsing Ma Bridge has a main span of 1,377 m and an overall length of 2,160 m, making it the world’s longest suspension bridge. The authors note that the main cables of a suspension bridge are the most crucial components, and small variations in the cable tensions affect the internal force distributions in the deck and towers influencing bridge alignment. Liu, Sana, and Rao (1999) use an autoregressive with exogenous inputs (ARX) model of a cantilever aluminum beam to extract vibration signatures, and employ multilayer, feed-forward neural networks (MFFNN) to locate damage and estimate its size. Table 1. Neural network models for SHM of Civil Structures Author Rytter and Kirgegaard (1997)

Modelling Two Neural Network MLP (multilayer perceptron RBF (radial basis function)

Data source FEM modelling

Type of structure 4 story RCC building

Results MLP could predict behavior when trained with past data but RBF cannot

Loh and Huang (1999)

Time series modelling using nonlinear auto-regressive moving-average (NARMA) model

Seismic response of FEM model under Peak ground acceleration of El centro and Kobe earthquake

half-scale five-story steel frame structure

Could predict the structure’s behavior during earthquakes when properly trained

Faravelli and Pisano (1997)

feed-forward neural network (FFNN) trained with a back-propagation algorithm

FEM modelling

2- D nine-bay truss structure

Predicts with sufficient accuracy but performance of the network varies depending on which element damage initiates first

Chan et al. (1999)

Multilayer feed-forward network with a bottleneck layer in the middle (Auto associative)

Frequency measures from Tsing Ma suspension bridge in Hong Kong

detect changes of cable tension on the Tsing Ma suspension bridge

combination of the auto associative network and novelty analysis has the ability of discerning between changes caused by damage and natural variations of the system

1. 2.

Liu, Sana, and Rao (1999)

Neural autoregressive with exogenous inputs (NARX)

Vibration data using piezoelectric patches as sensors

cantilever aluminum beam with holes

Damage location is estimated well whereas damage size while extents differs

Liu and Sun (1997)

Five neural networks are used to monitor local dynamic characteristics, elongation curves as inputs and stiffness reduction as outputs

finite element model of the bridge consisting of 30 uniform beam element

simply supported three-span bridge

damage influences the extreme values of the elongation curves in different ways, depending on the damage location in relation to the elongation curve

Hermann and Streng (1997)

Neural network with different learning rates, network types, reduction techniques of network topologies, and dimension analysis

FEM modelling of Six bar truss

six-bar truss structure subjected to both horizontal and vertical loads

success of the neural network analysis, the training patterns should include a sufficient number of characteristics that distinguish between undamaged and damaged states of the structure

Hanagud and Luo (1997)

three-layer feed-forward neural networks,

frequency response functions as inputs and delamination information as outputs

composite plate under dynamic load

delamination and the stiffness loss are successfully identified for different damage cases

Jenq and Lee (1997)

back-propagation neural network with an adaptive learning rate

finite element model to predict hole defect sizes and locations

glass fiber reinforced plastic (GFRP) composite laminated beams

averaged errors for predicting the hole diameter and location are 7% and 6%, respectively

Nakamura et al. (1998)

FFNN with Two hidden layers, one with 15 nodes and the other with 10 nodes, are used.

Velocity transducers are located at each story, and each story has its own network

seven-story steel moment resisting frame building

Although the method locates the story where damage occurs, the actual damaged connection of the story is not identified with the method

Masri et al. (2000)

Two different neural networks are investigated in this study. The first network approximates the four system accelerations as a function of the input force excitation, four displacements, and four velocities. The second network maps the displacements, velocities, and accelerations to the system force excitation.

Vibration signals

Undefined mechanical systems

First network produces the output errors increased by about 18% and 44% for the two damage cases investigated. The second network produces a similar result although a smaller dispersion in the prediction errors is observed

Barai and Pandey (1997)

Two different neural network, first network is a conventional multilayer perceptron(MLP) neural network, and time delay of the dynamic response is introduced in the second timedelay neural network

Vibration signals

railway bridge

Performance of the timedelay neural network is found to be generally better than that of the conventional MLP neural network.

Feng and Bahng (1999)

neural network with a standard backpropagation algorithm

finite element model was constructed to predict baseline vibration characteristics of the bridge, and the predicted responses were compared with the vibration test data taken from the scale model

CFRP jacketed columns

output pattern consists of correction coefficients of element stiffness for the columns and was found consistent

Choi and Kwon (2000)

Two separate neural networks were developed for damage localization. The first network determined whether the

FEM modelling of Truss Bridge identified eight truss

steel truss bridge

Damage diagnosis using the neural network system was performed successfully, but the

damage was located either to the left or to the right of the bridge’s midpoint. Binary output of first network act as input for second

members subjected to high stress level

performance of the second network was sensitive to the number of nodes chosen for network construction

Maseras-Gutierrez et al. (1998)

piezoceramic sensors and neural networks to detect impact in composite materials

piezoceramic sensors

Chang et al. (2000)

Iterative Neural network with initial training data sets consisting of assumed structural parameters as target outputs and their corresponding dynamic characteristics as inputs

FEM model of T beam

340 × 340 × 2.5 mm test specimen was a laminate consisting of a carbon fiber fabric and a toughened epoxy resin clamped-clamped T beam

Results from the networks showed that the impact location and magnitude could be predicted with acceptable error.

Verified using Experimental and numerical data

Table 2. Genetic Algorithms model for SHM of Civil Structures Author

Modelling

Data source

Type of structure

Results

Ruotolo and Surace (1997a and b)

Genetic optimization technique.

FEM model to obtain associated frequencies

Beam, to determine crack location and depth.

genetic optimization proved to be prohibitive because increase in complexities in the large structure so this method was limited to small structures such as beams only

Ruotolo and Surace (1998)

Compared simulated annealing, genetic algorithms, and eigensensitivity analyses

FEM model

Four story steel frame structure

Simulated annealing detects all damage cases with the fewest runs.

Mares et al. (1999)

a cost function as a function of the difference between measured and analytical transmissibility and minimize the cost function through a genetic algorithm

Numerical simulation of the dynamic behavior of a four-story building

four-story building

concluded that damage in braces is more easily identified that damage in columns through this model

Krawczuk et al. (2000)

genetic algorithm to identify and locate damage in a laminated composite beam

Numerical model two cases of delamination are considered, first with a delamination equal to 35% of the total beam length, and the second case with delamination 15%

genetic algorithm approach correctly determines damage for both cases

Friswell et. al (1998)

two level genetic algorithm approach for detecting damage using vibration data

Vibration Data

The method is demonstrated on a simulated beam example and an experimental plate example

algorithm is robust to systematic errors in the measured data, demonstrated in simulation by the addition of a discrete mass and experimentally by using a crude model for the plate

Maity and Tripathy (2005)

genetic algorithm (GA) to detect and assess the structural damage from changes

A laboratory tested data has been used

The technique has been applied to a

The outcomes show that this method can detect and estimate the amount

in natural frequencies

to verify the proposed algorithm

cantilever beam and a plane frame

of damages with satisfactory precision.

Pawar and Ganguli (2005)

genetic fuzzy system, it combines the uncertainty representation characteristics of fuzzy logic with the learning ability of genetic algorithm

Frequency analysis

composite matrix cracking in a thinwalled hollow circular cantilever beam

observed that the success rate of the genetic fuzzy system in the presence of noise is dependent on crack density

Chou and Ghaboussi (2001)

Genetic Algorithms to identify the changes of the characteristic properties of structural members such as Young's modulus and cross-sectional area, which are indicated by the difference of measured and computed responses

Laboratory data

Na et al. (2011)

Genetic algorithm using the structural flexibility matrix with dynamic analyses

numerical analyses using OpenSees

Shear Building

The validity of the proposed damage evaluation method is demonstrated through numerical analyses using OpenSees.

Betti et al. (2015)

Artificial neural networks and genetic algorithms for structural damage identification

Accelerometers reading under vibration of structure and FEM model

A reduced scale three-storey steel spatial frame

Results of the experimentation were compared with the results of the optimization algorithm in order to verify its ability to match the actual damage.

Mohan et al. (2014)

Proposed particle swarm optimization (PSO) and genetic algorithm (GA) in damage assessment of structures

Frequencies analysis

beam, plane and space truss

The methodology presented establishes the PSO as robust and competent tool over GA for crack identification

Rao et al. (2005)

Genetic Algorithms for locating and quantifying the damage in structural members using the concept of residual forces

Experimental data

plane truss, a cantilever Euler– Bernoulli beam and a portal truss

Approach was verified with analytical solution

He and Hwang (2006)

Adaptive real-parameter genetic algorithm with simulated annealing, is proposed to detect damage occurrence in beam-type structures.

Finite element Modelling using ANSYS

Beam type structures with different boundary conditions

it is demonstrated that the proposed algorithm is efficient in flexural stiffness damage identification for beam-type structures under free of noise condition

Perera et al. (2007)

Genetic algorithm for damage detection multi-objective optimization problem

Simulated beams and experimental data from the vibration tests of a beam

Beam Model

promising for locating and quantifying damaged elements and considerably improves predictions based only on modal flexibility parameters

proposed method is able to detect the approximate location of the damage

3. Genetic Algorithms Ruotolo and Surace (1997a and b) solved the inverse problem of dictation of position of crack and

depth of the crack in a beam using genetic optimization technique. A finite element model is created to obtain the associated frequencies and then cost function represented associated and modal frequencies. Although performance of genetic algorithm was similar to that of annealing scheme and both converging to global solution in 10,000 evaluation still genetic optimization proved to be prohibitive because increase in complexities in the large structure so this method was limited to small structures such as beams only. Ruotolo and Surace (1998) compared simulated annealing, genetic algorithms, and eigensensitivity analyses for detecting 4 damage scenarios in a finite element model of a four-story steel frame structure. Simulated annealing detects all damage cases with the fewest runs. Mares et al. (1999) construct a cost function as a function of the difference between measured and analytical transmissibility and minimize the cost function through a genetic algorithm. The authors numerically simulate the dynamic behavior of a four-story building with three degrees of freedom (DOFs) on each floor: one in-plane rotation and two orthogonal lateral displacements. They refer to these simulations as measured data. In this paper two-step optimization procedure has been used to obtain the global minimum of cost function. First, a genetic search is performed on the cost function so that the downhill of the global minimum is determined quickly. Then, a classical gradient-based algorithm is run to refine the solution. By performing this two-step optimization, the accuracy of the solution is improved, and global minima are successfully identified, thus avoiding local minima. Different damage scenarios are considered and simulated by reducing the stiffness of columns and/or braces in the building model. It is concluded that damage in braces is more easily identified that damage in columns through this model. Krawczuk et al. (2000) applied a genetic algorithm to identify and locate damage in a laminated composite beam. Author has defined damage as delamination and has validated their method with a numerical model. Author has used an objective function, θ, which is based on changes in natural frequencies and the Damage Location Assurance Criterion (DLAC). The position and size of the damage is estimated by maximizing this objective function. For these two cases of delamination are considered, first with a delamination equal to 35% of the total beam length, and the second case with delamination 15% of the total length of the beam. Only the first 4 natural frequencies are considered in the objective function. The genetic algorithm approach correctly determines damage for both cases. These results are compared with those from a neural network in which the first 4 natural frequencies are used as inputs, and the onset location of delamination, delamination end location, and the number of the delaminated layers are used as outputs. Friswell et. al (1998) have applied two level genetic algorithm approach for detecting damage using vibration data. The objective is to identify the position of one or more damage sites in a structure, and to estimate the extent of the damage at these sites. The genetic algorithm is used to optimize the discrete damage location variables. For a given damage location site or sites, a standard eigen sensitivity method is used to optimize the damage extent. This two-level approach incorporates the advantages of both the genetic algorithm and the eigen sensitivity methods. The method is demonstrated on a simulated beam example and an experimental plate example. A combined genetic algorithm and eigen sensitivity method has been used to identify the location and magnitude of damage from measured vibration data. Essentially, the genetic algorithm is used to identify the damage located and the eigen sensitivity is used to identify the damage extent. Damage at one and two sites have been successfully located in the simulated example of a cantilever beam. Damage was also successfully location in an experimental cantilever plate. The algorithm is robust to systematic errors in the measured data, demonstrated in simulation by the addition of a discrete mass and experimentally by using a crude model for the plate. One feature of the numerical example given is that genetic diversity was reduced as the generation number increased.

Alternatively, this could be viewed as the convergence to a uniform ``optimum'' population, although it would not be guaranteed that this population would be the global optimum. This may be a feature of the small number of genes associated with each member of the population. Maity and Tripathy (2005) used genetic algorithm (GA) to detect and assess the structural damage from changes in natural frequencies. A method is presented to detect and assess the structural damage from changes in natural frequencies using Genetic Algorithm (GA). Using the natural frequencies of the structure, it is possible to formulate the inverse problem in optimization terms and then to utilize a solution procedure employing GA to assess the damages. The technique has been applied to a cantilever beam and a plane frame, each one with different damage scenario to study the efficiency of the developed algorithm. A laboratory tested data has been used to verify the proposed algorithm. The study indicates the potentiality of the developed code to solve a wide range of inverse identification problems in a systematic way. The outcomes show that this method can detect and estimate the amount of damages with satisfactory precision. Pawar and Ganguli (2005) applied the genetic fuzzy system to detect the matrix crack in thinwalled composite structures based on changes in natural frequencies. In the present paper author has implemented a model of composite matrix cracking in a thin-walled hollow circular cantilever beam using an effective stiffness approach. Such structures are used to model connecting shafts and helicopter tail boom, for example, because of their high stiffness-to-weight ratios and excellent crashworthiness characteristics. The effect of variation in crack density on the fundamental frequency, for various combinations of composite is studied. Using these change in frequencies due to matrix cracking, a genetic fuzzy system for crack density and crack location detection is generated. The genetic fuzzy system combines the uncertainty representation characteristics of fuzzy logic with the learning ability of genetic algorithm. It is observed that the success rate of the genetic fuzzy system in the presence of noise is dependent on crack density (level of damage), number of 90 plies, angle of constraining layer, and noise level. It is found that the genetic fuzzy system shows excellent damage detection and isolation performance, and is robust to presence of noise in data. Chou and Ghaboussi (2001) considered static measurements of displacements at few degrees of freedom (DOFs) as a parameter to identify the changes of the characteristic properties of structural members such as Young's modulus and cross-sectional area, which are indicated by the difference of measured and computed responses. In order to avoid structural analyses in fitness evaluation, the displacements at unmeasured DOFs are also determined by GA. This proposed method is able to detect the approximate location of the damage, even when practical considerations limit the number of on-site measurements to only a few. Na et al. (2011) have introduces a new damage evaluation method that identifies the structural damage in a shear building based on a genetic algorithm using the structural flexibility matrix with dynamic analyses. The proposed method enables the deduction of the extent and location of structural damage, even when there is insufficient data on the dynamic characteristics and insufficient accurate measurements of the structural stiffness and mass. The validity of the proposed damage evaluation method is demonstrated through numerical analyses using OpenSees. Betti et al. (2015) have considered a combined approach based on artificial neural networks and genetic algorithms for structural damage identification. A reduced scale three-storey steel spatial frame was instrumented by a series of 12 accelerometers and progressively damaged by cutting one of its columns just above the first storey. Accelerations induced by ambient vibrations were recorded as the frame was progressively damaged, and the deepness of the cut was taken as the entity of the damage. At every damage level the modal properties (natural frequencies and modal

shapes) of the steel frame were evaluated through a neural network based approach. Subsequently, two error functions that measure the differences between the experimental results and those calculated from a finite element model of the steel frame were defined and a genetic algorithm was employed for damage detection. Results of the experimentation (where damage was known as both location and extent) were compared with the results of the optimization algorithm in order to verify its ability to match the actual damage. Mohan et al. (2014) have proposed particle swarm optimization (PSO) and genetic algorithm (GA) in damage assessment of structures and there efficiency has been tested on structures like beam, plane and space truss. The results show the effectiveness of PSO in crack identification and the possibility of implementing it in a real-time structural health monitoring system for aircraft and civil structures. The methodology presented establishes the PSO as robust and competent tool over GA for crack identification using changes in natural frequencies. Rao et al. (2005) proposed a method of locating and quantifying the damage in structural members using the concept of residual forces. To describe the damage in a structure, finite element (FE) models are parameterised by structural stiffness reduction parameters. The damage parameters are determined by minimising a global error derived from dynamic residual vectors, which are obtained by introducing a simulated “experimental” data into the eigenproblem. An eigenvalue prediction algorithm along with normalised residual function is employed to formulate the objective function. Two-point crossover binary coded genetic algorithm (GA) with tournament selection approach is adopted in minimising the objective and optimum set of stiffness reduction parameters are predicted. Current structural defect-identification scheme is verified and assessed using an analytically derived plane truss, a cantilever Euler–Bernoulli beam and a portal truss. He and Hwang (2006) proposed an algorithm, which combined an adaptive real-parameter genetic algorithm with simulated annealing, is proposed to detect damage occurrence in beam-type structures. The proposed algorithm uses the displacements of static response and natural frequencies of modal analysis, which are obtained by finite element software ANSYS. There are three different kinds of beam structures to verify the performance of the proposed algorithm. These three cases have different boundary conditions and different damage scenarios. From the results, it is demonstrated that the proposed algorithm is efficient in flexural stiffness damage identification for beam-type structures under free of noise condition. Even under the case of noise, the results show that the searched solutions are still in reasonable precision. Perera et al. (2007) has proposed an objective function dependent on modal flexibility is combined with another function able to predict damage location. The problem is formulated as a multiobjective optimization problem of finding the Pareto optimal damage distribution that simultaneously minimizes the two objective functions. Then, a genetic algorithm suitable for solving general multiobjective optimization problems is applied. The effectiveness of the proposed algorithm is illustrated for simulated beams and by using directly experimental data from the vibration tests of a beam. It has been verified that the proposed procedure is very promising for locating and quantifying damaged elements and considerably improves predictions based only on modal flexibility parameters. 4. Support Vector Machines (SVM) Li and Yu (2015) in this paper, author proposed an on-line SVM (OLSVM) for big data stream classification, and applied this classier for structural health monitoring of building. This OLSVM overcomes the problems of normal SVM, such as slow training, huge dimension of the kernel, and low classification accuracy. The main contribution of this method is we use recursive method to

calculate the kernel. The experimental results demonstrate that our approach has good classification accuracy while the training data are big and obtained on-line. In this paper, the velocity and position estimations are evaluated in a shaking table. The accelerometer used is Summit Instruments 13203B, which is mounted on the SDOF mechanical structure. The sensitive axis of the accelerometer is mounted parallel to the ground to measure the structure acceleration. A linear magnetic encoder (LM15) position sensor with a resolution of 50μm is used for verifying the estimated position data. The building structure base is mounted on the shaking table. Accuracy of OLSVM classifier is reported as 96.2%.

Meruane and Heylen (2011) have used hybrid real-coded Genetic Algorithm with damage penalization to locate and quantify structural damage. Here, the set-up of the Genetic Algorithm operators and parameters is addressed, providing guidelines to their selection in similar damage detection problems. The performance of five fundamental functions based on modal data is studied. In addition, this paper proposes the use of a damage penalization that satisfactorily avoids false damage detection due to experimental noise or numerical errors. A tri-dimensional space frame structure with single and multiple damages scenarios provides an experimental framework which verifies the approach. The method is tested with different levels of incompleteness in the measured degrees of freedom. The results show that this approach reaches a much more precise solution than conventional optimization methods. A scenario of three simultaneous damage locations was correctly located and quantified by measuring only a 6.3% of the total degrees of freedom.

References Barai, S.V., and Pandey, P.C. (1997) “Time-Delay Neural Networks in Damage Detection of Railway Bridges,” Advances in Engineering Software, Vol. 28, pp. 1–10. Bulut, A., Singh, A. K., Shin, P., Fountain, T., Jasso, H., Yan, L., and Elgamal, A., 2005, “RealTime Nondestructive Structural Health Monitoring Using Support Vector Machines and Wavelets,” Proc. SPIE, 5770, pp. 180–189. Betti, M, Facchini, L., and Biagini, P. (2015), Damage detection on a three storey steel frame using artificial neural networks and genetic algorithms. Meccanica, 10.1007/s11012-014-0085-9, 875886. B.-H. Koh, P. Dharap, S. Nagarajaiah, and M. Phan, “Real-Time Structural Damage Monitoring by Input Error Function,” Aiaa J., vol. 43, no. 8, pp. 1808–1814, Aug. 2005. Chang, C.C., Chang, T.Y.P, Xu, Y.G., and Wang, M.L. (2000) “Structural Damage Detection Using an Iterative Neural Network,” Journal of Intelligent Material Systems and Structures, Vol. 11, pp. 32–42. C. Farrar, K.Worden (2013) Structural health monitoring - A machine learning perspective, John Wiley & Sons, 2013. Chou, J. H., Ghaboussi, J. (2001) Genetic Algorithm in structural damage detection, Computers & Structures, vol. 79, no. 14, pg 1335-1353 Chan, T.H., Ni, Y.Q., and Ko, J.M. (1999) “Neural Network Novelty Filtering for Anomaly Detection of Tsing Ma Bridge Cables,” Structural Health Monitoring 2000, Stanford University, Palo Alto, California, pp. 430–439.

Choi, M.Y., and Kwon, I.B. (2000) “Damage Detection System of a Real Steel Truss Bridge by Neural Networks,” Smart Structures and Materials 2000: Smart Systems for Bridges, Structures, and Highways, Proceedings of SPIE, Vol. 3,988, Newport Beach, California, pp. 295–306. Farrar, C.R., Sohn, H., Hemez, F.M., Anderson, M.C., Bement, M.T., Cornwell, P.J., Doebling, S.D., Lieven, N., Robertson, A.N., and Schultze, J.F. (2003) “Damage Prognosis: Current Status and Future Needs,” Los Alamos National Laboratory report, LA-14051-MS. Faravelli, L., and Pisano, A.A. (1997) “Damage Assessment Toward Performance Control,” Structural Damage Assessment Using Advanced Signal Processing Procedures, Proceedings of DAMAS ‘97, University of Sheffield, UK, pp. 185–198. Feng, M., and Bahng, E. (1999) “Damage Assessment of Bridges with Jacketed RC Columns Using Vibration Test,” Smart Structures and Materials 1999: Smart Systems for Bridges, Structures, and Highways, Proceedings of SPIE, Vol. 3,671, pp. 316–327. Friswell, M. I., Penny, J. E. T, Garvey, S. D. (1998) A combined genetic and eigensensitivity algorithm for the location of damage in structures, Computers & Structures vol. 69, no. 5, pg. 547– 556. G.Park, et.al., Time series predictive models of piezoelectric active sensing for SHM applications, Health Monitoring of Structural and Biological Systems,Vol.7650, article id. 765004, 12 pp.12, 2010. He, R. S., Hwang, S. F., (2006), Damage detection by an adaptive real-parameter simulated annealing genetic algorithm, Computers & Structures, vol 84, no. 31-32, pg 2231–2243. Hermann, H.-G., and Streng, J. (1997) “Problem-Specific Neural Networks for Detecting Structural Damage,” Structural Health Monitoring, Current Status and Perspectives, Stanford University, Palo Alto, California, pp. 267–278. Hanagud, S., and Luo, H. (1997) “Damage Detection and Health Monitoring Based on Structural Dynamics,” Structural Health Monitoring, Current Status and Perspectives, Stanford University, Palo Alto, California, pp. 715–726. Jenq, S.T., and Lee, W.D. (1997) “Identification of Hole Defect for GFRP Woven Laminates Using Neural Network Scheme,” Structural Health Monitoring, Current Status and Perspectives, Stanford University, Palo Alto, California, pp. 741–751. K. Worden, G. Manson (2007) the application of machine learning to structural health monitoring, Philosophic Transactions A DOI: 10.1098/rsta.2006.1938 Krawczuk, M., Ostachowicz, W., and Kawiecki, G. (2000) “Detection of Delaminations in Cantilevered Beams Using Soft Computing Methods,” European COST F3 Conference on System Identification and Structural Health Monitoring, Madrid, Spain, pp. 243–252. Loh, C.H., and Huang, C.C. (1999) “Damage Identification of Multi-Story Steel Frames Using Neural Networks,” Structural Health Monitoring 2000, Stanford University, Palo Alto, California, pp. 390–399.

Liu, P., Sana, S., and Rao, V.S. (1999) “Structural Damage Identification Using Time- Domain Parameter Estimation Techniques,” Structural Health Monitoring 2000, Stanford University, Palo Alto, California, pp. 812–820. Liu, P.L., and Sun, S.C. (1997) “The Application of Artificial Neural Networks on the Health Monitoring of Bridges,” Structural Health Monitoring, Current Status and Perspectives, Stanford University, Palo Alto, California, pp. 103–110. Masri, S.F., Smyth, A.W., Chassiakos, A.G., Caughey, T.K., and Hunter, N.F. (2000) “Application of Neural Networks for Detection of Changes in Nonlinear Systems,” Journal of Engineering Mechanics, July, pp. 666–676. Maseras-Gutierrez, M., Staszewski, W., Found, M., and Worden, K. (1998) “Detection of Impacts in Composite Materials Using Piezoceramic Sensors and Neural Networks,” Smart Structures and Materials 1999: Smart Structures and Integrated Systems, Proceedings of SPIE, Vol. 3,329, pp. 491–497. Mares, C., Ruotolo, R., and Surace, C. (1999) “Using Transmissibility Data to Assess Structural Damage,” Damage Assessment of Structures, Proceedings of the International Conference on Damage Assessment of Structures (DAMAS 99), Dublin, Ireland, pp. 236–245. Maity, D. and Tripathy, R.R. (2005). Damage assessment of structures from changes in natural frequencies using genetic algorithm. Structural Engineering and Mechanics, 19, 21–42. 24. Merunane, V. and Mahu, J. (2014), Real time structural damage Assement using Artifical Neural Networks and Antiresonant Frequencies, Shock and Vibration, 10.115/2014/653279, 1-14. Meruane, V., Heylen, W. (2011) An hybrid real genetic algorithm to detect structural damage using modal properties, Mechanical Systems and Signal Processing,vol. 25, no. 5, pg. 1559–1573 Mulchandani, H. K., & Singh, A. P. (2015, December). Prediction of live storage of water resources using neural network time series model, case study on Mettur dam. In HYDRO 2015 INTERNATIONAL: 20th International Conference on Hydraulics, Water Resources and River Engineering (Vol. 20). HYDRO 2015 INTERNATIONAL. Mulchandani, H. K., Kumar, M., & Rawat, S. (2016, August). Application of direct displacement based design for shallow and deep foundations. In Sixth International Conference on Recent Advances in Geotechnical Earthquake Engineering and Soil Dynamics August 1 –6, 2016, IIT Roorkee Extension Centre, 20 Knowledge Park II, Greater Noida, India. Nakamura, M., Masri, S.F., Chassiakos, A.G., and Caughey, T.K. (1998) “A Method for NonParametric Damage Detection Through the Use of Neural Networks,” Earthquake Engineering and Structural Dynamics, Vol. 27, pp. 997–1,010. Pawar, P.M. and Ganguli, R. (2005). Matrix crack detection in thin-walled composite beam using genetic fuzzy system. Journal of Intelligent Material Systems and Structures, 16, 395–409. P.Taha, J. Lucero (2005) Damage identification for structural health monitoring using fuzzy pattern recognition, Engineering Structures, Vol.27, 1774-1783, 2005.

Perra, R., Ruiz, A., and Manzano, C. (2007), an evolutionary multiobjective framework for structural damage localization and quantification, Engineering Structures, 10.1016/j.engstruct.2007.01.003, 2540-2550 Perera, R., Ruiz, A, and Manzano, C. (2009), performance assessment of multi criteria damage identification genetic algorithms, Computers & Structures, 10.1016/j.compstruc.2008.07.003, 120127. Rao, M. A., Srinivas, J., Murthy, B.S.N (2004) Damage Detection in vibrating bodies genetic algorithms, Computers & Structures, doi:10.1016/j.compstruc.2004.01.005

Na, C., Kim, S. P., Kwak, H. G (2001) Structural damage evaluation using genetic algorithm, Computer & Structures, doi:10.1016/j.jsv.2011.01.007 Ruotolo, R., and Surace, C. (1997a) “Damage Assessment of Multi-Cracked Beams Using Combinatorial Optimisation,” Structural Damage Assessment Using Advanced Signal Processing Procedures, Proceedings of DAMAS ‘97, Univeristy of Sheffield, UK, pp. 77–86. Ruotolo, R., and Surace, C. (1997) “Damage Assessment of Multiple Cracked Beams: Results and Experimental Validation,” Journal of Sound and Vibration, Vol. 206, No. 4, pp. 567–588. Ruotolo, R., and Surace, C. (1998) “Diagnosis of Damage in a Steel Frame,” Proceedings of the International Modal Analysis Conference, pp. 609–615. Rytter, A., and Kirkegaard, P. (1997) “Vibration Based Inspection Using Neural Networks,” Structural Damage Assessment Using Advanced Signal Processing Procedures, Proceedings of DAMAS ‘97, University of Sheffield, UK, pp. 97–108. S.Tesfamarian, Z.Liu (2010) Earthquake induced damage classification for reinforced concrete buildings, Structural Safety, Structural Safety, Vol.32, 154-164. S.Thenozhi, W.Yu, A.López, X.Li (2012) Structural Health Monitoring of Tall Buildings with Numerical Integrator and on-line Classification, Mathematical Problems in Engineering, Vol.2012, Article ID 212369, 15 pages. S.Doebling, C.Farrar, M.Prime (1998) Summary review of vibration-based damage identification methods, The Shock and Vibration Digest, Vol.30 91-105, 1998. Shimada, M., Mita, A., and Feng, M. Q., 2006, “Damage Detection of Structures Using Support Vector Machines Under Various Boundary Conditions,” Proc. SPIE, 6174, pp. 61742K. S.C. Mohan , Amit Yadav , Dipak Kumar Maiti , Damodar Maity , (2014) "A comparative study on crack identification of structures from the changes in natural frequencies using GA and PSO", Engineering Computations, Vol. 31 Iss: 7, pp.1514 – 1531 Worden, K., and Manson, G., 2007, “The Application of Machine Learning to Structural Health Monitoring,” Philos. Trans. R. Soc. London, Ser. A, 365, pp. 515–537. Worden, K., and Lane, A. J., 2001, “Damage Identification Using Support Vector Machines,” Smart Mater. Struct., 10, pp. 540–547.