Damage Classification on the Z24-Bridge using Structure Health Monitoring (SHM)

Damage Classification on the Z24-Bridge using Structure Health Monitoring (SHM)

Background

Civil engineering includes buildings, bridges, furnace, pipelines and transitions lines. [1] Since they have built and operated, during their lifetimes, infrastructures are subjected to destructive natural and harsh environmental conditions as well as the loadings from human activities. Since any of these phenomena will cause decrease or damage the functional lifespan and strength of the structures, maintaining safe civil infrastructures for daily using becomes important to the well-being of all of us. [2]

Project Aim

Z24-Bridge is located in Switzerland, before it was damaged artificially, it has been monitored around 1 year. An approach of monitoring the structure, which called structural health monitoring (SHM), this monitoring has been used widely in the field of Civil Engineering. The aim of this project is to perform Structure Health Monitoring (SHM) of the Z24-Bridge and SHM is performed by using classifiers to classify the damaged and undamaged data. This project will look at different classifiers and evaluate their performance for SHM.

Project Objective

Basic Objectives:

1. Review literature on classification techniques.

2. Review literature on SHM and Z24-Bridge.

3. Apply a classifier to toy data.

Advanced Objectives:

4. Apply a classifier to the Z24 data (real data).

5. Evaluate the results to see if SHM is possible.

6. Investigate the effect of temperature on Z24 and SHM for Z24

Literature Review

SHM in general

Bridge Structural Health Monitoring (SHM) has rapidly become one of the main interests in civil engineering field. It can detect the changes to structural properties (i.e. stiffness) caused by damage (i.e. corrosion) as well as the structural responses changes (i.e. acceleration responses) to ambient motions. [2] Modal analysis algorithms applied to the vibration responses acquired through SHM provide the modal properties (i.e. natural frequency, modal shape and damping ratio) that will change with the changes in stiffness indicating possible existence of damage. [3]

Structural Health Monitoring (SHM) aimed to give a diagnosis of the “state” of different materials at each point in the life cycle of the structure. [3] SHM is depending on the repeated observation of the features, which is sensitive to a certain type of structural damage. Among many of the possible features, dynamic characteristics (especially natural frequencies), are often selected, since they depend on the local stiffness and the global of the structure as well as its boundary conditions. [4]

Farrar and Worden [7] defined SHM as a statistical pattern recognition (SPR) process consisting of four parts:

(1) Operational evaluation;

(2) Data acquisition, cleansing and merging

(3) Feature extraction;

(4) Development of the statistical models

This report will investigate classifiers (by using SHM) on toy data with the intention of using them on the real Z24 bridge dataset.

Z24-Bridge

Z24-Bridge, which is located in Switzerland, this bridge is a highway overpass of the national highway A1, and linking Zurich and Bern. [2] Z24-Bridge is a pre-stressed bridge with a total length of 58m’ three-span structure (2 side spans of 14 m and a main span of 30 m) This bridge is crossed the national highway A1 at a slightly inclined angle. A two-cell closed box constitute the upper-structure, and these two unit closed box beams with tendons in the three webs. The tendons over the main piers are more distributed than in the middle of the spans. Both the main piers are concrete diaphragms, that are fully connected to the superstructure. Three abutment columns were fixed at both ends. Both the ends of the bridge decks were extended to protect the anchor head [2]. (Shown as Figure 1). Although with the bridge, there were no known structural problems, it also needed to be demolished due to a new railway request next to the highway.

From 11 November 1997 to 11 September 1998, Z24-bridge has been monitored. The aim of the Structural Health monitoring (SHM) system is to provide both vibration and environmental data. [2]

Figure 1: Z24 Bridge, Switzerland [2]

KNN/SVM Classifier Model

Data classification is the process of organizing data by related categories to effectively protect and use data. [12] The classification process makes data more effectively and easy to retrieve and locate. Data classification is important in terms of risk management, data security and compliance. [11]

Nate [12] mentioned that data classification involves tagging the data, therefore, makes the data easily trackable and searchable the main advantages. It also eliminates multiple data duplication, which reduces the backup costs and the storage, as well as speed up the search process.

KNN Classifier Model

The KNN algorithm is one of the most versatile and robust classifiers, this algorithm is often used as a benchmark for more complex classifiers such as Support Vector Machines (SVM) and Artificial Neural Networks (ANN). [12] Although KNN algorithm is simplicity, it can outperform the classifier which is more powerful. Also, it can used in a variety of applications, such as data compression and economic forecasting as well as genetics. [13]

KNN classifier is based on grouping of closest data training points in the feature space under consideration. By majority of voting to the nearest neighbor points to complete the classes. K value in KNN algorithm indicates Euclidian. [13] The KNN classifier is also instance-based and a non- parametric learning algorithm.

Support Vector Machines (SVM)

In general, Support Vector Machines is considered to be a classification approach, but it can be used for both types of regression problems and classification. It can easily to deal with the multiple continuous and the categorical variables. In multidimensional space, SVM classifier can construct hyperplanes to separate different classes. [15] SVM classifier generates an optimal hyperplane by using an iterative manner, and this is use for minimizing the errors. The core idea of SVM algorithm is to find out the maximum marginal hyperplane (MMH) and divides the dataset into classes. [5]

An SVM model is to represent the examples as points in space, and the mapping divides the examples of the individual categories by a widely clear gap. New examples will be mapped to the same space and predict which belongs to a category based on which edge they fall on. [6]

Work Progress

Figure 3: plot by using KNN classifier [Note 1]

Figure 4: plot by using SVM classifier function

First to set up 200 points between 2-4 by random, and then using the KNN classifier function to figure it out, and in this function I keep the number of K is 1.

XY1 = 2+ rand(100,2);

XY2 = 3+ rand(100,2);

XY = [XY1;XY2];

Classify =[zeros(100,1);ones(100,1)];

group = [repmat(1,100,1); repmat(2,100,1)];

gscatter(XY(:,1),XY(:,2),group,’rg’,’+*’);

legend(‘Training group 1’, ‘Training group 2’);

hold on;

sample = 2+ 2*rand(100,2);

% Classify the sample using the nearest neighbor classification

c = knnclassify(sample, XY, group,1,’cityblock’,’nearest’);

gscatter(sample(:,1),sample(:,2),c,’mc’); hold on;

legend(‘Training group 1′,’Training group 2’, …

       ‘Data in group 1′,’Data in group 2’);

The MATLAB code by using SVM classifier function:

figure;

plot(XY1(:,1),XY1(:,2),’r+’);

hold on;

plot(XY2(:,1),XY2(:,2),’g*’);

SVM = svmtrain(XY,Classify,’showplot’,true);

svmclassify(SVM,sample,’showplot’,true);

These two plotting methods are completely different. KNN classifier calculate all of the test point to divided the classes. But SVM classifier was different, it just uses the support vectors to find a linear separation (divided the classes), And then judge the testing points’ location to get the classes. These three points which has been circled in both figures need to be in the second class, therefore, SVM classifier got the wrong results. By comparing the results, KNN algorithm is better than SVM algorithm.

Error & Conclusion

KNN has some advantages such as: it can detect the non-linear or the linear distributed data, and it is non-linear automatically as well as it can perform a perfect trend with a lot of data points. [13]

On the negative side, KNN needs to be carefully to adjust, and the choice of the K value and the used metric (distance) are critical. If there are many testing points and training points, KNN will might has performance problems [16]. KNN is very sensitive to the bad features, therefore, the selection of feature is also important. KNN is also sensitive to the outliers, therefore, removing the outliers before using KNN will improve the results.

By using the Kernel, SVM can be used in linear or non-linear fashion. If you have a limited set of points in many dimensions, SVM might be a better choice as it could find the linear separation. Comparing with KNN, SVM is more suitable for outliers, since it only use the support vectors to find a linear separation. SVM also needs to be adjusted, the use of a kernel and the cost “C” as well as its parameters are all critical hyper-parameters to the algorithm.

In general, there are two points we can get: [16]

-          if lots of the points are in lower dimensional space, KNN might be a suitable classifier algorithm.

-          if a few points are in a higher dimensional space then SVM classifier algorithm is probably better.

Project Management

                  Figure 3: Gantt Chart

After finding the project objectives, I spend 2 weeks on finding the sources. To complete the literature review about Z24 bridge, Structure Health Monitoring, Data Classification using in SHM, and SVM/KNN Classifier Model. In this semester, I need to build a simple simulated model:

200 points have been setting by random, and divided into 2 classes, therefore, 100 separately. 50 of each team will be regard as the training point, and the others will be the testing points.

Regarding Class A as damage data, Class B as undamaged data. By using K-nearest neighbors algorithm (KNN) classifier function and Support Vector Machines (SVM) classifier function to plot the figure separately. Comparing the figures and then analysis the results.

For my further project, I will be using KNN classifier and SVM classifier separately on the real Z24 bridge dataset, and then comparing the results as well as analyzing the error. Next, I will pay more attention on investigating the damage causing effects (environmental conditions), such as wind, humidity of surrounding soil and the most important, temperature. Since the temperature may influence the material properties and the boundary conditions. Specific investigation will apply in my final dissertation.

Self-review

I have divided my individual project into two main parts, and up to now, I have finished the first part. In the first part, the interim model needs to be established. There are several problems appear at this time:

1. At the first time, I set up the data by myself, finish the KNN code without using KNN functions. The result looks simply and the testing points looks regularly which is lacks of randomness.
2. After finishing the code, I found the results looks different. After meeting my supervisor, and I found out I need to combine both codes together. Since I set up the points in random, the result will change every times after running the code.
3. In the first meeting with my supervisor, she told me how I can complete the objectives, but unfortunately, I can not understand it clearly. But after searching the method, I realized what I need to do in final. Every meeting was useful and necessary for me.
4. My basic objectives are: Review literature on classification techniques; Review literature on SHM and Z24-Bridge; Apply a classifier to toy data. Up to now I have completed these objectives. In the first few meetings with my supervisor, she told me some academic words which I did not understand at that time, but when the project processing, I gradually understand what I need to do in this individual project.

Appendices

   Note 1: KNN function will not plot the line, I added the line which from the SVM plotting figure for comparison.

Reference

1. Reynders, E. and Roeck, G. (2018). Vibration-Based Damage Identification: The Z24 Bridge Benchmark. [online] SpringerLink. Available at: https://link.springer.com/referenceworkentry/10.1007/978-3-642-35344-4_72 [Accessed 15 Oct. 2018].
2. Silver.neep.wisc.edu. (2018). Bridge Z24 – Switzerland. [online] Available at: http://silver.neep.wisc.edu/~msallen/18_kul_Z24Bridge_Summary.pdf Z24 – Switzerland [Accessed 24 Oct. 2018].
3. Iste.co.uk. (2015). [online] Available at: http://iste.co.uk/data/doc_xqjujdlhnfls.pdf [Accessed 24 Oct. 2018].
4. Bisht, S. (2010). Methods for Structural Health Monitoring and Damage Detection of Civil and Mechanical Systems. [online] Vtechworks.lib.vt.edu. Available at: https://vtechworks.lib.vt.edu/handle/10919/33672 [Accessed 25 Oct. 2018].
5. Learning, M. and code), U. (2017). Understanding Support Vector Machine algorithm from examples (along with code). [online] Analytics Vidhya. Available at: https://www.analyticsvidhya.com/blog/2017/09/understaing-support-vector-machine-example-code/ [Accessed 30 Oct. 2018].
6. Farrar CR and Worden K. An introduction to structural health monitoring. Philos T R Soc A 2007; 365: 303-315.
7. Reynders E, Wursten G and Roeck GD. Output-only structural health monitoring in changing environmental conditions by means of nonlinear system identification. Struct Health Monit 2014; 13(1):82-93.
8. Farreras-Alcover I, Chryssanthopoulos MK and Andersen JE. Regression models for structural health monitoring of welded bridge joins based on temperature, traffic and strain measurements. Struct Health Monit 2015; 14(6):648-662.
9. Donald, S. (2003) Development of Empirical Possibility Distributions in Risk Analysis, PhD Thesis, Department of Civil Engineering, University of New Mexico, Albuquerque, New Mexico, USA.
10. Sotnikov, I. (2006). Data Classification: What It Is, Why You Should Care and How to Perform It. [online] Blog.netwrix.com. Available at: https://blog.netwrix.com/2018/03/15/data-classification-explained-what-it-is-why-you-should-care-and-how-to-perform-it/ [Accessed 26 Nov. 2018].
11. LORD, N. (2015). What is Data Classification? A Data Classification Definition. [online] Digital Guardian. Available at: https://digitalguardian.com/blog/what-data-classification-data-classification-definition [Accessed 19 Nov. 2018].
12. Farquad, M. and Bose, I. (2012). Preprocessing unbalanced data using support vector machine. Decision Support Systems, 53(1), pp.226-233.
13. Zakka, K. (2016). A Complete Guide to K-Nearest-Neighbors with Applications in Python and R. [online] Kevinzakka.github.io. Available at: https://kevinzakka.github.io/2016/07/13/k-nearest-neighbor/ [Accessed 14 Nov. 2018].
14. Zhai J, Barreto A (2006) Stress detection in computer users based on digital signal processing of noninvasive physiological variables. 2006 international conference of the IEEE engineering in medicine and biology society 1355-1358.  https://doi-org.sheffield.idm.oclc.org/10.1109/iembs.2006.4397662
15. Farquad, M. and Bose, I. (2012). Preprocessing unbalanced data using support vector machine. Decision Support Systems, 53(1), pp.226-233.
16. KIM, J. (2008). Comparing Image Classification Methods: K-Nearest-Neighbor and Support-Vector-Machines. [online] Wseas.us. Available at: http://www.wseas.us/e-library/conferences/2012/CambridgeUSA/MATHCC/MATHCC-18.pdf [Accessed 15 Nov. 2018].