Despite the recent advances in sensor technologies and data acquisition systems, interpreting 4 measurement data for structural monitoring remains as challenge. Furthermore, due to the 5 complexity of the structures, materials used and uncertain environments, behavioral models are 6 difficult to build accurately. This paper presents novel model-free data-interpretation methodologies 7 that combine MPCA with each of four regression-analysis methods -Robust Regression Analysis 8 (RRA), Multiple Linear Analysis (MLR), Support Vector Regression (SVR) and Random Forest (RF) -for 9 damage detection during continuous monitoring of structures. The principal goal is to exploit the 10 advantages of both MPCA and regression-analysis methods. The applicability of these combined 11 methods is evaluated and compared with individual applications of MPCA, RRA, MLR, SVR and RF 12 through four case studies. Result showed that the combined methods outperformed non-combined 13 methods in terms of damage detectability and time to detection.

12In vibration-based structural health monitoring, changes in the natural frequency of a structure 13 are used to identify changes in the structural conditions due to damage and deterioration. 14 However, natural frequency values also vary with changes in environmental factors such as 15 temperature and wind. Therefore, it is important to differentiate between the effects due to 16 environmental variations and those resulting from structural damage. In this paper, this task is 17 accomplished by predicting the natural frequency of a structure using measurements of 18 environmental conditions. Five methodologies -multiple linear regression, artificial neural 19 networks, support vector regression, regression tree and random forest -are implemented to 20 predict the natural frequencies of the Tamar Suspension Bridge (UK) using measurements 21 taken from three years of continuous monitoring.

Interpreting measurement data to extract meaningful information for damage detection is a challenge for continuous monitoring of structures. This paper presents an evaluation of two model-free data interpretation methods that have previously been identified to be attractive for applications in structural engineering: moving principal component analysis (MPCA) and robust regression analysis (RRA). The effects of three factors are evaluated: (a) sensor-damage location, (b) traffic loading intensity and (c) damage level, using two criteria: damage detectability and the time to damage detection. In addition, the effects of these three factors are studied for the first time in situations with and without removing seasonal variations through use of a moving average filter and an ideal low-pass filter. For this purpose, a parametric study is performed using a numerical model of a railway truss bridge. Results show that MPCA has higher damage detectability than RRA. On the other hand, RRA detects damages faster than MPCA. Seasonal variation removal reduces the time to damage detection of MPCA in some cases while the benefits are consistently modest for RRA.Keywords: Moving principal component analysis; robust regression analysis; damage detection; damage detectability; time to damage detection; seasonal temperature variation.Evaluating two model-free data interpretation methods for measurements that are influenced by temperature IntroductionRecently, the collapse of civil structures such as the I-35W bridge (USA, 2007) [1-2] and the Paris Airport (France, 2004) [3] has decreased public confidence in the safety of structures.Thus, to ensure that structures behave according to design criteria, it is useful to monitor their performance. Monitoring for possible damage is referred to as structural health monitoring (SHM). Due to advances in sensor technology, data acquisition systems and computational power, the number of structures that are monitored is growing. Thus, large quantities of measurement data are retrieved every day and much more will be available in the future.Extracting useful information from this data to detect damage is a challenge for SHM. This task is even more difficult when measurement data are influenced by environmental variations, such as temperature, wind and humidity. Brownjohn et al [4] studied the thermal effects on performance on Tamar Bridge and showed that thermal effects dominate the measured bridge behaviour. Catbas et al. [5] observed that the peak-to-peak strain differential due to temperature over a one-year period is more than ten times higher than the strain due to observed maximum daily traffic.Generally, there are two classes of data interpretation methods in SHM: model-based methods and model-free methods. These two classes are complementary since they are appropriate in different contexts. Strengths and weaknesses of both classes have been summarized in the ASCE state-of-the-art report on structural identification of constructed systems [6].Model-based data interpretation...

To identify physical parameters of a large structural system, the computational challenges in dealing with a large number of unknowns are formidable. A divide-and-conquer approach is often required to partition the structural system into many substructures, each with much lesser unknowns for more accurate and efficient identification. Furthermore, in view of the ill-conditioned nature of inverse analysis, it is highly beneficial to adopt non-gradient based search methods such as genetic algorithm (GA). To this end, this paper presents a GA-based substructural identification strategy for large structural systems. As compared to some recent work on substructural identification, the proposed strategy presents two significant improvements: (a) the use of acceleration measurements to directly account for interaction between substructures without approximation of interface force, and (b) the use of an improved identification method based on multi-feature GA. In numerical simulations, the mass, damping and stiffness parameters of a 100-storey shear building, involving 202 unknowns, are identified with very good accuracy (mean error of less than 3%) based on incomplete acceleration measurements with 10% noise. In addition, an experimental study on a 10-storey small-scale steel frame further validates the superior performance of the proposed strategy over complete structural identification.2

Measurement system configuration is an important task in structural health monitoring in that 4 decisions influence the performance of monitoring systems. This task is generally performed using 5 only engineering judgment and experience. Such approach may result in either a large amount of 6 redundant data and high data-interpretation costs, or insufficient data leading to ambiguous 7 interpretations. This paper presents a systematic approach to configure measurement systems 8 where static measurement data are interpreted for damage detection using model-free (non-physics-9 based) methods. The proposed approach provides decision support for two tasks: (1) determining 10 the appropriate number of sensors to be employed and (2) placing the sensors at the most 11 informative locations. The first task involves evaluating the performance of measurement systems in 12 terms of the number of sensors. Using a given number of sensors, the second task involves 13 configuring a measurement system by identifying the most informative sensor locations. The 14 locations are identified based on three criteria: the number of non-detectable damage scenarios, the 15 average time to detection and the damage detectability. A multi-objective optimization is thus 16 carried out leading to a set of non-dominated solutions. To select the best compromise solution in 17 this set, two multi criteria decision making methods, Pareto-Edgeworth-Grierson multi-criteria 18 decision making (PEG-MCDM) Laory, I., Bel Hadj Ali, N., Trinh, T., and Smith, I. configuring measurement systems based on the data-interpretation methods used for damage 25 detection. The approach is also able to accommodate the simultaneous use of several model-free 26 data-interpretation methods. It is also concluded that the number of non-detectable scenarios, the 27 average time to detection and the damage detectability are useful metrics for evaluating the 28 performance of measurement systems when data are interpreted using model-free methods. 29"Measurement System Configuration for Damage Identification of Continuously Monitored Structures." J of Bridge Engineering, Vol. 17, 2012, SPECIAL ISSUE: Nondestructive Evaluation and Testing for Bridge Inspection and Evaluation, pp 857-866