DEVELOPMENT OF AN INNOVATIVE METHOD FOR CRACK MONITORING Jónatas Valença PhD Student, ISISE, Institute Polytechnic of Coimbra, Portugal
[email protected]
Daniel Dias-da-Costa Research Assistant, ISISE, University of Coimbra, Portugal
[email protected]
Eduardo Júlio Professor, ISISE, University of Coimbra, Portugal
[email protected]
KEYWORDS: NDT; Image processing; Crack localisation; Crack characterisation; Monitoring.
ABSTRACT Monitoring the deterioration of concrete structures is the best way to assure a cost-effective maintenance. Normally, non-destructive tests (NDT) are used when any sign of alarm is observed throughout visual inspections. One of the most important anomalies in concrete structures is cracking. Therefore, accurate cracking detection, mapping and measuring is vital. Nevertheless, to undertake this task with traditional methods is time-consuming. In this paper, the authors describe the development of an innovative method, CONCRACK, using digital image processing, aiming to automatically detect, map and measure concrete cracking. Push-off specimens tested until failure were used to validate the proposed methodology. It was concluded that CONCRACK presents several advantages in relation to traditional methods: speed, efficiency, comprehensive information and reliability, since data is automatically processed and therefore it is not subjected to human error. INTRODUCTION The capability of accurately assessing the conservation state during the life time of a structure is of major importance. In fact, monitoring the structural health is the most effective way to optimise the interventions needed to keep / restore safety and serviceability conditions, defining a cost effective program. Structural health monitoring is composed by three stages. First, scheduled visual inspections are undertaken. Whenever anomalies are detected, the second stage is engaged by selecting adequate nondestructive tests (NDT). Finally, a detailed plan of intervention must be defined in accordance to the results from the previous stages. Cracking is commonly the first sign of structural faults. Consequently, to accurately detect, map and quantify the evolution of cracking is of utmost importance. However, most existing approaches are empirical regarding this parameter. For instance, the crack pattern is frequently defined by direct sketch from visual inspection and the crack width is evaluated by means of a measuring magnifier or using a crack width ruler. Therefore, most of the traditional methods to characterise and monitor cracking still remain: (i) time-consuming; (ii) subjective; and (iii) dependent of user‟s errors (Barazzeti & Scaioni, 2009). Recently, the development of digital optical equipments and methods has gradually widened their field of application. In this paper, an innovative method is introduced, CONCRACK – Image Processing of Cracking in Concrete Surfaces. The method was developed with the following objectives: (i) to automatically detect cracks; (ii) to automatically map the cracking pattern; and (iii) to automatically measure cracks‟ widths, by using digital image processing. METHODOLOGY FOR DETECTION AND CHARACTERIZATION Recently developed algorithms are capable of detecting discontinuities in images by analysing variations of the intensity of pixels. Considering cracks in concrete surfaces as discontinuities on an image, some applications devoted to their characterisation were developed (Lange et al, 2006; Dare et al, 2006; Sinha & Fieguth, 2002; Yamaguchi et al, 2008). Nevertheless, these only provide good results under controlled conditions and typically only a single crack, on a single stage, is analysed.
An innovative methodology specially developed for monitoring cracking in concrete is herein presented combining digital image processing and mathematical morphology (MM) (Kowalczyk et al, 2008). By means of a combined global-local approach, crack detection and characterisation is significantly enhanced on a concrete element surface. The different procedures of CONCRACK are summarised in the flowchart presented in Fig. 1. These include the following global steps: (1) Image Acquisition; (2) Global Analysis; (3) Selection of Areas with Discontinuities; (4) Local Analysis; and (5) Global Reconstitution. Immediately after the first step, a Global Region of Interest (GROI) is established and the second step is undertaken. At the end of this step, Local Regions of Interest (LROI) can be defined (third step) and a localised study is performed (fourth step). Finally, these LROIs are reconstituted into a global image which summarises the results (last step).
Image Acquisition
Selection of Areas with Discontinuities
Global Analysis
Local Analysis
Global Reconstitution
Fig. 1 Flowchart with CONCRACK global steps. At both Global and Local Analysis, respectively steps 2 and 4, several phases allow enhancing the results. These include: (i) pre-processing; (ii) processing; and (iii) post-processing: - during pre-processing, the contrast between discontinuities and background of the image is enhanced; - two independent algorithms have been selected for comparison in the processing phase: (i) image binarization applying Otsu‟s threshold method (Otsu, 1976); and (ii) a manual threshold adopted to remove the circles marked on the surface of the specimen. These marks are used for a photogrammetric project outside the scope of this work; - post-processing includes MM operations to clean discontinuity areas and to fill closed regions. Additionally, at step 4, the operations related to crack characterisation are applied. A sequence of three main steps allows evaluating the crack‟s length, width and area (see Fig. 2): (1) localisation of the crack on the surface of the structure; (2) definition of the boundary edges; and (3) assessment of the crack‟s length and corresponding width at each point of the boundary. This procedure characterises the crack profile (length, width and area) and is fully automatic for any selected boundary edges. The length of the crack is defined by the average length of both edges, whereas the crack width is defined by the minimum distance from each pixel to the opposite edge, according to Eq. 1: di min di , j (1) where: di , j represents the distance from pixel „i‟ to the edge „j‟.Crack Crack boundaries-f2
boundaries-f2 d1,1
di,j 2
2
d1,j
1
1
i =1,…, lenght boundary1 j =1,…, lenght boundary2
(a)
(b)
(c)
(d)
Fig. 2 Evaluation of the crack width at each point along the length of the discontinuity. 2
1
CASE STUDY Monitoring push-off specimens until failure (Fig. 3) was selected as case study to validate the proposed method. The surface of the specimen was painted in white to enhance crack identification. Sixty-six stages have been acquired during a complete test, with a total duration of nearly 30 minutes. From these, seven stages were considered relevant for monitoring and are herein addressed.
Fig. 3 Push-off specimen: (a) initial stage; (b) intermediate stage; (c) last stage: failure of the specimen. Image acquisition The images of the specimen were acquired with a Canon 400D camera using a 18 mm focal length, placed at 2.50 m, orthogonally to the surface of the specimen, and by recording images with 3872×2592 pixels format. Protection boards covering natural light sources lead to a homogeneous and diffuse light pattern which was kept unchanged during the test. A tripod guaranteed the stability of the camera. Parallelism between the surface of the specimen and the image plane was ensured by tape measuring the distance at various points, whereas a bubble level verifying both horizontal and vertical planes of the camera. Calibration and validation In this section the spatial resolution, calibration and validation of the method is performed. Two Crack Width Rulers (CWR) were placed at both sides of the specimen (Fig. 3a). Each ruler was composed by a set of lines with an increased thickness ranging from 0.1 to 4.0 mm. At each CWR, a LROI window was defined to evaluate the longitudinal pixel intensity profiles for the binary image. The thickness of each line was given by the length of the profile containing zero values. It should be highlighted that, due to the resolution of the images, the four thinnest lines (varying from 0.1 to 0.4 mm) could not be detected. A total of two measures for each width were obtained, one for each CWR, per analysed stage. Since seven stages were considered, the corresponding fourteen values obtained constitute a measure of precision. Therefore, the average and the total RMS precision are, respectively, 0.25 and 0.33 pixel. Precision was kept unchanged by considering only the left or the right placed CWR. This emphasises the parallelism between image and structural plane. The widths represented on the ruler have known values and can be used to evaluate the accuracy of the method. The corresponding spatial resolution of the project was 0.231 mm/pixel on average corresponding to 0.013 mm/pixel for standard deviation. These values of precision and spatial resolution allowed having an accuracy always less than 10%, reaching 4% with increasing thickness. Considering the known values for the CWR lines as ground truth, an absolute average and total RMS accuracy of 0.055 mm and 0.062 mm, respectively, was achieved.
RESULTS The results obtained from a push-off specimen are evaluated with CONCRACK and detailed in the following sections. Global analysis The Global Analysis step was applied to the complete surface of the specimen, identified as the GROI. For the Global Analysis, a reduced number of stages can be considered. This is related to the fact that the LROI can be almost exclusively defined by the final stage where the entire crack map is depicted. Nevertheless, three states are chosen in order to additionally compare the capabilities of the processing algorithms: (1) initial state, just before applying the load; (2) an intermediate stage, 11 minutes after starting the test; and (3) failure stage, corresponding to the end of the test, immediately before unloading. Both processing algorithms are identically applied to the GROI, followed by post-processing composed by cleaning and filling bounded regions (MM procedure). The obtained cracks‟ maps are depicted in Fig. 4.
Fig. 4 Binary images: (a) initial stage; (b) intermediate stage; and (c) final stage. The main observations are the following: - the Otsu‟s method is able to detect the crack pattern properly, even in early stages (see Fig. 4 for intermediate stage); - a manually defined threshold is able to eliminate the circular targets. Nevertheless, detail is lost concerning mapping of cracks; - the optimization of the manual procedure is adjusted to each case, not being suitable for all stages, rendering the development of a robust tool unfeasible. On the contrary, the Otsu‟s method proved to be a stable algorithm, since it performs identically at all stages.
Selection of areas with discontinuities Both algorithms were applied to identify the LROIs (see Fig. 5) from the failure stage. Despite losing some detail of the cracks, the manual method allows a clear visualization of the cracking pattern.
Fig. 5 Definition of the LROIs. Local analysis Since the Otsu‟s method remains stable and fully automatic for all stages considered, simultaneously keeping an adequate sensibility for detecting cracks, this algorithm was selected for the Local Analysis to be performed at the five LROIs defined, for the seven defined stages. As an example, the results of the LROI#1 are represented in Fig. 6, in terms of RGB image, binary image and boundaries. These last can be used to measure the width, length and area of the discontinuity according to the method developed.
Fig. 6 Local Analysis of LROI #1. Considering the LROIs detected, it is possible to automatically monitor the evolution of the width of any identified crack. Fig. 7a shows the location of five selected cracks. For each one, the Crack Mouth
Opening Displacement (CMOD) is monitored (Fig. 7b). Relevant information is provided by comparing Fig. 7b and Fig. 6: (i) first, crack 1 is the main crack until stage 3 is reached; (ii) afterwards, crack 2 becomes the active crack, presenting a significant opening after this stage, whereas crack 1 slightly closes at stage 5; (iii) the CMOD of the remaining cracks gradually increases until failure, although at a smaller rate than the CMOD of crack 2; (iv) finally, at the last stage, the CMOD of all cracks decreases due to unloading.
10
width (mm)
crack 1 8
crack 2
6
crack 3 crack 4
4
2 0 0
5
10
15
20
25
30
Time (min)
Fig. 7 Monitoring of crack mouth opening: (a) selected areas; (b) CMOD. Global reconstitution Fig. 8 contains the crack map, for three stages, resulting from the Global Reconstitution. This is depicted by merging each LROI with the remaining area of the GROI.
Fig. 8 Crack map depicted from the Global Reconstitution: (a) initial stage; (b) intermediate stage; (c) final stage. Comparison between boundary edges defined in the Global Reconstitution with boundary edges defined in the Global Analysis is represented in Fig. 9. The crack map obtained from the former is sharper with the boundary edges of each crack accurately defined. This is an advantage for crack characterisation. Nevertheless, it is stressed out that the crack map obtained from the Global Analysis is already satisfactory. Therefore, Local Analysis and Global Reconstitution steps can be disregarded if only a crack map is needed, neglecting crack characterisation. Fig. 10 contains the crack pattern obtained by two different users concerning three stages of the test. Comparing Fig. 8 with Fig. 10 leads to the following conclusions: (i) CONCRACK gives similar results
to different users; (ii) in traditional methods, users tend to assume cracks close to each other as a single crack; this does not happen using CONCRACK; (iii) in traditional methods, small differences between results obtained by different users tend to appear in areas where the presence of a crack is not obvious; regarding this issue, CONCRACK omits dubious discontinuities. Furthermore, it must be emphasised that each user needs nearly 1 hour to perform a detailed sketch of the three stages, whereas CONCRACK needs a couple of seconds to perform the same task. Additionally, CONCRACK can monitor the width of any crack at a given point whereas this task is not feasible with the traditional measuring methods.
(a)
(b)
Fig. 9 Crack map depicted in the Global Analysis superimposed with the map obtained in the Global Analysis: (a) complete crack map; and (b) detail.
Fig. 10 Crack maps defined from a sketch by two different users.
FINAL REMARKS AND CONCLUSIONS One of the most common sign of structural faults in concrete structures is cracking. This parameter is usually assessed by close visual inspection, being the crack width evaluated by means of magnifiers or crack width rulers. Therefore, characterisation and monitoring cracking is significantly: (i) timeconsuming; (ii) subjective; and (iii) dependent on errors of the user. In this paper, an innovative methodology, named „CONCRACK‟, is proposed which is capable of fully eliminating the drawbacks of traditional methods. The method was developed to automatically detect, map and measure cracks, being also possible to easily monitor the evolution of cracking with time. The global cracking pattern can be recorded at an unlimited number of pre-defined stages. The critical areas are defined by means of a global analysis. Afterwards, a localised detailed analysis is performed, followed by a reconstitution of the global surface, which allows obtaining the entire crack pattern and characterise the evolution of any crack. It can be stated that CONCRACK is a cost-effective, robust, non-contact and user-friendly measurement method, suitable for monitoring cracking in concrete structures. The method can assist the definition of optimised interventions, allowing an efficient approach to maintain structural safety and serviceability conditions. Relatively to traditional methods, the main advantages of CONCRACK are: speed, efficiency, comprehensive information, and reliability, since data is automatically processed. ACNOWLEDGMENTS The authors acknowledge the financial support of the Portuguese Science and Technology Foundation (FCT) and of the company EC+A – Projectos Lda, Ph.D. Grant number SFRH/BDE/15660/2007. REFERENCES Barazzetti, L. and Scaioni, M. (2009), “Crack measurement: Development, testing and applications of an automatic image-based algorithm,” ISPRS Journal of Photogrammetry and Remote Sensing, Vol. 64, No. 3, 285-296. Dare, P., Hanley, H., Fraser, C., Ridel, B. and Niemeier, W. (2002), “An operational application of the automatic feature extraction: The measurement of cracks in concrete structures,” The Photogrammetric Record, Vol. 17, No. 99, 453-464. Kowalczyk, M., Koza, P., Kupidura, P. and Marciniak, J. (2008), “Application of mathematical morphology operations for simplification and improvement of correlation of images in close-range photogrammetry,” The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXVII Part B5. Beijing. Lange, J., Benning, W. and Siering, K. (2006), “Crack detection at concrete construction units from photogrammetric data using image processing procedures,” ISPRS Commission VII Mid-term Symposium Remote Sensing: From Pixels to Processes, Enschede, Netherlands. Otsu, N. (1979), “A threshold selection method from gray-level histogram,” IEEE Transactions on System Man Cybernetics, Vol. SMC-9, No. 1, 62-66. Sinha, S. and Fieguth, P. (2006), “Segmentation of buried concrete pipe images,” Automation in Construction, Vol. 15, No. 1, 47-57. Yamaguchi, T., Nakamura, S., Saegusa, R. and Hashimoto, S. (2008), “Image-based crack detection for real concrete surfaces,” IEEJ Transactions on Electrical and Electronic Engineering, Vol. 3, No. 1, 28-135.