Joint Transportation Research Program JTRP FHWA/IN/JTRP-99/ll Final Report OPTICAL IMAGING METHOD FOR BRIDGE PAINTING MAINTENANCE AND INSPECTION Luh-Maan Chang Yassir Abdelrazig Po-Han Chen April 2000 Indiana Department of Transportation Purdue University Final Report FHWA/IN/JTRP-99/11 Optical Imaging Method for Bridge Painting Maintenance and Inspection By Luh-Maan Chang Principal Investigator Yassir Abdelrazig and Po-Han Chen Research Assistant School ofCivil Engineering Joint Transportation Research Program Project No: C-36-56TT File No: 7-4-45 SPR-2197 In cooperation with the Indiana Department ofTransportation Federal Highway Administration U.S. Department ofTransportation School ofCivil Engineering Purdue Uni\^ersity April 2000 TECHMCALRf.WjRTSTaMjaRDTITLEPACE I. ReportNo. 2. GovernmentAccession.No. 3.Rcdpicni'tCatalog.No. FHWA/IN/JTRP-99/ll •1.TilleandSubtitle 5. ReportDate April 2000 OpticalImagingMethodForBridgePaintingMaintenanceAndInspection i 6. PerformingOrganiialionCode 7.Aulhor(s; 8. PerformingOrg^inizaUon Rep'^rt.So. Luh-Maan Chang, YassirAbdelrazig, andPo-Han Chen FHWA/I.N7JTRP-99/]: 9. PerformingOrganizationNameandAddress 10.UorktnitNo. JointTransportation Research Program 1284Civil Engineering Building Purdue University WestLafayette, Indiana 47907-1284 11. ContractorGrant.No. SPR-219: 12. SponsoringAgencyNameandAddress 13. TypeofReportandPeriodCovered IndianaDepartmentofTransportation State Office Building FinalReport 100Nonh Senate Avenue Indianapolis, IN46204 14. SponsoringAgencyCode 15. Supplementary'Notes Prepared in cooperation with the IndianaDepartmentofTransportation andFederal Highway Administration. The term "quality" is defined as the conformance to predetermined requirements or specifications. These requirements may be set in terms of the end result required or as a detailed description of how work should be executed. Recently, there has been increasing interest in quality assurance in the construction industry. Quality assurance includes design and planning, sampling, inspection, testing, and assessment to ensure that end products perform according to specifications. This research proposes a new quality assessment model for highway steel bridges and more specifically for coating rust assessment. The research proposes a hybrid model using image processing and neural networks for defect recognition and measurement- The basic concept ofthe model is to acquire digital images ofthe areas to be assessed and analyze those images to recognize and measure defect patterns. Neural networks are incorporated into the model to learn from example and simulate human expertise to automate the process for future use. The model is supplemented with a statistical quality assessment plan to use the model efficiently and obtain consistent and reliable results. The statistical plan will determine the number and locations of assessment images to be taken. Moreover, the plan will address the risks associated with the estimated assessment Finally, theplan will assistmaking thefinal acceptance/rejectiondecisionbased on the predefinedcriteriaforacceptance and rejection. 17. KeyWords IS. DistributionSutement Back-PropagationAlgorithm, gray level, hybridmodel, Norestrictions. Thisdocumentisavailabletothepublicthroughthe inspection,neural learning, painting, thresholding NationalTechnical InformationService, Springfield.V.A22161 19. SecurityClassif.(ofthisreport) 20. SecurityClassif.(ofthispage) 21.No.of Pages 22. Price Unclassified Unclassified 190 FormDOTF1700.7i8-69) Digitized by tine Internet Arciiive in 2011 witii funding from LYRASIS members and Sloan Foundation; Indiana Department of Transportation http://www.archive.org/details/opticalimagingmeOOchan Ul TABLE OF CONTENTS Page ABSTRACT ii TABLE OF CONTENTS iii LIST OF TABLES v LIST OF FIGURES vi CHAPTER IINTRODUCTION 1 1.1 Background 1 1.2 Problem Statement 3 1.3 Research Objectives 6 1.4 Research Methodology 7 CHAPTER 2 PRIOR RESEARCH EFFORTS 10 2.1 Visual Quality Assessment 12 2.2 Computerized Visual Imaging for Bridge Coating Assessment 14 CHAPTER IMAGE PROCESSING 3 16 3.1 Introduction 16 3.2 Image Segmentation 18 3.3 Image Thresholding 18 3.3.1 Threshold Selection 25 3.3.2 Numerical Example 31 3.4 Pattern Recognition for Thresholding 37 3.4.1 Numerical E.\ample 40 3.4.2 Algorithms Comparison 44 CHAPTER 4 NEURAL LEARNING 45 4.1 Introduction 45 4.2 Back-Propagation Algoritlim 49 4.2.1 Back-Propagation Training Algorithm 52 4.2.2 Example Using Training Algorithm 57 4.3 Summary 63 CHAPTER THE HYBRID MODEL 5 65 5.1 The Hybnd Model 65 IV 5.2 Images Classification and Measurement 71 5.3 Model Application Example 73 5.4 Model Reliability 80 5.4.1 Error calculation 82 CHAPTER 6 MODEL APPLICATION FOR COATING ASSESSMENT 88 6.1 Steel Bridge Coating 88 6.1.1 Coatings Failure 90 6.1.2 Coating Assessment 91 6.1.3 Coating Maintenance 93 6.1.4 Corrosion 93 6.2 Statistical Assessment Plan 95 6.2.1 Sampling Acceptance Plans 95 6.2.2 Acceptance/Rejection Risk 96 6.2.3 Hybrid Model Acceptance Sampling Plan 97 6.3 Model Implementation 102 6.3.1 Data Acquisition 102 6.3.2 Pre-Processmg 103 6.3.3 Analysis .^ 106 6.3.4 Recognition 108 CHAPTER 7 IMPLEMENTATION 114 7.1 Implementation Suggestions 114 7.2 Image Acquisition 114 7.3 Images Transfer and Storage 118 7.4 Image Analysis 119 CHAPTER 8 CONCLUSIONS AND RECOMMENDATIONS 125 8.1 Conclusions ofthe Research 125 8.2 Recommendations for Future Work 126 LIST OF REFERENCES 129 Appendix A. Statistical Sampling 132 A.l Single Sampling Plan 132 A.2 Attribute Double Sampling Plan 135 Appendix B. Input and Output 138 B.l Training Images Set 138 B.2 Example Image Input Data (Histogram Format) 139 B.3 Example Image Output Data (Histogram Format) 145 B.4 Example Thresholded Image 151 B.5 Example Training Results 152 Appendix C. Digital Camera 153 C.l System Requirements 153 C.2 Specifications 153 Appecdix D. Implementation Procedures 155 D.l Sampling 155 D.2 Image Processing 161