Large Data Handling Challenges in Inspection of Infrastructure Assets Krishnan Balasubramaniam Professor, Mechanical Engineering Dean, Industrial Consultancy & Sponsored Research (ICSR) Head, Center for Non-Destructive Evaluation (CNDE) Indian Institute of Technology Madras Chennai INDIA www.cnde-iitm.net
Big Data??????????? 0011100101010100101010010101010101001011 01011101 U mean LARGE amount of data?
Big-Data Lots of information, Need to make sense, Cannot be handled by current means Evolving Target
Nondestructive Evaluation Non-destructive means of making quantitative measurements that has relevance to the material state and/or process parameters. Without decreasing the performance capability. Employs sensing methods using acoustic and/or electromagnetic spectrum.
NDE for Infrastructure Assets
Case Study 1: Safety of Rails
Rail Inspection Problems Railroad Number Miles Class I 7 93,921 Regional 23 12,804 Local 533 32,393 Total U.S. 563 139,118 NEED FASTER and RELIABLE INSPECTION MEANS
Ultrasonic Flaw Detection
Click to play
Volume of data At 4 mm data spacing at 10 kmph Each Channel 600*0.05 MB per second. 16 channel system will accumulate 480 MB/s For 1 km of railroad 480x60x6/1000 = 173GB Data Analysis must be performed online and ImR defects must be marked. Desired speed of inspection is 100 kmph Max Demonstrated speed of inspection is 60 kmph.
Data Processing Approach Data Compression & Feature Extraction Decision Processes Break data into small chunks. Predetermined rules Parallel Processing Look up tables approaches Heuristic Extract key features Approaches Scale features Hardware Processing uprocessors FPGA DMA Flash memory Currently 25 kmph USFD system has been demonstrated in India by IITM
Track Management System
Rail Defect Monitoring
Case Study 2: Pipeline Inspection Europe Pipeline Networks US has 305,000 Miles of Natural Gas Pipelines India has more than 45,000 km of pipes and exepcted to increase to 120,000 in 5 years. Failures can lead to loss of product, environmental impacts, and catastrophes http://www.theodora.com/pipelines/wor ld_oil_gas_and_products_pipelines.ht ml
ipig for country pipe inspection 5-10 kmph Sensors MFL Ultrasonics Onboard Computing storage Offline Evaluation
MFL Data Sample 10 m
BigData For a 12 inch pipe using MFL only Per km data volume is 2 GB For a 36 inch pipe using MFL only Per km data volume is 18 GB For a 12 inch pipe using MFL + UT Per km data volume is 100 GB For a 36 inch pipe using MFL+UT Per km data volume is 900 GB
Current limitations Typical run is 50-100 km per day Data is pre-processed & Stored on-board Data is analysed off-line Data transfer to Data Analysis Center takes 4-8 hours or more. Preliminary automated Analysis and initial report in 3 days. Detailed manual data analysis and final report in 15-45 days.
Data Handling Approach On-board Data Compression using wavelet based segementation and feature extraction. Off-line ISSUE: are we loosing information. Parallel processing of the data using GPUs Reduced automated analysis time by 90% Automation assisted Rule based and ANN based detailed analysis Reduced analysis time by 60% and increased reliability.
Automated Defect Recogntion 10 m
Pipe Integrity Management System http://www.atpuk.co.uk/index.asp http://www.infosys.com/industries/ene rgy/industryofferings/documents/pipelineintegrity-management.pdf
Summary Maintenance of infrastructure requires techniques that will result in large volumes of data that will require extracting the requisite information in very short time. On-line compression is often irreversible and can lead to lost information. Data must be interfaced to comprehensive Asset Management Software/Concepts New paradigms such as Structural Health Monitoring of infrastructure will pose new challenges.