presented by, gokul narayanan, p.e – asec, inc. locke
TRANSCRIPT
SEPTEMBER 4 - 6, 2019
ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING FOR INSPECTION OF TRANSMISSION LINES
Presented by,Gokul Narayanan, P.E – ASEC, Inc.Locke Brillhart – Kleinfelder, Inc.
SEPTEMBER 4 - 6, 2019
Determining conductor issues and severity of damage for a 51-mile transmission line by ground inspection
Time & Money QualitySafety
SEPTEMBER 4 - 6, 2019
As an alternative to a ground inspection of all 51-miles of transmission line Kleinfelder utilized UAS/Drones to inspect all conductors
Completed in 1/3 of the estimated time
Saved over $50,000 in overall budget
Gathered 4 times as many photos than ground
inspection w/bird’s eye view
Deployed only 2 technicians and no
bucket trucks
SEPTEMBER 4 - 6, 2019
CODE FIELDOBSERVATIONS
STRUCTURE NUMBERS COMMENTS
BIR 4 155,189,259,331 Bird flying/perchedBDC 5 109,110,301,302,306 Broken/Damaged
conductorBSW 1 Between 324 & 325 Shield strands frayedDIS 29 10,14,23,33,42, 53,83,87,112,114,120,121,130,136,137,141,
142,144,146,147,151,160,193,196,208,303,324,329,350Discoloration around
damper clamp
MIS 4 126,129,135,306 Missing dampersDIM 2 140,143 Damper distance not
typicalDDE 5 1,11,15,24,284 Discolored at Deadend SC 5 169,249,250,251,307 Soiled (by birds)
SPA 1 90 Spalling concreteTotal 56
SEPTEMBER 4 - 6, 2019
Machine Learning algorithms use computational
methods to “learn” information directly from data without
relying on a predetermined equation as a
model.
SEPTEMBER 4 - 6, 2019
1000’s of Images
Improveswith more
Data
HomogeneousSmall Corpus
ROITraining
VSManual
Tenets of AIFairnessAccountabilityTransparencyEthics
AccuracyHumanVSMachine
SEPTEMBER 4 - 6, 2019
Filters & Reductions
Math Transforms
Good 0.9Damaged 0.1
Deep Neural Net
Labeled Image
Convolution
SEPTEMBER 4 - 6, 2019
Good
• 91 Good• 2 Suspect
• 24 Suspect• 6 Good
Suspect
• 4 Unknown• 51 Good• 22 Suspect
Unknown
200Test
Photos
SEPTEMBER 4 - 6, 2019
Efficiency
Quali
ty
Traditional Approach
UAS/Human Review
Artificial Intelligence
Efficiency