a data driven approach to railway intervention planning derek bartram supervisors dr. m. burrow...
Post on 22-Dec-2015
214 views
TRANSCRIPT
A Data Driven Approach to A Data Driven Approach to Railway Intervention PlanningRailway Intervention Planning
Derek BartramDerek Bartram
SupervisorsSupervisorsDr. M. BurrowDr. M. BurrowProf. X. YaoProf. X. Yao
ContentsContents
Current TechnologiesCurrent Technologies– StructureStructure– Problems / IssuesProblems / Issues
ProjectProject– AimsAims– Comparison To Current TechnologiesComparison To Current Technologies– System DesignSystem Design
Progress To DateProgress To Date Progress ProblemsProgress Problems
Typical Track DeteriorationTypical Track Deterioration
Angle 1 < Angle 2 < Angle 3Angle 1 < Angle 2 < Angle 3
Geometry MeasurementsGeometry Measurements
Geometry MeasurementsGeometry Measurements
Top heightTop height Web heightWeb height Web thicknessWeb thickness Ballast thicknessBallast thickness Ballast SD sizeBallast SD size Corrugation wavelengthCorrugation wavelength GaugeGauge TwistTwist CantCant
Existing Technologies : Existing Technologies : Decision Support SystemsDecision Support Systems
Expert System : Inference EngineExpert System : Inference Engine
If (ballast_type == granite) If (ballast_type == granite)
then minimum_thickness = 50mmthen minimum_thickness = 50mm If (ballast_type == sandstone)If (ballast_type == sandstone)
then minimum_thickness = 200mmthen minimum_thickness = 200mm
If (ballast_thickness < minimum_thickness)If (ballast_thickness < minimum_thickness)
then replace_ballastthen replace_ballast
Expert System : Inference EngineExpert System : Inference Engine
If (ballast_thickness < 50mm &&If (ballast_thickness < 50mm &&
ballast_type == granite)ballast_type == granite)
then replace_ballastthen replace_ballast
If (ballast_thickness < 200mm &&If (ballast_thickness < 200mm &&
ballast_type == sandstone)ballast_type == sandstone)
then replace_ballastthen replace_ballast
Decision Support Systems : Decision Support Systems : Problems / IssuesProblems / Issues
Expert system only as good as the rule baseExpert system only as good as the rule base
Simplified modelsSimplified models
Possible rule / intervention flawsPossible rule / intervention flaws
Large track segmentsLarge track segments
AimsAims
Improved deterioration modellingImproved deterioration modelling
Improved intervention planningImproved intervention planning
Improved localised fault detectionImproved localised fault detection Improved total life-cycle costingImproved total life-cycle costing
Static Vs Dynamic SolutionsStatic Vs Dynamic Solutions
Static solutionStatic solution Guaranteed good behaviour initiallyGuaranteed good behaviour initially Never improvesNever improves
Dynamic solutionDynamic solution Initial behaviour potentially badInitial behaviour potentially bad Requires high quality existing datasetRequires high quality existing dataset Improves with timeImproves with time
My Project : Assumptions (1)My Project : Assumptions (1)
The various possible faults for track are The various possible faults for track are identifiable by unique combinations of identifiable by unique combinations of
track component deteriorationtrack component deterioration
My Project : Assumptions (2)My Project : Assumptions (2)
For each type of failure, the solution to the For each type of failure, the solution to the problem is not related to other failure problem is not related to other failure
typestypes
My Project : Assumptions (3)My Project : Assumptions (3)
Once a track sections starts failing with a Once a track sections starts failing with a particular failure type, it will continue to particular failure type, it will continue to
fail with the same failure typefail with the same failure type
My Solution : TasksMy Solution : Tasks
Classify the various failure typesClassify the various failure types
Provide a mechanism for classifying unclassified Provide a mechanism for classifying unclassified track sectionstrack sections
Produce a deterioration model for each failure typeProduce a deterioration model for each failure type
Determine best intervention for each failure typeDetermine best intervention for each failure type
My Solution : Data ProcessingMy Solution : Data Processing
Handle missing dataHandle missing data
Segment dataSegment data
Build data runsBuild data runs
Make absolute values relativeMake absolute values relative
My Solution : Failure TypesMy Solution : Failure Types
Plot last data recording of each run in Plot last data recording of each run in
n-dimension spacen-dimension space
My Solution : ClassificationMy Solution : Classification
We know sets of individual data points and We know sets of individual data points and associated failure typesassociated failure types
Failure type does not change until Failure type does not change until interventionintervention
Decision treesDecision trees Evolutionary algorithmsEvolutionary algorithms
My Solution : ClassificationMy Solution : Classification
Decision treesDecision trees
My Solution : ClassificationMy Solution : Classification
Evolutionary AlgorithmEvolutionary Algorithm
My Solution : Work My Solution : Work DeterminationDetermination
For each run in failure type {For each run in failure type {Calculate fitness of subsequent interventionCalculate fitness of subsequent intervention
}}
Calculate average of fitness's for each Calculate average of fitness's for each intervention typeintervention type
Choose intervention with best average fitness Choose intervention with best average fitness
My Solution : Work My Solution : Work DeterminationDetermination
Fitness metricFitness metric
Length of time before next interventionLength of time before next intervention
Next failure typeNext failure type
My Solution : Deterioration ModellingMy Solution : Deterioration Modelling
Simple modelSimple model
Enhanced simple modelEnhanced simple model
Evolutionary model buildingEvolutionary model building
Progress To DateProgress To Date
Classify the various failure typesClassify the various failure types
Provide a mechanism for classifying unclassified Provide a mechanism for classifying unclassified track sectionstrack sections
Produce a deterioration model for each failure typeProduce a deterioration model for each failure type
Determine best intervention for each failure typeDetermine best intervention for each failure type
My Solution : ProblemsMy Solution : Problems
Large number of missing values in geometry Large number of missing values in geometry datadata
Inconsistent / missing? work history dataInconsistent / missing? work history data
Data anomaliesData anomalies
ConclusionsConclusions
Long term improvements over static solutionsLong term improvements over static solutions
Deterioration modelsDeterioration models Intervention planningIntervention planning
CostingCosting
Thank you for listeningThank you for listening
Questions?Questions?