Upload
lala
View
55
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Analyzing Hybrid Track Configurations Through Design of Experiments Techniques. INFORMS 2012 Samuel L. Sogin, C. Tyler Dick, Yung-Cheng Lai, & Christopher P.L. Barkan. Outline. Overview Previous Findings Design of experiments (DOE) Hybrid track capacity factors Response surface model - PowerPoint PPT Presentation
Citation preview
Analyzing Hybrid Track Configurations Through Design of Experiments Techniques
INFORMS 2012Samuel L. Sogin, C. Tyler Dick,
Yung-Cheng Lai, & Christopher P.L. Barkan
Outline
Overview
Previous Findings
Design of experiments (DOE)
Hybrid track capacity factors
Response surface model
Preliminary findings
Questions
Railroad Capacity Analysis – 2
Capacity Upgrade Map
One track baseline
Siding extensions
Fill-in Sidings
Full Two-Mainline-Track
Incr
ease
d C
apac
ity
Incr
ease
d C
ost
Connect Sidings
Railroad Capacity Analysis – 3
Siding Extension
Fill-in Sidings
Second Track
A B
A B
A B
Single Track Sequence
Single Track BaselineA B
Partial Second TrackA B
Railroad Capacity Analysis - 4
5
Comparing Single Track & Double Track
Single Track Most delays are caused
by trains taking turns using the single track bottleneck sections
Very unlikely that trains will perform close to the minimum run time
Delays can follow a log-normal distribution
Speed of the passenger train is not a major factor
Double Track Most delays are caused
by overtaking trains or by crossovers movements
Likely that trains will perform close to the minimum run time
Delays can follow a exponential distribution
Speed differential between trains is a major factor by causing more overtaking situations
Freight Train Delay Distributions
0 20 40 60 80 100 120 140 1600.00
0.25
0.50
0.75
1.00
Freight Train Delay Per 100 Train Miles (min)
Cum
ulati
ve F
requ
ency
Double Track
Single Track
Railroad Capacity Analysis – 6
Freight trains interacting with 79 mph passenger trains
Freight Delay Distributions
0 20 40 60 80 100 120 140 1600.00
0.25
0.50
0.75
1.00
Freight Train Delay Per 100 Train Miles (min)
Cum
ulati
ve F
requ
ency
Double Track
Single Track
Railroad Capacity Analysis – 7
Freight trains interacting with 79 mph passenger trains
Freight trains interacting with 110 mph passenger trains
Railroad Capacity Analysis - 8
DOE – More Information From Less Data
Design of Experiments (DOE): Statistical tool to systematically determine runs in a
experiment design matrix Gain more information than simply varying 1 factor
at a time Consider 3 factors at low, medium, and high levels 27 () runs are needed to cover all permutations
(Full Factorial) Could we gain knowledge of the key drivers of the
process without doing all 27 runs? (Partial Factorial – 16 runs)
Could be used to analyze more factors effecting train performance than typical 2-3 factor analyses
Hybrid Track Response Surface Design
Factor Low Medium HighTotal Trains Per Day 40 48 56
% Slow Trains 0.25 0.5 0.75
% Double Track 0.325 0.528 0.730
Freight Speed (mph) 30 50 70
Passenger Speed (mph) 79 90 110
Double Track Grouping Siding ExtensionAlternateSplitGroup
Railroad Capacity Analysis - 9
50 runs to test all interactions and curvature factors
Hybrid Track Configurations DOE
Railroad Capacity Analysis - 10
Full Factorial: 952 Runs JMP Partial Factorial: 50 Runs
Alternate Double Track Sections (Alternate)
Double Track Progression
Railroad Capacity Analysis - 11
Extend Sidings
Build In (Split)
Build Out (Grouped)
Turnout & Crossover Management
Railroad Capacity Analysis – 12
Rail Traffic Controller
Developed by Eric Wilson from Berkeley Simulation
Software
Emulates a dispatcher controlling train movements across
a network based on train priority
Integrated train performance calculator
Inputs: track, signals, trains, and schedule
Output: delay, average velocity, on time performance
Route Characteristics
245 miles long
10,000 ft. sidings
10 miles between siding centers
(8 miles between turnouts)
2.0 miles between signals
2-block, 3-aspect signaling
1 Origin-Destination Pair
0% grade & curvature
Railroad Capacity Analysis -14
Train Characteristics
Freight Train 1 Freight Train 2 Passenger TrainPower x3 SD70 Locomotives x5 SD70 Locomotives x2 P42 LocomotivesNo. of Cars 115 hopper cars 75 cars 7 Horizon CarsLength (ft.) 6,325 5,659 740Weight (tons) 16,445 5,900 800Max Speed (mph) 30-50 51-70 79-110
30 milesbetween stops
Train schedules are randomized over the 24 hour period All delays are due to mainline train interactions
Railroad Capacity Analysis -15
Analysis Guide
The subsequent analysis shows snapshots of the profiles from the response surface models for both freight and passenger delay
Each panel shows the change in delay for each variable
12 panels total Steeper lines indicate the trains are more sensitive to
a change in the factor level Parallel lines indicate independent effects Non-parallel lines indicate interaction effects
Railroad Capacity Analysis -16
Pass
enge
r D
elay
Frei
ght
Del
ayGrouping
% Double Track
Traffic Level
Freight Speed
Passenger Speed % Freight
Alternate 50% 48TPD 50 mph 110 mph 75 %
73%
Pass
enge
r D
elay
Frei
ght
Del
ayGrouping
% Double Track
Traffic Level
Freight Speed
Passenger Speed % Freight
Alternate 50% 40 TPD 45 mph 110 mph 75 %
60 mph
Pass
enge
r D
elay
Frei
ght
Del
ayGrouping
% Double Track
Traffic Level
Freight Speed
Passenger Speed % Freight
Extend 75% 48 TPD 45 mph 110 mph 75 %
Alternate
Pass
enge
r D
elay
Frei
ght
Del
ayGrouping
% Double Track
Traffic Level
Freight Speed
Passenger Speed % Freight
Extend 75% 48 TPD 45 mph 110 mph 75 %
Split
Pass
enge
r D
elay
Frei
ght
Del
ayGrouping
% Double Track
Traffic Level
Freight Speed
Passenger Speed % Freight
Extend 75% 48 TPD 45 mph 110 mph 75 %
Group
Preliminary Findings
Major Effects Traffic Level % Double Track Freight Train Speed
Minor Effects Progression strategy Passenger Train
Speed Traffic Composition
Interesting Results By being the high priority train, the passenger trains
are less sensitive to the identified factors than freight trains
Adding double track improves train delay in a near-linear manner
Progression strategy effect is modest Priority trains can see significant delay reductions by
improving non-priority train speeds
Future Work Refine the response surface model to incorporate only
significant terms Validate the model against data not used in fitting the
model Analyze the delay distributions to explore the transition
from Lognormal to Exponential Expand the ranges covered in the identified factors to
improve confidence limits Add resolution to the progression from single to double
track while holding some minor variables constant
Rairoad Capacity Analysis – 23 23