RELATIONSHIPS BETWEEN MAINTENANCE & SERVICE PLANNING Lexcie Lu, Transit Analyst Dorchester, Mass. © 1980, 2002 Steve Zabel/Joe Testagrose © 2003 Robert

RELATIONSHIPSBETWEEN

MAINTENANCE & SERVICEPLANNING

Lexcie Lu, Transit Analyst

Dorchester, Mass.

© 1980, 2002 Steve Zabel/Joe Testagrose

© 2003 Robert Mencher

Assisted by

Lexcie Lu TRB #04-2697 – The Relationship between Maintenance and Service Planning

Slide 2 of 18

Outline

1. Problem Statement2. Social Externalities of

Maintenance3. Decision Support Framework4. Discussion

a. Inspection v.s. Repair Costsb. Consumer Perception of Service

Failuresc. Stochasticity of Transit Operationsd. Effect of Reduced Maintenance on

MDBF

5. Conclusions/Recommendationsa. Maintenance Standards Fault

Managementb. Maintenance Schedulingc. Engineering Research


Slide 3 of 18

Problem Statement

“How do we maximize the benefits provided

by fixed resources on a crowded transit corridor, considering such

user benefits as frequency, timeliness, reliability, safety, and agency traincrew, maintenance,

capital costs?”

or

“Should you run your EMU’s till they drop in their tracks?”


Slide 4 of 18

Hidden Cost of Maintenance

TrackPossession

Trains Detoured, Cancelled

Lost RevenueLost Goodwill (read “Future Revenue”)

Unrecoverable Fixed Costs(e.g. Rolling Stock Lease)

• British Rail (West Coast Route Modernization)

• “Big Dig” in Boston– Avoided I-93

clousure

• Chicago CTA – Green Line

rebuild lost ridership

– Cermak: weekendconstruction

“Train”Possession

Trains Annulled

Lost RevenueLost Goodwill (e.g Overcrowding)

Unrecoverable Fixed Costs(e.g. Crew Guaranteed Pay)

• Airlines – off peak maintenance

• Rush hour elevator– All vehicles

used!

© 2001 Lexcie


Slide 5 of 18

OverMaintenance(asset underuse)

High residual

component life (High

costs)

Maintenance Planning is a Balancing Act

UnderMaintenance

(asset overuse)

Low utilization

Overcrowding

Loss of Benefits

Complaints

Low Ridership

Christmas Trees

High labor costs Indifference

Poor ride quality,

poor appearan

ce

Lost Revenue

Loss of Goodwill

High costs“Deferred

maintenance”

Low reliability

Unsafe conditio

ns

Line Chiefs have to walk a tightro

pe!

Complaints

DelaysLoss of Benefits

Loss ofRevenue

DecisionSupport

Tool


Slide 6 of 18

Decision Support Framework

• Four models and an assumption– Passenger Spill Model

– Probabilistic Maintenance Impact Model

– Operating Cost Model– Wait-Time Savings Model– Life-cycle Cost Assumption

• Cost-Benefit Analysis Framework

Cost(Operations & Maintenance) = Cost(Failures) + Cost(Spill) + Cost(Wait-Time) + Cost(Crew) + Cost(Equipment Life-cycle)

• Compare different O&M scenarios– Some costs dominates others– Some benefits swamps costs… and vice-

versa!

Read about this in the paper


Slide 7 of 18

Orange Line Results

• Risk and Reward– More maintenance = less service

• ‘Protect’ set is a scheduled cancellation– Generates little value, so run all you’ve got– Determine headways based on 6am

availability

• Do inspections at night (less demand)• Do repairs in rush

– It’s already broken, don’t send it out!


Slide 8 of 18

Discussion:Inspection v.s. Repair

• Deferred maintenance leads to additional repair maintenance in the longer term

or does it?

Classical Relationship: Repair & PM Expenses

0 1 2 3 4 5 6 7 8 9 10

Planned Number of Protect/PM Sets

Eco

no

mic

Co

st (

Fle

et,

Th

ou

san

ds)

PM expense Repair expense TOTAL

CONCEPTUAL

Source: Haven (1980), p.15

The curve might be remarkably flat…


Slide 9 of 18

Trainset Utilization Economic Value Analysis

0 1 2 3 4 5 6 7 8 9 10

Planned Number of Protect/PM Sets

Eco

no

mic

Co

st (

Fle

et,

Th

ou

san

ds)

Frequency loss E($) Rush hour failures

Traincrew Passenger spill

Mileage-related maintenance PM expense

Repair expense TOTAL

Inspection + Repair + Operating + Customer

Costs• If PM takes a trainset out of rush

use, think hard about how you might break the job into two sessions, or defer it!

Sources: Railcar Maintenance Model, MBTA Budget 2001

MBTA Orange Line

Must recalibrate for each line, each transit authority,

each maintenance operationand operating plan

CONCEPTUAL


Slide 10 of 18

© 2003 Chuchubob

Consumer Perception of Service Failures

• Correct measure is:– “How long did I wait until the train

came?”(% of passengers waiting more than 1 hdwy)

– “Did Transit make me late for my meeting?”(% of passengers arriving within their own “comfort buffer”)

• Not:– Average wait time– MDBF– Schedule adherence– # of dropped trips

Is there an inherent disutility

of failure (i.e. image problem),

or do riders understand that

failures are inevitable?


Slide 11 of 18

Customer Requirement/ Performance Interaction

• Passengers will tolerate different amounts of delay, depending on trip purpose

• On-time performance distribution will vary, depending on railcar reliability (& other things)

Passenger Ontime Requirement Distribution

0.00

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.10

0.11

0 2 4 6 8 10 12 14 16 18 20

Maximum Tolerable Minutes Delay

Per

cen

tag

e o

f P

asse

ng

ers

Trip Ontime Performance Distribution (Normal)

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

-6 -4 -2 0 2 4 6 8 10 12 14 16 18 20

Headway Adherence (Minutes Delay)

Per

cen

tag

e o

f Tr

ips

Trip Ontime Performance Distribution(Erratic)

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

-6 -4 -2 0 2 4 6 8 10 12 14 16 18 20

Headway Adherence (Minutes Delay)

Per

cen

tag

e o

f Tr

ips

Cat

chin

g an

Am

trak

tra

in

Mee

ting

a da

te

Den

tist

appo

intm

ent

Job

inte

rvie

w

Com

mut

e

Goi

ng t

o sc

hool

Schedule allows 50% of trains to arrive on-time

(MBTA practice)

If trains are unreliable, mechanical/longer delays will increase

Ask me about this:Not in the paper

CONCEPTUAL

Rai

l exc

ursi

ons


Slide 12 of 18

Lost Productivity Associated with

Mechanical Delays• If customers have to leave earlier for the

same appointment to have 95% confidence of arriving on-time, precious time is lost.

• To calculate the social disbenefit of reliability loss, the correct measure is “increase in average buffer time”

• Decrease in on-time performance (to within four minutes) from 85% to 78% increased average buffer by… 47 seconds!

Contribution to Average Buffer Requiredfor 95% Confidence, by Type of Customer

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

0 2 4 6 8 10

Maximum Tolerable Minutes Delay

Co

ntr

ibu

tio

n t

o A

vera

ge

Bu

ffer

(M

inu

tes)

Normal

Erratic

Lost productivity (between the two curves)

CONCEPTUAL


Slide 13 of 18

Stochasticity of Transit Operations

• What’s really causing the delays?

• Cancelling one train due to maintenance difficulties is better than sloppy headways– Except for people on that train, and one

after– ‘Spreading out’ will further reduce delays

Effect of Operating Scenario on Average Passenger Wait Time

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

2.50 2.60 2.70 2.80 2.90 3.00 3.10 3.20

Average Passenger Wait (Minutes)

Hea

dw

ay C

oef

fici

ent

of

Var

iati

on

Base Case No Hdw y Mgmt (Allow Bunching) Annull 0700 Annull Any Train (No Intervention)

Ask me about this:Not in the paper

Ineq

uit

y o

f D

ela

ys

Quantity of Delays

Source: Headway Degradation Model


Slide 14 of 18

Transit Headway Degradation Model

• Straightforward 4 steps (30 run average)– Sectional Runtime Schedule– Perturbations (Normally Distributed)

• Simulated ‘Bunching’ (later trains run later)

– Schedule + Perturbations + Constraints = Schedule Adherence & Headway Data

• e.g Train Blocking, Minimum Separation

– Demand + Headway = Passenger Wait Times

• Determine the sources of delays– Forced cancellation much more

manageable than general headway degradation

Average Delays Under Different Operating Scenarios

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

Ideal Base Case No Hdw y Mgmt

(Allow Bunching)

Annull 0700 Annull Any Train

(No Interv ention)

Operating Scenarios

Pas

sen

ger

-Min

ute

s D

elay

p

er R

ush

Ho

ur

Bunching Cancellation

– How to distribute delaysequitably?

– Average delay the correct measure?

– How do riders plan trips?


Slide 15 of 18

Effect of Weather on Failures

• Does maintenance really reduce failures?

… or does warmer weather?

• Can you do anything about this?– Correl(MinTemp<25ºF, Failures) > 0.6

MBTA Orange Line Failure Data, by Type of Fault (July 2002)

0

1

2

3

4

5

6

0207

01

0207

02

0207

03

0207

04

0207

05

0207

06

0207

07

0207

08

0207

09

0207

10

0207

11

0207

12

0207

13

0207

14

0207

15

0207

16

0207

17

0207

18

0207

19

0207

20

0207

21

0207

22

0207

23

0207

24

0207

25

0207

26

0207

27

0207

28

0207

29

0207

30

0207

31

Date

Fai

lure

s p

er D

ay

Air Brake Half Door Side Door No Door Indication

Total = 30 events, 10 of which are Air

MBTA Orange Line Failure Data, by Type of Fault (January 2003)

0

1

2

3

4

5

6

0212

24

0212

25

0212

26

0212

27

0212

28

0212

29

0212

30

0212

31

0301

01

0301

02

0301

03

0301

04

0301

05

0301

06

0301

07

0301

08

0301

09

0301

10

0301

11

0301

12

0301

13

0301

14

0301

15

0301

16

0301

17

0301

18

0301

19

0301

20

0301

21

0301

22

0301

23

Date

Fai

lure

s p

er D

ay

Air Brake Half Door Side Door No Door Indication

Total = 40 events, 20 of which are Air

Source: MBTA MCRS

Source: MBTA MCRS


Slide 16 of 18

Inclement Weather

•3 interpretations for cold weather

– Unpreventable failures?

– Preventable by inspection?

– Requires operating & design precautions to mitigate its effects

Effect of Inclement Weather on Air, Brake and Door Systems Failures

y = 0.33x + 2.65

R2 = 0.41

0

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

0 2 4 6 8 10 12 14 16 18 20 22 24 26

Minimum Temperature, Degrees below 25¨F

Nu

mb

er o

f F

ailu

res

(in

th

e p

ast

7 d

ays)

•Research needed to determine % of failures actually preventable through maintenance

•Failure mode data required

Read about this in the paper

Sources: MBTA MCRS, Taunton MA Weather Data Center, Erik Wile

© 2002 Lexcie


Slide 17 of 18

Conclusions & Recommendations

• Light Maintenance Standards: The Old Way– Valve rebuild every x days;

Bearings greased every y miles

• Fault Management: The New Way– Opportunistic inspection (preventative)– Measure maintenance effectiveness through

railcar-related passenger minutes lost

• Some Standards Should Remain: Safety– Articulated joints on MBTA Type 7 cars,

dismantled, cleaned and greased every 6 years

• Maintenance Scheduling: Service Impacts– Maint. not ‘constraint’ – consumer of asset time– Scheduling to evaluate impact on service levels– Determine serv. & maint. schedule concurrently

• Engineering Research: Test, Get Data– Collect failure mode data accurately– Do post-mortems for in-service failures– Test things; test it again; test it different ways


Slide 18 of 18

Acknowledgements

• Research assisted by individuals from: Massachusetts Bay Transportation Authority, MIT Department of Civil & Environmental Engineering, and Reebie Associates

• Funded haphazardly by a variety of sources

Lexcie Lu, Transit Analyst, ([email protected])26 Grant St., Unit #3, Dorchester, Mass. 02125-1223

Freight Train in Harmony

All photos © 2001 Lexcie Lu unless otherwise specified

© 2003 Lexcie

Documents

RELATIONSHIPS BETWEEN MAINTENANCE & SERVICE PLANNING Lexcie Lu, Transit Analyst Dorchester, Mass. © 1980, 2002 Steve Zabel/Joe Testagrose © 2003 Robert