Optimal Contro To Save Fuel I Hha09 Rev4

14
Optimal Control of Heavy-Haul Freight Trains to Save Fuel Paul K. Houpt, Pierino G. Bonanni, David S. Chan, Ramu S. Chandra, Krishnamoorthy Kalyanam, Manthram Sivasubramaniam – GE Global Research, Niskayuna NY USA James D. Brooks, Christopher W. McNally, GE Transportation, Erie PA USA (correspondence to 1 st author at [email protected]) Summary: Trip Optimizer is a locomotive control system enhancement applicable to diesel-electric hauled freight that can achieve double-digit fuel savings. Energy savings derive from managing train momentum, with anticipation of its effects, to reduce the net energy outlay by the train as it completes a trip. GE’s system has two major components: the first is a planning system that derives an optimal way to drive the train (throttle together with a corresponding speed trajectory versus distance) subject to speed restrictions along the route and locomotive operating constraints; the second is a dynamic control system that executes the plan closed- loop, correcting for modeling errors from various sources and assuring proper train handling consistent with railroad requirements. To compute a plan, information about the track to be traversed (grade and curvature versus milepost), the power consist makeup (number and type of operational locomotives) and load (tonnage, train length etc) is required together with updated speed restrictions, work crew locations, and other constraints that may vary from day-to-day. This paper first gives an overview of the Trip Optimizer system in operation as implemented on GE Evolution locomotives. Next, key components in the architecture are briefly described, including how the system is operated with the aid of graphic interfaces. Results of pilot testing of the production system on various revenue service trains on Class 1 railroad’s territories are then summarized to demonstrate actual fuel savings in the 4-13% range while achieving acceptable train handling. Index Terms: Fuel optimal train control, freight train automation, energy saving, train cruise-control 1. Background and Design Objectives AAR 2007 “Railroad Facts” show that fuel burned by North American Class 1 railroads in diesel- electric freight service exceeded 4.1 Billion gallons, accounting for 13% of overall operations expense at 2007 fuel prices. While the current hiatus in fuel prices has provided some welcome operating cost relief, long-term trends in fuel prices are nearly certain to cause the expense and percentages to increase. To improve efficiency, performance of locomotive components like the diesel prime mover and electrical power conversion in the traction system have made an incremental impact over the past 20 years. Even hybrid technologies to © International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 1

description

Trip Optimizer is a locomotive control system enhancement applicable to diesel-electric hauled freight that can achieve double-digit fuel savings. Energy savings derive from managing train momentum, with anticipation of its effects, to reduce the net energy outlay by the train as it completes a trip. GE’s system has two major components: the first is a planning system that derives an optimal way to drive the train (throttle together with a corresponding speed trajectory versus distance) subject to speed restrictions along the route and locomotive operating constraints; the second is a dynamic control system that executes the plan closed-loop, correcting for modeling errors from various sources and assuring proper train handling consistent with railroad requirements. To compute a plan, information about the track to be traversed (grade and curvature versus milepost), the power consist makeup (number and type of operational locomotives) and load (tonnage, train length etc) is required together with updated speed restrictions, work crew locations, and other constraints that may vary from day-to-day. This paper first gives an overview of the Trip Optimizer system in operation as implemented on GE Evolution locomotives. Next, key components in the architecture are briefly described, including how the system is operated with the aid of graphic interfaces. Results of pilot testing of the production system on various revenue service trains on Class 1 railroad’s territories are then summarized to demonstrate actual fuel savings in the 4-13% range while achieving acceptable train handling.

Transcript of Optimal Contro To Save Fuel I Hha09 Rev4

Page 1: Optimal Contro To Save Fuel I Hha09 Rev4

Optimal Control of Heavy-Haul Freight Trains to Save Fuel

Paul K. Houpt, Pierino G. Bonanni, David S. Chan, Ramu S. Chandra, Krishnamoorthy Kalyanam, Manthram Sivasubramaniam – GE Global Research, Niskayuna NY USA

James D. Brooks, Christopher W. McNally, GE Transportation, Erie PA USA (correspondence to 1st author at [email protected])

Summary: Trip Optimizer is a locomotive control system enhancement applicable to diesel-electric hauled freight that can achieve double-digit fuel savings. Energy savings derive from managing train momentum, with anticipation of its effects, to reduce the net energy outlay by the train as it completes a trip. GE’s system has two major components: the first is a planning system that derives an optimal way to drive the train (throttle together with a corresponding speed trajectory versus distance) subject to speed restrictions along the route and locomotive operating constraints; the second is a dynamic control system that executes the plan closed-loop, correcting for modeling errors from various sources and assuring proper train handling consistent with railroad requirements. To compute a plan, information about the track to be traversed (grade and curvature versus milepost), the power consist makeup (number and type of operational locomotives) and load (tonnage, train length etc) is required together with updated speed restrictions, work crew locations, and other constraints that may vary from day-to-day. This paper first gives an overview of the Trip Optimizer system in operation as implemented on GE Evolution locomotives. Next, key components in the architecture are briefly described, including how the system is operated with the aid of graphic interfaces. Results of pilot testing of the production system on various revenue service trains on Class 1 railroad’s territories are then summarized to demonstrate actual fuel savings in the 4-13% range while achieving acceptable train handling.

Index Terms: Fuel optimal train control, freight train automation, energy saving, train cruise-control

1. Background and Design Objectives

AAR 2007 “Railroad Facts” show that fuel burned by North American Class 1 railroads in diesel-electric freight service exceeded 4.1 Billion gallons, accounting for 13% of overall operations expense at 2007 fuel prices. While the current hiatus in fuel prices has provided some welcome operating cost relief, long-term trends in fuel prices are nearly certain to cause the expense and percentages to increase. To improve efficiency, performance of locomotive components like the diesel prime mover and electrical power conversion in the traction system have made an incremental impact over the past 20 years. Even hybrid technologies to recover braking energy have been developed, but this option awaits improvements in cost and life expectancy of batteries to achieve wide penetration in the locomotive fleet. In this context, GE and other suppliers have set out to develop system level control strategies to reduce energy consumption, focusing on how the train can be driven for fuel (and emissions) use reduction while satisfying operating constraints of schedule, the rolling stock and track infrastructure. Among key design objectives were:

Computation of optimal driving profiles (speed, throttle [notch] as functions of time or distance) to minimize fuel use with no impact on schedule

Closed-loop operation to maximize consistency in fuel savings and schedule objectives and reduce driver workload

Simple setup and operation by crews with minimal training

Flexibility to modify objectives in route to adapt to changes (route switch, new slow orders, alternate arrival time)

Applicable to all classes of freight service from unit material trains to high HPT (horsepower per ton) premium services

Fuel benefits obtained from every equipped locomotive, incrementally

Use actual locomotive performance characteristics for planning and controls

What has emerged is a control system GE calls “Trip Optimizer,” which has progressed from short breadboard system demonstrations on 15 subdivisions of multiple railroads to commercial pilot programs on two Class 1 railroads in daily revenue service, operating over a wide range of terrains, tonnage and power configurations.

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 1

Page 2: Optimal Contro To Save Fuel I Hha09 Rev4

2. Trip Optimizer Overview

A simplified block diagram of Trip Optimizer is shown in Figure 1. The crew initiates a trip request based on train symbol or other code for the train being driven. A GE off-board system using a satellite link provides track database information about each trip. The GE off-board is linked to the railroad either manually or directly and stores all currently available trips. Once this information is mapped to a road number and received by the lead locomotive, a route is generated and passed to the trip planner.

On-Board Optimal Trip Plan Generation Hardware Implementation

Dispatch Directive

GPS Receiver

On Board Computer (CMU)

EvolutionLoco Controller

Location, speed

Optimized Driving Plan

RequestedTrip move Trip

Planner

TripPlanner

Optimized Speed &

Throttle Plan

Driver

SpeedRegulator

SpeedRegulator

+

-

ThrottleCommand

Location &Model Estimator

Location &Model Estimator

Track data Loco/train makeupSpeed limits

Loco DataLocos +

Train

Grade + drag

Data LinkNew Driver

Display

Min timeTimetable

UpdatedModel data

Sat +GPS

Antenna

DriverDisplay

On-Board Optimal Trip Plan Generation Hardware ImplementationOn-Board Optimal Trip Plan Generation Hardware Implementation

Dispatch Directive

GPS Receiver

On Board Computer (CMU)

EvolutionLoco Controller

Location, speed

Optimized Driving Plan

RequestedTrip move Trip

Planner

TripPlanner

Optimized Speed &

Throttle Plan

Driver

SpeedRegulator

SpeedRegulator

+

-

ThrottleCommand

Location &Model Estimator

Location &Model Estimator

Track data Loco/train makeupSpeed limits

Loco DataLocos +

Train

Grade + drag

Data LinkNew Driver

Display

Min timeTimetable

UpdatedModel data

Sat +GPS

Antenna

DriverDisplay

Figure 1 - Trip Optimizer Conceptual Diagram

The trip planner algorithm utilizes information about the locomotive performance, including efficiency, train data, such as length, weight, and road numbers along with trip information such as origin/destination, slow orders, and preferred routing. The planner produces an optimal speed trajectory and the corresponding expected notch levels, expressed a function of distance along the route. Optimization in the plan generation exploits information about train physics and terrain ahead to manage momentum in the most fuel-efficient way, subject to time objectives (typically minimum time) and speed limit constraints. Resulting speeds are typically not constant and avoid unnecessary braking wherever possible.

After a plan is created, and clearance authorization obtained, the engineer will depart under manual control until a critical speed, e.g. 10 mph, is

reached. Then to follow the optimal plan, the Trip Optimizer control system’s speed regulator can be engaged. Initiation is via key-presses and master control handle confirmation by the operator. Once engaged, the regulator will make closed-loop corrections to optimal throttle notch plan to follow the speed specified in the trip plan to compensate for small modeling errors and external disturbances such as wind.

Location navigation as derived from GPS is used in conjunction with the plan and speed regulator. Onboard algorithms use available locomotive speed data to compensate for satellite dropout and also estimate key train model parameters to validate the trip information received. Severe departures detected result in automatic recalculation of the optimal plan.

Trip optimizer employs an active graphic display HMI (human machine interface) of terrain and situational awareness information to assist the operator with setup, engagement and other operational aspects. In-route changes are incorporated via track switch prompts that ask the operator about desired track at control points. A new plan is generated if needed that conforms to turnout speeds and the new track characteristics. Because Trip Optimizer does not automatically brake in the speed regulator, the operator is notified well in advance of manual braking which is flagged in the plan generation process. Trip Optimizer provides the logic to aid the operator’s transition to braking with warning times of up to two minutes. The system is automatically disabled and an alarm sounded by a supervisory subsystem if the operator does not respond. Supervisory logic is provided to disable the automatic operation when serious errors occur such as: extended loss of GPS; an over-speed is impending that was not in the plan (due to various errors); the train is off the intended route; prolonged airbrake use; and other detected locomotive failures.

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 2

Page 3: Optimal Contro To Save Fuel I Hha09 Rev4

3. System Components

Trip optimizer is organized around the major subsystems shown in Figure 1. This section provides more detail on key Trip Optimizer sub-systems: the Trip Planner, Speed regulator and Human Machine Interface (HMI).

3.1 Trip Planner

The purpose of the planner is to compute a target driving recipe or “profile” which prescribes how the train should be driven from a starting location to a desired end location. The output of the planner is a set of speed and notch (throttle/brake) points which if followed will achieve desired quantitative objectives for the trip, including target arrival time at the destination with minimum fuel use and satisfy all equipment and track operating constraints. Input data to the planner includes information on the power consist, the load being hauled (weight, train length, number of cars, weight distribution), the track route starting and end points, and track description (grade, curvature and standing speed limits as functions of footage along the route). Other input data includes temporary slow orders or other operating restrictions relevant to the current run. Trips with multiple stops to do work (e.g. pickups and setouts) can also be accommodated in a single plan or can be handled as separate plans running from stop-to-stop.

Data for the planner is obtained from both on-board sources (e.g. track and known locomotive characteristics) and off-board sources via satellite radio links to the customer’s manifest and work orders. Some manual entry updates are also available to the crew at all times through the HMI. Various communication interfaces can be accommodated depending on customer infrastructure and preferences.

Both the planner and the speed regulator, which runs the locomotive to follow the plan, are based on simplified equations of motion for the train that are derived from basic laws of physics and energy balances. Models account for effects of grade and

curvature along the track, locomotive tractive effort, braking characteristics and other factors that influence train acceleration such as drag. All models are validated for consistency with observed data and are parameterized so that changes or errors in assumptions from the manifest can be detected and corrected.

Computing the plan is based on solution to a large optimization problem, set up to achieve desired objectives. Algorithms used for the planner are designed to run very fast compared to the time horizon of interest. For example, Figure 2 shows the solution obtained from the Trip Optimizer planner for a 200 mile trip over rolling terrain. This case was for a 4000 ton train operating at a horsepower per ton of approximately 4, typical in premium services. Note the large percentage of the route that is completed without braking, a byproduct of the fuel saving objective in the optimization. Plans by design calculate where braking is required and this information is used within the speed regulator and HMI to alert the operator to switch to manual operation with Trip Optimizer’s motoring only operation. For future generations of the product, the braking calculation will be used to allow automatic operation to be retained even through braking events.

Sp

ee

d (

mp

h)

Th

rott

le s

ett

ing

Gra

de

(%

)

Distance (miles)

0 20 40 60 80 100 120 140 160 180 2000

20

40

60

80

0 20 40 60 80 100 120 140 160 180 200-5

0

5

10

0 20 40 60 80 100 120 140 160 180 200-1

0

1

Sp

ee

d (

mp

h)

Th

rott

le s

ett

ing

Gra

de

(%

)

Distance (miles)

0 20 40 60 80 100 120 140 160 180 2000

20

40

60

80

0 20 40 60 80 100 120 140 160 180 200-5

0

5

10

0 20 40 60 80 100 120 140 160 180 200-1

0

1

Figure 2 - 200 Mile Optimized Trip on Rolling Terrain

Solution to this planning problem required approximately:

900 spatial steps

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 3

Page 4: Optimal Contro To Save Fuel I Hha09 Rev4

1816 decision variables (notch) 5440 constraints 2.25 seconds to converge to required

tolerance on a typical office computer

Speed of the planner is vital because planning with Trip Optimizer is not static. Re-plans can be initiated en-route for numerous reasons, including addition or removal of temporary slow orders, diversion from a main track route to a secondary route, stops added to do work, or change in planned meets and passes that require a siding diversion. If a stop is required due to traffic ahead, and no other changes have occurred, the currently executing plan can be resumed. Otherwise the stop provides an opportunity for the crew, in coordination with dispatch, to update changes in objectives and a new plan is computed

One of the very useful byproducts of the fast planning computation is the ability to generate fuel use / travel time trade-off curves such as Figure 3, which is calculated for the 200 mile example above.

3 3.5 4 4.5 5 5.5 6 6.5 70.6

0.7

0.8

0.9

1

1.1

1.2

1.3

1.4

1.5

1.6x 10

4

14% fuel benefit

Fu

el C

on

sum

ed (

lb)

Travel Time (hrs)

16 min incremental travel time

3 3.5 4 4.5 5 5.5 6 6.5 70.6

0.7

0.8

0.9

1

1.1

1.2

1.3

1.4

1.5

1.6x 10

4

14% fuel benefit

Fu

el C

on

sum

ed (

lb)

Travel Time (hrs)

16 min incremental travel time

Figure 3 – Fuel Travel Time Tradeoff

Each point on the curve has a corresponding plan like Figure 2. While most operators and railroad management will choose minimum time as the objective, there is a high sensitivity of fuel use to travel time. In this example, a 16 minute delay in a 3.5 hr trip yields a 14% incremental fuel saving from the min-time solution, on top of the

improvement compared to manual operation (not shown here). The curve represents a “snapshot” of entitlement for this train on this day taking account of all the prevailing train operating conditions and constraints.

Trip Optimizer’s planner has enormous flexibility to achieve complex requirements and operating rules of a railroad customer and/or operator preferences permitted by the railroad. A simple example shows some of the flexibility possible. Consider the small problem in Figure 4, with speed restrictions shown. Figure 5 shows the optimal plan solution.

A B

295 303.5 304 304.5 305 307.5 310

70

60

5040

70

60

5040

Start

End

Figure 4 - 15 Mile Simplified Planner Problem

Results are shown as a function of distance, but the corresponding time to complete the trip is 15:25 (minutes: seconds) and a total of 788 lbs of fuel are required. Astute operators may argue that a faster time might be achieved by delaying the speed reduction (relaxing some constraints, it is easy to find a plan that is 20 seconds faster at a cost of some extreme braking that would result in poor train handling). The optimal plan is seen to avoid braking to save energy, but has a sustained idle duration between mileposts 300 and 305.

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 4

Page 5: Optimal Contro To Save Fuel I Hha09 Rev4

295 300 305 31040

60

80

Spe

ed (

mph

)

295 300 305 310-5

0

5

10

Eff

ectiv

e N

otch

295 300 305 310-1

0

1

Gra

de (

%)

Distance (mi)

295 300 305 31040

60

80

Spe

ed (

mph

)

295 300 305 310-5

0

5

10

Eff

ectiv

e N

otch

295 300 305 310-1

0

1

Gra

de (

%)

Distance (mi)

Figure 5 – Optimal Plan with ‘long’ Idle Stretch

Since prolonged idle may result in undesirable slack-action, particularly over some terrains, the planner optimizer can be constrained to avoid idle in finding a solution as shown in Figure 6. In this example, adding this constraint requires the plan to add a small amount of braking to stay below the 45 mph speed restriction before milepost 305, but the additional fuel cost is only 2 lb (above the 788 lb) or 0.25%. Adding constraints to achieve desired objectives via the planner can be made active only at specified locations or over the entire route.

295 300 305 31040

60

80

Spe

ed (

mph

)

295 300 305 310-5

0

5

10

Eff

ectiv

e N

otch

295 300 305 310-1

0

1

Gra

de (

%)

Distance (mi)

295 300 305 31040

60

80

Spe

ed (

mph

)

295 300 305 310-5

0

5

10

Eff

ectiv

e N

otch

295 300 305 310-1

0

1

Gra

de (

%)

Distance (mi)

Figure 6 – Optimal Plan tuned to Avoid Idle

3.2 Speed Regulator Subsystem Functions

There are three inter-related functionalities used in Trip Optimizer to execute the plan as shown in Figure 1.

Speed regulator-manipulates the throttle closed-loop to follow the plan when automatic mode is engaged. It functions like “cruise-control” on a highway vehicle, but follows the prescribed varying speed plan from the optimizer. Errors in speed that result from modeling errors for train track and environment (e.g. wind, manifest errors), result in corrections to the optimally planned notch. This assures schedule compliance that is baked into the optimal plan.

The current implementation of Trip Optimizer allows the speed regulator to be active only when motoring: braking is not applied automatically. Over a typical trip, 50-70% of the trip miles can be driven automatically in this fashion depending on the subdivision terrain and train makeup. In computing the plan, regions where braking will be required are identified, and displayed to the driver through the HMI. The speed regulator prompts for and makes a bump-less handoff to the driver where braking is required. When conditions allow automated operation again, the HMI prompts the operator to re-engage automatic operation. While controlling to the planned speed, the system accounts for typical operating rules such as maximum notch/DB levels, power braking restrictions, and maximum “allowed notch above speed” rules.

Train Handling—Assurance of acceptable train handling is critical to any freight train control system that is expected to operate hands-off. Minimum fuel driving strategies turn out to also promote good train handling. As the example in Figure 6 showed, it is possible to create plans that are likely to have better likelihood of producing acceptable train handling. A hierarchy of rules determines how the planned throttle is modified to achieve acceptable train handling. Rules depend jointly on what is coming from the planner, the estimated train state, local track terrain,

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 5

Page 6: Optimal Contro To Save Fuel I Hha09 Rev4

locomotive health and other data to assure proper handling for all terrains and consists. Validation of train-handling performance is done through a combination of off-line simulation in tools like TOES and operator reports in field trials and pilots (see below).

A key benefit of closed-loop operation with the speed regulator is narrowing the distribution of travel-times and accuracy in following speed reductions. Figure 7 compares manual against automatic operation and the distribution of under-speeds (negative values) and over-speeds in transitioning from line speed to various slower speeds with the regulator active. The data is compiled from three runs over an entire subdivision on a North American railroad as part of pilot studies conducted in 2008 all with similar train makeup and HPT. Similar reductions to speed variation have been seen throughout field testing of Trip Optimizer.

10%

90%

25%

67%8%Manual Control

AutoControl

Over-speed (mph) - 5 - 1.5 0 +1.5 + 5

Nu

mb

er o

f re

du

cti

on

s

10%

90%

25%

67%8%Manual Control

AutoControl

Over-speed (mph) - 5 - 1.5 0 +1.5 + 5

Nu

mb

er o

f re

du

cti

on

s

10%

90%

25%

67%8%Manual Control

AutoControl

Over-speed (mph) - 5 - 1.5 0 +1.5 + 5

Nu

mb

er o

f re

du

cti

on

s

Figure 7 – Performance of Speed Regulator vs. Manual

Estimation--Performance benefits from Trip Optimizer are dependent on knowing the various train and track parameters used in the planner optimizer and speed regulator. Track data-bases are vetted through an off-line process, though developing tools to assist in track data-base construction was a significant development effort. Train data extracted from the manifest may be wrong 20% of the time. Provision has therefore

been made to provide on-line algorithms that observe train behavior compared to model predictions built on available data. When significant errors are detected, estimates of the impact on fuel entitlement are used to decide if a re-plan should be created on the fly or delayed to a future stopping point. Decision criteria to replan are flexible and vary by railroad preferences.

3.3 Human Machine Interface (HMI)

Setup & Results Summary--Standard Smart Display screens on Evolution locomotives are used to provide a human machine interface to Trip Optimizer. Together with associated function keys that are located below the on-board display, the HMI provides the means by which the operator sets-up, initializes, engages and disengages automatic operation and shuts down the system. Figure 8 shows a typical Trip Optimizer setup screen. Operators can request data to be downloaded by train symbol or other shorthand and proceed to make last minute edits to the power consist, e.g. change locos in consist different from the manifest, flag isolated units, set DB cutout etc. Future features are being considered to allow other editing capability for data supplied in the manifest and track data. Setup confirmation and review screens (not shown) are also provided before a trip, and summary statistics screens are provided to the operator at the end of the trip.

ChangeLength

Yes Save Changes

PageDown

PreviousCars

ChangeEmpty

PageUp

ChangeLoaded

PageDown

PageUp

End SmartThrottle

Cancel

L12525-0

Use Arrow Keys To Select Correct Mode For Each Locomotive,Then Press F7 To Continue.

HORN BELLSANDPARK

BRK ON

PRK BRK ON

Trip Optimizer – Locomotive Setup

<reserved for aar>

Locomotive Position Power Mode

GE 2010 1 Running

GE 2005 2 Isolated

GE 2015 3 Running

GE 2901 4 DB Cutout

New Power Mode

ATC

40 1201008060

0.0088Rear

90ER

90BP

72BC

140Main

2Flow

200200200155

10

4050

60

70

800

20

30 0Distance

2010GE

Cntr

Reverser

0Effort Klb

IdleThrottle

2:3 0

Consist Klb

0 180

Running

Isolated

DB Cutout

Figure 8 – Sample Trip Optimizer Setup Screen

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 6

Page 7: Optimal Contro To Save Fuel I Hha09 Rev4

Running Screen--Figure 9 Trip OptimizerRunning Screen shows the running screen for Trip Optimizer that appears after departure tests are completed and the proper setup of the train allow the operator to proceed. To minimize “heads down time” for the operator viewing the screen, only essential data to manage Trip Optimizer is provided. Situational data of the standard AAR type is provided in the upper 20% of the screen. In the center is a new rolling strip map with distance traversed established from GPS data, graph of terrain (grade), train on terrain, civil speed limit and, in a different color, temporary slow orders. Under the rolling map is current MP location, track being followed and destination for this trip. About 6 miles are displayed on this example, which is railroad configurable. Automatic status is displayed on the box over the rolling display and the current actual notch being generated by the speed regulator in the box to the right. The light area on the terrain to the right of the train is a region where manual (braking) operation will be required as inferred from the optimal plan. A sequence of warnings to the operator to take over are provided, as the system reverts to manual. When automatic mode can again be resumed, appropriate prompts will be made to the driver.When automatic operation is permitted, the operator presses the appropriate key and moves the throttle to Run 8. The speed regulator will then pickup like the cruise control on a car and modulate the throttle to follow the plan. At any time, the operator can disengage automatic operation by moving throttle out of Run 8 position or pressing the a key, making disengagement straightforward and intuitive.

AirBrakes

ConfirmAuto

ConfirmThrottle

End ofTrain

UpdateTrack

AutoControl

ManualControl

DistanceStart

ConsistManager

DistanceSetup

AutoStart/Stop

ScreenControls

EndTrip

Exit

L12550-0Ready

40 1201008060

47 MPH88Rear

90ER

90BP

0BC

140Main

2Flow

200200200155

10

4050

60

70

800

20

30

EOTMOVE

BATTDEAD

CS TTP19

HORN BELLSANDPARK

BRK ON

BRAKEWARN

PCSOPEN

WHEELSLIP

ALERTER20

UNITALARM

TripOptimizer

AUTOSTOPMM:SS

PARKBRK ON

LeadInd Brk

Cut InAuto Brk

Cab Signal

AUTO CONTROLACTIVE

HORN BELLSANDAUTO

N4

Current MP: 101.2Arrival In: 01:45 Arrival Time: 13:45 EDT

Track: MAIN1Destination: WILLOW SPRINGS

25

5060

50

103 104 105

Speed

Terrain UP: Cut OutCNW: Cut Out

101 102

0Distance

2010GE

Fwd

Reverser

30Effort Klb

N8Throttle

2:3 60 K

Consist Klb

0 180

Figure 9 Trip Optimizer Running Screen

4 – Pilot and Performance Test Results

Trip Optimizer has progressed from a prototype system in 2006 that ran four short-term, supervised pilots on 15 subdivisions to a complete production system now running around the clock in revenue service at two Class 1 railroads without GE supervision.

4.1 Evolution Locomotive Implementation

Trip Optimizer has been implemented as a production version in the hardware shown in Figure 1. Standard locomotive displays used in the SDIS architecture on EVOs are used for the HMI. For later application to non-GE power, other architectures are being considered. Only the lead power needs to be equipped to gain all the benefits of Trip Optimizer.

4.2 Pilot Test Methodology

Overview--A pilot is a key first step in understanding the benefit of Trip Optimizer over a particular subdivision and in preparing the system to run there. Prior to beginning a pilot, work is done to prepare track databases, identify the expected train types and configurations, and coordinate delivery of trip data with the railroad. Runs without Trip Optimizer active are made to collect data used to validate all aspects of the track database. Train handling analysis is carried out in

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 7

Page 8: Optimal Contro To Save Fuel I Hha09 Rev4

off-line simulations for the expected trains to ensure acceptable operation. This includes simulation comparison of train forces with manual operations based on historical event recorder data for similar trains on the same route.

Each pilot begins with manned runs wherein GE personnel ride with each Trip Optimizer equipped train to ensure operation is as intended and collect valuable data for system validation. Supported runs are used to provide crew training and get detailed feedback from each crew covering ease of use, transition from auto to manual, train handling, screen layouts and information displayed. All feedback is integrated in a database to assess gaps and identify enhancements for future product development.

Fuel Use Assessment - Actual test runs are selected in collaboration with the railroad to cover a tonnage and HPT range that is representative of their operations and for which benchmark manual operator runs are available. The same measurement methodology as the customer is used to compute fuel expended with and without Trip Optimizer. For all results discussed here, fuel use was predicted from records of time at notch and fuel-flow at notch summed up for all the power in the consist on a particular run. Procedures are vetted for consistency with railroad practices.

Train Handling Assessment-No train force couplers were available for actual in-train force measurements in any of our Pilot studies, and applying to a large number of trains would be logistically and cost prohibitive. Instead we relied on two methods of validation: (1) post-run analysis of event recorder data from Trip Optimizer trains with a third party simulation tool (similar to and validated against TOES train simulator developed by the AAR); (2) anecdotal subjective reports from crews and their supervisors on the frequency and magnitude of run-ins or other anomalies observed of excessive buff and draft forces in operation. Both methods consistently show, at a minimum, there is no negative impact on train handling with Trip Optimizer deployment compared to crews in the baseline. More

subjective feedback from train crews in the pilot indicate Trip Optimizer is performing better than the baseline from a run-in perspective.

4.3 Current Pilot Status

Several long-term pilots are currently underway at two Class 1 US railroads. Trains are running over six subdivisions containing more than 700 miles of track with tonnages up to 10,500 tons and varied distributed power configurations. Trip Optimizer is running without GE supervision on five of these six, with the last soon to follow. Over 50,000 trip miles have been run as of early 2009 with an average of 60% of these miles in automatic control. It is important to note that on only about 74% of the total miles was automatic available due to various operational factors, so that on average crews have been able to keep Trip Optimizer engaged about 81% of the time where it could be used. These totals are being added to daily at an average rate of 240 miles in automatic control, or 410 trip miles per day.

Fuel Saving results—The common normalization metric used for fuel expenditure has been in gross ton-miles/gallon where more is better or its reciprocal where less is better. Results using gallons per gross ton-miles for the most recent pilot runs completed in 2008 and early 2009 are summarized in Table 1 and Table 2 (actual railroads and subdivisions are not identified for proprietary consideration to the lines at their request). Trains dispatched in both populations ranged in HPT from just under 1.0 to 4.0. Terrains ranged from flat to mountainous, so that this sample includes both the middle and extremes of the population. Operators were representative of the population, both experienced and inexperienced. Savings of fuel ranged from 4.6 to 13% for these pilots; the wide variation in savings reflects the broad differences among territory, train type and railroad operation that were selected to benchmark Trip Optimizer capability.

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 8

Page 9: Optimal Contro To Save Fuel I Hha09 Rev4

Table 1Fuel Savings for Railroad A over 3 Subdivisions

Subdivision

Fuel Use Reduction From Baseline

Number of Trip Opt Test Runs With Valid Comparison Family

Alpha -7.8% 38Beta -13.0% 48Gamma -4.6% 47Ave / Total -8.6% 133

Railroad A

Subdivision

Fuel Use Reduction From Baseline

Number of Trip Opt Test Runs With Valid Comparison Family

Alpha -7.8% 38Beta -13.0% 48Gamma -4.6% 47Ave / Total -8.6% 133

Railroad A

Table 2Fuel Savings for Railroad B over 4 Subdivisions

Railroad B Subdivision

Fuel Use Reduction from Baseline

Number of TO Test Runs With Valid Comparison Family

CHI -5.9% 26NU -8.3% 19ETA -6.5% 21LAMBDA -8.2% 21AVE/Total -7.1% 87

Railroad B Subdivision

Fuel Use Reduction from Baseline

Number of TO Test Runs With Valid Comparison Family

CHI -5.9% 26NU -8.3% 19ETA -6.5% 21LAMBDA -8.2% 21AVE/Total -7.1% 87

Train Handling Analysis—Using the pilot runs for guidance, data was grouped for a total of four similar trains ranging in tonnage from about 4800 to 6800 tons with lengths from 6800 to 7400 feet. Looking through event recorder records where Trip Optimizer was in automatic, approximately 78 total miles were selected and partitioned into a total of 32 “segments” where the train speeds were similar between Trip Optimizer and manual control. These segments ranged in length from under a mile to more than 10 miles. Segments were picked to span the variation in terrain over the subdivisions selected. For each of the selected segments, a TOES dynamic simulation was constructed according to available manifest data, first with the manual field data of notch (and speed) and then with corresponding trip optimizer throttle time history and speed that were recorded. Resulting buff and draft force extremes were captured and are analyzed. Over the 32 segments, the in-train forces were shown to be statistically the same. Trip Optimizer averages 10 kips higher in draft and 2 kips higher in buff with the exact same number of run-in events as manual operation over the same segments. Analysis using this methodology continues to build confidence that

there are no unexpected train handling surprises. These simulation results, coupled with extensive crew feedback from post-run interviews and more than two years of field tests of Trip Optimizer, give confidence to assert that train handling with Trip Optimizer is equivalent to good manual operation.

5 Summary and Conclusions

Trip Optimizer has been shown to be a viable on-board control system for GE Evolution series locomotives to save fuel. By focusing on a closed-loop approach using GPS and an optimized driving plan, savings can be obtained with no compromise to operating schedule. Repeatability in operations reduces operator variability in achieving up to 13% or more fuel saving based on more than 50,000 miles of pilot testing in revenue service. Pilot test feedback from crews and their supervisors suggest that it is easy to set up and use with the provided HMI and graphics display design. The system requires minimal training to rapidly adopt in revenue service. Train handling has been shown, both in detailed simulation analysis and field reports of handling anomalies to be equivalent to good manual operation. While the existing product is a motoring-only design, crews found the cues for transition in and out of regions requiring manual braking intuitive and straightforward to use. Moreover, fully automatic operation could be sustained in 80% of the route distance where braking wasn’t required where other factors (e.g. traffic) did not impede operation. Extensive flexibility has been engineered into the product to not only generate fuel-efficient plans at the start of a journey but to flexibly re-plan as objectives and constraints change during the real world execution of a trip.

© International Heavy Haul Association Specialist Technical Session, Shanghai, June 22-25 2009 9