CSX Corp: Track Changes
Keith Triplett was clearly agitated.
In his 26th year on the job, the dispatcher for freight train operator CSX Corp. was worked up about losing one-third of the capacity at a key rail yard in his territory. Every day, the main track gets stopped up when a Chicago train pulls in, offloads its crew and then awaits new instructions. For hours on end.
Triplett has to figure out where to shunt trains bearing down on that track. Five more trains in a couple of hours, seven more behind those. He calls this "the Willard Shuffle,'' because the problem occurs at the rail yard in Willard, Ohio.
It's a dance Triplett thought should end with the advent of the company's new efficiency initiative, the One Plan.
"You have something to do with One Plan?" he calls out to a visitor inquiring about the program at CSX' operations center in Jacksonville, Fla., which controls movements on 80% of the carrier's rail network. "Why do we have to do all the switching off the main track at Willard?"
This is the kind of problem the One Plan is supposed to resolve. For the first time, CSX is putting in place a series of software programs that are designed to end the improvisational artistry of fabled dispatchers such as Roy Thigpen and Charlie Grady. These legends, whose idea it was to create a single dispatch center out of 33 spread across the country, could take any set of circumstancessuch as the Willard blockagefigure out choices in their heads and remedy the problem.
The new math-driven approach intends to make CSX a regularly scheduled railroad, albeit for freight instead of passengers. Instead of making trains as long as possible and running the fewest crews possible, the rail operator has made disciplineoverall efficiencyits basic goal since the fall of 2003. One network, one schedule, one plan.
"The company's long-term success is dependent on demonstrating the ability to provide reliable service to our customers," chairman and CEO Michael Ward said in January, when he discussed CSX' status at the end of last year.
Using historical statistics and forecasts of traffic for the immediate future, the software will figure out what "blocks" of traffic can be expected between any two points at any given time, the most effective sequence of blocks to send down any piece of track, and what each rail yard can handle effectively. It then creates an intelligent and repeatable schedule that maximizes throughput and minimizes waste. Every day.
The building blocks are, in fact, called blocks. A block can be as few as two train cars moving from point A to point B. The trick, in railroad efficiency, is to have the most blocks move the most distance without "pulling pins" and transferring the blocks to other trains headed for other destinations.
This is no mean task. CSXand dispatchers like Triplettmust juggle 20,000 carloads a day, involving 4,000 locomotives and 100,000 freight cars, to satisfy customers. These loads must move across 23,000 miles of track and deal with the idiosyncrasies of 166 rail terminals, 70 ports, 45 automobile distribution centers, 127 active coal mines and 105 coal-fired power plants.
Using code from the University of Florida, software specialist MultiModal Applied Systems and other sources, the One Plan uses an "optimal blocking model" to figure out the size of blocks it must deal with and the volumes each yard can handle. Then, using algorithms that track each class of freight, it creates "traffic block sequences" and uploads them to a mainframe that can keep track of details on how traffic should flow.
Then, MultiModal's MultiRail software figures out which blocks should be assigned to what train, each day of the week. And CSX' Intelligent Train Scheduling System determines feasible train schedules that it can publish and stick to.
Which is all well and good, in theory. But Ward, speaking to financial analysts in January, noted that the company's operating performance was "not where we want it to be yet.
"We remain very optimistic that the One Plan and the opportunity it holds for the company is going to deliver great results for our shareholders going forward," he said.
But better service to customers has yet to arrive. Railroad industry analyst Donald Broughton of A.G. Edwards in St. Louis says the One Plan "is not yet working. We see no evidence of it working yet.''
Nor is the company gaining kudos for its safety. Last month, the company agreed to pay $1 million to the state of New York for failure to fix hundreds of warning signals.
That isn't stopping CSX. Demand for rail transportation is strong, and so are the company's results. CSX' net income from continuing operations more than doubled in 2004, to $418 million, from $194 million in 2003, on revenue of $8.0 billion, up from $7.4 billion. Its core earnings from rail operations were up for the fourth quarter in a row, and revenue for the 11th consecutive quarter, Ward noted.
But that doesn't mean One Plan is contributing to the gains.Dwell Time'">
Dwelling On 'Dwell Time'
One key measure of whether the One Plan is working, he says, is "dwell time." This is the time a freight car spends in a rail yard before being attached to a new train and sent on its way.
In December, dwell times went up every week in the CSX system, according to railroad performance measures kept by the Association of American Railroads. For the week ended Dec. 10, dwell times were 27.5 hours. Then, 28.0 hours the next week, then 30.6 hours, then 41.3. Dwell times dropped at the start of the year, but went back above 30 hours in the last week of January and every week in February.
Broughton calls the performance "abysmal.'' Not only are dwell times up, but train speeds are down. CSX will only be able to get additional revenue by increasing its pricesuntil it improves efficiency, the analyst says. Otherwise, CSX cannot handle any more traffic, curtailing revenue growth.
One area where the pinch can already be felt is in CSX' automotive freight business. After CSX ran its blocks through the MultiModal software, "we came to the conclusion that we had to restructure how automotive business was handled,'' says MultiModal president Carl Van Dyke.
Before the software ran its algorithmic analysis, most auto freight was processed through a rail yard in Cincinnati. But autos are expensive and easily damaged. Which means they must be handled carefullyand slowly.
The software identified unused capacity at Louisville, Ky. Now, Louisville handles auto traffic coming out of auto plants in Southern states and heading for the West. A yard in Walbridge, Ohio, takes output from Michigan and Ontario plants. A Columbus, Ohio, yard also helps out.
Has it helped?
Clarence Gooden, CSX' chief commercial officer, reported in January that the company's auto revenue in the fourth quarter was $218 million, down 4.4%, or $10 million, from a year earlier. Light-vehicle production in the U.S. was up 3% from the previous yearbut downtime at CSX-served plants was "more than four times that of the fourth quarter of 2003,'' Gooden said.
One place where One Plan has clearly come through is in disaster recovery, according to transportation applications director Kathleen Brandt and assistant vice president Dharma R. Acharya, an operations researcher. CSX is based in Florida and, last fall, three major hurricanes disrupted operations, almost back to back to back.
"Because it is an outdoor sport, there's a lot of variability,'' says John L. West, president of CSX Technology.
In the past, such sport would have knocked CSX' roads out of commission for two months. Instead, it took just two weeks to fix rails and resume normal, scheduled operations.
Undetermined, to date, is whether CSX will achieve two major benefits touted by proponents of the One Plan: saving 1.5% of the miles its cars must travel each year through better routing, and cutting by 5% the number of times each car must be handled as it moves through rail yards toward a destination.
Part of this uncertainty is because the first phase of the One Plan concentrated on scheduling CSX' main long-distance traffic. In the second phase, just now beginning, the efficiency drive will focus on local railroad operations.
That means a lot of idiosyncratic knowledge, held in the head of managers like Thigpen, has to be built into the scheduling system. For instance, until now, models have continually shown that more traffic should be moved through the Atlanta yard.
But managers knew that wasn't the case, Brandt says. Why? Getting over the hump. In yards like Atlanta, cars are pushed over humps and thus slide into place on a new train.
In Atlanta, hump A has a physical limit: It can only handle 42 cars at a time. That makes it hard to create long trainsa factor not taken into account by the system out of the box.
Problems Around the Bend
Getting that kind of detail right is crucial, because CSX' corps of dispatchers is aging. Less rail yard experience is being brought to bear on the management of train movements.
Over the next five years, 70 of the 350 people working in the dispatch center will retire, according to director C. Brock Lucas. Only half of their replacements will have any prior railroad experience. They must take 12 weeks of classroom and 30 weeks of on-the-job training to get up to speed. And most won't have "the light go on" about how the job really works until they're in it for a year, Lucas says.
Some of the effect can be countered by higher productivity. Where determining the location of a train used to take 16 keystrokes, now it may take only four or five.
But regardless of which or how many keystrokes are entered, software will never solve problems that aren't built into its algorithms.
Which brings back to Triplett and his daily shuffle.
Every day, the 378 comes out of Chicago at 6:15 a.m. and heads to Buffalo. By 3 p.m., it reaches Willard.
But then it sits on the main track, blocking it. There is no further "call" for its services. Until hours later. If the "call" doesn't come until 7 p.m., the train will still sit there until about 8:30. The new crew has to come in, get its job briefing, check out the equipment, make sure it's got the proper blocks in tow and then take off.
Meanwhile, Triplett has to figure out how to move eastbound and westbound trains through Willardand switch loadsusing just two other tracks.
"This is nothing new," he says, out of frustration. "We run this train every day."
This is another problem not solvable by software. Software can't replace hardware like railroad signals or switches.
For instance, the day before Triplett ranted in Jacksonville about the Willard Shuffle, a 37-year-old man died on impact when he drove his truck across a CSX crossing in Pompano Beach, Fla. According to a law enforcement official, the guard gates were not down when the train came through.
In Willard, the culprit, according to Triplett, is a piece of equipment called a "hand-thrown crossover" switch that went out of service a couple of years ago. The crossover was torn up in a storm and never replaced, he claims.
Why is that important?
With the crossover switch, the 378 and two other trains like it, the 386 and 390, would not have to be parked in the middle of the main track. They could be switched into temporary storage on other rails in the yard.
Instead, Triplett continues to do the Willard Shuffle, as many as three times a day.
In the meantime, Ward remains "very optimistic" that shareholders can expect great results from the One Planat some point in the future.. Base Case">
CSX Corp. Base Case
500 Water St.,
Phone: (904) 359-3200
Business: Largest freight train operator on the East Coast.
Chief Technology Officer: John L. West
Financials in 2004: $8 billion in revenue; $418 million in net income from continuing operations; net profit margin 4.6%.
Challenge: Run freight trains on a fixed schedule, throughout network; minimize improvisation; maximize efficient use of trains, rail yards and personnel.
- Increase the number of cars on its rail lines on any given day, from 236,289 at end of February 2005.
- Improve average speed of a train, which fell to 20.3 mph in 2004, from 21.1 mph in 2003.
- Reduce amount of time a train spends in a terminal between runs, which rose to 28.7 hours in 2004, from 25.3 hours in 2003.