Data Mining: Maintaining great numbers

Oren Summer, president of FleetNet America, envisions his company becoming a clearinghouse for fault codes for anything that resides in a vehicle’s ECM.

Jim Coffren’s use of data mining often confirms the old 80/20 rule – that 80 percent of an output comes from 20 percent of the inputs. It’s said, for example, that 20 percent of workers do 80 percent of the work or account for 80 percent of the waste. It’s surprising, however, when empirical data confirms such an old-fashioned rule of thumb.

After fuel surpassed labor as the top expense for Beloit, Wis.-based Blackhawk Transport, Coffren began looking especially closely at the greatest area of fuel waste – engine idling.

“For a long time, we’ve always made the assumption that idle fuel is a factor of idle percentage time,” says Coffren, maintenance manager for the 125-truck carrier. “That has a general truth to it.” And he found that 20 percent of drivers with the highest idle percentage time accounted for 80 percent of the fleet’s idling fuel costs.

But Coffren probed deeper and found that drivers were just part of the problem. On a weekly basis, Coffren uses Excel spreadsheets to analyze data captured by onboard computers and other software applications. By sorting and pivoting the data, he identifies trucks that consume the most fuel, and then finds out why. In addition to idle percentages, Coffren analyzes each vehicle’s total fuel consumed and rate of fuel consumption.

In some cases, trucks with lower idle percentages consume more fuel idling than trucks with higher idle percentages. This helps identify what trucks need maintenance, such as reconfiguring the engine’s parameters to idle more efficiently. By monitoring miles per gallon and fuel consumption, Coffren also has identified vehicles that are not spec’d properly or matched with the right application.

“You find out that 10 to 15 percent of the fleet, no matter how good or bad the driver is, still consumes the majority of idle fuel costs,” Coffren says. “If you go out and address this 10 to 15 percent of the fleet, you can get a 60 percent improvement by focusing on a few units.” To identify such cause-and-effect relationships among fuel costs, Coffren uses an add-in package for Excel called descriptive statistics that helps him monitor the outcome of corrective actions.

Managers like Coffren mine data gleaned from electronic control modules, enterprise software systems and other sources to unearth hidden paths to higher return on their companies’ asset investments. Maintenance executives at fleets of all sizes are leveraging data not only to improve preventive maintenance (PM) but also to predict and plan for failures, breakdowns and other high-dollar events.

Starting with standards
Any meaningful data mining effort begins with inputs that are clear, precise and consistent. Remember: Garbage in, garbage out. This challenge grows not just with a fleet’s size but also with geographical diversity. Even within the same fleet operation, managers at different facilities might use different metrics, for example.

In maintenance, the heart and soul of data analysis is the Vehicle Maintenance Reporting Standards (VMRS) coding system, which helps fleet maintenance management systems capture a wide range of details in an easy format for analysis. Managed by the Technology and Maintenance Council of the American Trucking Associations, VMRS helps standardize the tracking of product and component reliability and costs.

In most computerized maintenance systems, technicians often rely on VMRS codes when they enter equipment repair orders. For example, Lexington, S.C.-based Southeastern Freight Lines (SEFL) developed its own maintenance software system to enable technicians to document repair orders using VMRS.

At SEFL, every time a technician creates a repair order, he selects the reason for repair – PM, a capital improvement, a modification, recall, inspection, warranty, etc. He also identifies the part(s) repaired or replaced by unique VMRS codes. For uniformity, VMRS codes supersede part numbers from manufacturers and suppliers. Technicians also apply failure and labor codes to each repair, says David Foster, director of maintenance for the 2,600-truck company.

Having all repair information properly coded from the start makes the data simple to mine, Foster says. Maintenance managers can export data easily as text files from SEFL’s database to programs such as Microsoft Access and Excel to run queries and perform in-depth analysis.

“We were just in a meeting with operations sharing information concerning equipment abuse,” Foster says. “We were trying to point out exactly what equipment abuse costs us, and what type of abuse it was. We were able to do that because of the way our system is set up with VMRS.” As a result of SEFL’s recent analysis of equipment abuse, management is making a concerted effort to get forklift drivers to reduce abuse to interior trailer sidewalls and doors, Foster says.

The very nature of VMRS lends itself to drill-down types of analysis, says Oren Summer, president of FleetNet America, a company that manages breakdown services for fleets of all sizes and configurations.

FleetNet built its own maintenance management system that incorporates VMRS codes and captures hundreds of other data elements on equipment failures and breakdowns. Using its software, FleetNet can prepare various types of reports for customers on failures – by system and by system assembly components, such as the fan hub in an air cooling system.

Bob Flesher, managing director of vehicle maintenance for FedEx Ground in Moon Township, Pa., says FleetNet’s slicing and dicing of data using VMRS codes has helped him prevent failures from happening.

“With this help, we have reduced in-route breakdowns by 79.52 percent over the last four years – impressive when you consider that we’ve had double-digit growth during the same time period,” says Flesher.

While VMRS helps identify trends and simplify reporting, determining the root cause of a failure requires human judgment. But by using historic failure data that’s coded properly, managers can identify root causes quickly.

A broader perspective
Although VMRS is the bedrock of maintenance data mining, it’s not enough to isolate problems at the component or part level. By its very nature, data mining analyzes numerous variables – some of which might not seem relevant at first.

Because FleetNet handles about 140,000 breakdown calls a year, it gathers significant data on all types of system and component failures. These failures can be sorted, grouped and layered in many ways, such as by location, customer size and type of operation – long or short haul – and equipment make and model.

As an example of what data mining can reveal, Summer says turbo failures were rampant during the hot summer months. By comparing turbo failures across multiple dimensions, such as by area and type of operation, Summer says he found instances where engine RPMs or fuel settings were not appropriate for the type of application. With this information, customers can prevent future breakdowns by changing their engine parameters and spec’ing their equipment differently.

SEFL recently began measuring shop productivity on a monthly basis, tracking such metrics as direct labor percentage, the number of PM schedules performed, expenses for parts and the number of federal DOT inspections. For deeper analysis, the company groups shops of similar size and shifts together in a spreadsheet program.

Comparing this data highlighted key differences in how shops document their activities. Without digging down to find these differences, the top-line productivity results for each shop would be misleading, Foster says.

“One of the challenges of comparing a lot of shops is how much traffic is going through there,” Foster says. A shop in Orlando may lead a similar-sized shop in south Georgia – but the traffic through Florida may be three times as much. With more traffic flow, Florida has more opportunity to do minor repairs and inspections. On the other hand, Georgia has to work harder to bring in equipment for repair, he says.

Identifying such differences in shop operations helps SEFL evaluate its current and future staffing requirements at each facility, Foster says. “What was true years ago may not be the case today.”

Data analysis tools like Access and Excel – both part of the widely used Microsoft Office suite – often are sufficient to mine data at small and large fleets. But some large fleets use more sophisticated data mining applications to compare and analyze data throughout the company, no matter where the data originates.

Pacific, Wash.-based Gordon Trucking uses a business intelligence application called Cognos. The 1,150-truck company pulls data from multiple sources into a data warehouse that incorporates its FleetRx maintenance software, spreadsheets and safety and dispatch systems, says Kirk Altrichter, Gordon Trucking’s vice president of maintenance. Unlike Access or Excel, Cognos can incorporate data from multiple platforms simultaneously without having to create separate files or databases.

Altrichter has been using Cognos only a few months, but the software already has simplified maintenance reporting greatly. “The biggest thing we are working toward is predictive maintenance,” he says. “It is a lot easier with Cognos to do that.”

Succeeding in failures
One of the most powerful applications of data mining in maintenance is pinpointing the useful life of vehicle components and systems. By being able to predict failures, managers can maximize uptime and minimize costs by replacing components proactively during scheduled PM intervals.

The availability – and cost – of extended warranties for specific vehicle systems makes predictive analysis especially useful. Using a simple analysis of historical repair data from its Transman software from TMT Software, Venezia Bulk Transport Inc. has saved thousands of dollars by determining which extended warranties to buy – and which not to buy.

“We can go back and look at trucks that have not had (the extended warranty) and look at what we spent in two years,” says J.P. Venezia, warranty director for the 420-truck carrier based in Royersford, Pa. “A lot of times, we spent half as much as the extended warranty would cost.”

Vehicle systems and components often are subject to various forces of wear and tear, and failures rarely happen in an orderly, predictable pattern. But while a turbo failure may seem like a random event, interesting patterns can emerge when the event is compared with historic failure patterns of similar systems and system components. Specifically, these patterns are best described in terms of a probability distribution (or model) and used as a basis for managerial decisions to replace components proactively.

The Weibull distribution often is used by engineers to model failures of physical systems, such as vehicles, in which the number of failures increases with time. Southeastern Freight Lines uses a Weibull analysis software package to match its historic failure and repair data to a Weibull distribution. By doing so, the company knows the mean time to failure of specific components and systems, Foster says.

“(The software) will show you the mean time to failure in time, miles or other measures such as fuel consumed,” Foster says. “It will also tell you, based on the pool of components being used, the ‘B life’ of those still operating based on the history you input.”

The B life represents the time in which a certain percentage of failures for a part will have occurred. For example, suppose the B life of Brand X tire is that in 50,000 miles, 20 percent of the tires will fail.

“This helps us in several ways,” Foster continues. “One is to compare different suppliers’ products’ lifecycles, and to determine when to replace these components. We want to be proactive in servicing or replacing these components without sacrificing usable life. We build our PM schedules around this information.” During the past 15 years, analysis tools such as this have played a vital role in helping SEFL cut its maintenance expenses in half by understanding which practices work and which ones do not, he says.

Gordon Trucking uses Cognos to analyze failure rates of components and vehicle systems. Using historic failure data retained in its FleetRx maintenance software, management can establish upper and lower control limits, Altrichter says. Control limits are determined by statistical criteria; points beyond the control limits indicate that assignable causes likely are present.

“We’ve made decisions on products that weren’t holding up to expectations or changed maintenance cycles on certain things to pre-empt breakdowns,” Altrichter says. The company recently had instances of some turbos going out at 350,000 miles.

“It makes sense to replace them ahead of time if 60 percent of them are going out on the road where it costs three times as much,” Altrichter says. “We do a pre-emptive strike 10,000 to 20,000 miles prior to the trend. We have seen some benefit from that.”

Chasing the big picture
Besides analyzing specific failure trends of vehicle systems and components, fleets use data mining to get a clear view of the big picture – lifecycle costs for each type of asset. Uncertain with the performance of 2007 engines in Class 8 tractors, fleets are developing new strategies to lower their lifecycle costs, such as changing the timing of vehicle purchases and disposals.

J.P. Venezia says one of the most revealing data mining projects he has done is to compare the cost per mile of all types of vehicles and engines in the fleet. Using the company’s Transman software, Venezia can plot a vehicle’s maintenance costs by both time and mileage.

Over a five-year period, the cost per mile generally doubles at a certain point in time, Venezia says. He uses this information to multiply a truck’s CPM by the number of miles traveled in a month, and to compare the result to a truck’s monthly payment. “When you back into the numbers, the cost per mile is more than your monthly payment,” Venezia says. “It really makes sense to look.”

Republic Services uses the Dossier fleet maintenance management system from Arsenault and Associates to collect and analyze enough data to paint a highly accurate picture of cost trends in its vehicles’ 10-year lifecycle.

“We know that in year 6 we have to invest heavily in the truck, be it another transmission, cylinder or steelwork,” says Jerry Wickett, vice president of purchasing and maintenance for the Fort Lauderdale, Fla.-based provider of solid waste collection, transfer and disposal services. “In the sixth year, costs flatten out and settle back down.”

Republic Services analyzes its vehicles on a cost-per-hour basis for 150 different hauling operations in 21 states. Having this data segmented by region helps determine annual maintenance budgets for each company.

“In fair-weather states like California, costs run lower than in Las Vegas because of extreme heat,” Wickett says. “(Dossier) points out if somebody is above average, and we start digging in and find out why costs are higher.” It may be due to equipment in a fleet being older than average or a recent acquisition of vehicles from another company, he says.

It’s not surprising that large, diverse fleets like SEFL or Republic Services mine data, but Blackhawk Transport operates a small fraction of the vehicles those operations do. Although data mining may sound scary and complicated, Coffren says the concept is really quite simple – to use existing information to find practical ways to make better managerial decisions. And you certainly won’t benefit from data mining if you don’t try it.


TECHNICIANS IN YOUR TRUCKS
Vehicle electronics help forecast, prevent breakdowns

Mining failure data to predict the life of components can reduce costs and improve vehicle uptime, but fleets increasingly are adopting what might be called “near-real-time data mining.” For years, fleet managers have retrieved data, such as fault codes, from vehicle systems and components as they return to the yard or come in for preventive maintenance. In recent years, however, downloading data remotely from vehicle electronic control modules (ECMs) has become more affordable as the cost of wireless communications has gone down.

But can fleets really make effective use of all this vehicle data?

“The key is to be able to capture live data and to capture it as consistently and as efficiently as possible,” says Jim Coffren, maintenance manager for Blackhawk Transport, a 125-truck carrier based in Beloit, Wis. For a few dollars extra per vehicle, Blackhawk receives weekly reports on driver and vehicle performance data through its mobile communications provider. This data is captured from each vehicle’s ECM, and with a click of a button, Coffren imports it into a custom spreadsheet for analysis.

Some companies are finding new ways to use wireless communications with the vehicle’s ECM to predict failures and take proactive measures – in real time. FleetNet America President Oren Summer says that a customer of his company’s breakdown management service now is using a device to monitor the amperage of its alternators. The device is connected to the fleet’s mobile communications system, but sends the readings directly to FleetNet.

If amperage falls below a certain level, FleetNet routes the driver to a repair shop to check the alternator. FleetNet plans to provide this same type of monitoring service for all vehicle systems monitored by the ECM. It intends to collect vehicle fault codes directly from customer vehicles through interfacing with mobile communications and expects to announce one such partnership later this year, Summer says.

“Once we do that, predictability becomes very viable,” he says. “In a couple of years, we will know more about the engine than the maintenance facilities. With more information, we can make a lot of things happen.”

Summer says his vision for FleetNet is that it will become a clearinghouse for fault codes for anything that resides in the ECM, from engine data to transmissions and tire pressure. Depending on the customer agreement, FleetNet will either react to fault codes as it would with any breakdown or repair request, or provide the codes to customers in an easy-to-understand format.


Case Study 1
Gordon Trucking, Pacific, Wash.

Gordon Trucking recently enhanced its maintenance data mining efforts by implementing a data mining and warehouse solution from Cognos. The solution pulls data from a myriad of sources into one location for analysis. One of its recent data mining successes is to pre-empt breakdowns from turbo failures. The 1,150-truck company made a decision to replace turbos 10,000 to 20,000 miles before the failure trend.

“We’ve seen some benefit from that on a series of trucks that were starting to go south,” says Kirk Altrichter, vice president of maintenance. “We found a pretty tight band of where they were failing.”

Case Study 2
Blackhawk Transport, Beloit, Wis.

With 125 trucks, Blackhawk Transport burns up to $35,000 a month in fuel idling. Through in-depth analysis of idling, the company found that $20,000 of this cost was due to mechanical problems and driver behavior from a very limited number of trucks. By improving the performance of those vehicles and drivers, the company is saving between $300,000 and $500,000 a year in fuel.

“The key is to understand technology and techniques, and to ask yourself every day, ‘Is this the best we can do, and is this a fight we want to fight?’ ” says Jim Coffren, maintenance manager.

Case Study 3
Southeastern Freight Lines, Lexington, S.C.

Since Southeastern Freight Lines designed its own maintenance management system 15 years ago, the 2,600-truck carrier has cut its costs in half by analyzing what works and what doesn’t across all areas of maintenance. Shop productivity now is at 85 percent systemwide, but SEFL isn’t satisfied: The company recently began comparing metrics across all its facilities to find even more areas for improvement.

“We have a great IT staff here,” says David Foster, director of maintenance. “They are very good about taking us to the next level and trying to help us maximize the information we have to work with.”

Case Study 4
Venezia Bulk Transport, Royersford, Pa.

After implementing a fleet maintenance management system from TMT Software six years ago, Venezia Bulk Transport Inc. now has a full set of data to analyze vehicle lifecycle expenses among all of its truck and engine vendors on a cost-per-mile (CPM) basis, allowing it to optimize the timing of trade-ins and purchases. The 420-truck carrier also uses the software to determine which extended warranties are worth the investment.

“Beyond a certain time and mileage, the cost per mile was more than what we felt it should be,” says J.P. Venezia, warranty director.

Case Study 5
Republic Services, Ft. Lauderdale, Fla.

To keep maintenance costs from increasing in the past four years, Republic Services has used the Dossier system from Arsenault and Associates to determine the “soft spots” in its maintenance program. The company analyzes road calls in depth to understand what failed and why. Management also has a firm understanding of lifecycle costs – as measured in cost per hour – for vehicles operating in any region of the country.

“After about 10 years, it is time to say goodbye,” says Jerry Wickett, vice president of purchasing and maintenance. “There is major work coming.”