Isn’t predicting future maintenance needs what many, if not all, fleet managers desire? We’re not just talking about getting advance warnings via remote diagnostic technology or how ongoing fluid analysis programs can extend drain intervals for engine oil, transmission oil, and engine coolant. We’re talking about giving fleets the ability to know when—based on accumulated data and analytical tools—specific components, such as water pumps or wheel hubs, need to be replaced before they fail, thus leaving a truck (not to mention its revenue-generating cargo) stranded on the side of the road.
Does that capability exist today? To some degree, yes, say experts. Yet at the same time, developing such “predictive models” is no simple task, stresses Michael Riemer, vice president-products and channel marketing for maintenance software firm Decisiv.
METHODS & MODELS
It is not simple because there are multiple factors that contribute to various failures in a fairly complex environment such as a heavy-duty truck,” he explains.
“That being said, though, being able to capture mileage, engine hours, gas usage, idling time, and other ‘metered’ truck activities are, at a minimum, a good starting place,” Riemer notes. “Next, you need to have enough data to create a statistically relevant model. This means that you have to have enough trucks and enough data about those trucks as well as the ability to analyze the various possible contributing factors. Finally and most importantly, you need a method for capturing all this data—the telematics data, the repair data, etc. —and the tools to then develop a predictive model.”
This last piece is the most immature in the current truck maintenance ecosystem, he points out. “Our research shows that access to a complete set of information about all service and repair events across internal and external locations is still very difficult to find and manage,” Riemer says.
Traditionally, he explains, many of the telematics systems, OEM warranty and repair data, intelligence gleaned from unique algorithms tied to fault codes, VMRS (vehicle maintenance reporting standards) codes, repair management systems, and driving behavior (hard braking, fast acceleration, hard turning) all remain in separate information silos with little or no ability to cross-mingle such data sets.
“Thus, we are still very early in terms of being able to provide true predictive models,” he emphasizes. “This does not mean that some OEMs are not already using their proprietary knowledge about their trucks or engines, or that fleets, especially very large ones, can collect bits and pieces of data to create more intelligent preventive maintenance schedules. But that is not truly predictive maintenance.”
Still, Alex Ognjanac, vice president-sales and marketing for telematics provider Isotrak, contends that trucks continue to evolve and provide not only actual or historical data but true predictive data as well.“
This is vital [because] one breakdown can cost a tremendous amount of money in lost or perishable inventory or missed deadlines,” he says. “[Truck] manufacturers and [telematic] application providers alike are focusing more attention on other elements of a vehicle that have traditionally not been given much attention, such as engine temperature, oil pressure, tire pressure, and more—with all of those leading to [vehicle] efficiency improvements and savings for the carrier.”
Riley Asher, vice president-fleet services for Clarke Power Services, a member of the WheelTime Network, adds that the “starting point” from which to develop a predictive maintenance approach is to use past maintenance history to predict future performance.
“That is here and now capability,” he stresses. “However, data management is a key component of this approach. If data is collected and reported by make, model and age, then the life of each component can be predicted in the future based on a histogram of performance.”
That way, Asher notes, fleets can use such data to change components prior to failure and avoid costly breakdowns, thus reducing unpredicted or unplanned maintenance events."
An unplanned maintenance event is more costly to a fleet not only because of the cost of the repair but also because of the interruption of the load of freight, loss of use of the assets (truck, trailer, and driver), and the negative impact on profit,” he explains.
Data, of course, is perhaps the key piece of the predictive maintenance puzzle, contends Conal Deedy, product manager for communications and electronics at Volvo Trucks."
More data is needed from each vehicle, and large vehicle populations are needed to develop the complex algorithms that are required to get to the vision of true predictive maintenance,” he says. “Sensor, system and failure mode data are key pieces as well. OEMs and large fleets will likely take a lead in this field, and I believe the industry will start with a limited set of components and then move out to the other components.”
Deedy expects that over time, some components will need to have sensors and usage data added to their design to enable robust predictive maintenance.In the simplest terms, Clarke’s Asher explains, telematically delivered data by itself is valuable if it can help craft a predictive maintenance solution.
“The use of histograms can take a few years of data collection, along with maintenance reports and analysis to produce predictive models,” he points out.
“The big idea [in trucking] has been to move to more aircraft-type maintenance management where almost all of the components are changed prior to failure using telematically derived data. While the technical hardware exists to be applied to trucking, the cost of the sensors and applications is not rapidly migrating into the trucking industry as a whole.”
There are examples where real-time data collected and communicated via telematic systems is making a difference, Asher explains.
“Hard braking events and engine over-speed are two examples that are resulting in data gathering and planning actions by fleets,” he says. “These actions often result in inspections to detect potential damage and create a way to build ‘if/then’ correlations between an event and required ‘predictive’ action.”
Mike McQuade, chief technology officer at Zonar, also believes that predictive maintenance modeling can be done with existing telematics systems; however, enhanced data collection and algorithms must be implemented in those systems to provide the input required to create predictive models that are accurate."
Today, we are just touching the tip of the iceberg,” he says. “Massive amounts of raw data are available and can be coupled with fleet feedback data to create a variety of models including predictive maintenance.”
Each fleet’s data needs are different, cautions McQuade. Also, tracking vehicle and component failure rates across brands is difficult because much of that data is proprietary.
“If models of component failures were released to the public, the process of predicting maintenance needs would be simpler,” he says. “Another possibility is to use a crowd-sourcing approach where telematics companies aggregate data for a wide selection of fleets. With the proper [data] reporting from fleets, predictive models for critical components could be produced within one to two years. [Then] some of the predictive information could be modified at the fleet level over time to increase accuracy.”
In all likelihood, however, where parameters can be monitored, McQuade thinks it is easier to record and analyze data for simpler components that perform less work on trucks.
"It would take far more analysis for the much more complicated components [such as] predicting a transmission failure, which is far more challenging than, say, an air dryer—assuming you know how often the air dryer is running and what the [exterior climate] conditions are,” he explains.
As sensors in the vehicle or additional situational and operational data can be integrated, telematics systems will rely less and less on driver intervention to confirm certain types of information,” explains Monica Truelsch, director of marketing for TMW Systems.
“It’s certainly conceivable that engine data can trigger predictive maintenance warnings, flagging a potential clutch or transmission failure or timing belt break before they occur so that a driver can get service before experiencing a catastrophic breakdown on the road,” she says
.“The key piece, though, is that telematics data is transmitted in digital form, not as analog sound waves. Software applications can then read the data and incorporate it directly into databases and trigger other actions based on the information contained in that data without requiring a human to re-key the data into an application,” says Truelsch. “Although an important use of telematics is the exchange of information between people, it also makes possible machine-to-machine communications that offer tremendous opportunities for [maintenance] efficiency.”
Decisiv’s Riemer adds to that another important caveat and that is data consistency. “Without a consistent use of VMRS codes across the industry, it will be very difficult to be able to track at a consistent, detailed level the specifics of a service and repair event,” he cautions. “Even within a fleet there can be significant [data] inconsistencies within internal and external locations.”
Of course, there’s going to be a price tag attached to developing such predictive maintenance systems, even though many of the components for such systems may already exist.
“There is definitely cost involved [as it] requires a great deal of data to be captured and analyzed to produce predictive algorithms,” Volvo’s Deedy says. “These algorithms must then be validated to ensure accuracy. These steps are essential to gain customer acceptance.”
“Larger fleets may have the capability to develop the software needed in-house, but midsize and smaller fleets will need to purchase it from telematics and maintenance software suppliers,” contends Chris Ransom, director of sales engineering for Networkfleet, a Verizon company.
Although most of the data regarding major components is already being produced, the predictive software is not yet cost-effective to develop and implement,” he adds. “As more fleets demand this type of information and as more advanced analytics become available, these types of predictive maintenance systems will pay for themselves by increasing uptime, avoiding larger repairs, and increasing the useful life of each vehicle.”
Ransom also points out that it’s currently easier to limit the cost of predictive maintenance models to what he calls “larger components” since there is already a large amount of historical data to analyze. With new types of sensors and engine computers becoming available, however, pushing such analytical capability down to smaller components in the near future may also be possible.
“Low hanging fruit” can provide instant returns as well, stresses Zonar’s McQuade. “Basic trending can provide you with an understanding that a vehicle at least needs to be brought into a shop to be reviewed by an experienced fleet technician,” he says. “The electronic equivalent of a driver reporting a truck is driving ‘funny’ can be done more analytically now. If you consider the costs of something like a turbocharger failure, the costs of more detailed telematics data for vehicle performance monitoring become insignificant.”