How predictive maintenance can improve asset utilization


Asset utilization is often cited as a crucial metric for evaluating the efficiency of a trucking carrier, but it’s really a category of measurements than a precise statistic in itself. Asset utilization can refer to anything from the percentage of seated versus unseated trucks in a fleet to how efficiently drivers exploit their hours of service, trucks per loaded mile, and the minimization of unplanned maintenance-related downtime. This week in Techpertise, we’re going to be looking at how data analytics-powered predictive maintenance initiatives can help fleets reduce and even eliminate downtime from unplanned maintenance and roadside breakdowns.

The driver shortage gets a lot of attention from industry media outlets, but it’s not an isolated problem—sectors like manufacturing, timber, mining, and construction are all raising wage in an effort to find workers. Similarly, the maintenance technicians fleets rely on to keep their trucks on the road are also in short supply. A shortage of maintenance personnel means that the opportunity costs of unplanned downtime are higher than normal, because even a relatively inexpensive repair job can take days longer than it should for lack of technicians. According to TMW, current industry estimates place the cost of an out-of-service truck at $800-1,000 per day.

FreightWaves spoke to Renaldo Adler, principal of asset maintenance, fleets and service centers at TMW Systems, about TMW’s predictive maintenance project leveraging PeopleNet data.

“The ideal situation is to be able to fix a component before you leave—to replace it before it fails,” said Adler. “From the analytics side we’re working with PeopleNet and Vusion and their data science team that’s getting all their data—all the fault data from there…we’re analyzing it for failures and various conditions on that vehicle. We’re predicting that you may have a high voltage load, and we’re able to predict a high probability that this part will fail in the next three days, for example.”

“We’re looking at 48 of the signal values coming off the truck that range anything from voltage to coolant temperature, pressure, engine speed, truck speed, torque, load,” said Adler, “anything that can be monitored from the centers, we’re pulling them and putting them in the model. It’ll take 3 of those signal values to indicate there’s a probability of some kind of fault.”

Adler said that after-treatment and cooling systems were the really low-hanging fruit in the predictive maintenance effort, and that heavy duty engines and transmissions were mostly reliable and already had predictable rates of failure. “Everything that’s been added to it over the years to de-rate the engine—that’s where the problems are,” said Adler.

When those systems fail, at some point the truck’s engine will start shutting down and the driver won’t have as much power. The truck will still run in a limited way, but the driver will have to pull over and do a re-gen of the after-treatment system. In other cases, high coolant temperatures can damage other, more critical components. 

TMW says its Predictive Maintenance Analytics product can reduce road breakdown events by 70%, and reduce overall maintenance costs by 25%. Because the service is offered to customers already using PeopleNet devices, there is no additional cost—the Predictive Maintenance Analytics represents TMW’s effort to leverage the vast amounts of data it’s already capturing through the internet of things. One million vehicle days of data has allowed Adler and his team to predict ‘red lamp’ critical faults up to 72 hours in advance. When vehicle data indicates an increased probability of failure, a dashboard alert appears within the TMT Fleet Maintenance Software containing the unit number, diagnostic trouble codes, description, leading performance variable, other key signal values, and probability of failure.

One of the decisions that has to be made in designing the algorithms that make critical fault predictions is whether a fleet wants its model weighted toward false positives (a predicted failure that didn’t occur) or false negatives (a failure not predicted by the model). Ultimately, the sensitivity of the model will be adjusted depending on how critical the potential failure would be. “To decrease one type of error, the other increases. We trade one type of error for another in order to adjust the model to best address the business need,” Adler said.