Photo: gettyimages.com/metamorworks
The secret sauce to predictive maintenance is amalgamating vehicle data points across millions of vehicle miles. AI and machine learning then take over to predict failures on components such as brakes, starter motors, alternators, emissions systems, and more.
“Five thousand miles or every six months, whichever comes first.”
Remember that vehicle maintenance edict? It was in place for decades, until we gained greater insights into duty cycles and vehicle component wear to get smarter and more flexible on managing fleet maintenance.
Telematics, AI, and machine learning are making maintenance even smarter today — new systems are using data combined with statistical analyses to better predict when a vehicle needs servicing or when components will fail. This is the promise of predictive maintenance, and the goal is to optimize vehicle health, mitigate downtime and expense, and extend the life of the asset.
Issues around vehicle maintenance are amplified today by the vehicle supply shortage, which is forcing fleets to hold current vehicles in much longer cycles. Yet activating predictive maintenance isn’t just a push of a button. How should you view predictive maintenance in the context of your fleet needs, and where should you start?
Understand Downtime Costs
Fleets first need to define their causes and costs of downtime to determine how and where predictive maintenance can help.
“Each fleet has a calculation for their downtime, and they know that for every hour those vehicles are not doing their jobs, the company is losing a certain amount of money,” says Ben Auslander, head of sales and marketing for Pitstop, a cloud-based predictive maintenance platform.
A sales fleet may still get along just fine by grounding a sedan and having the driver wait for the repair. However, mission-critical applications such as food service, medical supply, and last-mile delivery, along with many others, can’t afford to suffer the lost business revenue for even a few hours off the road and a poor customer experience.
Every fleet will have different, specific needs for predictive maintenance. A delivery fleet with a lot of stops and starts will have urgency around tires, brakes, and batteries. On the opposite end of the spectrum, a tree service will see heavy engine wear and long idling to power equipment, but fewer on-road miles. A mining truck goes through brakes three times as quickly. These needs will dictate which components to monitor, Auslander says.
Fleets running larger, more expensive trucks have higher repair and downtime costs. Those trucks’ frequent problem areas such as diesel exhaust fluid and diesel particulate filter systems can be measured and analyzed through predictive maintenance. “Anytime (those fleets) have an engine lockout situation, they’re in real trouble, and it’s a costly problem for them,” he says.
Even in the more benign cases, adding hard and soft costs around driver salaries, vehicle repairs, and lost revenue can add up to hundreds of dollars per day, per vehicle in downtime.
Collect Vehicle Data
The next step toward a predictive maintenance plan is to collect data on the areas you’d like to address. For most fleets, the gateway to that data is their telematics system, which must be enabled to pull engine diagnostic trouble codes (DTCs) from vehicles’ OBD-II ports.
Data from DTCs is only one factor in predictive maintenance. Accelerometer data that measures speed, stopping distance, and hard braking has a direct correlation to maintenance and component wear. “We’re absolutely seeing trends around hard braking to understand that the brakes on a particular vehicle will fail sooner than others,” Auslander says.
The secret sauce to predictive maintenance is amalgamating those data points across millions of vehicle miles, hundreds of duty cycles, and a myriad of driving conditions. But while fleets regularly use DTCs and driving behavior data for better fleet maintenance and safety, few fleets have the data science expertise to make confident decisions to predict parts failures. And they wouldn’t have access to data sets outside their organization. That’s where third parties come in.
Connect to Maintenance Platforms
Starting a conversation with your telematics provider is a good first step, Auslander says. The provider may be able to tailor a predictive analytics tool that can be accessed through your system’s dashboard.
Larger telematics service providers (TSPs) such as Geotab, Samsara, and Verizon Connect have marketplaces of third-party applications — think apps on an iPhone for fleet vehicles — that address diverse areas of fleet such as ELD (electronic logging device) tracking, video telematics, route planning, and maintenance. The benefit to these marketplaces is the seamless transfer of data from the vehicle and through the telematics system, to be used by these apps.
Fleetio, a cloud-based maintenance management system known for robust preventive maintenance and scheduling tools, also offers integration with many telematics platforms. The Fleetio system uses inputs based on a vehicle’s daily average usage to predict when maintenance is due.
With an ever-increasing percentage of factory-embedded modems, automotive data services startups such as Motorq, Wejo, Otonomo, and Smartcar are connecting to and standardizing vehicle data to leverage solutions for automakers, consumers, dealers, and fleets. Each platform has a different business model and brings data to end users in different ways. Developers (in-house and third-party) are connecting to these platforms for a variety of applications, including predictive maintenance.
Your fleet management company (FMC) can also provide guidance based on your fleet needs and their knowledge of these systems. The FMC may be able to configure a data feed through its fleet-facing dashboard as well.
Fleets can access Pitstop’s predictive maintenance platform through Geotab’s marketplace and Smartcar’s APIs, as well as many other TSPs, OBD-II plugin devices. Pitstop uses proprietary algorithms and AI to analyze vehicle data to predict failures. That data, combined with routine maintenance scheduling and recall notices, can help fleet managers better plan downtime.
Pitstop’s competitor, Uptake, uses artificial intelligence to predict failures and improve uptime in industries such as oil and gas, rail, mining, and manufacturing, as well as over-the-road trucking.
Configure your System
Platforms such as Pitstop work with fleets to understand where predictive maintenance can drive the most value and on which components have the highest parts failure accuracy, then tailor the system accordingly.
“We can see trends four to six weeks in advance,” Auslander says. “As you get closer to the failure event, the accuracy of the model improves. For instance, we’re achieving 95% accuracy to predict battery failure. And as you get closer to the actual failure, (the system) can amplify the message.”
Using anonymized and aggregated data from hundreds of vehicles and duty cycles, the Pitstop system can also see trends on brakes, starter motors, alternators, emissions systems, and clutches. Other components’ failure rates are harder to define because many — say, wheel alignments — are more directly tied to driving environment, which varies greatly.
Nonetheless, “Because of AI and machine learning the system is in a constant state of improvement and learning,” he says. As more and more data become available, “The (prediction) model that’s 94% accurate today could be up to 96% next year.”
The ability to predict parts wear and failures allows for scheduling of services and then grouping those services on a vehicle or multiple vehicles. For instance, the system would decide the right time and place to bring in a truck to replace the battery and clutch, change the tires, and adjust the transmission.
With an automated warning, the repair facility is better prepared for the vehicle’s arrival. “And therein lies how you really go after downtime,” says Auslander.
Auslander cautions that configuring a system with the right data inputs is a process and “ROI doesn’t happen immediately.”
Auslander also brings up the point that predictive maintenance data, along with analysis from real-time sensor data, is extremely valuable to automakers, which are only using the data to improve the component and mitigate warranty repairs — not decide when it should be replaced.
Some question predictive models, particularly the trust put in them that a component should be replaced before it trips a DTC. Could it create more downtime and expense than necessary by causing vehicles to be pulled out of service before they need to be?
Replies Auslander: “I leave that to how fleets want to manage their vehicles. And remember, each fleet’s cost of downtime will also dictate some of their behavior. What does it cost (a fleet) to be on the side of the road? If the cost of a vehicle’s downtime is $400 an hour, is it too much to worry about replacing a starter motor for $150 as part of scheduled maintenance?”
Originally posted on Fleet Forward
By Chris Brown
Source: https://www.automotive-fleet.com
