Model Behavior: Driving More Value from Industrial Machines

Welcome to the third installment of Model Behavior. This ongoing series provides an inside look at Uptake’s industrial data scientists and the challenging work they do on behalf of our customers. Why is this topic so important? Because tough problems don’t solve themselves. See how model Uptakers build and deploy Uptake’s models — and how both of those elements must work together in harmony to create a world that always works.

In heavy industry, disruption impedes success. On the front lines, people and machines must be dependable to do the job. That’s non-negotiable.

The good news is that new technology empowers industrial businesses to always know the road ahead. Failure, it turns out, is now an optional outcome. But the best path forward isn’t always obvious. And more often than not, the answers don’t come easy.

That’s where AI comes into play. By turning data into meaningful insights — and putting that intelligence into practice — people and machines are empowered to do their best work. Businesses gain the benefit of running more efficiently than ever before.

To see for yourself, check out episodes one and two of Model Behavior to learn how Uptake’s industrial data scientists accomplish amazing feats, like keeping wind turbines turning and locomotives running.

Now Here’s the Catch

Uptake industrial data scientist Ted Bakanas uses his skills to help keep trucks on the road. This is no small task. Today’s trucking fleets face the constant challenge of getting from point A to point B safely and on time. It means they have to count on the availability and reliability of their trucks, and to ensure the safety of their drivers every leg of each route.

A common failure that trucks are susceptible to occurs with the Diesel Particulate Filter (DPF). The DPF is an important component that’s designed to remove diesel particulate matter, or soot, from the exhaust gas of a truck’s diesel engine.

Like any filter, DPFs are prone to clogging over time. This is problematic for a few reasons.

First, when the DPF clogs, it places additional strain on the rest of the engine. If the truck cannot pass exhaust fully through its system, then in essence it cannot breathe. All sorts of self-regulating systems — from air intake on through to the EGR — can start to malfunction.

Second, if the issue persists for long enough or grows severe enough, then the engine can derate itself. This means the truck will deliberately constrain its engine power, and then the driver will have to take it directly to a shop.

Over the past 20 years, U.S. emissions regulations have tightened for light- and heavy-duty diesel-powered trucks. It’s part of the reason why DPFs have become a standard component. While DPFs are certainly great for the environment, they can be a headache for the average fleet to manage.

Why? Because not only must DPFs be kept clean, there are numerous factors that affect how quickly soot builds up in them: excessive braking, stop-and-go traffic, operator error and frequent short journeys where the engine doesn’t reach an ideal operating temperature. All of these factors make it extremely difficult to know exactly when a DPF will need to be cleaned.

At Uptake, we tune our DPF model using two unique signals:

  • Differential pressure: Sensors at both the inlet and outlet of the DPF measure the differential pressure of the aftertreatment system. A clogged DPF causes the pressure to be much higher on the front end — where exhaust is having difficulty passing through the filter — than on the back end.
  • Temperature: When soot builds up inside a clogged DPF, forced regeneration needs to happen. This triggered event forces up the temperature of the aftertreatment system in an attempt to burn off some of the soot. It requires the driver to pull over and initiate a self-cleaning process for the truck, which can take up to an hour to complete. Forced regeneration becomes less effective the more clogged the DPF is.

By correlating these signals and detecting anomalous patterns early on — or, in other words, when the engine heats up and tries to clear itself, but an insufficient amount of exhaust is able to clear the filter — our model can accurately predict future DPF failures. This can give fleet owners and operators weeks of lead time before those failures occur.

Armed with advance warning, fleets can proactively schedule their trucks to come into the shop for planned DPF maintenance. They effectively minimize the risk of those vehicles breaking down on the side of the road while upholding their commitment to driver safety and on-time shipments.

Ask Me About My Failures

Check out this video to hear Ted explain the catch in his own words:

Think you have what it takes to become an industrial data scientist at Uptake? Check out our Careers Page to view our open positions.

New to Our Model Behavior Series?

Here are some helpful definitions to know:

A machine learning model is a codified structure and digital representation of an industrial asset that physically exists in the real world. It organizes the data generated by the asset and standardizes how that information relates to each other — and, importantly, how all of it relates to the physical properties of the actual asset. A model targets a specific component of a specific asset, growing increasingly accurate over time with each new data point.

At Uptake, our machine learning models are the output of our AI engines, which are a tool that automates the development and deployment of our models. Our engines use valuable insights gained from different industrial use cases to come up with approaches that solve common sets of problems.

A catch is when one of our models — running in a live production environment for our customers — generates an alert that correctly warns of a failure or performance issue that’s bound to happen. Or, in other words: A catch means our model did what it set out to do.

A failure happens when an asset or one of its components shuts down unexpectedly in its operating environment, causing problems like unplanned downtime, lost production, service disruptions, unsafe working conditions and more. Failures are what models try to predict and prevent from happening.

Check out the fourth installment of Model Behavior here.