Predictive analytics is transforming many of today’s connected industrial organizations by proactively identifying asset efficiency and performance issues before they impact operations. However, a lack of understanding surrounding a key enabling technology – machine learning – has led some businesses down a path of disappointment. Because organizations often fail to properly evaluate the viability of these sorts of technologies within the context of their operations early enough in the process, many end up wasting resources and end up frustrated when an initiative inevitably falls short of the desired outcome. A firm grasp of the role machine learning plays in predictive analytics early in the planning process is a key aspect of setting yourself up for success.

In a nutshell, machine learning is when you teach a computer or system to predict some kind of outcome (see this IoT Business News article if you need a more in-depth explanation). The ability to do this hinges primarily on having the right dataset. Therefore, understanding what data you have available is critical to understanding the potential outcomes and setting expectations across the business.

Let’s review three real-world scenarios in the fleet management and transportation space where predictive analytics seemed like the best path toward a measurable outcome – but on closer inspection, wasn’t always the case.

1. Predict Engine Failure 

Goal: Keep vehicles on the road longer and more effectively schedule planned maintenance.  

The Data Assessment: Engines are one of the most critical parts of the vehicle and one of the most intricate. Major components, including the engine, are typically embedded with large numbers of sensors and can generate a significant amount of data related to their performance. For example, it’s common to capture lots of information about operating conditions when something goes wrong. With enough historical data – from engines operating normally as well as engines exhibiting problems – organizations can use predictive analytics with machine learning to forecast future engine failures. Which means they can proactively schedule maintenance, better set technicians up for success, streamline repair processes, and more.

The Bottom Line: Any asset that has lots of sensors capturing data related to the outcome you’re aiming for – along with plenty of examples of good and bad operating conditions – is prime for machine learning. (You can learn some tips for getting started with your predictive project and how to find the right data this video.)

2. Optimize Inventory Management

Goal: Predict bumper damage to have sufficient parts on site for repairs.

The Data Assessment: Unlike engine data, a transportation business isn’t likely to have much information on their vehicle bumpers (which aren’t typically outfitted with sensors). Without data specific to this outcome, machine learning won’t be much help. In this case, you might instead determine inventory demand for bumper parts using alternative data and traditional methods, such as forecasting. This can allow fleet operators to better manage supply chains and costs.

The Bottom Line: Sometimes the cost or complication of adding sensors to a device to meet one business goal isn’t a wise investment. Instead, working with available datasets and more conventional approaches can provide the outcome you’re looking for.

3. Hyper-focused Component Repair

Goal: Predict a clogged fuel filter to reduce repair expenses.

The Data Assessment: If a business does not have past data examples of this specific condition, it is impossible to predict it. The only option at this point is to collect many examples of this situation – along with lots of examples of normally operating fuel filters – and then rerun the machine learning process so it can recognize normal vs abnormal characteristics. As more and more data for both good and bad fuel filters builds up over time, identifying when a clog is on the horizon becomes possible.

The Bottom Line: If the right data isn’t readily available and the outcome is a priority, steps can be taken to set baselines and evolve the data model. The key is recognizing that the reliability of the model will improve over time.

I hope these examples provided you with a glimpse into predictive analytics and the role machine learning plays, as it pertains to fleet operations. Obviously, there are major implications across a variety of industrial applications.