Every trucking fleet exists to make money, and sustaining itself in the market requires managers to keep freight hauling competitive and to seek methods to lower operational and maintenance costs.
Over the years, managers of successful fleets have figured this out by giving driver benefits to keep churn rates low and by sending trucks to the maintenance garage anticipating a potential breakdown. However, with the proliferation of technology, fleets are now gravitating towards data analytics and machine learning that can help predict their maintenance needs, equipment failure, and even refine driver behavior to improve truck safety.
FreightWaves discussed these issues with Rebecca Grollman, data scientist at Bsquare, to understand how data can be leveraged – irrespective of the size of the data set. “Before we start out, it is important to see if the collected data is actually of high quality. If the quality is not good, there is not much that you can do, even if you have a lot of it. Quality of data is more important than quantity,” said Grollman.
It helps fleet managers to have a clear idea of the questions they want to answer before data collection begins. This is critical because truck fleets generate several data streams from everyday operations – be it from the trucks or the back office. The importance of figuring out the issues that matter and devising means to collect data specific to that cannot be overstated.
For instance, a trucking company might have thousands of data points on the exact colors and paint jobs of all the trucks in its fleet. However, all that will be worth nothing if the company ultimately wants to predict when its trucks will need to schedule a maintenance visit to the garage.
Grollman explained that with relevant historical data, company management can look at predictive analytics and root-cause analysis – helping them pinpoint where their equipment failures originate and follow it up with measures that will stem such future scenarios.
For companies that are just a few months into their operations, data analytics might be a hard sell, as they lack historical data to drive meaningful insights. However, Grollman insisted that such companies can look towards anomaly detection, as its prerequisite does not include substantial data sets.
“Even if you have only been collecting data for a few months, it should be enough to gain insights on normal operating parameters. It helps with understanding what to expect with the data that you’re collecting on a daily or monthly basis,” said Grollman. “You may be able to see some trends and seasonality using anomaly detection. You can start to pick out different anomalies in your data and even make correlations to things that those anomalies indicate.”
For instance, data can point out a spike in tire pressure. This could be because there is a problem with the tire, or perhaps one of the sensors on the truck is malfunctioning. These are anomalies and figuring out a way to work on them will help weed out operational issues. Over time, with a considerable amount of historical data, machine learning algorithms can be used to push decisions. If the insights are not well-defined at the start, it will help to keep iterating on the data until there is definitive meaning.
“Apart from collecting quality data, it is important to have domain expertise to make sense of the data. Companies should discuss the possibilities with a subject matter expert and understand the filters to use on the data, how data streams relate to each other, and what can be expected from them,” said Grollman.
“For example, there might be a number that comes up which indicates median tire pressure, but if I don’t have an idea on the reasonable number, it would be of no use. For small companies, being able to have this collaboration and understanding the data that they are collecting would actually make a big difference,” she said.