The Institute is preparing me to engage in the next generation of how businesses are run.
When you think about traditional American industries, it is hard not to conjure up the Freight Rail industry. After all, the industrial development of the United States and its westward expansion during the 1800s was done on the back of the railroad. There is a reason the rail industry is featured in the classic board game Monopoly – all the way back to the original 1935 version. But do not be fooled by the industry’s longevity. It is embracing the modern age and the wealth of data that comes with it.
Through automated detectors, railroads are collecting data and leveraging it to make for a safer, more efficient transportation network.
The United States rail network, consisting of over 140,000 miles of track, is lined with many varieties of sensors. Each sensor measures the performance of train parts, often picking out different symptoms of wear and damage to critical safety components. Some sensors will measure visual aspects, such as a wheel’s geometry or the sway of a railcar in motion. Others measure things humans simply cannot, such as acoustics or temperature of a failing bearing. The goal of these sensors is to reduce derailments and accidents, which hurt not only the bottom line but the people involved.
The cutting-edge trackside sensors are machine vision systems. Many railroads have erected complex rigs of lighting and high-speed cameras to capture the train in images as they pass by. By carefully standardizing how these images are captured, the railroads are left with a full-length capture of the train. Standardizing the images is a difficult task—and that is not even considering environmental effects like snow, flooding, and obscuring graffiti. There are also the complex tasks of transporting the data from remote locations and storing them in data centers. Only once a railroad has set up its vision system, accounted for environmental effects, and established a data flow can data scientists start to leverage the technology. From then forward, with each train passing through the machine vision system, the railroad’s data scientists gain new valuable data.
There are some immediate practical benefits, such as seeing which rail cars are passing by your sensor. Knowing where your sensor is provides the geographical location data for the rail car. Additionally, if the box car’s side door became ajar in transit, a human could easily spot that by looking at a still image. But with the sheer volume of train passes in a day, doesn’t it make sense to automate that process? That is where machine learning comes into play.
What if we could teach a program to identify discrepancies between what a normal rail car looks like and what a broken rail car looks like?
Let’s use the side door example. The machine vision system identifies the open side door as disparate from a typical side door and notifies the train’s crew. This example is easy to spot and of relatively low consequence. However, what about locations on a rail car that are not easy to spot by humans, such as the undercarriage? And what if the images captured were not within our visual light spectrum, such as infrared? We could spot things like missing pins, screws, and bolts or identify parts that are heating up when they should be keeping cool.
If we can find them before they break, we will prevent damage and save lives.
In Fall 2022, I completed Dr. Aric LaBarr’s module on Machine Learning. I also attended a lecture on Anomaly Detection Using Principal Component Analysis (PCA) with Nic Larsen. While these tools are not the same approaches used in computer vision systems, they provide another lens through which data can be leveraged to make predictions. These foundations will guide business decisions for data scientists and in turn like-minded businesspeople across industries worldwide.
So, I task you with considering your walk of life, industry or passion. How can you use data to do things humans simply cannot? How can you leverage this technology to make your industry safer and more efficient?
Columnist: Cameron Bumgarner