Machine Learning on the Edge
Haunted by Phantom Events
The trucking industry is undergoing a silent data revolution. Since the Electronic Logging Device (ELD) mandate was passed, an industry that was long deprived of digital connectivity now has a connected device in every cab.
With new technology come new problems. Most vehicles will give faulty sensor readings once in a while. The false sensor readings make the vehicle appear to be moving while it is stationary, creating false driving periods and other issues for drivers.
While ELD vendors generally accept this as an unavoidable problem, at KeepTruckin, our team of data scientists and engineers has been hard at work on solving this issue using artificial intelligence. Over the last few months, we’ve released new updates to our products to automatically detect false sensor readings in most cases, eliminating the problem in more than 98% of situations.
What Are Phantom Events?
Regulation mandates that ELDs (at KeepTruckin, we call these devices “Vehicle Gateways”) detect truck movements above 5 mph and automatically log such status changes in the driver log.
Many vehicles have faulty sensors that can cause bad readings, sometimes making a stationary vehicle seem like it is moving. This creates non-existent hours-of-service (HOS) violations. Non-existent HOS violations cause significant pain for drivers, who must manually correct the logs. Moreover, drivers can be put out of service for this malfunction or be forced to find workarounds by reducing their work shifts. In our experience, most vehicles show such aberrations from time to time, making phantom events a pernicious problem across the industry.
The first step was to understand exactly what was going on.
Why Do They Happen?
Phantom events are false positives from the Gateway Unit signaling that the vehicle is moving while the vehicle is actually not. The speed sensor has a reading of greater than 5 mph and creates a change in status in the electronic log. Specifically, the driver log will change status from off-duty to → driving or on-duty to → driving. This can cause apparent violations that register on the driver’s logs and must be corrected manually.
Why do such events occur? There are a few reasons why false readings happen: (1) power take-off, and (2) defective sensor or bad transmission protocol.
- Power take-off: A secondary power generator for some auxiliary function of the truck, such as turning a cement mixer drum or raising the bed of a dump truck, causes a spike in the speed sensor when it is engaged.
- Defective sensor or bad transmission protocol: A common cause of short, invalid driving events can occur from malfunctions in sensors or bad transmission protocol operating on the truck.
How to Contain Phantom Events?
Phantom events haunt the entire ELD industry, but KeepTruckin is not sitting idly on the sidelines waiting for customers to complain. We decided that phantom events would undermine one of our core missions — to provide reliable safety and compliance solutions — and we acted to build a solution to contain them.
Given what we knew about phantom events — that they caused brief spikes in vehicle speed sensors — we formulated a strategy using many independent sensors to verify that the truck really had crossed 5 mph. This is commonly referred to as sensor fusion.
Next, we trained the model. In machine learning, one of the most important ingredients is data. More data usually lead to better models. Using KeepTruckin’s industry-leading virtual fleet of over 300,000 connected vehicles, we could gather data on thousands of these rare events. This allowed us to train a highly accurate model that could detect whether the increase in speed was really happening, or if it was just the result of a faulty sensor read.
Machine Learning on the Edge
After the model was ready, we deployed it directly on the Gateway Unit to ensure that it would work without a cellular connection. For compliance reasons, the driver log has to work anywhere in the United States, even in locations without an internet connection. Because the classifier validates a basic function of the Gateway Unit (that is, knowing whether the vehicle is moving or not), KeepTruckin cannot wait for the Gateway Unit to go back online to generate the driver log. If a law enforcement official makes a traffic stop, the driver has to provide the driver log in any circumstance.
The model achieved great success: it reduced the occurrence of phantom events by 98%. Although we have not eliminated the problem completely, we have reduced its occurrence dramatically and confined it to the most severe cases.
Conclusion
ELDs are bringing about change to the logistics industry, and KeepTruckin’s products have the most advanced hardware and software capabilities for both local and long-haul fleets. Compliance is near and dear to long-haul trucking companies, and HOS logs are a crucial piece of that equation.
All electronics, including sensors, cables, and circuit boards, can fail. The problem of false speed readings affects the entire ELD industry. As a leader in this industry, KeepTruckin has taken a proactive stance toward false speed readings by building a machine learning model deployed on the edge. This model runs on the ELD and validates the detection of vehicle movements using about ten independent sensors. The use of multiple sensors minimizes the risk of a single sensor triggering a status change while the vehicle is not moving.
This is just one more example of how KeepTruckin is leveraging cutting-edge AI technology and our industry-leading network to provide the best service to our customers. Safety is at the heart of our mission, and helping fleets with compliance is a big part of that. Our tools reduce the amount of time spent on logs, so that your fleet can concentrate on the most important tasks at hand.
