Distributed fiber optics detect everything everywhere at the same time. When combined with artificial intelligence and machine learning, it has the potential to serve as a bridge between the physical and digital worlds.
Existing underground fiber-optic telecommunications cable networks accessible via street manholes are assisting a team from NEC Labs America (Princeton, NJ) in improving wireless communications systems and the Internet of Things (IoT).
“Hundreds of millions of fiber-optic cables are already there for communications purposes,” says Shaobo Han, a NEC Labs America researcher who works on the design and implementation of machine learning and signal-processing algorithms for real-world sensing applications. “We’re turning it all into a’thinking’ device, using the same cable that’s already there.”
Han’s team can monitor and trace the passage of practically any object above ground using existing cable networks without installing extra sensors or connecting to wireless networks (see video and Fig. 1). The fiber sensing system offers optical and auditory data by utilising thin-film optoelectronics as an integrator that transmits optical power from lasers to the optical fiber. Physical characteristics like as acoustic aberration and temperature can be discovered by analysing its scattering.
Researchers can use fiber optics to detect two types of objects: moving and stationary.
Landmarks are objects that do not move, such as cables and manholes. Potholes or pavement deformations on the road surface can also be detected by the subsurface fiber sensing system. Vehicle traffic and construction equipment in use are examples of moving things.
Manholes can aid in determining the position of detected acoustic events, much like a lighthouse serves as a reference point to navigate the location of ships on the ocean.
“When we use fiber sensing to detect events, we know how far away it is from the sensor, how many metres away,” Han explains. “We don’t always have GPS coordinates because the cable isn’t always in a straight line.” It may traverse many blocks of the street. It is vital to create a mapping between the physical and digital worlds.”
Geolocalization, as Han points out, is required for sensing. Previously, each manhole had to be manually visited, “maybe even blocking street traffic and then knocking out the manhole power basically to create something recognisable.” This manual field surveying is time-consuming and labor-intensive.
The fiber sensing technique proposed by the researchers provides options. Among these is the utilisation of ambient traffic noise on the road surface to help with automated manhole localisation (see Fig. 2).
“We learn from ambient noise,” Han explains. “Normally, when we connect the sensor to a cable, we immediately detect road traffic.” We collect data over time using static items like manholes and fiber sensing equipment to analyse how infrastructure and cars interact.”
The way an automobile interacts with a road surface differs from that of a manhole, and thus distinguishes the manhole using fiber sensing data.
AI and machine learning implementation
In real time, the researchers developed a machine-learning model to detect auditory events above ground and determine the routes and variations in ambient sounds created by diverse objects. It effectively eliminates the necessity for physical labour.
The technology can also detect faults and other issues with wireless communications connections faster and more precisely than traditional techniques, and at a fraction of the normal maintenance expenses.
“AI and machine learning are used a lot in distributed fiber-optical sensing,” Han explains, describing a current method for creating an event, recording the data, and time stamping it.
The researchers conducted an experiment in which they discharged a rifle and let off a firecracker. “We could train a neural network to distinguish between a gunshot and a firework using a weakly supervised machine-learning model.” “It’s called weakly supervised learning because the interesting event is already present in the data,” Han explains. “The neural network only needs to learn how to classify.” These physical factors are computed by AI into high-level, significant events, such as a car passing by, an earthquake, a machine operating and excavating in the ground, or floods.”
Their fiber-optic-based solution might also assist smart cities and IoT progress, as well as communications service providers who employ subterranean fiber optics. It has the potential to assist city planners better analyse and optimise things like traffic patterns, as well as provide law enforcement with more specific information to investigate traffic incidents.