Smart cities powered by connected sensors promise to transform everything from public transportation, to public health monitoring and energy grids. But the Internet of Things (IoT) will also require new ways to ensure timely and stable data flow among millions of connected devices – particularly for applications performing critical functions.
New research out of Ghent University suggests anomaly detection through fog computing may be the answer to helping ensure the reliability of delay-sensitive and data-intensive applications that are driving growth in the IoT and making cities smarter.
Traditional anomaly detection approaches for identifying IoT problems are significantly impacted by latency, and therefore, not suitable for the real-time communication requirements of many IoT applications. However, scalable, low-latency anomaly detection has become more feasible with 5G networks—due to their significantly greater data transmission capacity—as well as with new paradigms like Software-defined Networking (SDN), Network Function Virtualization (NFV) and edge computing.
“Current anomaly detection approaches for IoT only focus on the centralized cloud management aspects,” said José Santos, a researcher at Ghent University’s imec IDLab. “We propose a distributed fog computing-based anomaly detection approach that goes beyond the current state-of-the-art by considering not only cloud requirements but also low power wide area network (LPWAN) constraints.”
The researchers are part of the City of Things project, a collaboration between imec, the City of Antwerp, Belgium and the Flemish region. The team is working to create a large-scale living laboratory where researchers and companies can test and develop smart city applications, products and services. Santos’ team worked with local governments to create a large testing ground in the heart of Antwerp, including the Port of Antwerp, to test their novel approach to anomaly detection.
As an initial proof-of-concept, the team mounted air-quality sensors on the roofs of delivery cars used by Bpost, the postal service of Belgium. Because the cars are continuously moving around the city, their roof-mounted sensors can gather real-time air-quality data across the entire city more efficiently compared to the limited reach of static sensors, which would require far more sensors to deploy and maintain. The test was to monitor air quality in Antwerp to detect high amounts of organic compounds in the atmosphere and alert local citizens of air pollution in real-time.
Figure 1: Inside view of the multi-radio sensor,
and the air quality sensor mounted on a Bpost car.
In a traditional anomaly detection system, all data samples from connected devices are transported to the centralized cloud layer, where the anomaly detection operations are executed. In the researchers’ approach, anomaly operations are performed on multiple computational nodes, also called fog resources, located closer to the sensors to allow for better data transmission.
Each IoT sensor sends its data samples to nearby fog resources, which reduces bandwidth for data transmission. The researchers use a Birch clustering algorithm to process the data, as the algorithm offers high performance and low convergence time when analyzing large data sets. After the Birch algorithm clusters data samples, the researchers employ a Robust Covariance (RC) algorithm to identify air quality outliers in the clustered data samples.
Figure 2: High-level view of the fog-based anomaly detection approach.
After completing anomaly detection, fog resources send alerts to the cloud and then to the IoT sensors if unusual events are detected in the data, enabling faster response times if necessary. Outcomes of the anomaly operations are also sent to the cloud layer to combine results from the different fog resources in order to provide a broad view of network behavior.
In testing, the results showed that Birch clustering and the RC outlier anomaly detection mechanisms can be performed by fog resources close to IoT sensors. The tests proved the system could send timely alerts when unusual events are detected on a low power network.
“The team’s test demonstrated that fog computing provides an efficient manner of dealing with the stringent requirements introduced by IoT use cases, such as the air quality test,” said Bruno Volckaert, a researcher at Ghent University’s imec IDLab. “The fog computing approach can achieve a substantial reduction in network bandwidth usage and latency when compared with traditional centralized cloud solutions for anomaly detection.”
The team also evaluated multiple LPWAN technologies against its air quality application and determined that IEEE 802.11ah, DASH7 and LTE-M could all serve as wireless communication enablers for the use case.
Going forward, the researchers plan to examine the selected LPWAN technologies in the City of Things testbed and extend the air quality use case to several tens of cars. They also plan to see if the integration of 5G technologies and distributed management approaches could further improve their system’s efficiency and bandwidth usage.
As the development of smart cities continue, the researchers’ system could be a critical part of a smart city architecture. Fog computing can greatly reduce the costs of transmitting large amounts of data by reducing bandwidth needs, so this unique approach could eventually be a key pillar in smart city foundations.
For more information about fog computing, visit the IEEE Xplore Digital Library.
