An innovative system to monitoring ice road conditions based on Wireless Sensor Networks

Authors: Mirko Gremes (Algorab) - Antonio Francescon (CREATE-NET)

Wireless Sensor Networks (WSN) are known to be affected by scalability, cost, and maintainability constraints that may limit their applicability in practical applications. This article describes the hands-on experience gained by the authors in developing an innovative solution to a real-life challenge, the remote detection of ice formation on mountain road surfaces. A combination of multihop routing and nomadic routing, together with inferred detection of ice formation, solved the scalability and cost issues. On the other hand, the lessons learned concerning maintainability of the system clearly indicated areas of further improvement for optimizing the services and guaranteeing profitability. 

The formation of ice layers over road surfaces directly impacts road traffic and drivers' safety, since even a thin ice layer may drastically reduce tires grip on road surface. Such loss of friction, combined with the difficulty to timely detect such threat, often leads to car accidents. In the Alps, weather conditions favoring the ice formation are typically satisfied in winter, but depending on climatic and topographic characteristics of the site, ice may even appear in autumn and spring time, when, being ice unexpected, the risks for road traffic increase.

The ice monitoring system in Trentino (an Italian north-eastern province in the Alps) is mainly based on on-field checks along potentially dangerous roads made by skilled personnel, evaluating the situation and trying to predict its evolution in the next hours. Such estimation is clearly driven by personnel experience and the situation met at the moment, which can vary a lot during time, especially in autumn-winter nights.

In order to reduce the overall costs of the service, the local government of Trentino (PAT -Provincia Autonoma di Trento) asked Algorab and CREATE-NET to collaborate for the set up of an automated service for ice-formation risk detection which triggers human on-site intervention only when an actual danger is present. Instead of using the canonical approach, based on some standalone (and expensive) meteorological stations installed in critical points (which grant a very fine spot analysis, but only an estimation for the surrounding area), a solution based on Wireless Sensor Networks (WSN) technology has been considered.

The proposed WSN is made of a large network of cheap intercommunicating battery-powered nodes, equipped with environmental sensors and performing a simpler but more reliable (being based on actual measurements) risk evaluation. The ice-risk evaluation algorithm is based on the dew point evaluation (i.e. the temperature at which water vapor present in the air starts condensing into water, which is mainly function of temperature and humidity, at constant pressure) and its comparison with the road temperature: a simple operation which even sensor nodes can easily compute. Moreover, formula parameters can be retrieved with cheap, precise and low power consuming sensors.

Eight test sites have been identified to realize a self organizing, permanently connected multi-hop WSN. Every cluster is supervised by a gateway, which coordinates the sensing activity of the network nodes, performs ice-risk analysis and is in charge of the data concentration and transmission via GPRS to an Operational Center. Moreover, if a potential risk is detected, an alarm SMS is sent by the gateway directly to PAT personnel. Sensor nodes operating in the described scenario are said being working in "Multi-Hop Mode".

The road is divided into variable length segments (based on connectivity), each monitored by one Node uniquely associated with the segment, thus allowing data geo-referencing. The WSN size depends on radio coverage and scalability factors, posed by the technology used and the topography of the territory: in our test, it never stretched over 2 Km.

For extending the service geographical coverage in a scalable and cost-effective way, additional isolated Sensor Nodes (said working in "Nomadic Mode") are installed in spots known to frequently freeze (e.g. bridge edges). Mobile users (currently only PAT personnel) collect data from those Nodes, convey it to Gateways and even share it among themselves. Mobile users thus act as vectors for physically transporting any data across different geographical locations. As proof of concept, a set of isolated sensor nodes has been installed along the road linking two close multi-hop test clusters and tested, with promising results.

At the Operational Center data incoming is processed again and finally stored and made available to the PAT personnel via web, so all measurements' trends are stored and available online for further analysis (e.g., for defining new predictive models).

One of the strongest advantages of the Multi-hop Mode is its quasi-real-time monitoring service. Through periodical and/or on demand polling, the gathered alarms are conveyed to both Operational Centers and any passing-by driver. The analysis of network partitions allows easily localizing the troubled Nodes.

On the other hand, the Nomadic Mode allows monitoring specific places, which would be otherwise not covered by traditional technologies due to costs issues.

The resulting system is easily extendable and features failure resilience, since the Nodes can operate in either mode based on the reception of a specific synchronization beacon sent by the gateway: upon beacon reception, the Node switches into Multi-hop Mode; otherwise it switches into Nomadic Mode.

For properly operating, the Multi-hop Mode requires connectivity among Nodes. Any connectivity failure leads to network partitions, especially when the failed Node is located the closest to the Gateway. In case of partition, only the portion still connected with the Gateway operates properly, while the entire network beyond the failure is unreachable. The Nomadic Mode failure resilience property lessens, but does not completely remove this problem.

Being the Nodes powered by batteries, they need to operate with a given duty cycle and subject to a synchronism mechanism, whose precision is fundamental, since any synchronism loss translates into a temporary failure.

On the other hand the Nomadic Mode heavily depends on mobile users' mobility patterns, which might be limited in rural areas; and for energy saving reasons, only alarms are transmitted, whence no temperature and humidity map can be computed.

Moreover tests have emphasized problems of vandalism, rising the need of keeping device invisible as much as possible (e.g., via miniaturization or camouflage), and maintenance of the system, with simple procedures for device fixing/replacement.

At the cost of three traditional meteorological stations, the proposed system allows detecting alarms on eight road tracts over 120 locations, and collecting a quite detailed temperature and humidity map: a scalable and cost-effective solution for PAT.