thitivong/ iStock / Getty Images Plus

Technology Perspective

Pinpointing pollutants using the Internet of Things

Researchers are building large networks of wireless sensors to monitor our environment and health 

4 July 2019

Smart technologies have the potential to transform lives by helping build efficient and healthy environments. The development of tiny chips and sensors that can be incorporated into devices could help carefully monitor many aspects of daily life, from gauging exercise levels to measuring air quality.

Technology specialists at KACST are well placed to embed this ‘Internet of Things’ (IoT) into modern Saudi culture, with access to state-of-the-art laboratories and testing centres to conduct their research. In a project built on a long-standing collaboration with researchers at the University of California, KACST scientists are taking the first steps in creating a low-cost communication network for IoT applications using mobile sensors to measure air quality. This project is part of Saudi Arabia’s pioneering Vision 2030 governmental programme.

“Building low-cost networks of reliable sensors, including those that can be incorporated into personal devices like mobile phones, will provide us with high-quality data and unprecedented details about our local environment,” says Hatim Bukhari, co-director of the Center of Excellence for Telecom Applications (CETA), a joint center between KACST and University of California San Diego (UCSD). “This will help us make informed decisions on personal, community, national and international levels, to improve urban planning and air quality, for example.” 

The team is creating technologies that enable mobile environmental measurements and improve the reliability and efficiency of IoT systems. A central aim is to build a robust mobile sensing platform for measuring air quality. This system includes low-cost sensors that can accurately measure common pollutants, including nitrogen dioxide, ozone, carbon monoxide and carbon dioxide. This can help identify localised hot spots of excessive pollution, and keep people informed about their daily exposure to pollutants. 

“We have now gathered several rounds of air quality data, which has been used to generate accurate calibration models for the low-cost sensors,” says Bukhari. “Sensor calibration is a vital part of data collection, The  sensor needs to be ‘trained’ to recognise influencing factors, such as localized temperature shifts and cross-contamination from other pollutants, in order to filter out the most valuable information.” 

Optimizing sensors

Usually, sensor packages are calibrated in the field alongside a high-spec reference instrument gathering high-quality data at a desired location. The two datasets are then compared and used to build a calibration model and train the low-cost sensors. However, the low-cost sensors may perform poorly when they are moved to different locations. Poor data quality from low-cost gas sensors is of particular concern, because they are very sensitive to changing environmental and atmospheric conditions, and to cross-contamination. 

Tajana Rosing and her UCSD team recently demonstrated the importance of building calibration models using data from diverse locations1. They have built a generalised calibration model based on neural networking, which has harnessed the power of multiple sensor packages rotated around multiple sites (each collecting information on individual gases). The model splits calibration into two stages — one corrects for sensor-to-sensor variation, while the other combines data gathered by similar sensors across all training environments. In this way, new sensors are trained using readily-calibrated sensors, reducing costs and enhancing network reliability.

While the technology within the sensors itself is paramount, the infrastructure around the sensors is also important. If wireless communications within and between sensors are disrupted or inconsistent, this interferes directly with the results received, resulting in inaccurate or inefficient data collection and analyses. The complexities of ensuring efficient communication with mobile-based sensing, and indeed with fixed sensor networks, is a key challenge for IoT development. 

For all IoT systems, there is a fundamental trade-off between battery lifetime and the amount of data that a device can transmit. Bukhari’s team is exploring methods of processing data on the sensors themselves, so that they are no longer required to send data wirelessly to a server. This can save power and enhance the overall security of the system. The team is currently running simulations to trial IoT deployments with the aim of maximising battery life. They have also developed a system to facilitate wireless communication between distributed fixed sensors using an adaptive drone swarm. 

“Further, we have been working on a control algorithm for mobile environmental monitoring using drones,” says Bukhari. “This enables the drones to communicate with one another and carry sensors throughout the environment along the paths selected to collect the most useful information in an efficient manner.”

A versatile system for diverse scenarios

The optimization of control algorithms, particularly in robotics-based or drone-based platforms, is further complicated by the requirements of different systems. A good example would be a group of autonomous vehicles working together to clear up following an accident at a chemical plant. The robots or drones  would first require access to real-time pollutant data in order to prioritise cleaning of the worst affected areas first. 

“The more robots or drones that are measuring the environment or searching a disaster scenario, the more information can be gathered in a timely manner,” says Bukhari. “However, with a large number of autonomous vehicles it is more complicated to determine the best possible set of actions for each one.” 

This problem can be approached in various ways, from calculating plans based solely on the activity of a robot's neighbours, to calculating a series of actions for the whole group based on short timeframes. There are many circumstances in which a situation may be persistent over time, requiring continual monitoring and repeated actions – for example, cleaning an area that is continually accumulating pollutants. Data gathered by persistent monitoring can feed into models that predict future conditions, or predict conditions at other, unmeasured locations.

Michael Ostertag and colleagues in the IoT project team based in San Diego are working to improve autonomous robotic sensing systems for persistent monitoring2. To measure a site comprehensively, control algorithms are needed to ensure each robot covers an optimal set path at the correct velocity, with time to pause and collect viable data at precise points. Ostertag’s team has developed a velocity controller that has outperformed existing controllers in initial drone model trials. 

All cities could one day benefit from the IoT technologies currently being developed at KACST and UCSD. The project’s overall results and insights will feed into urban planning, highlighting the most polluted zones and ensuring resources and traffic management plans are implemented appropriately. 

“Ultimately, embracing the future of IoT will enable first-class environmental monitoring, allowing for better measurement of real-time atmospheric conditions, pollution and weather prediction, and efficient search capabilities during natural disasters,” says Bukhari. “It is crucial to keep these larger goals in mind as we work through the initial set-up and maintenance of our IoT platforms, ensuring they remain robust and valuable tools long into the future.” 

References

  1. Vikram, S., Collier-Oxandale, A., Ostertag, M., Menarini, M., Chermak, C., Dasgupta, S., Rosing, T., Hannigan, M., & Griswold, W.G. Evaluating and improving the reliability of gas-phase sensor system calibrations across new locations for ambient measurements and personal exposure monitoring. Atmospheric Measurement Techniques Discussions (2019)