J1.1
Evaluation of sensor placement techniques

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 19 January 2010: 3:30 PM
B308 (GWCC)
Ian H. Griffiths, RiskAware Ltd, Bristol, United Kingdom; and I. Bush

Presentation PDF (1.8 MB)

Sensors for detecting toxic airborne materials are important assets that can be expensive and often limited in availability. Chemical and biological (CB) sensors need to positioned with care to provide the maximum information to allow timely identification of an accidental or intentional release of dangerous materials. The aim is to minimise the casualties and effects, and this should be the key goal of any placement strategy. This application is of particular relevance to homeland security and defence applications where authorities wish to put in place protective measures.

There are two main approaches to sensor placement: rules-based placement and computer optimisation. The former involves placing sensors according to heuristic rules. The latter generally involves running multiple dispersion simulations, then using an optimiser to determine the best placement. This approach can be computationally intensive, and so its application may need to be reserved for special cases. We have conducted a comprehensive study into CB sensor placement to determine the relative benefits of rules-based and computer optimisation, and how these approaches may be improved. We have implemented a rules-based method that uses a simulated annealing algorithm to automate placing of sensors against prescribed rules. We have also implemented a computer optimisation approach that uses a sequential optimiser to place sensors to minimise the number of casualties across several thousand Monte Carlo sampled hazard plumes. We will briefly describe both approaches and some recent developments on each.

Our evaluation study is composed of two main elements: a simulation-based study and field trials data from the FUSION Field Trial 07 (FFT07) experiment. The former involved modelling a large numbers of scenarios generated using Monte Carlo simulation, then comparing the performance of the rules-based and computer optimisation techniques. The study ranged from simple test cases, designed to provide a deep understanding of the optimal placement strategies, through to realistic operational cases that allowed the fundamental insights to be confirmed in real-world situations. Within each, we considered numerous combinations of sensor types (both biological and chemical) and material releases, as well as variations in the overall scenarios. To date, over 1000 scenarios have been considered, with a total in excess of 3 million simulations. Further scenarios are being modelled. The results of the simulation-based study are being validated using field trials data from the FFT07 experiment.

This presentation will summarise the full results of the evaluation. We aim to determine the relative benefits of rules-based and computer optimisation for sensor placement, and the minimum number of sensors required in different scenarios as well as the effectiveness of deploying more sensors.