Systems approach to evaluating sensor characteristics for real-time monitoring of high-risk indoor contaminant releases
Introduction
The sudden release of a toxic agent indoors can pose an acute health threat to building occupants. Consequently, it is of interest to devise indoor air monitoring systems that can detect, locate, and characterize accidental or deliberate toxic gas releases. However, developing such a system is complicated by several requirements. To limit the adverse impacts of the release, the monitoring system must detect and characterize releases in real time. Furthermore, the system must be robust against sensor error, such as false positive or negative measurements.
Much effort is being directed to the development of toxic chemical sensors. However, relatively little attention has been devoted to identifying and selecting sensors characteristics that will optimize the performance of sensor systems designed to protect occupants in indoor environments. Moreover, journal publications on improving chemical protection within indoor environments almost exclusively focus on optimizing the performance of sensors individually (e.g., electronic noses and robotic systems for plume tracking). The proceedings of a recent Indoor Air conference (2002) contained only two papers out of 726 that discussed how to design monitoring systems that protect building occupants from toxic releases. The proceedings of the 2004 IEEE Sensors conference contained no papers on this topic.
Commercial buildings commonly have monitoring systems for fire protection, security, and control of heating, ventilating and air conditioning (HVAC) equipment. Advances in microprocessor computing, networking, sensor technology, and artificial intelligence have helped foster a movement toward automated control of building systems and intelligent building systems (Piette et al., 2001; Kintner-Meyer et al., 2002; Liu and Kim, 2003; Jablonski et al., 2004). Despite rapid advances in these areas, systems that incorporate real-time information about indoor air pollutants have thus far been restricted mainly to ventilation control and energy management utilizing carbon dioxide sensors (Fisk and De Almeida, 1998).
Sohn et al. (2002a) demonstrated a Bayesian interpretation scheme for real-time reconstruction of an indoor contaminant release. That study used synthetic data from a single, small building. A follow-up study used real tracer-gas data in a three-story building (Sohn et al., 2002b). In both of these investigations, the sensors were assumed to be capable of reporting continuous concentration measurements.
This paper advances the earlier work by focusing on trigger- or alarm-type sensors, rather than continuous-output devices. A case-study approach is employed, using data from one of 12 tracer-gas experiments conducted at a three-story, 660 m3 building at the Dugway Proving Ground, Utah (Sextro et al., 1999). Through a series of examples, we examine how well various sensor systems, each system consisting of sensors with different sensor characteristics (threshold level, response time, and accuracy), reconstruct the release event. These examples demonstrate the importance of a systems perspective in selecting sensors with desirable sensor characteristics.
In addition to those already cited, only a few other published studies are closely related to the subject of this paper. Bayesian methods have been proposed for interpreting data from accidental radioactivity releases into outdoor air (Smith and French, 1993; Politis and Robertson, 2004); and for improving uncertainty estimates in Lagrangian photochemical air quality models (Bergin and Milford, 2000). Federspiel (1997) proposed a Kalman filter method to infer emission source strengths based on measured concentrations in a multizone building. A few studies have explored optimal sensor placement within a building for monitoring high-risk release events (Arvelo et al., 2002; Whicker et al., 2003).
Section snippets
Approach
We consider the following problem. A finite quantity of a contaminant is released, over a short duration, somewhere in a building (this may include its indoor air intake vents). A network of threshold or alarm-type sensors operates to detect the contaminant. We seek to understand how sensor characteristics such as threshold level and response time affect the ability of a sensor interpretation algorithm to quickly detect and characterize the contaminant release.
The indoor environment of the
Case study: sudden tracer release in a three-story building
This section explores the potential utility of the two-stage Bayesian interpretation algorithm for designing a sensor system that consists of single-level threshold sensors. Several examples are constructed using data from a single case study. The objectives are: (1) to test how well Bayesian interpretation works with threshold sensors; and (2) to explore how the system's performance varies with the sensor threshold level, response time, and error. The emphasis here is not on optimizing the
Results and discussion
To demonstrate data interpretation using threshold sensor data, we first investigate the ability of the sensor system to estimate the release location, mass, and duration. Next, we investigate the effect of changing the threshold level and response time characteristics, and lastly, the effect of changing the sensor error in conjunction with these characteristics.
Conclusion
Real-time environmental monitoring systems have the potential to help protect building occupants in the event of high-risk pollutant releases. The premise of this paper is that the selection of sensor characteristics is best performed from a systems perspective. Here, we have demonstrated—albeit for a limited set of circumstances—that a network of single-level threshold sensors can be used to determine the location and magnitude of the release within a Bayes Monte Carlo framework. More
Acknowledgements
This work was supported in part by a fellowship from the National Science Foundation, and by the Office of Chemical Biological Countermeasures, of the Science and Technology Directorate of the Department of Homeland Security, and performed under US Department of Energy Contract No. DE-AC03-76SF00098. We thank David Brown, Greg Foltz, William Swansiger, Norm Davis, Anthony Policastro, William Dunn and Edward Zellers for providing information on chemical sensors. We also thank Richard Sextro,
References (18)
- et al.
Application of Bayesian Monte Carlo analysis to a Lagrangian photochemical air quality model
Atmospheric Environment
(2000) COMIS—an international multizone air-flow and contaminant transport model
Energy and Buildings
(1999)- et al.
Sensor-based demand-controlled ventilation: a review
Energy and Buildings
(1998) - et al.
Analysis of an information monitoring and diagnostic system to improve building operations
Energy and Buildings
(2001) - et al.
On Monte Carlo methods for Bayesian inference
Ecological Modelling
(2003) - et al.
An enhanced multizone model and its application to optimum placement of CBW sensors
ASHRAE Transactions
(2002) - et al.
Updating uncertainty in an integrated risk assessment: conceptual framework and methods
Risk Analysis
(1995) Estimating the inputs of gas transport processes in buildings
IEEE Transactions on Control Systems Technology
(1997)- et al.(2004)
Cited by (50)
A comparative experimental study of two multi-robot olfaction methods: Towards locating time-varying indoor pollutant sources
2022, Building and EnvironmentCitation Excerpt :At present, extensive research has been conducted, and substantial progress has been made. Typical SSN methods include the quasi-reversibility and pseudo-reversibility methods [21–23], the regularization method [18,24], the probability-based adjoint inverse modeling method [20,25], the Markov-chain-based inverse modeling method [2,26], the optimization-based methods [13,27,28], and the Bayesian probability methods [29,30]. In recent years, SSN research has made encouraging progress in identifying indoor time-varying sources.
Comparing calculation methods of state transfer matrix in Markov chain models for indoor contaminant transport
2022, Building and EnvironmentCitation Excerpt :Besides, characterizing harmful releases using sensor data and adopting proper measures can significantly reduce the exposure risk of persons. These inverse tracking problems can be solved by forward methods [18,19], backward methods [20,21], and probability methods [22,23]. In those methods especially those forward, the Bayesian interpretation is popular [24], in which a mass of contaminant release simulations are necessary to build a pre-established database.