Carl Malings

Value of information for spatially distributed systems: Application to sensor placement

This paper investigates how the value of information (VoI) metric can guide information collection and optimal sensor placement in spatially distributed systems. VoI incorporates relevant features to decision-making, such as uncertainty about the state of the system, precision of measurements, the availability of intervention actions, and the overall cost of managing the system. Spatially distributed systems also allow for information propagation, i.e. measurements collected at one location can be used to update knowledge at other related locations. In this paper, while restricting our attention to Gaussian random field and binary state models, we illustrate first how sensor placements depend on the decision-making problem to be addressed, as encoded in a problem-specific loss function, and second how the complexity of VoI computations is impacted by this loss functions characteristics. In doing so, we consider several loss functions and present computational techniques for evaluating VoI under them. Finally, we apply these techniques to efficiently optimize sensor placements by the VoI metric in two example applications.

Sensor Placement Optimization for Structural Health Monitoring

When designing a system for structural health monitoring, one must decide how to optimally deploy sensor systems to obtain information beneficial to the management of the structure. In this paper, we present a methodology for making these decisions based on a probabilistic graphical model representing the structure to be monitored, as well as actions undertaken by infrastructure managers in monitoring and maintaining the structure. A value of information metric is used to quantify the benefit of different proposed sensor placement schemes. A greedy algorithm heuristic is used to optimize sensor placements based on this metric. This methodology is demonstrated on an example structural health monitoring problem based on fiber optic strain gauge measurements of the Sherman and Joyce Bowie Scott Hall, under construction on the Carnegie Mellon University campus in Pittsburgh, PA, USA. doi: 10.12783/SHM2015/301

Surface heat assessment for developed environments: Probabilistic urban temperature modeling

Abstract Extreme heat waves, exacerbated by the urban heat island effect, have major impacts on the lives and health of city residents. Projected future temperature increases for many urban areas of the United States will further exacerbate these impacts. Accurate predictions of the spatial and temporal distribution of risk associated with such heat waves can support the optimal implementation of strategies to mitigate these risks, such as the issuance of heat advisories and the activation of cooling centers. In this paper, we describe how fine resolution simulations of historic extreme heat events are generated and used to train a probabilistic spatio-temporal model of the temperature distribution in an urban area. We further demonstrate how this model can be used to combine temperature data from various sources and downscale regional predictions in order to provide accurate fine resolution temperature forecasts. Applications of this model are presented for two urban areas: New York City, NY and Pittsburgh, PA, USA. Based on simulated temperature data from fine resolution forecasting models, we find that this probabilistic method can improve the prediction accuracies of urban temperatures, locally and especially in the short-term, with respect to other temperature forecasting and interpolation methods, such as the use of average city-wide temperature predictions and estimates at discrete weather stations.

Value-of-information in spatio-temporal systems: Sensor placement and scheduling

Abstract The management of infrastructure involves accounting for factors which vary in space over the system domain and in time as the system changes. Effective system management should be guided by models which account for uncertainty in these influencing factors as well as for information gathered to reduce this uncertainty. In this paper, we address the problem of optimal information collection for spatially distributed dynamic infrastructure systems. Based on prior information, a monitoring scheme can be designed, including placement and scheduling of sensors. This scheme can be adapted during the management process, as more information becomes available. Optimality can be defined in terms of the value of information (VoI), which provides a rational metric for quantifying the benefits of data gathering efforts to support system management decision-making. However, the computation of this metric in spatially and temporally extensive systems can present a practical impediment to its implementation. We describe this complexity, and investigate a special case of system topology, termed as a temporally decomposable system with uncontrolled evolution, in which the complexity of assessing VoI grows at a manageable rate with respect to the system management time duration. We demonstrate the evaluation and optimization of the VoI in an example of such a system.

Sensor network optimization using Bayesian networks, decision graphs, and value of information

Bayesian Networks (BNs) and decision graphs provide a useful framework for modeling the uncertain behavior of civil engineering infrastructures subjected to various risks, as well as the potential outcomes of risk mitigation actions undertaken by managing agents. These graphs can also guide optimal sensing and inspection of infrastructure by maximizing the value of information of sensing efforts. This paper presents a general framework for modeling infrastructure systems using BNs and for evaluating sensor placement metrics within this model. An example application of the use of the value of information metric in guiding optimal sensing in a system of infrastructure assets in the San Francisco Bay area subjected to seismic risk is then presented. A parametric study also investigates the sensitivity of the value of information metric to various parameters of the BN system model.