Sensor Sensibility: Applications to Analytics, Modeling and Simulation
Deploying sensors in the environment provides opportunities that previously were impossible, or at least cost prohibitive. Twenty years ago, monitoring a wide field meant a deployed field crew visiting areas of interest frequently, or perhaps aerial surveys building an image product on a regular basis. Yet neither of these resource and cost intensive activities could provide the kind of granular insight that is becoming available today. At best, these were coarse approximations. While they allowed for the monitoring of change, they failed to capture sufficient data to begin to understand cause and effect – change in the context of independent variables.
With the wealth of data now available thanks to sensor networks, it is becoming possible to analyze these granular changes in correlation inputs to the system, and to ask meaningful questions. Sensor-based data provides significant applications to analysis, modeling and simulation. To understand how sensor data enables each of these, it is important to understand the distinction between these similar but distinct activities.
Business organizations are realizing tremendous operational improvements through the use of analytics. Analytics are tools that facilitate exploration of data, in order to understand patterns that can inform decisions. Analytics as it is currently understood are the tools and technologies that provide capabilities for data analysis and reporting. Key performance indicators represent any number of factors – asset failure rates, operations and maintenance dollars spent, capital investiture, asset age, asset depreciation, and many others. Analytics allow for exploration of these key performance indicators, potentially correlated with other factors using statistical techniques. Analytic tools cannot exist in a vacuum – they need good data in order for the statistical analysis to be meaningful.
Modeling and simulation are related but distinct activities, more closely allied with engineering than with management science. A model is a simplified representation of a system of some interest, typically fixed at some point in time and space. The purpose of building a model is that it promotes understanding of how the real system works, but allows for simplifying assumptions to be made to limit the complexity. So-called random components to the model are typically modeled as stochastic variables. A simulation is the application of the model to see how it evolves over space and time, or over other independent variables. Simulations typically allow for the compression of time and space – again, a simplification, but a necessary one. Simulation of the performance of a proposed cathodic protection program over ten years does little to facilitate decision making if it requires ten years for the simulation to complete. Modeling, then, is the process of deconstructing the system, while simulation is the exercising of the model over various input conditions – both flip sides of the same coin.
So how do sensors facilitate these activities? In terms of analysis, the provision of granular data from sensors provides more input for patterns to emerge, and better fit for regression. A pattern may appear random with only fifty input points, but with five million points patterns become evident. The accuracy of the statistical representation that key performance indicators describe improves with more data. Sensors make obtaining more data more cost effective and practical. In terms of analytics – the tools that facilitate analysis and reporting – sensors not only make provision of data to the tools more cost effective and practical, but also make that data available to the analytics system in real-time. As conditions change in the real world, the analytics tools can factor those conditions into decision support.
In terms of modeling, sensor-based data provides an important capability to calibrate the model as it is being constructed. Recalling that modeling is the decomposition of a complex system into simpler parts, the most difficult part of modeling is ensuring that the model is a faithful representation of the system. With only a few data points, the model may be calibrated to fit at those points in any number of ways; at x = 0 and x = 1, both y = x^2 and y = x^3 have the same solution, but outside of these data points the solutions diverge.
Sensor-based data provide not only more data, but data under a variety of different conditions in time and space, and this variety provides an important capability to calibrate the model – to ensure that the model is realistic and that simplifying assumptions are not too simple. Sensor-based data improve the capability for simulation, by allowing the simulation to leverage real-world conditions as they change over time and as they change with respect to each other. Rather than having to approximate input conditions – or even model the input conditions themselves – real data can be used, so that the simulation results are more realistic. Better engineering decisions can be made with better simulations.
An example from each of these areas will serve to better illustrate the application of sensor data to the challenges and opportunities of informed infrastructure. Physical systems such as bridges respond to stressors in the environment – whether due to movement from erosion or shifting tectonics, corrosion from environmental pollutants, or loading from increased traffic. Sensors in place on the bridge gather data about how much the support structures of the bridge are moving, how quickly erosion is occurring or how much traffic is crossing the bridge hourly. Sensors also provide data about the relationship between these different components – how much has movement increased with the decrease in structural integrity due to corrosion. Gathering this data improves the model’s ability to represent these stressors and allows the simulation to yield better designs with improved materials.
Sensors in the context of an asset analytics solution allow for monitoring of the assets continuously, rather than only having insight into asset condition on a yearly inspection schedule. Analysis of asset performance (or degradation of performance) facilitates a reliability centered maintenance (RCM) or condition based maintenance (CBM) program – so that valuable O&M funding is focused on assets for which condition has deteriorated, rather than a blanket time based maintenance strategy where maintenance is performed on all assets at the same time, irrespective of whether or not maintenance is required. Yearly inspection cycles do not provide enough insight into how the asset is performing over time, but sensor based data does – minimizing risk and reinvestment costs.
Sensor networks have applications to analytics, modeling and simulation that inform intelligent infrastructure, both for ongoing maintenance and for improvements to engineering next generation solutions.