Pablo L. Durango-Cohen

Incorporating Maintenance Effectiveness in the Estimation of Dynamic Infrastructure Performance Models

In this paper, the authors demonstrate how intervention analysis can be used in conjunction with dynamic performance modeling to analyze the effect of maintenance activities on the performance of infrastructure facilities. Specifically, state-space specifications of autoregressive moving averages with exogenous inputs models are considered to develop deterioration and inspection models for infrastructure facilities, and intervention analysis is used to estimate transitory and permanent effects of maintenance (e.g., performance jumps or deterioration rate changes). To illustrate the methodology, the effectiveness of an overlay on a flexible pavement section from the AASHO Road Test is analyzed. Results show the effect of the overlay on improvements both on surface distress (rutting and slope variance) as well as on the pavements underlying serviceability. The results also provide evidence that the overlay changes the pavements response to traffic; that is, the overlay causes a reduction in the rate at which traffic damages the pavement.

Application of Statistical Process Control for Structural Health Monitoring of a Historic Building

AbstractThe authors apply a statistical process control framework to support structural health monitoring of the Grace Church building in Charleston, South Carolina. Specifically, they conduct a post-hoc analysis of displacement data acquired via remote monitoring of delamination between two wythes of brick in a clerestory wall. The framework consists of formulation and estimation of statistical models to explain the progression of the measurements under ordinary conditions and use of control charts to detect unusual events. One such event was excessive displacement in September 2011 that led the engineer of record to close the building to public access and order immediate repairs. The analysis also reveals a few unusual events that were not apparent from visual interpretation of the data, including a possible precursor to the aforementioned event.

A Bernoulli-Gaussian mixture model of donation likelihood and monetary value: An application to alumni segmentation in a university setting

Advances in computational power and enterprise technology, e.g., Customer Relationship Management (CRM) software and data warehouses, allow many businesses to collect a wealth of information on large numbers of consumers. This includes information on past purchasing behavior, demographic characteristics, as well as how consumers interact with the organization, e.g., in events, on the web. The ability to mine such data sets is crucial to an organizations ability to deliver better customer service, as well as manage its resource allocation decisions. To this end, we formulate a Bernoulli-Gaussian mixture model that jointly describes the likelihood and monetary value of repeat transactions. In addition to presenting the model, we derive an instance of the Expectation-Maximization Algorithm to estimate the associated parameters, and to segment the consumer population. We apply the model to an extensive dataset of donations received at a private, Ph.D.-granting university in the Midwestern United States. We use the model to assess the effect of individual traits on their contribution likelihood and monetary value, discuss insights stemming from the results, and how the model can be used to support resource allocation decisions. For example, we find that participation in alumni-oriented activities, i.e., reunions or travel programs, is associated with increased donation likelihood and value, and that fraternity/sorority membership magnifies this effect. The presence/characterization of unobserved, cross-sectional heterogeneity in the data set, i.e., unobserved/unexplained systematic differences among individuals, is, perhaps, our most important finding. Finally, we argue that the proposed segmentation approach is more appealing than alternatives appearing in the literature that consider donation likelihood and monetary value separately. Among them and as a benchmark, we compare the proposed model to a segmentation that builds on a multivariate Normal mixture model, and conclude that the Bernoulli-Gaussian mixture model provides a more coherent approach to generate segments.

Dynamic Learning Process for Selecting Storm Protection Investments

Increasingly aggressive weather events, such as hurricane-driven storm surges, threaten surface transportation systems and motivate defensive actions, including hardening. Decisions about the design and scale of hardening investments are informed by meteorological records. Historically based probabilities of severe storms are used in practice to define expected values of the intensity of weather assaults (e.g., the 100-year storm) and then to select defenses. The prospects of climate change and rising sea level suggest that assuming weather events are stationary may present added risks to surface transportation infrastructure, particularly in coastal environments. This paper proposes a dynamic, learning-based investment strategy, similar to the concept of real options, that updates estimates of storm surges on the basis of experience and recommends incremental hardening investments when observed trends indicate that additional defense is warranted. Monte Carlo simulation is used to compare and evaluate static (expected value-based) and dynamic investment strategies in the context of storm intensity patterns that are (a) known, (b) incorrectly estimated, and (c) nonstationary, with growing intensity. Results suggest that when the future is well described by past experience, the static, once-and-done decision strategy works well, but when the underlying storm generation process is unknown, or when it is changing (growing) in intensity, the learning-based dynamic strategy is especially advantageous. These results underscore the importance of flexibility in designing storm protection, of tracking weather events closely to detect emerging trends, and of data-driven decision strategies. This dynamic approach to decision making under uncertainty can be applied to other sources of uncertainty, for example, demand estimates.

Introduction to special issue on transportation infrastructure management

Transportation infrastructure provides mobility, and thus access to people, goods, services and resources. The availability and level of service of these systems determine quality of life and the continuity of economic and business growth, and, in turn, have served as motivation to develop models to support their management. In the last two decades, there have been significant developments in technologies to collect data related to infrastructure condition, as well as in theoretical and computational capabilities enabling the solution of ever more sophisticated and realistic resource allocation problems. As in other areas, these advances have, in turn, spurred research to improve models, for example to better capture alternatives and their impact (on users), and to address technical issues, such as accounting for serial dependence or unobserved heterogeneity. The motivation for the special issue is to show how researchers from around the world are addressing these issues. Specifically, the special issue includes papers that present interdisciplinary quantitative methods to address problems arising in monitoring, evaluating, repairing, and renewing transportation infrastructure systems with the objective of ensuring that they are capable of performing the functions for which they were designed and built in a reliable, sustainable and equitable fashion.

Coordination of Maintenance and Rehabilitation Policies for Transportation Infrastructure

This paper formulates the maintenance and repair (M&R) decision-making process as a quadratic program. The functional interdependencies in transportation infrastructure are captured in the model, and the conditions under which it is optimal to synchronize maintenance schedules for groups of components are found. This result suggests that an effective management process depends on attending to the interdependencies that link a system’s facilities.

Identification and estimation of latent group-level-effects in infrastructure performance modeling

As in other panel data analyses, the presence of unobserved heterogeneity is a critical issue in the estimation of infrastructure performance models. In the literature, this issue has been addressed by formulating variable intercept, fixed or random effects models under the assumptions that (1) heterogeneity stems from facility/individual-level effects, and that (2) the coefficients are constant and homogeneous across the population. In contrast, we present mixture regression as a performance modeling framework. The approach relies on the assumption that the underlying population is comprised of a finite set of classes/segments in unknown proportions. The segmentation basis is latent meaning that the criteria to establish the number and type of segments are related to unobserved heterogeneity manifested in facility performance/deterioration. The segments are characterized by a set of commonly specified regression equations, which allows for the identification and estimation of coefficients, i.e., group-level effects, that differ in terms of their level-of-significance, magnitude or sign. We also derive an instance of the Expectation-Maximization Algorithm to estimate the associated parameters, and to assign facilities to the population segments. To illustrate the framework, we analyze the performance of a panel of 131 pavements from the AASHO Road Test. The results suggest both observed and unobserved sources of heterogeneity in the panel. The heterogeneity is captured by differential group-level effects, which we estimate and interpret. We also discuss how these effects can be exploited in the development of resource allocation strategies. We also compare the mixture regression model to established benchmarks.

A clusterwise linear regression model of alumni giving

We present a clusterwise regression model to analyse alumni contributions to a private, PhD-granting university in the Midwestern USA. The model provides a framework to simultaneously segment a population, and to explain the effect of various factors on the mean annual value of donations. We contribute a different approach to marketing studies in the university fundraising context, where segmentation is often based on intuitive, albeit possibly biased criteria. Instead, in clusterwise regression, individuals are assigned to segments with the objective of maximising the within-segment variation explained by a set of regression models. Our main finding is that individuals in different segments display systematic, but unobserved differences in their responses, i.e., the coefficients in the segment-level regression models exhibit differences in their magnitude, sign and level of significance. We discuss how characterising such differences can support tailored solicitation strategies.

Estimating Pavement Performance Models using Advanced Technologies and Time Series Analysis

This paper proposes state-space specifications of autoregressive moving average models and structural time series models as a framework to develop and estimate performance models for transportation infrastructure facilities. Time series models in state-space form fit the maintenance optimization model of Durango-Cohen and are consistent with the latent performance modeling approach of Ben-Akiva and Ramaswamy. To illustrate the proposed framework, the paper developed and estimated performance models for an asphalt pavement using pressure and deflection measurements generated by sensors and falling weight deflectometers, respectively. Analysis of the results shows that the ensuing models are consistent with physical properties of flexible pavements. The results also indicate that state-dependence may be statistically significant and further reinforce the computational and statistical advantages of the proposed framework over Markovian transition probabilities.