HRHeraldo Rozas

Heraldo Rozas

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Heraldo Rozas has a total of 13 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • A Wasserstein Distributionally Robust Multiclass Support Vector Machine for Industrial Fault Diagnosis

    by Michael Ibrahim, Heraldo Rozas, Nagi Gebraeel

    Modern fault diagnosis models that rely on machine learning and AI tools have been beset by three key challenges: first, highly uncertain training data due to sensor noise and potentially erroneous labeling by human operators, second, large numbers of fault classes, and third, highly imbalanced training data (normal versus fault data signatures). While various works in the literature attempt to address these challenges, they often leave at least one unaddressed. Indeed, some efforts leverage distributionally robust optimization to robustify common classifiers against data uncertainty. However, they commonly only focus on binary models, which can amplify data imbalance issues and fail to learn correlations between classes in multiclass problems. Alternatively, the existing multiclass diagnostic models do not commonly focus on hedging against data uncertainty and erroneous labels. Finally, data imbalance issues are commonly addressed via data augmentation or ensemble models. However, the former can lead to low-quality synthetic data, whereas the latter can be computationally expensive. This article introduces a Wasserstein distributionally robust multiclass support vector machine model designed to mitigate uncertainties in both data and label information. We utilize the multiclass Crammer–Singer (CS) loss to capture correlations between multiple classes. This also makes it more resilient to data imbalance. We prove key regularity properties of the CS loss and derive a convex and tractable reformulation and upper bound for the linear and kernel versions of our proposed model, respectively. Using extensive numerical experiments, we show that our model outperforms popular existing approaches, especially in settings with severe data imbalance, even when the synthetic minority oversampling technique is used to balance the data.


  • On the Computation of Contextual Distributionally Robust Preventive Maintenance Intervals

    by Heraldo Rozas, Nagi Gebraeel, Weijun Xie

    The optimization of preventive maintenance (PM) intervals traditionally follows predict-then-optimize (PTO) frameworks. These involve two sequential steps: training a statistical model to estimate the failure time distribution (FTD) and then integrating it into an optimization model for deciding the optimal PM interval. However, PTO models may have poor out-of-sample performance if the fitted FTD differs significantly from the true distribution or fails to capture covariate effects. To overcome these issues, this paper introduces a contextual distributionally robust optimization (DRO) model for computing PM intervals. The proposed model integrates empirical failure time data directly into the optimization framework without assuming specific distributions. Our setting assumes that the component FTD is affected by covariates. Therefore, our formulation seeks to exploit covariate knowledge to compute efficient PM decisions conditional on the observed covariates. We formulate a DRO model that accounts for potential misspecifications of the empirical FTD. This DRO formulation aims to minimize the long-term maintenance cost rate by optimizing PM decision policies over an infinite space, where these policies map covariate information to optimal PM intervals. We demonstrate that the proposed DRO model admits tractable mixed-integer linear programming reformulations in various practical cases. The efficacy of our model is demonstrated through computational studies involving simulated and real-world failure time data.


  • Open Access

    Wasserstein Distributionally Robust Multiclass Support Vector Machine

    by Michael Ibrahim, Heraldo Rozas, Nagi Gebraeel

    We study the problem of multiclass classification for settings where data features $\mathbf{x}$ and their labels $\mathbf{y}$ are uncertain. We identify that distributionally robust one-vs-all (OVA) classifiers often struggle in settings with imbalanced data. To address this issue, we use Wasserstein distributionally robust optimization to develop a robust version of the multiclass support vector machine (SVM) characterized by the Crammer-Singer (CS) loss. First, we prove that the CS loss is bounded from above by a Lipschitz continuous function for all $\mathbf{x} \in \mathcal{X}$ and $\mathbf{y} \in \mathcal{Y}$, then we exploit strong duality results to express the dual of the worst-case risk problem, and we show that the worst-case risk minimization problem admits a tractable convex reformulation due to the regularity of the CS loss. Moreover, we develop a kernel version of our proposed model to account for nonlinear class separation, and we show that it admits a tractable convex upper bound. We also propose a projected subgradient method algorithm for a special case of our proposed linear model to improve scalability. Our numerical experiments demonstrate that our model outperforms state-of-the art OVA models in settings where the training data is highly imbalanced. We also show through experiments on popular real-world datasets that our proposed model often outperforms its regularized counterpart as the first accounts for uncertain labels unlike the latter.

  • Condition-Based Maintenance for Wind Farms Using a Distributionally Robust Chance Constrained Program

    by Heraldo Rozas, Weijun Xie, Nagi Gebraeel

    Operations and Maintenance (O&M) expenses account for up to 30% of the operating costs of wind farms. Condition-based maintenance (CBM) strategies, which incorporate predictive analytics into maintenance optimization, have been proven to be effective in reducing O&M costs in wind farms. Existing predictive CBM strategies for wind farms rely on the assumption that predictive analytics can accurately estimate the remaining lifetime distribution (RLD) of wind turbines, allowing for the direct implementation of stochastic programming or threshold-based policies. However, estimated RLDs can be inaccurate due to noisy sensors or limited training data. To address this issue, this paper develops a CBM strategy for wind farms that uses a Distributionally Robust Chance Constrained (DRCC) optimization model. Our formulation acknowledges that estimated distributions may be inaccurate and so seeks solutions that are robust against distribution perturbations within a Wasserstein ambiguity set. We show that the proposed DRCC optimization problem can be exactly reformulated as an integer linear program. We derive methods to strengthen the Big-M values of this reformulation, thereby enabling the DRCC model to be efficiently solved by off-the-shelf optimization software. The proposed strategy is validated through computational studies using real-world and synthetic degradation data, outperforming stochastic programming and robust optimization benchmark models.


  • An Integrated Detection-Prognostics Methodology for Components With Intermittent Faults

    by Michael Ibrahim, Heraldo Rozas, Nagi Gebraeel

    Some industrial components, such as valves, relay switches, and motors, occasionally experience intermittent faults (IFs) that usually disappear without any repair or intervention. This phenomenon occurs at a relatively low frequency even in components that are in an “as-good-as-new” state. However, an increase in the frequency of IFs often indicates the onset of degradation. We develop an integrated detection-prognostics model for components that exhibit IFs and whose degradation data are high-dimensional. We discuss the use of dynamic time warping (DTW) and a variational autoencoder (VAE) to perform feature engineering on the data. We then propose a hidden Markov model (HMM)-based monitoring strategy composed of two parts: (1) a detection model that tracks and flags changes in the intermittent fault frequency (IFF) and (2) a prognostic model that leverages how the transition probabilities of the HMM evolve with progressive degradation to compute the remaining life distribution (RLD) of the component. We examine the performance of our modeling framework using high-dimensional data generated from a vehicular electrical system testbed designed to accelerate the degradation of a vehicle starter motor.


  • A stochastic programming model for jointly optimizing maintenance and spare parts inventory for IoT applications

    by Jiachen Shi, Heraldo Rozas, Murat Yildirim, Nagi Gebraeel

    Service supply chain models typically use conservative maintenance and spare part management policies that result in significant losses due to redundancies. Conservatism without an improved understanding of risks, however, does not cushion against unexpected consequences. Risk scenarios associated with asset failure and inventory shortage are frequently observed in practice. Advances in Internet of Things (IoT) technology is unlocking new methods that attain significant prediction accuracy for these risk factors. IoT-enabled predictions on asset state of health can drive dynamic decision models that conduct maintenance and replenishment actions more efficiently while reducing risk. In this study, we propose a unified framework that utilizes IoT data to jointly optimize condition-based maintenance and inventory decisions. We formulate our problem as a stochastic mixed integer program that accounts for the interplay between maintenance, spare parts inventory, and asset reliability. We introduce a new reformulation that’s efficient for solving large-scale instances of the proposed model. The framework presented herein is applied to real world degradation data to demonstrate the benefits of our methodology in terms of cost and reliability.


  • Open Access

    Unifying Criteria for Calculating the Levelized Cost of Driving in Electro-Mobility Applications

    by Diego Arias-Cazco, Heraldo Rozas, Diego Jimenez, Marcos E. Orchard, Claudio Estevez

    The Levelized Cost of Driving (LCOD) calculates how much it costs to drive a vehicle, per kilometer, over the vehicle lifespan, and it is typically measured in USD/km. LCOD has been widely applied to assess and compare the performance of Electric Vehicles (EVs) and Internal Combustion engine Vehicles (ICVs). Unfortunately, there is no common methodology to establish how to compute LCOD. As authors use different criteria to determine which elements are relevant in LCOD, it is difficult to compare results. To overcome this issue, we re-introduce the LCOD concept, where the most relevant expenditures and incomes are included. This novel LCOD definition also incorporates key elements that are sometimes neglected, such as the second life of batteries in EVs, EV battery degradation models, and stochastic models, to characterize daily operation. The proposed methodology is applied to assess the cost of using EVs as taxis in the city of Quito, Ecuador.