Accurate prediction of damage in electrode materials is crucial to automating reliability management for lithium-ion batteries. Phase field-based simulations offer valuable insight into the chemomechanical processes and are nowadays often used as substitutes in lieu of physical microscopy data for developing fracture detection and tracking algorithms. Even for simulated phase-field videos, the data amount is still limited, making a machine learning task challenging and the traditional approaches less effective. This study proposes an attention-enhanced approach that applies physics-informed attention operations under deep learning-based image prediction models. Using synchronized stress and damage field data, the proposed method is able to focus effectively on critical fracture regions to improve spatio-temporal prediction accuracy despite the limited amount of training data. Experimental results demonstrate the robustness and effectiveness of the proposed attention-enhanced approach, which could offer a scalable solution for real-time fracture analysis and for vision systems for in-line metrology in the long run. Note to Practitioners—Machine learning approaches are finding increasing application in material science and engineering problems nowadays. One bottleneck encountered in battery reliability management is the limitation of data amount, not only from physical experiments but also from large-scale physics-based simulations. Researchers and practitioners argue for the use of attention mechanisms to alleviate data scarcity challenges but still face the conundrum that whilst data scarcity strongly requires focus, a lot of training data is needed to know where to focus. Our study resolves this conundrum by combining domain information with machine learning and identifying the critical regions for focus without requiring a large amount of training data. This approach mirrors how a skilled engineer would prioritize areas of concern in material degradation analysis. This study will help practitioners develop scalable and data-efficient solutions that can eventually lead to better battery management towards reliability and longevity.

An autonomous experimentation platform in manufacturing is supposedly capable of conducting a sequential search for finding suitable manufacturing conditions by itself or even for discovering new materials with minimal human intervention. The core of the intelligent control of such platforms is a policy to decide where to conduct the next experiment based on what has been done thus far. Such policy inevitably trades off between exploitation and exploration. Currently, the prevailing approach is to use various acquisition functions in the Bayesian optimization framework. We discuss whether it is beneficial to trade off exploitation versus exploration by measuring the element and degree of surprise associated with the immediate past observation. We devise a surprise-reacting policy using two existing surprise metrics, known as the Shannon surprise and Bayesian surprise. Our analysis shows that the surprise-reacting policy appears to be better suited for quickly characterizing the overall landscape of a response surface under resource constraints. We do not claim that we have a fully autonomous experimentation system but believe that the surprise-reacting capability benefits the automation of sequential decisions in autonomous experimentation.Note to Practitioners Autonomous systems should be able to go beyond repetitive automatic actions that are generally pre-programmed through a recipe. To decide what to do next on the fly differentiates autonomy from automation. Arguably, autonomy is the highest form of automation. To endow a manufacturing with autonomy, one necessary capability is for it to react properly to the “unexpected,” which are those observations disagreeing with its model’s anticipation. Are these bad measurements, an anomaly, or an indicator of model inadequacy? Should the observations be discarded or should the model be updated using the new observation? If latter, should model be updated gradually overtime or radically altered? Figuratively, upon observing the unexpected, we say that a manufacturing control system is “surprised” and ask the question of how it should react. Our investigation shares our current insights on this question.

Real-time density estimation is ubiquitous in many applications, including computer vision and signal processing. Kernel density estimation is arguably one of the most commonly used density estimation techniques, and the use of "sliding window" mechanism adapts kernel density estimators to dynamic processes. In this article, we derive the asymptotic mean integrated squared error (AMISE) upper bound for the "sliding window" kernel density estimator. This upper bound provides a principled guide to devise a novel estimator, which we name the temporal adaptive kernel density estimator (TAKDE). Compared to heuristic approaches for "sliding window" kernel density estimator, TAKDE is theoretically optimal in terms of the worst-case AMISE. We provide numerical experiments using synthetic and real-world datasets, showing that TAKDE outperforms other state-of-the-art dynamic density estimators (including those outside of kernel family). In particular, TAKDE achieves a superior test log-likelihood with a smaller run-time.

Dimensionality reduction is a crucial first step for many unsupervised learning tasks including anomaly detection and clustering. Autoencoder is a popular mechanism to accomplish dimensionality reduction. In order to make dimensionality reduction effective for high-dimensional data embedding nonlinear low-dimensional manifold, it is understood that some sort of geodesic distance metric should be used to discriminate the data samples. Inspired by the success of geodesic distance approximators such as ISOMAP, we propose to use a minimum spanning tree (MST), a graph-based algorithm, to approximate the local neighborhood structure and generate structure-preserving distances among data points. We use this MST-based distance metric to replace the euclidean distance metric in the embedding function of autoencoders and develop a new graph regularized autoencoder, which outperforms a wide range of alternative methods over 20 benchmark anomaly detection datasets. We further incorporate the MST regularizer into two generative adversarial networks and find that using the MST regularizer improves the performance of anomaly detection substantially for both generative adversarial networks. We also test our MST regularized autoencoder on two datasets in a clustering application and witness its superior performance as well.

Tracking multiple moving objects in real-time in a dynamic threat environment is an important element in national security and surveillance system. It helps pinpoint and distinguish potential candidates posing threats from other normal objects and monitor the anomalous trajectories until intervention. To locate the anomalous pattern of movements, one needs to have an accurate data association algorithm that can associate the sequential observations of locations and motion with the underlying moving objects, and therefore, build the trajectories of the objects as the objects are moving. In this work, we develop a spatio-temporal approach for tracking maritime vessels as the vessel’s location and motion observations are collected by an Automatic Identification System. The proposed approach is developed as an effort to address a data association challenge in which the number of vessels as well as the vessel identification are purposely withheld and time gaps are created in the datasets to mimic the real-life operational complexities under a threat environment. Three training datasets and five test sets are provided in the challenge and a set of quantitative performance metrics is devised by the data challenge organizer for evaluating and comparing resulting methods developed by participants. When our proposed track association algorithm is applied to the five test sets, the algorithm scores a very competitive performance.

It is time-consuming and expensive to take high-quality or high-resolution electron microscopy (EM) and fluorescence microscopy (FM) images. Taking these images could be even invasive to samples and may damage certain subtleties in the samples after long or intense exposures, often necessary for achieving high-quality or high-resolution in the first place. Advances in deep learning enable us to perform various types of microscopy image-to-image transformation tasks such as image denoising, super-resolution, and segmentation that computationally produce high-quality images from the physically acquired low-quality ones. When training image-to-image transformation models on pairs of experimentally acquired microscopy images, prior models suffer from performance loss due to their inability to capture inter-image dependencies and common features shared among images. Existing methods that take advantage of shared features in image classification tasks cannot be properly applied to image transformation tasks because they fail to preserve the equivariance property under spatial permutations, something essential in image-to-image transformation. To address these limitations, we propose the augmented equivariant attention networks (AEANets) with better capability to capture inter-image dependencies, while preserving the equivariance property. The proposed AEANets captures inter-image dependencies and shared features via two augmentations on the attention mechanism, which are the shared references and the batch-aware attention during training. We theoretically derive the equivariance property of the proposed augmented attention model and experimentally demonstrate its consistent superiority in both quantitative and visual results over the baseline methods.

Dimensionality reduction is considered as an important step for ensuring competitive performance in unsupervised learning such as anomaly detection. Non-negative matrix factorization (NMF) is a widely used method to accomplish this goal. But NMF do not have the provision to include the neighborhood structure information and, as a result, may fail to provide satisfactory performance in presence of nonlinear manifold structure. To address this shortcoming, we propose to consider the neighborhood structural similarity information within the NMF framework and do so by modeling the data through a minimum spanning tree. We label the resulting method as the neighborhood structure-assisted NMF. We further develop both offline and online algorithms for implementing the proposed method. Empirical comparisons using twenty benchmark data sets as well as an industrial data set extracted from a hydropower plant demonstrate the superiority of the neighborhood structure-assisted NMF. Looking closer into the formulation and properties of the proposed NMF method and comparing it with several NMF variants reveal that inclusion of the MST-based neighborhood structure plays a key role in attaining the enhanced performance in anomaly detection.

This paper is concerned with investigating super-resolution algorithms and solutions for handling electron microscopic images. We note two main aspects differentiating the problem discussed here from those considered in the literature. The first difference is that in the electron imaging setting. We have a pair of physical high-resolution and low-resolution images, rather than a physical image with its downsampled counterpart. The high-resolution image covers about 25% of the view field of the low-resolution image, and the objective is to enhance the area of the low-resolution image where there is no high-resolution counterpart. The second difference is that the physics behind electron imaging is different from that of optical (visible light) photos. The implication is that super-resolution models trained by optical photos are not effective when applied to electron images. Focusing on the unique properties, we devise a global and local registration method to match the high- and low-resolution image patches and explore training strategies for applying deep learning super-resolution methods to the paired electron images. We also present a simple, non-local-mean approach as an alternative. This alternative performs as a close runner-up to the deep learning approaches, but it takes less time to train and entertains a simpler model structure.

Anomalies are data points or a cluster of data points that lie away from the neighboring points or clusters and are inconsistent with the overall pattern of the data. Anomaly detection techniques help distinguish the anomalous observations from the regular ones, and thus provide the basis for developing a standard performance guideline for process control. The process of identifying anomalies becomes complicated in the absence of labeled training data as in supervised learning. Moreover, Euclidean distance between two points is less likely able to reflect the intrinsic structural distance imposed by the underlying manifold structure. In this paper, the authors propose a minimum spanning tree (MST)-based anomaly detection method. The merit of the method is that an MST provides a new distance measure, capable of capturing the relative connectedness of data points/clusters in a complicated manifold, and could be a better (dis)similarity metric, than the simple Euclidean distance, to identify anomalies in unsupervised learning settings. The proposed method is compared with 13 popular anomaly detection methods on 20 benchmark data sets, demonstrating a considerable improvement in its ability of identifying anomalies. Furthermore, the MST-based anomaly detection is applied to the data set from a hydropower turbine and demonstrates remarkable detection competence.

Multiresolution metrology devices coexist in today's manufacturing environment, producing coordinate measurements complementing each other. Typically, the high-resolution (HR) device produces a scarce but accurate data set, whereas the low-resolution (LR) one produces a dense but less accurate data set. Research has shown that combining the two data sets of different resolutions makes better predictions of the geometric features of a manufactured part. A challenge, however, is how to effectively match each HR data point to an LR counterpart that measures approximately the same physical location. A solution to this matching problem appears a prerequisite to a good final prediction. We solved this problem by formulating it as a quadratic integer program, aiming at minimizing the maximum interpoint distance difference among all potential correspondences. Due to the combinatorial nature of the optimization model, solving it to optimality is computationally prohibitive even for a small problem size. We therefore propose a two-stage matching framework capable of solving real-life-sized problems within a reasonable amount of time. This two-stage framework consists of downsampling the full-size problem, solving the downsampled problem to optimality, extending the solution of the downsampled problem to the full-size problem, and refining the solution using iterative local search. Numerical experiments show that the proposed approach outperforms two popular point set registration alternatives, the iterative closest point and coherent point drift methods, using different performance metrics. The numerical results also show that our approach scales much better as the instance size increases, and is robust to the changes in initial misalignment between the two data sets.

This paper studies the problem of detecting the presence of nanoparticles in noisy transmission electron microscopic (TEM) images and then fitting each nanoparticle with an elliptic shape model. In order to achieve robustness while handling low contrast and high noise in the TEM images, we propose an approach to fuse two kinds of complementary image information, namely, the pixel intensity and the gradient (the first derivative in intensity). Our approach entails two main steps: 1) the first step is to, after necessary pre-processing, employ both intensity-based information and gradient-based information to process the same TEM image and produce two independent sets of results and 2) the subsequent step is to formulate a binary integer programming (BIP) problem for conflict resolution among the two sets of results. Solving the BIP problem determines the final nanoparticle identification. We apply our method to a set of TEM images taken under different microscopic resolutions and noise levels. The empirical results show the merit of the proposed method. It can process a TEM image of 1024×1024 pixels in a few minutes, and the processed outcomes appear rather robust.

This paper presents a method that enables automated morphology analysis of partially overlapping nanoparticles in electron micrographs. In the undertaking of morphology analysis, three tasks appear necessary: separate individual particles from an agglomerate of overlapping nano-objects; infer the particle's missing contours; and ultimately, classify the particles by shape based on their complete contours. Our specific method adopts a two-stage approach: the first stage executes the task of particle separation, and the second stage conducts simultaneously the tasks of contour inference and shape classification. For the first stage, a modified ultimate erosion process is developed for decomposing a mixture of particles into markers, and then, an edge-to-marker association method is proposed to identify the set of evidences that eventually delineate individual objects. We also provided theoretical justification regarding the separation capability of the first stage. In the second stage, the set of evidences become inputs to a Gaussian mixture model on B-splines, the solution of which leads to the joint learning of the missing contour and the particle shape. Using twelve real electron micrographs of overlapping nanoparticles, we compare the proposed method with seven state-of-the-art methods. The results show the superiority of the proposed method in terms of particle recognition rate.

A surveillance sensor system is a network of sensors that provides surveillance coverage to designated geographical areas. If all sensors are working properly, a well-designed surveillance system can supposedly provide the desirable level of detection capability for the locations and regions it covers. In reality, sensors may fail, falling out-of-service. Motivated by the need to determine the ability of a surveillance sensor system to tolerate the failure of sensors, we propose a fault tolerance capability measure to quantify the robustness of surveillance systems. The proposed measure is a conditional probability, characterizing the likelihood that a surveillance system is still working in the presence of sensor failures. Case studies of the surveillance sensor system in a major US port demonstrate that this new measure differentiates different surveillance systems better than using the sensor redundancy measure, or the reliability measure.

Using multiple, heterogeneous sensors in surveillance systems is desirable, not only to tolerate sensor failures, but also to increase the accuracy of the event detection process and to provide complementary capability under different operating conditions. In the operation of multiple, heterogeneous sensors, we may encounter inconsistent sensor observations. Motivated by the need to make coherent decisions, we propose in this study a decision scheme to determine the right interpretations of sensor outputs when conflict arises. The proposed decision rule considers sensor heterogeneity in a surveillance system, while attempting to minimize the expected misclassification cost. Case studies of the surveillance sensor system in a major U.S. port demonstrate that the proposed decision scheme achieves a better robustness in the presence of sensor failures than the populark-out-of-ndecision fusion rule.

A linear system whose model matrix is of size n×p is considered structured if some p row vectors in the model matrix are linearly dependent. Computing the degree of redundancy for structured linear systems is proven NP-hard. Previous computation strategy is divide-and-conquer, materialized in a bound-and-decompose algorithm, which, when the required conditions are satisfied, can compute the degree of redundancy on a set of much smaller submatrices instead of directly on the original model matrix. The limitation of this algorithm is that the current decomposition conditions are still restrictive and not always satisfied for many applications. We present a mixed integer programming (MIP) formulation of the redundancy degree problem and solve it using an existing MIP solver. Our numerical studies indicate that our approach outperforms the existing methods for many applications, especially when the decomposition conditions are not satisfied. The main contribution of the paper is that we tackle this challenging problem from a different angle and test a promising new approach. The resulting approach points to a path that can potentially solve the problem in its entirety.

This paper presents a robust calibration procedure for clustered wireless sensor networks. Accurate calibration of between-node distances is one crucial step in localizing sensor nodes in an ad-hoc sensor network. The calibration problem is formulated as a parameter estimation problem using a linear calibration model. For reducing or eliminating the unwanted influence of measurement corruptions or outliers on parameter estimation, which may be caused by sensor or communication failures, a robust regression estimator such as the least-trimmed squares (LTS) estimator is a natural choice. Despite the availability of the FAST-LTS routine in several statistical packages (e.g., R, S-PLUS, SAS), applying it to the sensor network calibration is not a simple task. To use the FAST-LTS, one needs to input a trimming parameter, which is a function of the sensor redundancy in a network. Computing the redundancy degree and subsequently solving the LTS estimation both turn out to be computationally demanding. Our research aims at utilizing some cluster structure in a network configuration in order to do robust estimation more efficiently. We present two algorithms that compute the exact value and a lower bound of the redundancy degree, respectively, and an algorithm that computes the LTS estimation. Two examples are presented to illustrate how the proposed methods help alleviate the computational demands associated with robust estimation and thus facilitate robust calibration in a sensor network.

We examine optimal repair strategies for wind turbines operated under stochastic weather conditions. In-situ sensors installed at wind turbines produce useful information about the physical conditions of the system, allowing wind farm operators to make informed decisions. Based on the information from sensors, our research objective is to derive an optimal preventive maintenance policy that minimizes the expected average cost over an infinite horizon. Specifically, we formulate the problem as a partially observed Markov decision process. Several critical factors, such as weather conditions, lengthy lead times, and production losses, which are unique to wind farm operations, are considered. We derive a set of closed-form expressions for the optimal policy, and show that it belongs to the class of monotonic four-region policies. Under special conditions, the optimal policy also belongs to the class of monotonic three-region policies. The structural results of the optimal policy reflect the practical implications of the turbine deterioration process.

Manufacturing systems are inherently imperfect both statically and dynamically. Tolerance and maintenance design are two major tools to address the static and dynamic imperfection of manufacturing processes (i.e., inherent process imperfection and tooling deterioration, respectively). Yet, traditionally, tolerance and maintenance designs have been studied separately to address these two critical areas of manufacturing systems. This paper presents an integrated framework of tolerance and maintenance design for multistation manufacturing processes. Two nonlinear optimization problems are formulated to minimize the overall average production cost in the long run, which includes the tolerance cost of tooling fabrication, maintenance cost, and the overall loss of quality (as a part of the objective function or as a constraint function). The proposed methodology is illustrated, analyzed, and further discussed in the context of a multistation automotive body assembly process. Extensive numerical analyses are conducted to demonstrate the efficiency of the developed methodology. Given various cost components and time horizons, the integrated design scheme is compared with traditional design schemes in terms of cost efficiency, offering new insights into the interrelation between manufacturing process maintenance and tolerancing in the context of the product life cycle. Note to Practitioners-With intensified competition as a result of economic globalization, quality and cost have become crucial factors to the success of any manufacturing industry. Decisions in the process design phase, such as process tolerance assignment and maintenance planning, play a substantial role for overall manufacturing quality and costs. Tolerance of process variables determines the inherent variation level of a manufacturing process. Preventive maintenance oversees and controls process degradation and its resulting deterioration on product quality. Significant tooling and operational costs result from both tolerancing and maintenance activities. Traditionally, tolerancing and maintenance decision-making have been studied separately. Tolerancing was mainly conducted during the design stage; while maintenance policy was often determined after a manufacturing system was designed and installed. However, tolerancing of process variables and maintenance decision-making policy are interconnected in modern manufacturing systems. Intuitively, tight initial tolerances specified on process variables are able to reduce the frequency of conducting maintenance during production, since the process can accommodate more deterioration to reduce maintenance cost; but they take a toll on tolerance cost. On the other hand, loose initial tolerances specified on process variables can lower design cost but increase the frequency of maintenance during production. Hence, there is a critical need to strike a balance between the tolerance cost of tooling fabrication and the maintenance cost of tooling replacement. This paper presents a new framework to integrate tolerance design and maintenance planning for multistation manufacturing processes. Optimization problems are formulated to minimize the overall production costs including tooling costs, maintenance costs, and quality loss. The proposed framework is illustrated in the context of automotive body assembly processes. When compared to other separated designs, this integrated design methodology leads to more desirable system performance with a significant reduction in production cost

Distributed sensing, a system-wide deployment of sensing devices, has resulted in both temporally and spatially dense data-rich environments. This new technology provides unprecedented opportunities for quality and productivity improvement. This paper discusses the state-of-the-art practice, research challenges, and future directions related to distributed sensing. The discussion includes the optimal design of distributed sensor systems, information criteria, and processing for distributed sensing and optimal decision making in distributed sensing. The discussion also provides applications based on the authors' research experiences. Note to Practitioners—This paper is based on a panel discussion on the topic of the emerging technology of distributed sensing and the associated challenges and opportunities. The panel, constituted by a group of leading researchers and practitioners with expertise in operations and statistics, convened during the Institute for Operations Research and the Management Sciences (INFORMS) 2003 annual meeting in Atlanta, GA. This panel focused its discussion on the information layer technology of distributed sensing for quality and productivity improvements, which differentiates this panel from other similar panels that were formed in a different society. The panelists provided their visions about the state-of-the-art practice, research challenges, and future research directions, and also discussed potential applications based on their own experiences.

Variance component estimation algorithms, in conjunction with automated in-process measurement technology, can be effective tools for identifying and eliminating major sources of manufacturing variation in assembly processes. Whether a particular set of variation sources are diagnosable depends critically on how the sensor system is laid out. Diagnosability tests are mathematical in nature and provide little insight into why a particular sensor layout may be nondiagnosable or how to modify the layout to ensure diagnosability. This paper translates the mathematical diagnosability conditions into a set of more conceptually meaningful conditions that provide better insight into the reasons behind the nondiagnosability. Note to Practitioners - This paper was motivated by the problem of identifying and eliminating major sources of variation in discrete-part manufacturing, which are critical steps in improving product quality. The effectiveness of statistical algorithms for estimating sources of variation depends on whether the sensor system for measuring key product and process variables is laid out properly, so that a particular set of diagnosability conditions are satisfied. This paper translates the rather abstract mathematical conditions for diagnosability into a set of more intuitive and conceptually meaningful conditions. This provides practitioners with insight into why a sensor system may be nondiagnosable and how to add or adjust sensors in order to ensure diagnosability. The diagnosability characterization can also be used to enhance performance and reduce computational expense in numerical search strategies for optimizing sensor layout.

This paper presents a methodology for the optimal design of fixture layouts in multistation assembly processes. An optimal fixture layout improves the robustness of a fixture system against environmental noises, reduces product variability, and leads to manufacturing cost reduction. Three key aspects of the multistation fixture layout design are addressed: a multistation variation propagation model, a quantitative measure of fixture design, and an effective and efficient optimization algorithm. One of the challenges raised by this multistation design is that a high-dimension design space, which usually embeds a lot of local optimums, will have to be explored. Consequently, it makes a global optimality more difficult and, if an inefficient algorithm is used, may require prohibitive computing time. In this paper, exchange algorithms, originally developed in the research of optimal experimental design, are adopted and further revised to optimize fixture layouts in a multistation process. The revised exchange algorithm provides a good tradeoff between optimality and efficiency: it remarkably reduces the computing time without sacrificing the optimal value. A four-station assembly process for a sports utility vehicle sideframe is used throughout the paper to illustrate the relevant concepts and the resulting methodology. Note to practitioners-This paper was motivated by the problem of planning a fixture locator layout in a multistation assembly process. Existing approaches generally focused on planning fixture locator layouts on a single workstation. In a multistation production process, such as an automobile body assembly process, the fixture locating holes used on one station will be reused on different stations, which could cause a station-to-station coupling in variation propagation. In other words, dimensional variation could originate from fixture elements on every station, propagate along the production line, and accumulate on the final assembly. Station-wise fixture layout design may not necessarily lead to a good solution because it overlooks the variation coupling and propagation effect. In this paper, we modeled the variation propagation across multiple stations and provided a quantitative characterization of the performance of fixture layout with the presence of environmental noises. Then, we recommended an efficient computation algorithm to solve for the optimal fixture layout. Our results showed that the multistation layout design is different from a single station one; some intuitions gained from single-station design work may not be still valid. The current work is based on a two-dimensional, rigid panel assembly model. The extension to accommodate more sophisticated two-dimensional, complaint-part assembly processes is much needed in the future research.

This paper presents a methodology for optimal allocation of sensors in a multistation assembly process for the purpose of diagnosing in a timely manner variation sources that are responsible for product quality defects. A sensor system distributed in such a way can help manufacturers improve product quality while, at the same time, reducing process downtime. Traditional approaches in sensor optimization fall into two categories: multistation sensor allocation for the purpose of product inspection (rather than diagnosis); and allocation of sensors for the purpose of variation diagnosis but at a single measurement station. In our approach, sensing information from different measurement stations is integrated into a state-space model and the effectiveness of a distributed sensor system is quantified by a diagnosability index. This index is further studied in terms of variation transmissibility between stations as well as variation detectability at individual stations. Based on an understanding of the mechanism of variation propagation, we develop a backward-propagation strategy to determine the locations of measurement stations and the minimum number of sensors needed to achieve full diagnosability. An assembly example illustrates the methodology.