Search

Summary: Currently, structural health monitoring (SHM) of complex metallic structures based on the localization and identification of acoustic emission (AE) sources has become one of the most common condition monitoring method. However, existing methods are difficulty in accurately localizing and identifying AE sources generated by complex metallic structures that have been surface modified or machined. To overcome this problem, this paper presents a novel architecture named nearfield frequency space sparse decomposition (NFSSD) for localizing AE sources collected from complex metallic structures. Main contributions of the proposed NFSSD are to incorporate the decomposed subbands of AE signal in frequency into the traditional sparse decomposition (SD), which can extract more effective information and improve the identification of coherent AE sources. On this basis, NFSSD‐based AE feature extraction scheme is further proposed for improving the accuracy and stability of AE source localization for complex metallic structures. First, all frequency point estimates of the original AE signal used to divide the subbands are obtained, where each frequency corresponds to the center frequency of the subband. Furthermore, the spatial spectrum of each subband signal is solved over the entire spatial domain, and the spatial spectrum of the signal is obtained to estimate the location of AE source. Two experimental results of coordinate‐based AE source localization of complex metallic structures indicate that the proposed method has better AE source localization performance compared to conventional localization approaches. Specifically, the results show that the proposed approach can provide an effective theoretical reference for AE‐based SHM of complex metallic structures. [ABSTRACT FROM AUTHOR]

Summary: Nowadays, the feature extraction method of multichannel acoustic emission (AE) signal provides a solid research foundation for digital and intelligent fault diagnosis of the hoisting system. More specifically, AE signal collected from the hoisting system is generally characterized by nonlinear and non‐stationary, thus making the traditional intelligent fault diagnosis methods cannot accurately extract the inherent fault features. To alleviate this problem and improve the accuracy of multichannel fault diagnosis, a new fault diagnosis method for hoisting system based on differential search algorithm‐variational mode decomposition (DSA‐VMD) and improved deep convolutional neural network (IDCNN) is proposed in this paper. Specifically, the proposed DSA‐VMD and IDCNN method is divided into two main components: (i) the inside parameters (K, a) of VMD is optimized to effectively extract the multichannel AE fault feature via DSA‐VMD and (ii) the extracted multichannel fault components are fed into the designed IDCNN algorithm to accomplish fault identification automatically. Experimental results from the hoisting system demonstrate the effectiveness of the proposed approach. Additionally, the superiority of the proposed approach has also been verified in extracting fault information and fault identification compared to the other multichannel fault diagnosis methods. [ABSTRACT FROM AUTHOR]

The optimal stopping problem is concerned with finding an optimal policy to stop a stochastic process in order to maximize the expected return. This problem is critical in stochastic control and can be found in many different fields, such as operations research, finance, and healthcare. In this paper, we model the underlying stochastic process of the optimal stopping problem as a Markov decision process and propose a computationally efficient model-free value-based reinforcement learning approach, named Δ V-learning. The efficiency is improved by taking advantage of the unique structural properties of the optimal stopping problem into our algorithm design. We consider two types of the optimal stopping problems: the standard optimal stopping and the regenerative optimal stopping, which differ in their transition dynamics once the stopping action is executed. We conduct numerical experiments on our proposed method and compare its performance against existing reinforcement learning algorithms and rule-based policies. The results show that our Δ V-learning method is able to outperform the benchmark algorithms in all experiments. • Study the standard and regenerative optimal stopping problems. • Develop a new reinforcement learning method to solve the optimal stopping problems. • Incorporate the structural properties of optimal stopping into the algorithm. • Conduct experiments to demonstrate performance improvements. [ABSTRACT FROM AUTHOR]

• A CBM model with adaptive inspection for repairable multi-unit systems is proposed. • An environment is developed using the PH model to train the DRL agent. • DRL is implemented to solve a large-scale problem in the SMDP framework. Condition-based maintenance (CBM) optimization may turn intractable when a complex system with multiple units becomes an asset of interest. This paper aims to find a CBM policy for a multi-unit series system subject to stochastic degradation, where a new inspection is scheduled based on age and condition monitoring data upon each inspection. The novelty of this study lies in proposing a modified deep reinforcement learning (DRL) algorithm for the semi-Markov decision processes (SMDP) to find an opportunistic CBM policy for a multi-unit system with economic dependency over an infinite horizon, where a range of repair actions are allowed under an aperiodic inspection scheme. We also suggested a novel environment simulator that considers the simultaneous impact of age and covariates using the proportional hazards (PH) model and the system's reliability characteristics. DRL acts as not only a learning algorithm obviating the full specification of the model but also an approximate scheme producing a solution in a limited computation. The proposed algorithm is applied to a multi-unit hydroelectric power system with the damage self-healing property to demonstrate the higher performance of the DRL algorithm in cost reduction than alternative policies and explain how enhancing system reliability reduces costs during the learning process. [ABSTRACT FROM AUTHOR]

Socially Responsible Service Operations Management (SRSOM) can be defined as managerial activities to design and control (or optimize) various service operations that involve (or integrate) social responsibility toward legal, ethical, and environmental concerns. We propose a framework of research on SRSOM from the perspective of the service triad: target market, service concept, and service delivery system design choices. Unlike conventional service operations management, SRSOM focuses on: (1) non-voluntary citizens rather than voluntary customers as the main target group; (2) the provision of a general necessary service to citizens as the service concept; and (3) effectiveness and responsibility rather than efficiency as the service delivery system design choice. We also provide an overview of the related research on SRSOM published over the past 20 years, and attempt to place the seven papers shown in this collection in the proposed framework. While the current literature on SRSOM is limited, the number of papers has been steadily increasing. We hope that this paper will lead to further research on SRSOM among researchers and practitioners in related fields. [ABSTRACT FROM AUTHOR]

The article presents information related to pairs trading that have been popular as a statistical arbitrage strategy among major investment banks and hedge funds since its birth in the 1980s. It is noted that distance method, co-integration and stochastic spread are three quantitative methods that have been developed and applied to pairs trading in the literature.

In highly connected financial networks, the failure of a single institution can cascade into additional bank failures. This systemic risk can be mitigated by adjusting the loans, holding shares, and other liabilities connecting institutions in a way that prevents cascading of failures. We are approaching the systemic risk problem by attempting to optimize the connections between the institutions. In order to provide a more realistic simulation environment, we have incorporated nonlinear/discontinuous losses in the value of the banks. To address scalability challenges, we have developed a two-stage algorithm where the networks are partitioned into modules of highly interconnected banks and then the modules are individually optimized. We developed a new algorithms for classical and quantum partitioning for directed and weighed graphs (first stage) and a new methodology for solving Mixed Integer Linear Programming problems with constraints for the systemic risk context (second stage). We compare classical and quantum algorithms for the partitioning problem. Experimental results demonstrate that our two-stage optimization with quantum partitioning is more resilient to financial shocks, delays the cascade failure phase transition, and reduces the total number of failures at convergence under systemic risks with reduced time complexity. [ABSTRACT FROM AUTHOR]

• A deep partial transfer learning network is proposed for intelligent diagnosis. • Diagnostic knowledge is selectively transferred in presence of domain asymmetry. • Domain-asymmetry factors are automatically learned to weight distance metrics. • An adversarial network is constructed to adapt partial distributions of features. Applications of deep transfer learning to intelligent fault diagnosis of machines commonly assume symmetry among domains: 1) the samples from target machines are balanced across all health states, and 2) the diagnostic knowledge required by target machines is consistent with source machines. In reality, however, such assumptions cannot be justified as machines operate normally in most of the time with only occasional faults. As a result, the collected monitoring data from target machines contain massive healthy samples but a small number of faulty samples, and some health states experienced in source machines may never happen in target machines. Therefore, if sufficient labeled data are available with diverse health states from the source machines, only partial diagnostic knowledge can be transferred to a target machine in presence of domain asymmetry. In order to selectively transfer diagnostic knowledge across asymmetric domains, we propose an adversarial network architecture named deep partial transfer learning network (DPTL-Net). The DPTL-Net uses a domain discriminator to automatically learn domain-asymmetry factors, by which the source machine samples are weighted to block irrelevant knowledge in the maximum mean discrepancy based distribution adaptation. The performance of the DPTL-Net is demonstrated in two case studies, where the diagnostic knowledge is transferred across different working conditions of a planet gearbox and across different yet related bearings. The results show that the DPTL-Net achieves better diagnostic performance than other transfer learning methods due to its transfer capability in presence of domain asymmetry. [ABSTRACT FROM AUTHOR]

Abstract: This paper addresses a hub-and-spoke network problem for railroad freight, where a central planner is to find transport routes, frequency of service, length of trains to be used, and transportation volume. Hub-and-spoke networks, often found in air freight, have not been favoured by railways in the past. Such a structure could be profitable, however, if there exist concentrated freight flows on some service links. We formulate a linear integer programming model whose objective function includes not only the typical operational cost, but also cost due to the transit time spent by freight in the network. We then develop heuristic algorithms to solve large scale instances occurring in rail freight systems in France plus Italy; Germany; and a 10-country European network. By assuming that every node is equipped with consolidation capability, we let the final solution naturally reveal potential hub locations, the impact of several of which is studied by sensitivity analysis. [Copyright &y& Elsevier]

Abstract: We consider the carrier’s optimal bid generation problem in combinatorial auctions for transportation procurement. Bidders (carriers) employ vehicle routing models to identify sets of lanes (origin-destination pairs) based on the actual routes that a fleet of trucks will follow in order to maximize profit. Routes are constructed by optimally trading off repositioning costs of vehicles and the rewards associated with servicing lanes. The carrier optimization represents simultaneous generation and selection of routes and can incorporate any existing commitment. We employ both column generation and Lagrangian based techniques for solving the carrier optimization model and present numerical results. [Copyright &y& Elsevier]

Abstract: We consider a multiple-vehicle routing problem with split pick-ups (mVRPSP). This problem involves multiple suppliers, a single depot, and a fleet of identical capacity trucks responsible for delivering supplies from the suppliers to the depot. Any supplier may be visited by more than one truck, thus allowing split pick-ups. The problem is to determine, for each truck, which suppliers to visit and the size of loads to pick up so as to minimize the total transportation cost for the fleet, which depends on the number of trucks used and their routes. We develop a fundamentally new model for the mVRPSP, a deterministic dynamic program (DP). Although the most natural DP formulation results in a DP with uncountably-infinite state and action spaces, an optimality-invariance condition we establish leads to an equivalent DP with finite state and action spaces. This DP formulation leads to a new exact algorithm for solving the mVRPSP, based on a shortest path search algorithm, which is conceptually simple and easy to implement. [Copyright &y& Elsevier]

Continuous Unsupervised Domain Adaptation (CUDA) can alleviate deep learning models' performance degradation on out-of-distribution data. However, low stability, the erosion of past knowledge while adapting to new domains, remains a major challenge. Traditional proposed solutions such as (1) A pproximation, or (2) M emorization fail to harmonize system's computational load, memory resources and stability. In response, we introduce A pproximate and M emorize (A&M). Compared to traditional approximation methods, A&M utilizes factorized generative models to mitigate modal collapse, offering improved computational efficiency and training stability. Moreover, compared to traditional memorization approaches, A&M boosts memory efficiency by learning compressed representations of past information rather than storing it in the raw format. A&M shows 50% less forgetting for problems with constrained memory and an extended number of domains, making A&M ideal for real-world systems due to its scalable and parallel design. [ABSTRACT FROM AUTHOR]

Reinforcement learning (RL) for dynamic asset allocation is an emerging field of study. Total return, the common performance metric, is useful for comparing algorithms but does not help us determine how close an RL algorithm is to an optimal solution. In real-world financial applications, a bad decision could prove to be fatal. One of the key ideas of our work is to combine the two paradigms of the mean-variance optimization approach (Markowitz criteria) and the optimal capital growth approach (Kelly criteria) via the actor-critic approach. By using an actor-critic approach, we can balance optimization of risk and growth by configuring the actor to optimize the mean-variance while the critic is configured to maximize growth. We propose a Geometric Policy Score used by the critic to assess the quality of the actions taken by the actor. This could allow portfolio manager practitioners to better understand the investment RL policy. We present an extensive and in-depth study of RL algorithms for use in portfolio management (PM). We studied eight published policy-based RL algorithms which are preferred over value-based RL because they are better suited for continuous action spaces and are considered to be state of the art, Deterministic Policy Gradient (DPG), Stochastic Policy Gradients (SPG), Deep Deterministic Policy Gradient (DDPG), Trust Region Policy Optimization (TRPO), Proximal Policy Optimization (PPO), Twin Delayed Deep Deterministic Policy Gradient (TD3), Soft Actor Critic (SAC), and Evolution Strategies (ES) for Policy Optimization. We implemented all eight and we were able to modify all of them for PM but our initial testing determined that their performance was not satisfactory. Most algorithms showed difficulty converging during the training process due to the non-stationary and noisy nature of financial environments, along with other challenges. We selected the four most promising algorithms DPG, SPG, DDPG, PPO for further improvements. The modification of RL algorithms to finance required unconventional changes. We have developed a novel approach for encoding multi-type financial data in a way that is compatible with RL. We use a multi-channel convolutional neural network (CNN-RL) framework, where each channel corresponds to a specific type of data such as high-low-open-close prices and volumes. We also designed a reward function based on concepts such as alpha, beta, and diversification that are financially meaningful while still being learnable by RL. In addition, portfolio managers will typically use a blend of time series analysis and cross-sectional analysis before making a decision. We extend our approach to incorporate, for the first time, cross-sectional deep RL in addition to time series RL. Finally, we demonstrate the performance of the RL agents and benchmark them against commonly used passive and active trading strategies, such as the uniform buy-and-hold (UBAH) index and the dynamical multi-period Mean-Variance-Optimization (MVO) model. [ABSTRACT FROM AUTHOR]

• Incorporating portfolio constraints into recurrent reinforcement learning framework. • Reinforcement learning algorithm with continuous trading actions over multiple assets. • Simultaneous control of the portfolio constraints and policy network hyperparameters. • Hourglass shape network architectures emerge as a natural choice for asset management. Recurrent reinforcement learning (RRL) techniques have been used to optimize asset trading systems and have achieved outstanding results. However, the majority of the previous work has been dedicated to systems with discrete action spaces. To address the challenge of continuous action and multi-dimensional state spaces, we propose the so called Stacked Deep Dynamic Recurrent Reinforcement Learning (SDDRRL) architecture to construct a real-time optimal portfolio. The algorithm captures the up-to-date market conditions and rebalances the portfolio accordingly. Under this general vision, Sharpe ratio, which is one of the most widely accepted measures of risk-adjusted returns, has been used as a performance metric. Additionally, the performance of most machine learning algorithms highly depends on their hyperparameter settings. Therefore, we equipped SDDRRL with the ability to find the best possible architecture topology using an automated Gaussian Process (GP) with Expected Improvement (EI) as an acquisition function. This allows us to select the best architectures that maximizes the total return while respecting the cardinality constraints. Finally, our system was trained and tested in an online manner for 20 successive rounds with data for ten selected stocks from different sectors of the S&P 500 from January 1st, 2013 to July 31st, 2017. The experiments reveal that the proposed SDDRRL achieves superior performance compared to three benchmarks: the rolling horizon Mean-Variance Optimization (MVO) model, the rolling horizon risk parity model, and the uniform buy-and-hold (UBAH) index. [ABSTRACT FROM AUTHOR]

In industrial environments, individual sensor is easily affected by background noise, etc. In order to improve the reliability of anomaly detections, sensors are arranged at multiple measuring points to collect monitoring data of machines. However, under the coupling of vibration responses of multiple components of machines, the complex nonlinear relationship between monitoring data of multiple measuring points makes it difficult to achieve the best feature extraction and fusion effect, which reduces the accuracy of anomaly detection. To solve this problem, an unsupervised spatiotemporal fusion network augmented with random mask and time-relative information modulation is proposed. Firstly, we creatively propose random mask and modulated signal generation method based on mask index to learn the dependence of waveform and time dimension and achieve temporal dimension fusion of signals. Based on end-to-end training, modulated signals are also more conducive to spatial fusion. Then, to fully exploit the correlation between monitoring data of multiple measuring points and obtain the best spatial dimension fusion effect, a multi-head graph neural network based on self-attention weight matrix is carried out. Finally, we use transformer encoder to reconstruct the signal of each measuring point and obtain reconstruction error. Based on exponentially weighted moving average, anomaly detection threshold is obtained. Two anomaly detection experiments are conducted, and accuracy of 99.78%, 99% are achieved. [ABSTRACT FROM AUTHOR]

Recent years have witnessed the prominent advancements of deep learning (DL) in the arsenal of prognostics and health management. However, the prognostic uncertainty problem extensively existed in industrial devices is not addressed by most DL approaches. This article formulates a novel Bayesian Deep Learning (BDL) framework to characterize the prognostic uncertainties. A distinguished advantage of the framework is its capacity of capturing the comprehensive effects of two critical uncertainties: 1) epistemic uncertainty, accounting for the uncertainty in the model, and 2) aleatoric uncertainty, representing the impact of random disturbance, such as measurement errors. The former arises from the variability of the model weights, and the latter is characterized by selected lifetime distributions. We integrate both uncertainties by defining BDL as priors of lifetime parameters. A sequential Bayesian boosting algorithm is executed to improve the estimation accuracy and compress the credible intervals. The superior prediction performance of our framework is validated by a real-world dataset collected from hydraulic mechanisms of circuit breakers. [ABSTRACT FROM AUTHOR]

The successful applications of deep transfer learning to intelligent fault diagnosis testify to a positive correlation between transferable feature similarity and knowledge transferability across diagnostic tasks. This correlation makes feature similarity possible to assess diagnostic knowledge transferability. Therefore, researchers have attempted various measures for feature similarity, and distance metrics have been adopted as an objective measure for feature distribution discrepancy. However, the commonly used distance metrics cannot address the joint distribution discrepancy (JDD) due to the difficulty in fitting conditional distributions of target domain samples. To overcome the problem, we resort to explore cluster-conditional distributions instead and propose an optimal transport-embedded joint distribution similarity measure (OT-JDSM) that is implemented in two steps. First, a cluster-true label propagation spreads labels from a small number of labeled target domain samples to the whole. Second, the JDD of transferable features is produced via an efficient solution of optimal transport. OT-JDSM is demonstrated on synthetic examples and 144 transfer diagnosis tasks that are created by public and private bearing datasets. The results show that OT-JDSM of transferable features has a stronger correlation with diagnostic knowledge transferability than other distance metrics. Moreover, the OT-JDSM gain can quantify the transfer performance of diagnostic models on tasks. [ABSTRACT FROM AUTHOR]

This article studies a single-item periodic-review inventory problem with stochastic demand, uncertain price, and price search cost. At the beginning of a period, an inventory manager has to decide, considering the current inventory level, whether a price should be searched for at a non-zero cost. Once the price is known, she will have to decide the order size. For tractability the number of realizable prices is limited to two and (r,S1,S2)-type policies are considered, whereris the threshold for the price search decision andSiis the order-up-to level for pricepifori= 1, 2. Although the problem is significantly simplified, it still allows for price speculations by the inventory manager; i.e., she requests a quote but may not buy. The properties of long-run average costs are studies and optimization algorithms are presented. Numerical studies show the effectiveness of the proposed policy compared with classic (s,S)-type policy and its natural three-parameter extension. [ABSTRACT FROM AUTHOR]

Cloud manufacturing (CMfg) is emerging as an advanced service-oriented manufacturing model for the manufacturing industry. It is changing the way manufacturing enterprises organize production activities in that tangible and intangible manufacturing resources encapsulated as virtualized manufacturing services are provided to customers. In this background, CMfg task semantic modeling and description is proposed to address different challenges such as matching with manufacturing services. Firstly, some of the essential features of CMfg tasks are briefly discussed and classification as well as the modeling requirements of CMfg tasks is also investigated, and then this paper proposes the modeling framework of CMfg task to solve the general CMfg task ontology (GCMT_Ontology) construction and the task sub-ontology matching from GCMT_Ontology. In GCMT_Ontology construction process, the original CMfg task ontology (OCMT_Ontology) is built from the CMfg task description model, and then an ontology learning approach is advanced to perfect the GCMT_Ontology with graph-based semantic similarity algorithm. In task sub-ontology matching process, semantic feature values of concepts are calculated to extend the semantic content of the task document, and the extended concepts are input to find the sub-ontology by computing minimum connected sub-graph. Finally, the proposed framework is realized in a prototype system, and an example of cement grinding mill assembling task is carried out in a group enterprise to illustrate the application. The results show that the proposed method improves enterprise collaboration and interoperation effectively. [ABSTRACT FROM AUTHOR]

Purpose: To develop mathematical models to predict the evolution of tumor geometry in cervical cancer undergoing radiation therapy. Methods: The authors develop two mathematical models to estimate tumor geometry change: a Markov model and an isomorphic shrinkage model. The Markov model describes tumor evolution by investigating the change in state (either tumor or nontumor) of voxels on the tumor surface. It assumes that the evolution follows a Markov process. Transition probabilities are obtained using maximum likelihood estimation and depend on the states of neighboring voxels. The isomorphic shrinkage model describes tumor shrinkage or growth in terms of layers of voxels on the tumor surface, instead of modeling individual voxels. The two proposed models were applied to data from 29 cervical cancer patients treated at Princess Margaret Cancer Centre and then compared to a constant volume approach. Model performance was measured using sensitivity and specificity. Results: The Markov model outperformed both the isomorphic shrinkage and constant volume models in terms of the trade-off between sensitivity (target coverage) and specificity (normal tissue sparing). Generally, the Markov model achieved a few percentage points in improvement in either sensitivity or specificity compared to the other models. The isomorphic shrinkage model was comparable to the Markov approach under certain parameter settings. Convex tumor shapes were easier to predict. Conclusions: By modeling tumor geometry change at the voxel level using a probabilistic model, improvements in target coverage and normal tissue sparing are possible. Our Markov model is flexible and has tunable parameters to adjust model performance to meet a range of criteria. Such a model may support the development of an adaptive paradigm for radiation therapy of cervical cancer. [ABSTRACT FROM AUTHOR]

• Optimal scheduling of repair actions reduces maintenance costs of two-unit systems. • Identifying appropriate maintenance level is a critical factor in cost reduction. • Replacement policy enhances reliability while it can be highly expensive. • Allowing repair actions can result in the efficient use of component's life cycle. This paper proposes an opportunistic maintenance policy for maintaining mechanical systems consisting of two series units with economic dependency. Condition monitoring and age information is available for unit 1 and unit 2, respectively. A gamma process describes the stochastic deterioration of unit 1, and condition monitoring data is utilized to estimate its hazard rate using the proportional hazards model (PHM). A matrix-based algorithm is employed to derive the conditional reliability of a two-unit series system with a continuous-state deterioration process. The policy proposes preventive actions on scheduled intervals, based on lifetime and PHM parameters, to enhance system reliability and mitigate unexpected downtimes. Corrective actions rectify the failed unit, while the other unit can be maintained opportunistically. The problem is formulated as a semi-Markov decision process (SMDP) with an objective of minimizing the long-run expected average cost over an infinite time horizon. This paper relaxes a common assumption that repair brings the units to the as-new condition, and provides solutions for the maintenance scheduling of two-unit series systems, where repair is also allowed. [ABSTRACT FROM AUTHOR]

We study a joint replenishment and delivery scheduling (JRD) problem in which a central warehouse serves n-retailers in the presence of vague operational conditions such as ordering cost and inventory holding cost. In the proposed fuzzy set-based approach, an exact membership function is not assumed and instead can be approximated using piecewise linear functions based on alpha level sets because of their easy handling and efficiency. Subsequently, the fuzzy total cost is defuzzified by the widely used signed distance method to ranking fuzzy numbers. However, due to the JRD's difficult mathematical properties, efficient and effective solution procedures for the problem have eluded researchers. To find an optimal solution, an effective and efficient differential evolution (DE) algorithm is designed. After determining the appropriate parameters of the DE by parameter tuning test, the effectiveness of the DE is verified by numerical examples. We compare the DE with the available best approach and results show that DE can solve this non-deterministic polynomial hard problem in a robust way with a high convergence rate and low average error. [ABSTRACT FROM AUTHOR]

Abstract: The joint replenishment problem (JRP) has received considerable attention and all of the work on JRP is under explicit environment. In fact, the decision makers often have to face vague operational conditions. In this paper, a novel JRP model with fuzzy minor replenishment cost and fuzzy inventory holding cost is developed. More concisely, this model is a fuzzy dependent-chance programming (DCP) model. Subsequently, the technique of the traditional fuzzy simulation (FS) approach and differential evolution algorithm (DE) are integrated to design a hybrid intelligent algorithm named FSDE-I to solve this practical fuzzy JRP. Thirdly, another intelligent algorithm named FSDE-II using an improved FS approach is proposed to estimate the credibility more precisely. Finally, FSDE-I and FSDE-II are illustrated with numerical examples and the results show the effectiveness of FSDE-II. [Copyright &y& Elsevier]

Abstract: We propose a two-stage stochastic integer programming model for the winner determination problem (WDP) in combinatorial auctions to hedge the shipper’s risk under shipment uncertainty. The shipper allows bids on combinations of lanes and solves the WDP to determine which carriers are to be awarded lanes. In addition, many other important comprehensive business side constraints are included in the model. We demonstrate the value of the stochastic solution over one obtained by a deterministic model based on using average shipment volumes. Computational results are given that indicate that moderately sized realistic instances can be solved by commercial branch and bound solvers in reasonable time. [Copyright &y& Elsevier]

In responding to freight transportation fire incidents, first responders refer to the terials labeled on the freights and the Emergency Response Guidebook (ERG) for guidance on the initial response. However, when the burning goods are mixed or unknown, first responders require support on the appropriate response strategy. In this work, we propose an artificial intelligence (AI) enabled tool to aid first responders in the initial emergency response to fire incidents involving mixed or unknown materials. We used many different machine learning models, including MultiRocket, Omni-Scale CNN (OS-CNN), Ultra-Fast Shapelets (UFA), WEASEL+MUSE, Dynamic Time Warping (DTW), Time Series Forest (TSF), the Random Interval Spectral Ensemble (RISE), and InceptionTime, to develop a trained machine learning (ML) model integrated into the AI tool. The ML model identifies the hazard characteristics (e.g., toxicity and explosivity) of a given fire, based on chemicals found in the effluent. The effluent data was collected from samples of burning materials using a cone calorimeter and a Fourier Transform Infrared Spectroscopy (FTIR) gas analyzer. We discuss the methodologies behind developing the ML model and demonstrate its high classification accuracy with time series augmentations to enhance the training dataset. We also implemented Self-Supervised Learning, Domain-Agnostic Contrastive Learning (DACL), and Adversarial Perturbation-based Latent Construction (APLR) to further improve the model. The results show that MultiRocket outperforms other models in terms of accuracy and running time. Furthermore, we developed a user interface for our AI-enabled tool, which uses a data sample as input to provide not only predictions of the fire hazards but also actionable intelligence to the first responders for safe and effective fire suppression strategies. Our proposed toolkit benefiting from AI will pave the way to further research in accessible, easily-deployable AI-enabled tools in fire sciences and management. Environmental Implications We leverage machine learning to predict the hazard classification of materials that may not be considered hazardous by Emergency Response Guidebook (ERG) used by Transport Canada, but may release toxic substances when burning. For example, polyvinyl chloride is not considered as hazardous by ERG, but toxic and corrosive hydrochloric gas is produced when burning. We aim to provide more decision support to first responders in freight transportation fires involving non-hazardous or unknown materials to identify the scope of the emergency from the early stage of the fire, which is critical to protect the environment from the effects of the fire. • Applied machine learning to identify fire hazards from a combustion analysis dataset. • Time series classification and augmentations achieve high classification results. • Self-supervised learning achieves good results without hand-crafted augmentations. • Developed a software to help first responders identify emergency response guides. [ABSTRACT FROM AUTHOR]

Since classical deep learning (DL) techniques are hungry for massive data and suffer from domain shift, domain adaptation (DA) methods are broadly adopted in prognostics and health management (PHM) to align source and target domains. However, DA relies on target datasets collected in advance, which are not always available in practice. In this paper, a domain generalization (DG) approach, which learns from multiple source domains and generalizes well to unseen domains, is introduced for remaining useful life (RUL) prediction of bearings under unseen operating conditions. Specifically, we propose an adversarial out-domain augmentation (AOA) framework to generate pseudo-domains, thereby increasing the diversity of available samples. Hence, a generator is trained in an adversarial manner to generate augmented pseudo-domains by maximizing the domain discrepancy of the latent representations. In addition, we add manifold and semantic regularization to its objective function to ensure the consistency of the pseudo-domains. Trained with these available domains, a task predictor can improve the generalization in inaccessible target domain. Based on this, we provide a specific implementation of AOA-based RUL prediction for DG and validate its effectiveness and superiority using experimental datasets. [ABSTRACT FROM AUTHOR]

The tremendous success of deep learning and transfer learning broadened the scope of fault diagnosis, especially the latter further improved the diagnosis accuracy under multiple working conditions. However, most existing attempts assume that label distribution is domain-invariant despite taking into account the different feature distributions. This does not accommodate the diversity of fault mode distributions under different operating conditions and weakens the generalization to imbalanced domain adaptation (IDA) scenarios. Therefore, this work proposed a novel deep imbalanced domain adaptation (DIDA) framework for fault diagnosis of bearings, aiming at the challenging scenario where feature shift and label shift exist simultaneously under different working conditions. Specifically, DIDA overcomes the class-imbalanced label shift and achieves a fine-grained latent space matching by cost-sensitive learning and categorical alignment. Besides, margin loss regularization is introduced to further optimize classification boundaries and improve cross-domain generalization for IDA fault diagnosis tasks. Finally, we simulated the IDA protocols on experimental datasets and conducted case studies under multiple working conditions, thus validating the effectiveness and superiority of the proposed framework. • A novel deep imbalanced domain adaptation (DIDA) is proposed. • DIDA narrows both feature shift and label shift. • DIDA broadens fault diagnosis to IDA scenarios. • Experimental case studies verified validity and superiority. [ABSTRACT FROM AUTHOR]

The sheer amount of Municipal Solid Waste (MSW) produced from residential and commercial activities has posed a big threat to the modern society, especially to large cities by causing environmental issues such as greenhouse gas emission, leakage of leachate and contamination of air, soil and water resources. The urban managers are keen on improving MSW management to mitigate the environmental impacts and to improve the sustainability of their cities. This study proposes a systematic approach to build an integrated recycling and disposal network for MSW by explicitly considering the sustainability with an objective to maximize the total profit with a budget constraint. A multi-period mixed integer linear programming model was proposed to design the network optimally as well as to optimally operate the network. The optimization model involves decisions related to facilities selection, capacity level and location of facilities, allocation of solid waste to facilities, transportation of waste among facilities and distribution of recycled materials. This paper also extended the developed social life cycle assessment methodology to model the social impacts of the network. The model was applied to a real-world case in Tehran, Iran, and could provide the profit of USD 43.49 M over a five-year planning period. In order to show the impacts of sustainability and budget limitations, the model was compared with three models: I) the model without sustainability, II) the model without budget limitations and III) the model without sustainability and budget limitations. The models I and III led to the loss of USD -308.60 and -362.80 M, which are not desirable. This shows the consideration of sustainability will improve the profit in the long term. The model II could provide the profit of USD 99.73 M over a five-year planning period, indicating that limited budget have significant impact on the way the recycling network is operating. [ABSTRACT FROM AUTHOR]

• A multi-source transfer learning network is proposed for intelligent diagnosis. • Diagnostic knowledge is complemented to recognize unseen faults of machines. • Multiple subnets are proposed to adapt partial distribution of source-target pairs. • A multi-source knowledge fusion module is constructed to fuse results from subnets. Most of the current successes of deep transfer learning-based fault diagnosis require two assumptions: 1) the health state set of source machines should overlap that of target machines; 2) the number of target machine samples is balanced across health states. However, such assumptions are unrealistic in engineering scenarios, where target machines suffer from fault types that are not seen in source machines and the target machines are mostly in a healthy state with only occasional faults. As a result, the diagnostic knowledge from source machines may not cover all fault types of target machines nor address imbalanced target samples. Therefore, we propose a framework, called a multi-source transfer learning network (MSTLN), to aggregate and transfer diagnostic knowledge from multiple source machines by combining multiple partial distribution adaptation sub-networks (PDA-Subnets) and a multi-source diagnostic knowledge fusion module. The former weights target samples by counter-balancing factors to jointly adapt partial distributions of source and target pairs, and the latter releases negative effects due to discrepancy among multiple source machines and further fuses diagnostic decisions output from multiple PDA-Subnets. Two case studies demonstrate that MSTLN can reduce the misdiagnosis rate and obtain better transfer performance for imbalanced target samples than other conventional methods. [ABSTRACT FROM AUTHOR]

• Multiple impacts of wind conditions and resource constraints are integrated into operations & maintenance of wind farms. • A wind-centered maintenance policy is designed to reduce production losses and operational costs. • A two-level resource scheduling problem is solved by combining the MULTIFIT algorithm and a heuristic algorithm. • The comprehensive effect of operation costs and power generations is captured using the performance-based contracting. • The applicability is demonstrated via a case study from an onshore wind farm in Ontario. Operations & maintenance management is a cost effective approach to ensuring reliable and safe way to control and maintain wind farms. An essential objective is to improve the revenue level by balancing the increase of power generation and the reduction of operational costs. Despite prominent research advancements in wind farm operations & maintenance, no attempts have successfully integrated diverse impacts of wind conditions and resource constraints into the decision-making process. This paper develops a new framework by incorporating: (1) both positive (maintenance opportunity) and negative (maintenance delay) impacts of wind conditions, and (2) production losses during downtime due to resource constraints. A wind-centered maintenance policy integrating multiple maintenance opportunities is designed. Furthermore, a two-level resource scheduling problem with constraints in both repairmen and maintenance teams is solved using a combination of the Multifit algorithm and a heuristic algorithm. The net revenue is evaluated using the performance-contracting and optimized via the artificial bee colony algorithm. The applicability of the framework is demonstrated by a case study from an onshore wind farm in Ontario. Experimental findings illustrate that: (1) wind-centered opportunistic maintenance can effectively improve system revenue, and (2) the proposed framework shows a better performance when facing resource shortages or maintenance delays. [ABSTRACT FROM AUTHOR]