Machine Learning Approaches for Thermoelectric Materials Research

Thermoelectric (TE) materials provide a solid‐state solution in waste heat recovery and refrigeration. During the past few decades, considerable effort has been devoted towards improving the performance of TE materials, which requires the optimization of multiple interrelated properties. A fundamental understanding of the interaction processes between the various energy carriers, such as electrons and phonons, is critical for advances in the development of TE materials. However, this understanding remains challenging primarily due to the inaccessibility of time scales using standard atomistic simulations. Machine learning methods, well known for their data‐analysis capability, have been successfully applied in research on TE materials in recent years. Here, an overview of the machine learning methods used in thermoelectric studies is provided, with the role that each machine learning method plays being systematically discussed. Furthermore, to date, the scale of thermoelectric‐related databases is much smaller than those in other fields, such as e‐commerce, image identification, and speech recognition. To overcome this limitation, possible strategies to utilize small databases in promoting materials science are also discussed. Finally, a brief conclusion and outlook are presented.     

Structural changes in epoxy resin polymer after heating and their influence on space charges

Infrared spectroscopy was used to characterize changes in physicochemical properties of an epoxy‐amine resin as a function of temperature. The thermal degradation of the system was studied by means of thermogravimetric analysis. The effect of heat on evolution of space charges was investigated by the thermal step method and the thermally stimulated depolarization currents method. It was found that heat induced ring opening of epoxide groups and promoted formation of carbonyl groups. Space charge density was found to increase with increasing temperature while dipolar relaxation decreased. Copyright © 2003 Society of Chemical Industry     

A comparative study on the identification of the dynamical model of multi-mass electrical drives using wavelet transforms

This paper deals with dynamical modelling and modal identification of a special class of electromechanical systems: the multi-mass electrical drives. The aim of the work is to investigate the usefulness and the performance of the continuous wavelet transform (CWT) for the estimation of modal parameters, such as damping ratios and natural frequencies of these electromechanical systems. For this investigation, two models of elastic drive systems are considered: the two-mass and the three-mass models. The modal analysis identification using CWT is compared to the Hilbert–Huang transform technique and evaluated in terms of accuracy and ability to estimate modal parameters with different inputs, aptitude to distinguish several modes especially in the case of closely spaced modes and resistance to noisy conditions. In view of the results of the study, the advantages and shortcomings of each technique are presented and discussed.     

Study of relaxations in epoxy polymer by thermally stimulated depolarization current (TSDC) and dielectric relaxation spectroscopy (DRS)

Thermally stimulated depolarization current (TSDC) and dielectric relaxation spectroscopy (DRS) techniques were employed to study the relaxations and the conductivity phenomena in epoxy-based polymer. In addition to the primary α relaxation process associated with the glass–rubber transition, significant interfacial relaxation and ionic conduction process have been revealed. The ac conductivity is temperature and frequency dependent and shows a dc plateau at low frequencies. Above the glass transition temperature, dc conductivity is described by a Vogel–Tamman–Fulcher–Hesse (VFTH) equation while it shows Arrhenius behaviour at higher temperatures.     

基于三流DBN模型的听视觉情感识别

为更好地对听视觉情感信息之间的关联关系进行建模,提出一种三流混合动态贝叶斯网络情感识别模型(T_AsyDBN)。采用MFCC特征及基于基频和短时能量的局域韵律特征作为听觉输入流,在状态层同步。将面部几何特征和面部动作参数特征作为视觉输入流,与听觉输入流在状态层异步。实验结果表明,该模型优于有状态异步约束的听视觉双流DBN模型,6种情感的平均识别率从 52.14%提高到63.71%。     

基于三音素动态贝叶斯网络模型的大词汇量连续语音识别

考虑连续语音中的协同发音现象,基于词-音素结构的DBN(WP-DBN)模型和词-音素-状态结构的DBN(WPS-DBN)模型,引入上下文相关的三音素单元,提出两个新颖的单流DBN模型:基于词-三音素结构的DBN(WT-DBN)模型和基于词-三音素-状态的DBN(WTS-DBN)模型.WTS-DBN模型是三音素模型,识别基元为三音素,以显式的方式模拟了基于三音素状态捆绑的隐马尔可夫模型(HMM).大词汇量语音识别实验结果表明:在纯净语音环境下,WTS-DBN模型的识别率比HMM,WT-DBN,WP-DBN和WPS-DBN模型的识别率分别提高了20.53%,40.77%,42.72%和7.52%.     

Application of fuzzy TOPSIS in evaluating sustainable transportation systems

Sustainable transportation systems are the need of modern times. There has been an unexpected growth in the number of transportation activities over years and the trend is expected to continue in the coming years. This has obviously associated environmental costs like air pollution, noise, etc. which is degrading the quality of life in modern cities. To cope us this crisis, municipal administrations are investing in sustainable transportation systems that are not only efficient, robust and economical but also friendly towards environment. The challenge before the transportation decision makers is how to evaluate and select such sustainable transportation systems. In this paper, we present a multicriteria decision making approach for selecting sustainability transportation systems under partial or incomplete information (uncertainty). The proposed approach comprises of three steps. In step 1, we identify the criteria for sustainability assessment of transportation. In step 2, experts provide linguistic ratings to the potential alternatives against the selected criteria. Fuzzy TOPSIS is used to generate aggregate scores for sustainability assessment and selection of best alternative. In step 3, sensitivity analysis is performed to determine the influence of criteria weights on the decision making process. A numerical illustration is provided to demonstrate the applicability of the approach.The strength of the proposed work is its practical applicability and the ability to generate good quality solutions under uncertainty.     

基于改进异步DBN模型的听视觉融合情感识别

提出了一个改进的三特征流听视觉融合异步动态贝叶斯网络情感模型（VVA_AsyDBN）,采用面部几何特征（GF）和面部主动外观模型特征（AAM）作为两个视觉输入流,语音Mel倒谱特征（MFCC）作为听觉输入流,且视觉流的状态和听觉流的状态可以存在有约束的异步。在eNTERFACE’05听视觉情感数据库上进行了情感识别实验,并与传统的多流同步隐马尔可夫模型（MSHMM）,以及具有两个听觉特征流（语音MFCC和局域韵律特征LP）和一个视觉特征流的听视觉异步DBN模型（T_AsyDBN）进行了比较。实验结果表明,VVA_AsyDBN获得了最高识别率75.61%,比视觉单流HMM提高了12.50%,比采用AAM、GF和MFCC特征的MSHMM提高了2.32%,比T_AsyDBN的最高识别率也提高了1.65%。     

Unsupervised learning of arbitrarily shaped clusters using ensembles of Gaussian models

We propose a new clustering algorithm, called SyMP, which is based on synchronization of pulse-coupled oscillators. SyMP represents each data point by an Integrate-and-Fire oscillator and uses the relative similarity between the points to model the interaction between the oscillators. SyMP is robust to noise and outliers, determines the number of clusters in an unsupervised manner, and identifies clusters of arbitrary shapes. The robustness of SyMP is an intrinsic property of the synchronization mechanism. To determine the optimum number of clusters, SyMP uses a dynamic and cluster dependent resolution parameter. To identify clusters of various shapes, SyMP models each cluster by an ensemble of Gaussian components. SyMP does not require the specification of the number of components for each cluster. This number is automatically determined using a dynamic intra-cluster resolution parameter. Clusters with simple shapes would be modeled by few components while clusters with more complex shapes would require a larger number of components. The proposed clustering approach is empirically evaluated with several synthetic data sets, and its performance is compared with GK and CURE. To illustrate the performance of SyMP on real and high-dimensional data sets, we use it to categorize two image databases.     

Flexibility-based multi-objective approach for machines selection in reconfigurable manufacturing system (RMS) design under unavailability constraints

The reconfigurable manufacturing system (RMS) is a recent manufacturing paradigm driven by the high responsiveness and performance efficiencies. In such system, machines, material handling units or machines components can be added, modified, removed or interchanged as needed. Hence, the design of RMS is based on reconfigurable machines capabilities and product specification. This paper addresses the problem of machines selections for RMS design under unavailability constraints and aims to develop an approach to ensure the best process plan according to the customised flexibility required to produce all parts of a given product. More specifically, we develop a flexibility-based multi-objective approach using an adapted version of the well-known non-dominated sorting genetic algorithm to select adequate machines from a set of candidate (potential) ones, in order to ensure the best responsiveness of the designed system in case of unavailability of one of the selected machines. The responsiveness is based on the flexibility of the designed system and a generated process plan, which guarantees the management of machines unavailability. It is defined as the ability and the capacity to adapt the process plan in response to machines unavailability. Two objectives are considered, respectively, the maximisation of the flexibility index of the system and the minimisation of the total completion time. To choose the best solution in the Pareto front, a multi-objective decision-making method called technique for order of preference by similarity to ideal solution is used. To demonstrate the applicability of the proposed approach, a simple example is presented and the numerical results are analysed.