Search

The rail fastener is an indispensable component used to connect the rail and sleepers in the track structure. Real‐time recognition of the fastener defects plays a vital role in ensuring the safe and stable operation of rail transit. In this paper, an intelligent and innovative method is proposed to detect the fastener defects including the invisible defects appearing as bolt loosening and the visible defects such as the worn or completely missing fasteners by using axle‐box vibration acceleration and deep learning network. First, the dynamical relation between the fastener defects and the axle‐box vibration acceleration is investigated by using the first principle and the vehicle–track dynamical model. Then a defects recognition network is built based on the deep convolution neural network for track fasteners by using the frequency spectrum images of the axle‐box vibration. The results show that the proposed method achieves a classification accuracy of 98.27%. Finally, the track section where the fasteners are most likely to be damaged is investigated, and rail corrugation is found to be a key factor that causes fastener fatigue. [ABSTRACT FROM AUTHOR]

It is an interesting topic to interpret artificial neural networks (ANNs) by considering some change various approaches. This paper explores the relationship between the input and output units of the simplest ANN, a single layer perceptron for the binary classification problem, from the probability point of view. If the feature variables of datasets follow independent normal distribution and outputs are activated by sigmoid function or smooth Relu function, we advocate that the probability density function (pdf) of the output variable is an exponential family distribution. Furthermore, by introducing an intermediate variable, the pdf of the output variable can be written as a linear combination of three normal distributions with same spread but different centers. Based on these results, the probability of the predicted class label can be written as a standard normal cumulative distribution function (cdf). The originality of this paper comes with interesting theoretical results to provide ANNs with a new description of the relationship between input variables to output variables, which can enable ANNs to be understood from a new perspective. Extensive experiments based on one artificial synthesized dataset and ten real‐world benchmark datasets validate the reasonability of those results. [ABSTRACT FROM AUTHOR]

The driver authority decision for driver‐initiated takeover is closely related to vehicle driving safety. In this paper, a lateral driver‐automation driver authority decision method considering the safety of the intended functionality is proposed to improve vehicle driving safety. Based on systems‐theoretic processes analysis, functional safety analysis of driver initiative takeover is carried out to clarify the safety of the intended functionality issues caused by the unreasonable decision. Then, a method for defining the safe driving area is developed to assess the safety level of the driving area around the ego vehicle. Taking the driver‐vehicle state parameters and vehicle–vehicle state parameters as input, a deep neural network model is constructed to determine the intention of the driver to take over the vehicle. The lateral driver‐automation driver authority decision method is designed to make the driver‐machine take‐over decisions, which considers the safety of the intended functionality. The effectiveness of the proposed method is evaluated via numerical simulation and hardware‐in‐the‐loop experiments. The results show that the designed method not only improves the control ability of the driver to the maximum extent but also integrates the perception ability of the driver into the vehicle control system to improve further vehicle driving safety. [ABSTRACT FROM AUTHOR]

Fault section diagnosis (FSD) is significant for the power system dispatching. Artificial neural network (ANN)‐based FSD method has strong fault tolerance but it looks like a black box and lacks the interpretability to the diagnosis outputs. In addition, when the topology of power systems changes, the ANN structure needs to be reconstructed and retrained, and thus has low adaptive capability. In order to tackle these challenges, in this paper, an ANN‐based FSD method by constructing universal transparent diagnosis models is proposed. The diagnosis models are constructed for the transmission line, transformer, and bus types rather than for a specific power system section. They can express the logical relations among sections, protective relays (PRs) and circuit breakers (CBs) clearly and intuitively. In addition, fuzzy values are used to model the uncertainties of PRs and CBs, and to determine the inputs of diagnosis models. Furthermore, the differential evolution algorithm is employed to optimize the network parameters of diagnosis models. The proposed method is verified on the IEEE 30‐bus test system and an actual local power system in Jilin Province of China. [ABSTRACT FROM AUTHOR]

Technology computer‐aided design (TCAD) simulation is an important tool for the development of semiconductor devices. Based on coupled partial differential equations (PDEs) for behaviors, TCAD can calculate objects such as impurities, point defects, and electronic carriers in semiconductors. However, over recent years semiconductor devices have become increasingly miniaturized and complicated, resulting in much longer calculation times for TCAD. Machine learning is one technology that may be used to overcome this simulation cost problem. In this study, a neural network architecture is proposed that considers the structure of the coupled PDEs. Features representing each concentration distribution of the calculation objects are extracted by convolution operations and their reaction is modeled by channel attention. The performance of the proposed architecture and of conventional neural network models is evaluated using a simulation dataset generated by 1D coupled PDEs that models the diffusion and reaction of vacancies and interstitial atoms. In addition, the advantage of the method is discussed through the analyses of error correlations of the two predictions and attention coefficients. The machine learning method developed in this study will be applicable to other physics described by coupled PDEs and is expected to speed up the computation of simulations in various fields. [ABSTRACT FROM AUTHOR]

Accompanied by the fast growth of data and computing power, Deep Learning has developed at a tremendous speed. Other related areas including material science, physics, chemistry, medical science, and engineering also benefit from this well‐proven powerful tool. Activation function is an essential part of Deep Neural Networks, which has attracted a lot of attention, but there are few researches based on nanophotonics problems. In this work, comprehensive research is made on the effectiveness of several widely used activation functions. This research shows that among those fancy activation functions that are investigated, Tanhshrink performs the best, which can predict the spectra with a Root Mean Square Error <0.005 for over 99% of these randomly generated instances. Other traditional activation functions like Tanh and Sigmoid also show excellent outcomes in competition with novel ones. This work shows that although deep learning has already been a powerful tool to solve physics problems, there is still a lot of fundamental work to be done to achieve its maximum potential. [ABSTRACT FROM AUTHOR]

The study of reliability analysis of mixture model is essential in confirming the quality of devices, equipment, and electronic tube flops etc. In recent years, statisticians have developed more interest in mixture model research, notably in the last decade, without taking into account the issue of modeling the metrics of reliability of mixture models using artificial neural networks. In the present study, the influence of pertinent parameters on reliability metrics is studied. The effect of components and mixing parameters for failure function, reversed hazard rate function, mean time to failure, hazard rate function, mean inactivity time, mean residual life, reliability function, Mills Ratio profiles are plotted and discussed. A multi‐layer artificial neural network is developed using the numerical analysis results obtained using four different scenarios. The values extracted from the artificial neural network and the numerical findings of the reliability studies are extensively compared and examined. The deviation rates obtained for the developed artificial neural network model are obtained at values lower than 0.12%. The outcomes demonstrate that neural networks are a powerful and effective mathematical tool that can be used in the reliability analysis of mixing models. [ABSTRACT FROM AUTHOR]

In this study, a quantitative structure‐activity relationship (QSAR) model of anticancer activity against myeloid cell leukemia 1 (Mcl‐1) for a series of 41 tricyclic indole diazepinone derivatives is established. Three different modeling methods, multiple linear regression (MLR), partial least square (PLS), and artificial neural network (ANN) are investigated to perform a QSAR model with significant predictiveness. A clustering method is also used for dividing all compounds into training and external test (ET) sets. Component principal analysis is used to eliminate the redundancy between descriptors. The accuracy and predictability of the proposed models are proven by comparing their key statistical terms. The good results obtained with the internal and external validations (EV) show that the proposed models can predict high‐performance activities and that the selected descriptors are pertinent. This model is also validated using internal validation (IV), mainly using cross‐validation (leave‐many‐out (LMOCV)). The applicability domain (AD) is identified. Based on the SAR map analysis, a novel Mcl‐1 inhibitor with a good predicted activity using the best model is proposed, the interaction of the designed compound with the binding site of Mcl‐1 protein is evaluated and its docking score is found high. [ABSTRACT FROM AUTHOR]

In this work, artificial neural networks (ANNs) are used to recognize nano‐objects solely from the absorbance spectrum of a macroscopic sample. For this, ANNs with two recognition schemes are constructed. The first one is designed to recognize ensembles of dielectric scatterers. The second ANN model recognizes the dimensions of gold nanospheres in a mixture and the refractive index of a matrix. A challenge in the first scheme arises at and near the invisibility point, i.e., when the refractive index of nanoparticles is close to that of the medium. Of course, particle recognition in this regime faces fundamental physical limitations. However, such recognition near the invisibility point is possible, and this study reveals its unique properties. Interestingly, the recognition process for the refractive index in the vicinity of the invisibility point shows very small errors. In contrast, the errors for recognition of the radius grow strongly near this point. Another regime with limited recognition occurs when the extinction spectra are not unique and can correspond to different realizations of nanoparticle mixtures. The recognition schemes proposed and investigated herein can find their applications in sensing. [ABSTRACT FROM AUTHOR]