Papers by Keyword: Artificial Neural Network (ANN)

Paper TitlePage

Authors: Yong Li, Yun Yi Zhang, Ren Jie Gao, Shuai Tao Xie
Abstract: Jixi mine area is one of the early mined areas in China and it's a typical deep mine. Because of large deformation of underground roadway and dynamic disasters occurred frequently in this mine, five measurement points of in-situ stress in this mine was measured and then analyzed with inversion. Based on these in-situ stress measurement data, numerical model of 3D in-situ stress back analysis was established. According to different stress fields, related analytical samples of neural network were given with FLAC program. Through the determination of hidden layers, hidden nodes and the setting of parameters, the network was optimized and trained. Then according to field measurement of in-situ stress, back analysis of initial stress field was conducted. Compared with field measurement, with accuracy requirement satisfied, it shows that the in-situ stress of rock mass obtained is basically reasonable. Meanwhile, it proves that the measurement of in-situ stress can provide deep mines with effective and rapid means, and also provide reliable data to optimization of deep roadway layout and supporting design.
Authors: Zhi Biao Li
Abstract: In this paper, artificial neural network architecture is introduced to predict the Yin-Yang index of body constitution in traditional Chinese medicine (BCTCM). With pre-processing the inputting data by the median, the collected data is more consistent with the exact value of the characteristic parameters of BCTCM. Quasi-Newton algorithm is used to train the network model to accelerate the convergence speed of network training. Experiments show that, the result showed that they had good prediction accuracies for BCTTCM. The mean absolute error for 10 true measured points was 0.034. Therefore, the prediction model of BCTCM Yin-Yang index with BP neural network is doable.
Authors: Peng Fei Bao, Wei Dong Miao, Rong Xie, Yan Jun Shi
Abstract: Engineering analysis and simulation are time-consuming, and often trapped to computational burden, such as analyzing forging press. We herein employ surrogate modeling to reduce such computation cost while keeping high precision. This paper use a BP neural networks to building the surrogate model (BPNN-SM for short), and predicting the analysis results of mechanical structures with this model. The predicting process include confining design variables, sampling, building finite element model with business software ANSYS, constructing surrogate model to replace the original model and finally predicting data with the new model. In such process, we build a back-propagation neural network, and train it with sampling data from ANSYS results. We tested our methods with a mechanical structure design of hydraulic forging press. The experimental results verified the surrogate modeling.
Authors: Ge Yang, Si Lu Xie, Jing Huang
Abstract: A way to recognize printed characters based on BP network was proposed in this paper. It was implemented with C language. After a lot of experiments, the experimental results show that the character recognizer has good validity and correctness. The printed characters can be successfully recognized within the reasonable range of error rate.
Authors: Gang Li, Xing San Qian, Chun Ming Ye, Lin Zhao
Abstract: This paper focuses mainly on a clustering method for pruning Fully Connected Backpropagation Neural Network (FCBP). The initial neural network is fully connected, after training with sample data, a clustering method is employed to cluster weights between input to hidden layer and from hidden to output layer, and connections that are relatively unnecessary are deleted, thus the initial network becomes a PCBP (Partially Connected Backpropagation) Neural Network. PCBP can be used in prediction or data mining more efficiently than FCBP. At the end of this paper, An experiment is conducted to illustrate the effects of PCBP using the submersible pump repair data set.
Authors: Ke Wen Liu, Jing Wang
Abstract: For analyzing the accuracy of wind power prediction, an analyzing model combined with multi-leaner and dynamic weight distribution is proposed. With this method, Numerical Weather Prediction (NWP), Wind power data (historical) and weather data (historical) are structured into several sample sets, each set has a different weight value, which determined by the training errors, these sample set is trained by different learner algorithm with a weight too. Finally, using these models to predict the outputs. The experiments indicate the effectiveness of the method this paper proposed. Compared with Single model of Support Vector Machine and Artificial Neural Network, the combination method has better performance in both calculation accuracy and generalization.
Authors: Hai Yang Kong, Lan Xiang Sun, Jing Tao Hu, Yong Xin, Zhi Bo Cong
Abstract: Spectra of 27 steel samples were acquired by Laser-Induced Breakdown Spectroscopy (LIBS) for steel classification. Two methods were used to reduce dimensions: the first is to select characteristic lines of elements contained in the samples manually and the second is to do principal component analysis (PCA) of original spectra. Then the data after reducing dimensions was used as the input of artificial neural networks (ANN) to classify steel samples. The results show that, the better result can be achieved by selecting peak lines manually, but this solution needs much priori knowledge and wastes much time. The principal components (PCs) of original spectra were utilized as the input of artificial neural networks can also attain a good result nevertheless and this method can be developed into an automatic solution without any priori knowledge.
Authors: Yan Lou, Luo Xing Li
Abstract: Artificial neural network (ANN) and inverse method were employed in modeling the rheological behavior of the AZ80 magnesium. The hot deformation behavior of extruded AZ80 magnesium was investigated by compression tests in the temperature 350-450 and strain rate range 0.01-50 s-1. Investigation of flow stress curves and microstructure of the compression specimen illustrate occurrence of dynamic recrystallization. The inverse method of non-liner regression was used to determine the parameters of the suggested constitutive equation. The maximum relative errors at different temperatures and different strain rates between experimental and predicted flow stresses by ANN and inverse method were compared. The results show the ANN derives statistical models have better similar prediction ability to those of inverse method, especially at high strain rate. This indicates that ANN can be used as an alternative modeling tool for high temperature rheological behavior studies.
Authors: Fernando Parra dos Anjos Lima, Fábio Roberto Chavarette, Simone Silva Frutuoso de Souza, Adriano dos Santos e Souza, Mara Lúcia Martins Lopes
Abstract: This article presents the application and comparison of two techniques for intelligent computing to perform the analysis of the structural integrity of an aircraft structure. In this context, a ARTMAP-Fuzzy neural network and immunological negative selection algorithm are used in the identification and characterization of structural failure. The main application of these methodologies is to assist in the inspection of aircraft structures aiming at detecting and characterize flaws and decision making. To evaluate the methodology was performed modeling and simulation of signals from a numerical model using an aluminum beam. We performed a comparative analysis of methodologies, proving the efficiency of intelligent methods in the analysis of structural integrity. The results obtained by the method show efficiency, robustness and accuracy. To Evaluate the methodology was Performed modeling and simulation of signals from the numerical model using an aluminum beam. We performed a comparative analysis of methodologies, proving the efficiency of intelligent methods in the analysis of structural integrity. The results Obtained by the methods show efficiency, robustness and accuracy.
Authors: A. El Ouafi, R. Belanger, M. Guillot
Abstract: On-line quality assessment becomes one of the most critical requirements for improving the efficiency of automatic resistance spot welding (RSW) processes. Accurate and efficient model to perform non-destructive quality estimation is an essential part of the assessment. Besides the usual welding parameters, various measured variables have been considered for quality estimation in RSW. Among these variables, dynamic resistance gives a relative clear picture of the welding nugget formation and presents a significant correlation withseveral RSW quality indicators. This paper presents a structuredand comprehensiveapproach developed to design an effective dynamic resistancebased model for on-line quality estimation in RSW. The proposed approach examines welding parameters and conditions known to have an influence on weld quality, and builds a quality estimation model step by step. The modeling procedure begins by examining, through a structured experimental design, the relationships between welding parameters, typical characteristics of the dynamic resistance curves and multiple welding quality indicators. Using these results and various statistical tools, different integrated quality estimation models combining an assortment of dynamic resistance attributes are developed and evaluated. The results demonstrate that the proposed approach can lead to a consistentmodel able to accurately and reliably provide an appropriate estimationof the weld quality under variable welding conditions.
Showing 1 to 10 of 643 Paper Titles