International Scientific Journals
of Scientific Technical Union of Mechanical Engineering "Industry 4.0"

The increasing application of NoSQL database technology and the neural networks raises the question of how compatible and applicable are the NoSQL databases to the neural network prediction models. This paper examines the applicability of a traditional relational database for storing air quality data and compares it to a NoSQL database performing the same functions. The possibility of the NoSQL database to feed a neural network model for predicting the atmospheric air quality is evaluated. The tendencies in the data are studied, and some solutions for improving the air quality are proposed. An analysis and a comparison of the performance of both relational SQL and NoSQL database systems by using real-world data for the Air Quality Index in the city of London is assessed and their performance is compared. Bivariate analysis on the data in order to assess the quality of the neural network forecast is performed.

A mathematical method of linear surrogate parametric synthesis of frame surface non-coaxial eddy current probes with a uniform eddy current density distribution in the testing object’s zone is proposed. The metamodel of a frame eddy current probe with a planar structure of the excitation system is constructed. Acceptable accuracy of the created metamodel is obtained by using the decomposition of the extremum search space and using associative neural networks. Examples of the synthesis of such excitation systems using modern metaheuristic stochastic algorithms for finding the global extremum are considered. The numerical results of the obtained solution and graphic illustrative material of the density distribution of the eddy currents on the surface in the testing object’s zone are given.

This paper examines the possibility to use neural networks for approximately solving the MiniSum problem, a classic facility location problem. For this we first create a set of realistic MiniSum instances, based on the Bulgarian road network. Two standard neural network approaches – Hopfield networks and Boltzmann machines, are then applied to the instances. Since the quality of solutions is not satisfactory, the reasons for the poor performance are discussed. An improved neural network approach is then proposed. This approach has excellent performance on the MiniSum instances. It always finds solutions just several percent worse than the optimum, and is often able to find the exact optimum.

A multiparameter metamodel of the eddy current probe with the volumetric excitation structure is constructed. As variable parameters of the metamodel, the spatial coordinates of the testing zone, the radii of the excitation coils and the height of their location above the testing object were used. Due to the use of hybrid construction of multiple neural networks using decomposition of the search space, an acceptable metamodel’s error of the eddy current probe with volumetric excitation structure is obtained.

This study aimed to predict the protein content(PC) and canopy spectra in winter wheat were measured based on field test. Key spectral bands were chosen by principal component analysis (PCA) method, and the predicted models were built by Partial Least Squares Regression (PLSR) and Artificial Neural Network (ANN). The performance of the feed forward and cascade forward ANNs was compared with those of PLS regression models using root mean square error (R) and the correlation coefficient (2). The finest consequence by CFBP was related to topology of 8-8-1 with Levenberg-Marquardt (LM) algorithm, threshold function of TANSIG-TANSIG-PURELIN and the initial strategy. This arrangement output was R=0.0289 and 2=0.9881 at 14 epochs. The consequences of estimate for correlation values of PLSR model was 0.9783. The results of prediction for the two models were in order of ANN > PLSR with correlation values of 0.9881 and 0.9783, respectively. Therefore, NIRS shows the potential for predicting protein content with accuracies suitable for process control.

The paper presents the application of modern computational approaches in determining the appropriate chemical composition of titanium alloys . Discusses three types of problems for the selection of the chemical composition of alloy at set requirements for mechanical properties . Tasks are solved in a different formulation of the target criteria . The proposed solutions are based on software model using approximation with artificial neural networks and genetic optimization algorithm.

This article offers a neural network architecture for automatic classification of Inverse Synthetic Aperture Radar objects represented in images with high level of optimization. A full explanation of the procedures of two-layer neural network architecture creating and training is described. The neural network is experimentally simulated in MATLAB environment.