Authors

Neural network model for determining the regulations parameters in the technological process of ore raw materials processing

Machine learning methods are currently widely used to solve various production problems, the problems of defects diagnosing and predicting for items in mass production, in particular. One of the most important problems is defects diagnosing and predicting, basing on its solution the regulations for the technological processes parameters and raw materials used can be determined, that insures the minimum probability of defects and the highest possible quality of manufactured products. The solution of this urgent problem with the help of a neural network model is shown on the example of the technological process for manufacturing products from fine ore material. The proposed model is based on the neural network trained on the set of historical data including examples of manufacturing products with different sets of technological parameters and raw ore material. The predicted parameter is warping of the product in one of its sections. Designing and training of the proposed neural network structure allowed achieving the coefficient of determination R2 between the predicted and actual warpage values of 92%. The dependences for the warpage value on the most significant parameters of the technological process, including thermophysical and chemical power technological processes of raw materials processing were constructed by conducting computer experiments using the method of partial freezing for input parameters. Due to these dependencies, the regulations for the most significant parameters of the production process are determined, which ensures the product to be without violating the tolerance for the warpage value specified by the design documentation. Thus, a specific example shows the possibility of using neural network modeling to solve the problem of setting regulations for the production process parameters, which compliance ensures the minimum amount of rejects and, accordingly, a higher quality of a production batch. Read more...

Neural network models of the influence of the chemical composition of ore raw materials and parameters of melting processes on the properties of products

Studies of the influence of the chemical composition of ore raw materials on the properties of foundry products, as a rule, do not consider the features of production processes. The main reasons for this are that, firstly, the results of such research are usually limited to laboratory studies, secondly, the models do not take into account the specifics of the structure of ore raw materials, since these data can be obtained only after the completion of all thermophysical and chemical-energy-technological processes of heat treatment, thirdly, the use of models is difficult due to the need to promptly account for changes in the composition and values of systemic and external factors during heat treatment. The purpose of the paper is to design neural network models and tools that provide the possibility of adaptive structural and parametric adjustment to changes in the parameters of the analyzed processes of heat treatment of ore raw materials. Data on the chemical composition of ore raw materials and information on heat treatment processes are used as parameters of the projected neural network models. As a result of the research, the hypothesis about the possibility of indirect accounting for the influence of structural features of ore raw materials on the quality of products has been confirmed. Confirmation of this hypothesis will allow us to offer effective tools for operational management of thermophysical and chemical-energy-technological processes of heat treatment of ore raw materials. Read more...

Neural network forecasting of defects in the production of metallurgical products

The paper is devoted to the current problem of reducing the percentage of defective products produced at serial steelmaking plants. To study the patterns of defect formation, a neural network was created that predicts the formation of defects such as “crack” in castings produced by the open-hearth method. To train the neural network, statistical data on the chemical composition of ore raw materials and the corresponding values of the percentage of defects were used. The data was taken under the conditions of an ongoing serial production process, which led to a high degree of noise in the information both on the chemical composition of the ore raw material and on the mechanical properties based on the results of its heat treatment. Outliers of statistical information were detected and removed using the original author’s neural network technique. A neural network model for controlling thermophysical and chemical-energy-technological processes of thermal processing of ore raw materials was created on the basis of a perceptron-type neural network with sigmoid activation functions. By conducting virtual computer experiments on a neural network model, some important dependences of the probability of formation of the defects under study on the content of manganese, phosphorus, silicon, chromium and sulfur were identified. Based on the identified dependencies, practical recommendations have been developed to reduce the percentage of defects by adjusting the chemical composition of ore raw materials. Despite the relatively low accuracy of the developed neural network model, the application of the practical recommendations obtained made it possible to reduce the percentage of defective products manufactured in a large-scale production process by 2.51 times. Read more...

A neural network algorithm for identifying and removing outliers in noisy data sets

Outliers in statistical data, which are the result of erroneously collected information, are often an obstacle to the successful application of machine learning methods in many subject areas. The presence of outliers in training data sets reduces the accuracy of machine learning models, and in some cases, makes the application of these methods impossible. Currently existing outlier detection methods are unreliable. They are fundamentally unable to detect some types of outliers, while observations that are not outliers are often classified as outliers by these methods. Recently emerging neural network methods for outlier detection are free from this drawback, but they are not universal, since the ability of neural networks to detect outliers depends both on the architecture of the neural network itself and on the problem being solved. The purpose of this study is to develop an algorithm for creating and using neural networks that can correctly detect outliers regardless of the problem being solved. This goal is achieved by using the property of some specially created neural networks to demonstrate the largest training errors on those observations that are outliers. The use of this property, as well as the implementation of a series of computational experiments and the generalization of their results using a mathematical formula, which is a modification of the consequence of the Arnold – Kolmogorov – Hecht-Nielsen theorem, made it possible to achieve the stated goal. The use of the developed algorithm turned out to be especially effective in solving the problems of forecasting and controlling interdependent thermophysical and chemical-energy-technological processes of processing ore raw materials, occurring at existing serial metallurgical enterprises, where the presence of outliers in statistical data is almost inevitable, and without their identification and exclusion, the construction of neural network systems that are acceptable in accuracy models are generally impossible. Read more...