Reducing the Percentage of Defective Castings Using the Neural Network Modeling Method

The article is devoted to the current problem of reducing the percentage of defective products manufactured at serial steelmaking plants. To study the patterns of defect formation, a neural network was created that predicts the formation of defects such as “Cracks” in castings produced by the open-hearth method. To train the neural network, statistical data taken under the conditions of the current production process was used, which resulted in a high degree of information noise. Outliers of statistical information were detected and removed with the original author's neural network technique. By conducting virtual computer experiments on a neural network model, some important dependences of the probability of the defect formation on the content of manganese, phosphorus, silicon, chromium, and sulfur were identified. Based on that, practical recommendations have been developed to reduce defects by adjusting the chemical composition of the metal. Despite the relatively low accuracy of the created neural network model, the application of the practical recommendations made it possible to reduce the percentage of defective products manufactured at a serial plant by half.