Development of the method of early recognition of the slag of the steel ladle of the continuous casting machine

The article deals with the problem of early recognition of the slag of the steel casting ladle of a continuous casting machine. In this paper, the vibration method of slag recognition was investigated, since it is the most informative. A number of methods were tested for analyzing the vibration acceleration signal of the protective tube manipulator for timely cutting off of slag and preventing it from entering the intermediate bucket, such as averaging it using a moving average filter, entropy calculation, construction of the signal spectrum envelope, as well as the power spectrum envelope. The analysis of the results of the approbation showed that not all methods can be applied to solve this problem. The highest efficiency, equal to 93 percent, was provided by an approach based on the analysis of the power spectrum of the vibration acceleration signal. In addition, in this paper, a neural network method for detecting anomalies in the vibration acceleration signal using various autoencoder architectures is considered and tested. This approach was tested on both "synthetic data", where it confirmed its efficiency in detecting anomalies, and on real data, where an accuracy of 73 percent was achieved. Further research will be aimed at a more thorough elaboration of this method.