The CING system is a software product that provides efficient processing of enterprise data and allows you to increase the efficiency of rolling process management. Through the use of modern heuristic algorithms, CING is able to provide the calculation, prediction and adjustment of virtually any parameter of the hot-rolled mill, based on aggregate data on the technological process at the appropriate time.

Basic CING Applications:

• Reserving the functions of measuring instruments - transferring the calculated data to the automated control system of the enterprise in the event of a failure of the measuring device
• Predicting the strip thickness prior to rolling - informing the operator about the calculated result of the mill operation under specified conditions
• Correction of mill presets - calculation and provision of information about the optimal operation mode of the mill to achieve the specified value of the strip thickness.

CING has a modular structure, due to which it is possible to further expand and connect various auxiliary functions, for example: calculation of water consumption, mechanical properties of rolled products, prediction of roll wear, strip defects and strip temperature at the exit of the furnace. Using the CING system in conjunction with the hot rolled metal gauge will allow you to achieve increased speed, fault tolerance and reliability of the technological section in case of failure of some measuring devices, as well as to optimize the control accuracy indicators.

Modern metallurgical enterprises, as a rule, have developed systems for collecting data on the progress of the technological process and many control and management systems. When collecting data, a large number of parameters are recorded at a high frequency — hundreds of signals from various sensors and systems can be stored every second. This approach allows you to accumulate huge amounts of information.

A problem common to all modern enterprises is the low utilization of accumulated information: usually only a small part of the data collected is directly used to control the process flow. Thus, there is a low level of usability of information.

Since even the simplest statistical processing “manually” of such huge amounts of data causes difficulties, the problems of mathematical and probabilistic analysis in this case are a pressing problem. The arising need to process large amounts of accumulated information arose not only in metallurgy, but also in many other areas of industrial and scientific activity, which led to the active development of new mathematical methods, called Data Mining.

Such methods of in-depth data analysis as machine learning have found widespread use, including in metallurgy. This mathematical discipline uses sections of mathematical statistics, numerical methods, probabilistic methods, etc. to extract knowledge from data arrays.

The goal of machine learning is to automate the solution of complex problems of forecasting, management and decision-making. Machine learning is successfully used in various fields: from image recognition and speech to technical diagnostics, from document categorization to forecasting the progress of various processes.

In the metallurgical industry, artificial neural networks are widely used as one of the methods of machine learning. An artificial neural network (hereinafter referred to as a neural network or a neural network) is a system of simple elements interconnected (nodes, neurons). Each of these elements (neurons) is extremely simple: it receives and processes input signals (usually using threshold functions) and sends a signal to the following neurons. But combining a large number of similar neurons into one network allows you to receive complex structures that perform non-trivial tasks: the input signal activates the first neuron, it sends a signal to its neighbors, and such an activation wave propagates through the network, transforming in accordance with the task. Neural networks are used for pattern recognition, data classification and clustering, approximation and regression, prediction and optimization, and also as control or correction algorithms.

Neural networks are not programmed in the usual sense of the word, i.e. they are not based on rigid formulas and dependencies. Neural networks are trained on large volumes of data, finding connections and patterns in them. There is a huge number of types of neural networks of various types of neuron location, method of learning, nature of connections, time of signal transmission, and more. etc. The choice of a suitable type of network and its configuration (the number of neurons, the ways of combining them into layers, the types of threshold functions used, etc.) is a complex engineering task, especially since there are no hard criteria or recommendations for such a choice. With the right choice and skillful setup, neural networks are able to achieve high accuracy (for example, image recognition accuracy in some neural networks is comparable and even exceeds the capabilities of human vision - as in the case of recognizing handwritten numbers, where neural networks are able to achieve accuracy of 99.8% with a precision of human vision 98%) .

The advantages of neural networks are:

• Learning ability - i.e. independently find dependencies in the data;
• Ability to build non-linear dependencies, ability to adapt to changes in the course of the technological process (the network can be retrained when data changes or even create a network that adapts in real time);
• High speed due to parallel structure;
• Upon successful learning, the neural network is able to give correct results based on data that were missing from the training sample or were partially distorted or “noisy”.

In recent years, neural networks have been actively used in hot rolling mills to solve a variety of problems:

• Determination of slab temperature at furnace exit;
• Determination of rolling force to obtain the required thickness of rolled;
• Determination of the required amount of water to cool the strip;
• Determination of the optimal strip width at the exit from the roughing stand group to obtain the required exit width from the finishing group;
• Rolling force control and clearance value to compensate for overshoot from PID elements.

Neural networks are used to solve the above problems due to the significant number of factors affecting the desired performance. Thus, there is no possibility of creating a "hard algorithm" for calculating these quantities.

Brief description of the CING system, you can find out by following the link