Control system of the hottest intelligent ladle re

  • Detail

Intelligent ladle refining furnace control system in recent years, electric furnace/converter secondary refining continuous casting short process, as a trend, occupies a more and more important position in steelmaking plants. Ladle refining furnace (LF) has made great progress because of its characteristics of low investment and strong function. It provides both opportunities and challenges for LF automatic control technology

domestic lf automation technology has a low starting point and weak function. Generally, it only uses PLC for logic interlocking control and PID loop control, and uses industrial computer (IPC) to monitor equipment operation. Its technical and economic indicators are low, which is far from meeting the needs of the market. Important process variables such as arc power, arc length and refractory index cannot be directly detected. The consumption measurement of actual molten steel temperature, sampling analysis of molten steel composition and other factors restrict the application of process optimization technology. Only the operator's personal experience can be used to set the transformer voltage and secondary current. The multivariable nonlinear time-varying control object greatly reduces the control effect of the traditional PID electrode regulator

in order to improve the technical level of LF automation in China, achieve the effects of energy saving, consumption reduction and productivity improvement, we have developed an intelligent lf control system (ILF), and its principle is shown in Figure 1. Based on the compound artificial intelligence technology, the system performs heat balance calculation, molten steel temperature prediction, energy input dynamic optimization and intelligent electrode lifting control. It overcomes the shortcomings of the traditional lf control system and achieves satisfactory application results

Figure 1 ILF schematic diagram

1 ILF structure and function

ilf adopts IPC hardware structure, which is mainly composed of two IPC (regulator ipc1 and server ipc2). Two IPC are connected via Ethernet. Ipc1 is interpolated with analog input and output templates and digital input and output templates for signal input and output

two sets of IPC are respectively equipped with a set of LCD, mouse and keyboard. One set is placed on the console in the main control room for steelmaking workers, and the other set is used in the control cabinet in the electrical room for system commissioning and maintenance. The system is also equipped with necessary electric quantity transmitter, signal conditioning module and DC power supply, which are centrally placed in the control cabinet in the electrical room. In addition, the control cabinet is equipped with a set of on-site signal simulation device and power display meter for commissioning and equipment maintenance

ilf includes the following main functions: (1) data i/o; (2) Heat balance calculation and molten steel temperature prediction; (3) Dynamic optimization of energy input set point; (4) Intelligent electrode lifting control; (5) Database management and statistical process analysis; (6)MMI; (7) Network communication

2 heat balance calculation and molten steel temperature prediction

the main function is to calculate the balance relationship between the initial temperature of molten steel, energy input, energy loss and useful energy, and to predict the change of molten steel temperature in the whole smelting process. There are usually two methods: mechanism analysis and statistical calculation

the mechanism analysis method is mainly to establish the mechanism model. According to the smelting stage, the stress that the ladle, argon blowing and stirring, cooling water, soot, arc, charging and steel rubber limb samples can bear under constant elongation is becoming smaller and smaller, and the energy balance relationship of each water unit is further calculated. The advantage of this method is that the physical meaning is clear, and the calculation results have guiding significance for improving the operation system and smelting process. However, this mechanism analysis requires a large number of assumptions, and a large number of process data need to be provided on site. These data are not available in the actual production, which restricts the operation of the mechanism model

statistical calculation method is based on statistical analysis methods, such as linear regression analysis, to find the relationship between the forecast and various process variables through a large number of data. Its advantage is that the algorithm is simple and easy to implement; However, because these models can only reflect the linear relationship, and the relationship between molten steel temperature and electricity, alloy, time and other factors is complex, the accuracy of statistical models is usually not high

the system adopts the dynamic prediction model of molten steel temperature based on the organic combination of artificial neural network and expert system, as shown in Fig. 2 and Fig. 3

Figure 2 network topology structure

Figure 3 logical structure of molten steel temperature prediction model

first, the initial value of molten steel temperature prediction is obtained by using artificial neural network. The model and control problem in ILF can be reduced to a pattern recognition problem of time series. From the theoretical understanding, or from the practical consideration of calculation efficiency and storage space, the problem of time series pattern recognition can be solved. It is better to use the recurrent neural network with memory function, such as Elman or avalanche. However, it is difficult to train this kind of neural network and its performance is not stable enough. Under the current conditions, it is difficult to achieve the reliability required by practical engineering applications. After a lot of investigation and research, it is found that most of the artificial neural networks that have been applied in engineering abroad are still multilayer feedforward networks based on error back propagation, i.e. BP. This network is easy to master, and has been tested by a lot of practice, and its reliability is relatively high. BP is originally only applicable to static pattern recognition. In order to apply it to dynamic time series pattern recognition, the time delay network technology (TDNN) can be used, that is, the timing signal is input to the delay register. The shift register actually forms a FIFO queue. The length of the queue determines the width of a moving window intercepted from the timing signal. Every time a new signal is received, Together with the signals of the first n-1 sampling points stored in the shift register, Goldilocks can be called a static mode in the mobile window composed of the most complete variety of new material enterprises in the world, and then the artificial neural network is used to identify the static mode. In order to improve the operation speed of the algorithm, we use the generalized delta algorithm. The network topology is shown in Figure 2. Network training is carried out through the collected data, and the trained network is used to calculate the initial value of molten steel temperature

due to the complexity of smelting process, new and old ladle, molten steel baking degree, hot stop time and other factors have a certain impact on molten steel temperature. According to mechanism analysis and on-site experience, expert rules in the form of if-then are generated. During on-site application, forward reasoning is carried out according to the actual situation of each furnace, and the corresponding expert rules are called to correct the initial value of temperature prediction, so as to give an accurate prediction value of molten steel temperature

molten steel temperature shall be calculated once every minute from the first molten steel temperature measurement of each furnace. The dynamic prediction method of molten steel temperature, which combines artificial neural network and expert system, overcomes the shortcomings of mechanism analysis and statistical calculation at the same time. It has strong adaptability and high prediction accuracy

3 dynamic optimization of energy input set point

the dynamic optimization function of energy input set point refers to selecting reasonable decision variables (arc voltage and arc flow) to make the arc power input into the ladle furnace meet the process requirements under certain constraints

in the past, the power set point was only based on the static electrical circle diagram to formulate the power curve under various voltage levels, which was selected by the operator according to his own experience. This static analysis assumes that the resistance and reactance of the electrical circuit are unchanged and the three phases are independent. Due to the complexity of the smelting process, there are many random interference factors, and the coupling between the three-phase electrical variables is serious. In fact, this assumption is not tenable. In recent years, foreign countries have made some progress in using artificial intelligence technology to optimize the power set point. Generally speaking, there are two schemes: expert system and artificial neural network. The expert system scheme adopts the experience-based method, stores the power set point experience summarized from production practice in the form of rules of the expert system in the computer, and makes logical reasoning according to smelting objectives, smelting performance, detected electrical variables and other factors to obtain a more reasonable power set point. However, because these processes are more advanced, the expert system is only based on the detected or predicted current External factors, such as voltage value, fail to make full use of the dynamic changes of electrical characteristic parameters, such as resistance and reactance, which reflect the internal law of the electrical system, so the accurate optimal set point cannot be obtained. The artificial neural network scheme uses the artificial neural network to establish the corresponding function relationship between power and impedance, and optimizes the impedance value at the maximum power by calculating the partial derivative of the impedance, which is used as the set point of electrode lifting control. The optimization goal of this algorithm is to find the maximum secondary power, which is more suitable for electric arc furnace as primary smelting furnace. For LF, maximum power does not mean optimal power. It is also necessary to consider the requirements of production rhythm, molten steel temperature and smelting process, and seek the optimal power set point under these nonlinear and time-varying constraints

the system considers the following constraints:

(1) the apparent power is less than the allowable capacity of the transformer

(2) the working current shall not exceed the allowable current of the transformer

(3) arc length control

(4) high power consumption efficiency and thermal efficiency

(5) appropriate refractory index

(6) three phase power balance

(7) requirements of smelting process and production rhythm on temperature rise

ilf adopts the energy input set point dynamic optimization scheme organically combining industrial neural network and expert system in his keynote speech "industrial Internet of things helps realize smart factory". Firstly, the artificial neural network model is established to dynamically calculate the electrical parameters related to LF, such as resistance, reactance, etc. then, based on these parameters, the process parameters at different set points, such as secondary active power, arc power, arc length, arc voltage and refractory index, are calculated. Then, the expert system rules are invoked. According to the different characteristics of each stage in the smelting process, the smelting power curve is divided into three stages: slagging, composition adjustment and temperature rise before tapping. At different stages, the arc voltage and arc current set points are adjusted according to the temperature rise requirements of the predicted molten steel temperature, electrical characteristic curve, smelting technology and production rhythm, so as to realize the optimization of electric energy input. The algorithm flow chart is shown in Figure 4

Figure 4 power set point optimization flow chart

4 neuron network electrode lifting control

the purpose of electrode lifting control is to adjust the distance between the electrode end and the molten steel level in the ladle furnace through the proportional valve or servo valve of the hydraulic station to ensure that the state variable of electric quantity in the smelting process tracks the optimized input power set point. The traditional electrode lifting control is PID control based on impedance control, that is, it controls the electrode lifting according to the feedback information of current and voltage, and keeps the ratio between voltage and current to meet the preset impedance value. In recent years, the development of artificial intelligence technology has provided new impetus for the development of electrode lifting control technology. American neural application company launched IAF and SMI company launched smartarc. Compared with PID control system, the control system based on artificial intelligence technology has obvious advantages. It should be pointed out that the above intelligent electrode lifting controllers are all developed against the background of primary smelting furnace and cannot be completely copied for LF. Since argon blowing at the bottom of ladle furnace will cause fluctuation of molten steel slag level, strong disturbance caused by such fluctuation shall be considered in LF electrode lifting control, especially in refining slag melting period

the system adopts a composite electrode lifting control scheme based on artificial neural network and fuzzy control. Firstly, the furnace simulation model is used to predict the arc flow of the three-phase electrode, and then the output of the controller is adjusted based on the difference between the set point and the predicted value, which is sent to the hydraulic valve through the signal amplification board for the lifting control of the three-phase electrode

Copyright © 2011 JIN SHI