Method for controlling the powder feed rate in radio frequency plasma gasification separation

Abstract

The application discloses a method for controlling the powder feeding rate of radio frequency plasma gasification separation, detects the powder feeding rate of the powder to be separated, takes the powder feeding rate error and the first derivative of the powder feeding rate error as input values of a neural network, and outputs the lumped uncertainty affecting the powder feeding rate after learning of the neural network; and the lumped uncertainty is combined with an adaptive hyper-spiral strategy and a fractional order non-singular terminal sliding mode function to be used for the control law of a powder feeding motor; the application adjusts the rotating speed of the powder feeding motor in real time by monitoring the actual powder feeding rate in real time, feeds back the actual powder feeding rate to the control layer of the powder feeding motor, controls the rotating speed of the powder feeder servo motor on line, realizes the on-line detection and closed-loop control of the powder feeding rate, improves the accurate feeding of the powder to be separated, guarantees the uniformity, stability and accuracy of the powder feeding, and is helpful to further improving the metal separation and purification precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of separation and purification, in particular to a powder feeding rate control method for radio frequency plasma gasification separation. BACKGROUND

[0002] The radio frequency plasma spheroidization technology is suitable for separation and purification of impurities of high melting point materials (such as ceramics, tungsten, molybdenum, tantalum, niobium and other metal materials, and high melting point composite materials). The radio frequency plasma spheroidization technology uses plasma to heat raw materials at high temperature, so that the high melting point metal is melted, and the low melting point impurities are gasified and burned. The molten high melting point metal rapidly solidifies under the combined action of surface tension and extremely high temperature gradient to form spherical powder, and then separates from the impurities. The radio frequency plasma separation and purification has the following advantages: no electrode pollution, high material purity; large plasma flame volume, high temperature, and large processing capacity; can be used in inert atmosphere, oxidizing atmosphere, reducing atmosphere and other working environments. The above-mentioned characteristics make the radio frequency plasma become one of the choices for the production of industrial materials.

[0003] In the radio frequency plasma gasification separation process, the raw material powder mixed with impurities is rotated by the roller type powder feeder driven by the stepping motor. The powder feeding port above the roller type powder feeder is connected with the powder storage barrel. The stepping motor drives the powder feeding shaft in the roller type powder feeder to rotate. The powder in the groove on the surface of the powder feeding shaft enters the conveying pipeline through the powder outlet, and is blown into the radio frequency plasma generator by the gas feeding system for heating, separation and purification. The powder feeder needs to provide stable powder to be separated for the high-temperature plasma torch according to the powder feeding rate requirement, which is also one of the core technologies of radio frequency induction plasma separation and purification. During the process of conveying the powder to be separated by the powder feeder, the uniformity, stability and accuracy of the powder conveying will directly affect the metal separation and purification precision (especially for materials used in the field of aviation and aerospace with high purity requirements). The traditional powder feeder control system method such as PID cannot meet the high-performance powder feeding control demand, and improvement is urgently needed.

[0004] The present application provides a powder feeding rate control method for radio frequency plasma gasification separation to solve the above technical problems. SUMMARY

[0005] The powder feeding rate control method for radio frequency plasma gasification separation comprises:

[0006] S1: sensing the powder feeding rate in the transparent conveying pipeline by the photoelectric sensor , obtaining the powder feeding rate error of the set powder feeding rate and the actual powder feeding rate ;

[0007] S2: Online prediction of lumped uncertainty using RBF neural network; In the actual process of powder feeding control for radio frequency plasma gasification separation, there are many factors that affect the accuracy of powder feeding (such as sensor error, friction of the powder feeding actuator, structural gap, execution response lag time difference, etc.). These interference factors, namely lumped uncertainty, seriously affect the stability and accuracy of the control system. The lumped uncertainty can be predicted online using RBF neural network.

[0008] S3: Based on the lumped uncertainty predicted by the RBF neural network, the control law of the powder feeding motor is obtained by combining the adaptive superspiral strategy and the fractional non-singular terminal sliding mode function, and the speed of the powder feeding motor is controlled in real time.

[0009] The powder feeding rate control method for radio frequency plasma vaporization separation, in step S2, the input value of the RBF neural network (The input value of the RBF neural network is defined as...) (the transpose of the matrix) To account for powder feeding rate error, For the set target powder delivery rate, The network output value is the actual measured powder feeding rate. The details are as follows:

[0010] ;

[0011] ;

[0012] ;

[0013] ;

[0014] in, Represents the hidden layer of the network. 1 node For the input values ​​of the neural network, For the reason The vector formed For ideal network weights, The transpose matrix representing the ideal network weights. For network approximation error, It is a positive scalar representing the width of the Gaussian function. It is the center vector of the hidden layer.

[0015] The powder feeding rate control method for radio frequency plasma vaporization separation, step S3, is as follows:

[0016] S 3.1, the fractional-order non-singular terminal sliding mode function is:

[0017] ;

[0018] wherein , , , , , is a positive parameter to be tuned, wherein > 0, < 1, , is a Riemann-Liouville fractional calculus function, and , ;

[0019] S 3.2, in order to further improve the convergence speed and robustness of the fractional non-singular terminal sliding mode and reduce chattering, an adaptive super-spiral method is used to further optimize the non-singular terminal sliding mode function:

[0020] ;

[0021] ;

[0022] wherein and are adaptive parameters, the upper and lower limit values of , are artificially set or the empirical values of , are captured from the system, wherein:

[0023] ;

[0024] wherein represents the running time of the controller for controlling the motor speed;

[0025] ;

[0026] ;

[0027] wherein , , is a constant parameter, and > 0 , is a fractional non-singular terminal sliding mode function; it can be seen that the tracking performance is good and the convergence speed increases from to when is within the design range. Therefore, while suppressing the influence of noise, the control performance can be maximized; , The value of the parameter is reasonably selected according to the motor performance, and the greater the value, the greater the convergence speed (the powder feeding motor can adjust the powder feeding speed in a shorter time), but the motor has a certain bearing limit, and if the convergence speed is too large, it will lead to the risk of difficulty in execution and damage to the motor;

[0028] S 3.3: Based on the calculation results of the above steps, the fractional order non-singular terminal sliding mode control law based on RBF neural network and adaptive super-helix is as follows:

[0029] 。

[0030] The ideal weight value of the radio frequency plasma gasification separation powder feeding rate control method The adaptive law is as follows:

[0031]

[0032] In the formula, is the learning rate, is the fractional order non-singular terminal sliding mode function, is a vector composed of .

[0033] The working principle is as follows:

[0034] First, the actual powder feeding rate is detected, and the powder feeding rate error and the first derivative are calculated, The transpose matrix data of is used as the input value of the neural network, the neural network is operated to learn the lumped uncertainty of the powder feeding rate control system, the intermediate function related to is set, and in order to improve the convergence of (making tend to 0 quickly), the adaptive super-helix strategy is combined with the non-singular terminal sliding mode function, which can maximize the control performance while suppressing the influence of noise, and the fractional order non-singular terminal sliding mode control law based on RBF neural network and adaptive super-helix is obtained. The speed of the powder feeding motor is controlled in real time by using the control law.

[0035] The advantages are as follows: the RBF neural network, adaptive super-helix strategy, and fractional order non-singular terminal sliding mode function are combined, the actual powder feeding efficiency is monitored in real time, the lumped uncertainty is output, the adaptive super-helix strategy is used to optimize the fractional order non-singular terminal sliding mode function to improve the function convergence speed, and the system robustness and reduce chattering are improved. By Real-time control the rotating speed of the powder feeding motor (DC motor or AC motor) to make the actual powder feeding rate infinitely close to the ideal powder feeding rate, and provide stable powder to be separated for the high-temperature plasma torch according to the set ideal powder feeding rate requirement, so as to uniformly, stably and accurately transport the powder to be separated and purified, and ensure the high-purity state of the separated material. BRIEF DESCRIPTION OF DRAWINGS

[0036] The specific embodiments will be further described below in combination with the drawings, in which:

[0037] Figure 1 is the optical detection principle diagram of the powder flow involved in the specific embodiment 1 of the present application;

[0038] Figure 2 is the principle diagram of the photoelectric powder feeding rate control system involved in the specific embodiment 1 of the present application;

[0039] The following specific embodiments will further illustrate the present application in combination with the above drawings. DETAILED DESCRIPTION

[0040] The powder feeding rate control method for radio frequency plasma gasification separation comprises:

[0041] As shown in Figure 2 , the powder to be separated is transported to the radio frequency plasma separation and purification system by the purchased law for science and technology planetary servo integrated powder feeding motor (60 servo 400W set), wherein the commonly used roller shaft rotating speed is about 12 r / min, the carrier gas flow is about 7.7 L / min, the powder particle size is 250 mesh, and the corresponding powder feeding rate is about 5.9 g / min. However, in order to further improve the accuracy and stability of the powder feeding rate and improve the purity of the separated and purified product, the rotating speed of the powder feeding motor can be adjusted in real time by the following method, as follows:

[0042] S1: The powder feeding rate in the transparent conveying pipeline is sensed by the photoelectric sensor (of course, it can also be monitored in real time by other ways) , to obtain the powder feeding rate error between the set powder feeding rate and the actual powder feeding rate ;

[0043] The light emitted by the light emitting tube is received by the photoelectric sensor after passing through the transparent tube with powder gas flow. The light intensity received by the photoelectric sensor is related to the flow of the powder in the transparent tube. The greater the powder flow, the greater the light attenuation, and the smaller the light intensity obtained by the photoelectric sensor. In this way, after experimental calibration, the powder feeding rate flowing through the transparent tube can be calculated according to the output signal of the photoelectric sensor.

[0044] S2: Estimate the lumped uncertainty online by RBF neural network;

[0045] S3: Based on the lumped uncertainty predicted by the RBF neural network, the control law of the powder feeding motor is obtained by combining the adaptive superspiral strategy and the fractional non-singular terminal sliding mode function, and the speed of the powder feeding motor is controlled in real time.

[0046] Preferably, in the radio frequency plasma vaporization separation powder feeding rate control method, the input value of the RBF neural network in step S2... (The input value of the RBF neural network is defined as...) The transpose of the matrix formed To account for powder feeding rate error, For the set target powder delivery rate, The network output value is the actual measured powder feeding rate. The details are as follows:

[0047] ;

[0048] ;

[0049]

[0050] ;

[0051] in, Represents the hidden layer of the network. 1 node For the input values ​​of the neural network, For the reason The vector formed For ideal network weights, The transpose matrix representing the ideal network weights. For network approximation error, It is a positive scalar representing the width of the Gaussian function. It is the center vector of the hidden layer.

[0052] Preferably, the powder feeding rate control method for radio frequency plasma vaporization separation, step S3, is operated as follows:

[0053] S 3.1, the fractional-order non-singular terminal sliding mode function is:

[0054] ;

[0055] =1.9, =1.5, =0.1, =0.5, =0.86, =0.99 is the positive parameter to be tuned. , is the Riemann-Liouville fractional calculus function and , .

[0056] S 3.2, in order to further improve the convergence speed and robustness of the fractional non-singular terminal sliding mode and reduce chattering, an adaptive super-spiral method is used to further optimize the non-singular terminal sliding mode function:

[0057] ; ;

[0058] wherein, and is an adaptive parameter, , manually set upper and lower limit values or experience values from the system, wherein:

[0059] ;

[0060] wherein, represents the running time of the controller for controlling the motor speed;

[0061] ;

[0062] ;

[0063] wherein, = 30, = 60, = 1.5, = 0.5, is the fractional non-singular terminal sliding mode function; it can be seen that the tracking performance is good and when the design range is within the design range, the convergence speed increases from to ; therefore, while suppressing the influence of noise, the control performance can be maximized;

[0064] S 3.3: based on the calculation results of the above steps, the fractional non-singular terminal sliding mode control law based on RBF neural network and adaptive super-spiral is as follows:

[0065] ;

[0066] Further, the adaptive law of the ideal weight is as follows:

[0067]

[0068] wherein, is the learning rate and takes a value of 10, a fractional order nonsingular terminal sliding mode function, a vector composed of .

[0069] By using the above powder feeding rate control method, the powder feeding rate can be strictly controlled around the target rate, and the experimental data are as follows: set = 5.90 g / min:

[0070] Monitoring time Roller shaft rotation speed (r / min) V (g / min) 30 S 12 5.88 60 S 12.5 5.90 90 S 12.5 5.91 120 S 12.3 5.89 3 min 12.3 5.90 4 min 12.4 5.90 5 min 12.4 5.91 6 min 12.3 5.89 10 min 12.3 5.90 14 min 12.3 5.90 18 min 12.4 5.91 22 min 12.4 5.90 26 min 12.4 5.89 56 min 12.5 5.89 86 min 12.5 5.90 116 min 12.4 5.90 146 min 12.4 5.90 206 min 12.4 5.91 266 min 12.3 5.91 326 min 12.4 5.90 386 min 12.3 5.90 446 min 12.5 5.90 646 min 12.4 5.89 846 min 12.6 5.90 1046 min 12.5 5.90 1246 min 12.5 5.91

[0071] As can be seen from the above table, the speed of the roller shaft (powder feeding shaft) driven by the reducer-motor and the actually measured powder feeding rate do not have a linear relationship, so it is particularly necessary to monitor and adjust the speed of the motor in real time for accurately controlling the powder feeding amount; by using the above method to monitor the actual powder feeding rate in real time and through the RBF neural network, the adaptive hyper-spiral strategy and the fractional order nonsingular terminal sliding mode function, the speed of the powder feeding motor can be effectively controlled in real time and the powder feeding rate can be accurately adjusted.

[0072] The above-described embodiments only express several embodiments of the present application, and the description is relatively specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be noted that for ordinary skilled persons in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method of controlling the feed rate of powder for radio frequency plasma gasification separation, characterized by: The steps are as follows, S1 : measuring the actual feed rate of the powder to be separated and purified , obtaining a set feed rate and the feed rate error of the actual feed rate of the powder S2: estimating the lumped uncertainty affecting the powder feeding rate of the powder to be separated and purified by a neural network online; S3: based on the lumped uncertainty estimated by the neural network, combining the adaptive hyper-spiral strategy and the fractional order non-singular terminal sliding mode function to obtain the control law of the powder feeding motor, and controlling the speed of the powder feeding motor in real time; The input value of the neural network in step S2 is the transpose matrix of , that is, , , the powder feeding rate error, , the set target powder feeding rate, , the actually measured powder feeding rate, and the network output value is , and the specific process is as follows: ； ； ； ； wherein represents the i-th node of the hidden layer of the network, is the input value of the neural network, is the vector composed of is the ideal network weight, represents the transpose matrix of the ideal network weight, is the network approximation error, is a positive scalar, representing the width of the Gaussian function, is the center vector of the hidden layer; the operation method of step S3 is as follows,​​ S 3.1, the fractional order non-singular terminal sliding mode function is: ； wherein , , , , , is a positive parameter to be regularized, wherein > 0, < 1, , is a Riemann-Liouville fractional calculus function, and , ; S 3.2, the adaptive hyper-spiral strategy is used to optimize the non-singular terminal sliding mode function: ； ； In the formula, and are adaptive parameters, upper and lower limit values of , or empirical values of , are artificially set or captured from the system internally. ； denotes the controller run time controlling the motor speed; ； ； wherein , , , is a constant parameter, , is a fractional order nonsingular terminal sliding mode function;​ S 3.3: based on the calculation results of the above steps, the fractional order non-singular terminal sliding mode control law based on the neural network and the adaptive hyper-spiral is obtained: ； Ideal weights The adaptive law is as follows: ； wherein is the learning rate, is the fractional order nonsingular terminal sliding mode function, is the vector consisting of is the vector consisting of

Images