A method for controlling pressure in a transition bag during the production of ultra-thin amorphous alloys

Abstract

The present invention discloses a pressure control system and method for a transition ladle during the production process of ultra-thin amorphous alloys, including the following steps: First, collect the real-time pressure in the transition ladle through a pressure sensor, and compare the set value of the pressure in the transition ladle with the measured value to obtain a pressure measurement deviation. e 1. Then, establish a pressure prediction model for the transition ladle using an adaptive PSO-BP neural network. By comparing the set value of the pressure in the transition ladle with the predicted value, obtain a pressure prediction deviation. e 2. Then, according to the pressure measurement deviation e 1 and the pressure prediction deviation e 2, obtain the pressure deviation a combined based on the weight factor e , and obtain the pressure deviation change rate ec through differential operation. Finally, according to the pressure deviation a combined by the weight factor e and the pressure deviation change rate ec , obtain the control quantity U in different modes. The electro-hydraulic proportional valve receives the control signal U sent by the controller, changes the opening degree of the electro-hydraulic proportional valve, and fills argon into the transition ladle to realize the pressure regulation in the transition ladle.

Description

Technical Field

[0001] The invention belongs to the field of ultra-thin amorphous alloy production, and in particular relates to a method for controlling pressure in a transition bag during the production of ultra-thin amorphous alloy. Background Art

[0002] As an emerging soft magnetic material, ultra-thin amorphous alloys are widely used in the fields of electricity, machinery and chemical industry. The formation of ultra-thin amorphous alloys is different from traditional preparation technology. It uses ultra-rapid cooling solidification technology to rapidly cool from high-temperature liquid to solid state. The process is one-time forming and the speed is relatively fast.

[0003] The transition bag is a key equipment for the formation of ultra-thin amorphous alloys. In the production process of ultra-thin amorphous alloys, whether the pressure in the transition bag can be stably controlled is an important factor affecting the thickness of the ultra-thin amorphous alloy. The pressure in the transition bag directly affects the flow rate of the alloy melt at the nozzle, and then affects the thickness of the ultra-thin amorphous alloy. Since the pressure change in the transition bag has time hysteresis and is affected by multiple factors, the control effect is not good when the pressure in the transition bag is controlled by a traditional PID controller. Therefore, in view of the hysteresis in the pressure control process in the transition bag and the complexity of the control system, the present invention proposes a method for controlling the pressure in the transition bag during the production of ultra-thin amorphous alloys. Summary of the invention

[0004] In order to avoid the problem that large pressure fluctuations in a transition bag during the production of ultra-thin amorphous alloys may cause uneven thickness of the ultra-thin amorphous alloy and that a pressure setting curve in the transition bag cannot be changed according to the production site during actual production, the present invention provides a method for controlling the pressure in a transition bag during the production of ultra-thin amorphous alloys.

[0005] The present invention adopts the following technical scheme: a method for establishing a transition bag internal pressure prediction model, comprising the following steps, S100 ~ data preprocessing, processing abnormal data in the ultra-thin amorphous alloy production process, and improving the accuracy of the transition bag internal pressure prediction model through data normalization processing; S200 ~ determining the topological structure of the BP neural network, the transition bag internal pressure prediction model constructed by the BP neural network only includes one hidden layer, wherein the number of neurons in the input layer is 7, corresponding to the thickness of 5 different positions of the ultra-thin amorphous alloy, the pressure in the transition bag at the previous moment, and the temperature in the transition bag; the number of neurons in the output layer is set to 1, which is used to predict the pressure in the transition bag; S300 ~ adaptive PSO optimizes the weights and thresholds of the BP neural network; S400 ~ screening a training sample set, the training sample of the neural network is the offline data with the greatest similarity to the data under the preset working conditions, the data under the preset working conditions includes the preset ultra-thin amorphous alloy thickness, the preset transition bag internal pressure curve and the transition bag set temperature, and the similarity calculation formula is as follows:

[0006]

[0007] In the formula, It is an array of offline data; It is the array under the preset working condition.

[0008] Step S100 adopts the following method: S101-based on the collected actual production site data of ultra-thin amorphous alloys, the absolute value difference median method is used to screen out abnormal values ​​in the data to obtain abnormal data; S102-the neighboring data average value interpolation method is used for abnormal data points; S103-the actual production site data of ultra-thin amorphous alloys, including the thickness of ultra-thin amorphous alloys, the pressure in the transition package, and the temperature in the transition package, are normalized.

[0009] Step S300 adopts the following method: the weight and threshold of the BP neural network are used as the elements contained in the particles in the adaptive PSO algorithm, the particle length is determined according to the input, output and number of layers of the BP neural network, and the square sum of the errors between the predicted value of the pressure in the transition package output by the BP neural network and the original value is used as the fitness function of the adaptive PSO algorithm. The calculation formula is as follows:

[0010]

[0011] In the formula, fit is the fitness value; p j is the original value of the pressure in the transition bag; It is the predicted value of the pressure in the transition bag output by the BP neural network.

[0012] S301~By calculating the fitness value of each particle, the median fitness value is obtained, and the particles in the population are divided into particles in the superior zone and particles in the inferior zone according to the median fitness value; the calculation formula of the median fitness value is as follows:

[0013]

[0014] In the formula, X i is the fitness value of the i-th particle; N is the total number of particles; f 0.5 is the median of the particle fitness value. The particles in the superior zone have a fitness value greater than f 0.5 The particles in the difference zone have a fitness value less than f. 0.5 of particles.

[0015] S302~For particles in the superior zone, they are close to the optimal position of the population and are prone to fall into the local optimum. Therefore, a self-mutation update mechanism is designed for them, and n is set. t It can adaptively decrease with the increase of the number of iterations, which plays a role in balancing the exploration and development of particles in the superior zone. The calculation formula for updating particles in the superior zone is as follows:

[0016]

[0017] n t =1-t / t max

[0018] Where n t To control the parameters of variable step length; rand(0,1) is a random number uniformly distributed between [0,1]; represents the position of the particle in the superior area; t is the current iteration number; t max is the maximum number of iterations.

[0019] S303~For the difference area, the particle swarm first updates the speed and position of the particles according to the traditional PSO algorithm, and then updates the population in the difference area to its historical optimal position X g1 and its historical suboptimal position X g2 , and make full use of the difference calculation results between the two to perform local search near the historical optimal position of the population to enhance particle diversity. The specific calculation formula is as follows:

[0020] X′ g1 =X g1 +rand(-1,1)×d t ×(X g1 -X g2 )

[0021] d t+1 =1-t / t max

[0022] In the formula, rand(-1,1) is a random number uniformly distributed between [-1,1]; d t is the local scaling factor at the tth generation; t is the current iteration number; t max is the maximum number of iterations.

[0023] S304~Result X′ of local search g1 Using the greedy retention strategy, the calculation formula is as follows:

[0024]

[0025] Where fit(x) is the fitness value of x.

[0026] A method for controlling pressure in a transition bag during the production of an ultra-thin amorphous alloy, comprising the following steps:

[0027] S1~The real-time pressure in the transition bag is collected through the pressure sensor. The collected current signal is amplified and fed back to the PLC controller to obtain the pressure measurement value in the transition bag. The pressure setting value in the transition bag is compared with the measured value to obtain the pressure measurement deviation e1.

[0028] S2~PLC controller uploads data to WinCC host computer, WinCC host computer is connected to Matlab, the control signal U is calculated on Matlab and transmitted to PLC through the communication loop.

[0029] S3~The electric proportional valve receives the control signal U sent by the controller, changes the opening of the electric proportional valve, and realizes the pressure regulation in the transition pack.

[0030] Step S2 adopts the following method:

[0031] S21: Using the established transition bag internal pressure prediction model, a transition bag internal pressure prediction value is obtained, and the transition bag internal pressure setting value is compared with the prediction value to obtain a pressure prediction deviation e2;

[0032] S22~According to the pressure measurement deviation e1 and the pressure prediction deviation e2, the pressure deviation e based on the combination of weight factor a is obtained, and the pressure deviation change rate ec is obtained through differential operation;

[0033] S23~According to the combined pressure deviation e and pressure deviation change rate ec, the expert controller given below is used to obtain the incremental coefficient of the PID parameter, which is used to adjust the control ratio adjustment coefficient K p , integral adjustment coefficient K i , differential adjustment coefficient K d , and accordingly, the PID parameters are adjusted online. Finally, the control quantity U under different modes is obtained according to the combined deviation and deviation change rate of the pressure in the transition bag.

[0034] In step S22, a weight factor a is set to combine the prediction deviation and the measurement deviation, and the calculation formula is as follows:

[0035]

[0036] Where, e is the pressure deviation after compensation using the predicted value; e1 is the pressure measurement deviation; e2 is the pressure prediction deviation; a is the weight factor; is the predicted value of pressure in the transition bag; p i is the pressure setting value in the transition bag; p i ′ is the measured value of the pressure inside the transition bag.

[0037] In step S23, the calculation process of the expert controller is as follows:

[0038] Assuming that the current sampling is the kth time, the current deviation is e(k), and the deviation of the previous sampling time is e(k-1), and the deviation of the previous two sampling times is e(k-2), then the calculation formula for the two deviation increments can be obtained as follows:

[0039] Δe(k)=e(k)-e(k-1)

[0040] Δe(k-1)=e(k-1)-e(k-2)

[0041] Wherein, e(k) is the pressure deviation in the transition bag at time k; e(k-1) is the pressure deviation in the transition bag at time k-1; e(k-2) is the pressure deviation in the transition bag at time k-2; Δe(k) is the pressure deviation increment in the transition bag at time k; Δe(k-1) is the pressure deviation increment in the transition bag at time k-1.

[0042] 1) When |e(k)|>E max hour,

[0043]

[0044] Where U(k) is the control quantity at time k; U max is the maximum output value; U min is the minimum output value.

[0045] 2) When |e(k)|>E mid hour,

[0046] U(k)=U(k-1)+k1×(K p Δe(k)+K i e(k)+K d Δe(k-1))

[0047] In the formula, U(k-1) is the control quantity at time k-1; k1 is the gain coefficient, k1 is (1,1.5); K p is the proportional adjustment coefficient; K i is the integral adjustment coefficient; K d is the differential adjustment coefficient.

[0048] 3) When |e(k)| <E mid hour,

[0049] U(k)=U(k-1)+K p Δe(k)+K i e(k)+K d Δe(k-1)

[0050] 4) When e(k)Δe(k)<0 and e(k)Δe(k-1)>0 or e(k)=0,

[0051] U(k)=U(k-1)

[0052] 5) When e(k)Δe(k)<0 and e(k)Δe(k-1)<0,

[0053] U(k)=U(k-1)+k2K pe(k),|e(k)|>E mid

[0054] U(k)=U(k-1)+k3K p e(k),|e(k)|≤E mid

[0055] Wherein, k2 is the gain coefficient, k2 is (1,5]; k3 is the gain coefficient, k3 is (0,1).

[0056] 6) When |e(k)|≤E min hour,

[0057] U(k)=U(k-1)+k4×(K p Δe(k)+K i e(k))

[0058] In the formula, k4 is the gain coefficient, and k4 is (1.4, 2.5].

[0059] Compared with the prior art, the present invention has the following beneficial effects: in the process of controlling the pressure in the transition bag, an expert controller based on the deviation combination is introduced to modify the three PID parameters online, which solves the problem that the parameters in the PID control cannot be self-tuned, resulting in poor control effect and instability. An adaptive PSO-BP algorithm is used to establish a pressure prediction model in the transition bag, and then the pressure prediction deviation and measurement deviation in the transition bag are combined, and expert PID control is used to adjust the pressure in the transition bag. Compared with PID control and traditional expert PID control, the control method proposed by the present invention has better effect, which is mainly manifested in small overshoot and small steady-state error. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A schematic diagram of the structure of a transition package provided by an embodiment of the present invention;

[0061] Figure 2 A BP neural network topology diagram provided by an embodiment of the present invention;

[0062] Figure 3 A flow chart of the weights and thresholds of the BP neural network optimized by the adaptive PSO of the present invention;

[0063] Figure 4 This is a schematic diagram of the pressure compensation control principle in the transition bag of the present invention;

[0064] In the figure, 1 is liquid inlet, 2 is air inlet, 3 is pressure sensor, 4 is plug rod, 5 is shell, 6 is liner, 7 is nozzle, and 8 is nozzle package. DETAILED DESCRIPTION

[0065] The present invention is described in detail below in conjunction with specific embodiments.

[0066] A pressure control system for a transition bag in the production of ultra-thin amorphous alloys, the system comprising: a PLC controller, a pressure sensor, and an electric proportional valve. During the production of ultra-thin alloys, the pressure in the transition bag is collected in real time by the pressure sensor, and the collected current signal is fed back to the PLC controller through an analog input module; then the PLC controller runs an expert PID algorithm based on a deviation combination, and the control result is converted into a 4-20mA current to control the opening of the electric proportional valve, and argon is filled into the transition bag to achieve pressure regulation in the transition bag.

[0067] The PLC controller and the WinCC host computer realize data transmission through PROFIBUS-MPI, and realize data display and storage; the WinCC host computer connects with Matlab through OPC technology to realize data transmission, realize the transition bag pressure prediction model and the expert PID controller simulation based on deviation combination.

[0068] like Figure 1 The figure shows the structure of the transition bag. The transition bag is the last metallurgical container that needs to be passed before the production of ultra-thin amorphous alloys. The exterior of the transition bag is made of steel materials, and the interior is made of refractory materials. The transition bag is equipped with a pressure sensor to collect the pressure inside the transition bag in real time, and an electrical proportional valve is installed to control the pressure inside the transition bag. During the production process, argon gas needs to be filled from the air inlet to control the pressure inside the transition bag and maintain the flow rate of the molten liquid at the nozzle.

[0069] The present invention provides a pressure control system in the production of ultra-thin amorphous alloys, and the required equipment includes: PLC (SIEMENS-S7-200CN), pressure sensor (SMC-ISE30A-01-D), and electric proportional valve (SMC-ITV3050-044N). The pressure sensor is connected to the PLC controller through an analog input module, and the real-time collected pressure analog signal is transmitted to the PLC, and converted into a pressure value by the PLC. The analog output port of the PLC controller is connected to the current drive port of the electric proportional valve. When the PLC controller outputs 4-20mA current, the opening of the electric proportional valve is controlled to fill argon into the transition bag, so that the pressure in the transition bag changes from 5kPa to 50kPa, thereby realizing pressure regulation in the transition bag.

[0070] A method for controlling pressure in a transition bag during the production of an ultra-thin amorphous alloy, comprising the following steps:

[0071] S1: The real-time pressure in the transition package is collected by a pressure sensor. The collected current signal is amplified and fed back to the PLC controller to obtain the pressure measurement value in the transition package. The pressure setting value in the transition package is compared with the measured value to obtain the pressure measurement deviation e1.

[0072] S2: The PLC controller and the WinCC host computer upload data through PROFIBUS-MPI. The WinCC host computer is connected to Matlab through OPC technology. The transition package pressure prediction model and the expert PID algorithm based on deviation combination are run on Matlab to obtain the control signal U, which is transmitted to the PLC through the communication loop.

[0073] S21: Use the adaptive PSO-BP neural network to establish a transition bag pressure prediction model to obtain a transition bag pressure prediction value. Compare the transition bag pressure setting value with the prediction value to obtain a pressure prediction deviation e2.

[0074] S22: According to the pressure measurement deviation e1 and the pressure prediction deviation e2, a pressure deviation e based on a combination of weight factors a is obtained, and a pressure deviation change rate ec is obtained through differential operation.

[0075] S23: According to the combined pressure deviation e and pressure deviation change rate ec, the expert controller given below is used to obtain the incremental coefficient of the PID parameter, which is used to adjust the control ratio adjustment coefficient K p , integral adjustment coefficient K i , differential adjustment coefficient K d , and accordingly, the PID parameters are adjusted online. Finally, the control quantity U under different modes is obtained according to the combined deviation and deviation change rate of the pressure in the transition bag.

[0076] S3: The electric proportional valve receives the control signal U sent by the controller, changes the opening of the electric proportional valve, and realizes the pressure regulation in the transition pack.

[0077] The process of establishing the pressure prediction model in the transition package in S21 is as follows:

[0078] (1) Data preprocessing can process abnormal data in the production process of ultra-thin amorphous alloys and improve the accuracy of the pressure prediction model in the transition package through data normalization. The data preprocessing process is as follows:

[0079] (a) Based on the actual production site data of ultra-thin amorphous alloys collected, the median absolute difference method is used to screen out the abnormal values ​​in the data to obtain abnormal data. The processing steps are as follows: Find the median x of all data m ; Calculate the absolute deviation errorx of each data from the median mn ; Find the median absolute deviation error m ; Determine the parameter w and keep it in the range [x m -w×error m , x m +w×error m ] data within the range and record data outside the range.

[0080] (b) For abnormal data points, the average value of neighboring data is used for interpolation.

[0081] (c) Since the actual production site data of ultra-thin amorphous alloys include the thickness of ultra-thin amorphous alloys, the pressure in the transition package, and the temperature in the transition package, their units and orders of magnitude are different and need to be normalized.

[0082] (2) Determine the topological structure of the BP neural network. The transition bag pressure prediction model constructed by the BP neural network contains only one hidden layer, in which the number of neurons in the input layer is 7, corresponding to the thickness of the ultra-thin amorphous alloy at 5 different positions, the pressure in the transition bag at the previous moment, and the temperature in the transition bag; the number of neurons in the output layer is set to 1, which is used to predict the pressure in the transition bag; the calculation formula of the number of neurons in the hidden layer is as follows:

[0083]

[0084] Where k and l are the number of neurons in the input layer and output layer respectively; q is the number of neurons in the hidden layer; a is a constant, a=[1,10].

[0085] When the number of neurons in the hidden layer is 5, the performance of the transition bag pressure prediction model reaches the optimal performance.

[0086] (3) Adaptive PSO optimizes the weights and thresholds of the BP neural network. The weights and thresholds of the BP neural network are used as the elements contained in the particles in the adaptive PSO algorithm. The particle length is determined according to the input, output and number of layers of the BP neural network. The square sum of the errors between the predicted value of the transition bag pressure output by the BP neural network and the original value is used as the fitness function of the adaptive PSO algorithm. The calculation formula is as follows:

[0087]

[0088] In the formula, fit is the fitness value; p j is the original value of the pressure in the transition bag; It is the predicted value of the pressure in the transition bag output by the BP neural network.

[0089] The specific process of the adaptive PSO algorithm is as follows:

[0090] (a) By calculating the fitness value of each particle, the median fitness value is obtained, and the particles in the population are divided into particles in the superior zone and particles in the inferior zone according to the median fitness value. The calculation formula for the median fitness value is as follows:

[0091]

[0092] Where, X i is the fitness value of the i-th particle; N is the total number of particles; f0.5 is the median of the particle fitness value. The particles in the superior zone have a fitness value greater than f 0.5 The particles in the difference zone have a fitness less than f. 0.5 of particles.

[0093] (b) For particles in the superior zone, they are close to the optimal position of the population and are prone to fall into the local optimum. Therefore, a self-mutation update mechanism is designed for them, and n is set t It can adaptively decrease as the number of iterations increases, which plays a role in balancing the exploration and development of particles in the superior zone. The calculation formula for updating particles in the superior zone is as follows:

[0094]

[0095] n t =1-t / t max (5)

[0096] Where n t To control the parameters of variable step length; rand(0,1) is a random number uniformly distributed between [0,1]; represents the position of the particle in the superior area; t is the current iteration number; t max is the maximum number of iterations.

[0097] (c) For the differential zone, the particle swarm first updates the particle speed and position according to the traditional PSO algorithm, and then updates the population in the differential zone to its historical optimal position X. g1 and its historical suboptimal position X g2 , and make full use of the difference calculation results between the two to perform local search near the historical optimal position of the population to enhance particle diversity. The specific calculation formula is as follows:

[0098] X′ g1 =X g1 +rand(-1,1)×d t ×(X g1 -X g2 ) (6)

[0099] d t+1 =1-t / t max (7)

[0100] In the formula, rand(-1,1) is a random number uniformly distributed between [-1,1]; d t is the local scaling factor at the tth generation; t is the current iteration number; t max is the maximum number of iterations.

[0101] (d) The result of local search X g′1 adopts the greedy retention strategy, and the calculation formula is as follows:

[0102]

[0103] Where fit(x) is the fitness value of x.

[0104] (4) Screening the training sample set: The training sample of the neural network is the offline data with the greatest similarity to the data under the preset working conditions. The data under the preset working conditions include the preset ultra-thin amorphous alloy thickness, the preset pressure curve in the transition package, and the set temperature in the transition package. The similarity calculation formula is as follows:

[0105]

[0106] In the formula, It is an array of offline data; It is the array under the preset working condition.

[0107] Through training and verification of the model, the root mean square error of the transition bag pressure prediction model based on the adaptive PSO-BP neural network is 0.58, the average relative error is 0.19, the goodness of fit is 0.99, and the detection accuracy is high, which can provide a reference for the subsequent transition bag pressure control.

[0108] An offline database is established, and the offline data in the database is retrieved for similarity with the data under the preset working conditions. The offline data with the maximum similarity is obtained as the training sample of the prediction model. The temperature in the transition package, the pressure in the transition package at the previous moment, and the thickness of the ultra-thin amorphous alloy are used as inputs, and the pressure in the transition package is used as output to train the BP neural network. At the same time, in view of the problem of gradient disappearance in the training process of the BP neural network, an adaptive PSO-BP neural network is proposed, which makes two improvements on the basis of the traditional PSO algorithm: first, adaptive partitioning is performed through the median of fitness, and the population is divided into superior and inferior areas, and different update mechanisms are used respectively to make each particle participate in the competition; second, for the particles in the inferior area, the difference results of the historical optimal and suboptimal positions are used for local search, so that each particle can adjust the flight direction and speed in time to obtain the optimal solution.

[0109] The design of the pressure deviation combination in the transition package in S22 is as follows:

[0110] In the production process of ultra-thin amorphous alloys, the pressure in the transition package is affected by many factors, and the pressure setting value in the transition package cannot be changed according to the actual production process. Therefore, an adaptive PSO-BP neural network is used to establish a pressure model in the transition package to compensate for the pressure setting value in the transition package. The weight factor a is set to combine the prediction deviation and the measurement deviation. The calculation formula is as follows:

[0111]

[0112] Where, e is the pressure deviation after compensation using the predicted value; e1 is the pressure measurement deviation; e2 is the pressure prediction deviation; a is the weight factor; is the predicted value of pressure in the transition bag; p i is the pressure setting value in the transition bag; p i ′ is the measured value of the pressure inside the transition bag.

[0113] In the control method, in S23, the design of the expert controller is as follows:

[0114] Assuming that the current sampling is the kth time, the current deviation is e(k), and the deviation of the previous sampling time is e(k-1), and the deviation of the previous two sampling times is e(k-2), then the calculation formula for the two deviation increments can be obtained as follows:

[0115] Δe(k)=e(k)-e(k-1) (11)

[0116] Δe(k-1)=e(k-1)-e(k-2) (12)

[0117] Wherein, e(k) is the pressure deviation in the transition bag at time k; e(k-1) is the pressure deviation in the transition bag at time k-1; e(k-2) is the pressure deviation in the transition bag at time k-2; Δe(k) is the pressure deviation increment in the transition bag at time k; Δe(k-1) is the pressure deviation increment in the transition bag at time k-1.

[0118] The pressure in the transition package is required to have good stability and small overshoot during the control process, so the following controller adjustment requirements are obtained:

[0119] (1) When |e(k)|>E max This situation indicates that the absolute value of the deviation is very large. At this time, the controller input is set to the maximum (or minimum) output to quickly adjust the deviation so that the absolute value of the deviation decreases at the fastest speed.

[0120] (2) When |e(k)|>E mid When , it means that the deviation is also large, and the controller needs to implement a stronger control action to achieve the absolute value of the twist deviation to change in the direction of decreasing and quickly reduce the absolute value of the deviation;

[0121] (3) When |e(k)|≤E mid When , it means that although the deviation changes in the direction of increasing absolute value, the absolute value of the deviation itself is not very large. The controller needs to implement general control action, which only needs to reverse the change trend of the deviation and make it change in the direction of decreasing absolute value of the deviation.

[0122] (4) When e(k)Δe(k)<0 and e(k)Δe(k-1)>0 or e(k)=0, it means that the absolute value of the deviation changes in the direction of decrease, or it has reached a balanced state. At this time, the controller output remains unchanged;

[0123] (5) When e(k)Δe(k-1)<0 and e(k)Δe(k)<0, it means that the deviation is in the limit state. If the absolute value of the deviation is large at this time, |e(k)|>E mid , implement stronger control; if the absolute value of the deviation is small at this time, |e(k)| <E mid , implement a weaker control effect;

[0124] (6) When |e(k)|≤E min This situation indicates that the absolute value of the deviation is very small and is caused by the static error of the system. The integral action can be added to eliminate the steady-state error.

[0125] According to the above analysis, the following expert rules are designed:

[0126] Rule 1: When |e(k)|>E max When , the calculation formula is as follows:

[0127]

[0128] Where U(k) is the control quantity at time k; U max is the maximum output value; U min is the minimum output value.

[0129] Rule 2: When |e(k)|>E mid When , the calculation formula is as follows:

[0130] U(k)=U(k-1)+k1×(K p Δe(k)+K i e(k)+K d Δe(k-1)) (14)

[0131] In the formula, U(k-1) is the control quantity at time k-1; k1 is the gain coefficient, k1 is (1,1.5); K p is the proportional adjustment coefficient; K i is the integral adjustment coefficient; K d is the differential adjustment coefficient.

[0132] Rule 3: When |e(k)| <E mid When , the calculation formula is as follows:

[0133] U(k)=U(k-1)+K p Δe(k)+K ie(k)+K d Δe(k-1) (15)

[0134] Rule 4: When e(k)Δe(k)<0 and e(k)Δe(k-1)>0 or e(k)=0, the calculation formula is as follows:

[0135] U(k)=U(k-1) (16)

[0136] Rule 5: When e(k)Δe(k)<0 and e(k)Δe(k-1)<0, the calculation formula is as follows:

[0137] U(k)=U(k-1)+k2K p e(k),|e(k)|>E mid (17)

[0138] U(k)=U(k-1)+k3K p e(k),|e(k)|≤E mid (18)

[0139] Wherein, k2 is the gain coefficient, k2 is (1,5]; k3 is the gain coefficient, k3 is (0,1).

[0140] Rule 6: When |e(k)|≤E min When , the calculation formula is as follows:

[0141] U(k)=U(k-1)+k4×(K p Δe(k)+K i e(k)) (19)

[0142] In the formula, k4 is the gain coefficient, and k4 is (1.4, 2.5].

Claims

1. A method for controlling pressure in a transition bag during the production of ultra-thin amorphous alloys, characterized in that: The following steps are included: S1~The real-time pressure in the transition bag is collected through the pressure sensor, and the collected current signal is amplified and fed back to the PLC controller to obtain the pressure measurement value in the transition bag. The pressure setting value in the transition bag is compared with the measured value to obtain the pressure measurement deviation e1; S2~The PLC controller uploads data to the WinCC host computer, and the WinCC host computer is connected to the computer to calculate the control signal U, which is transmitted to the PLC controller through the communication loop; Step S2 adopts the following method, S21: Establish a pressure prediction model in the transition package, obtain a pressure prediction value in the transition package, compare the set value of the pressure in the transition package with the prediction value, and obtain a pressure prediction deviation e2; S22~According to the pressure measurement deviation e1 and the pressure prediction deviation e2, the pressure deviation e based on the combination of weight factor a is obtained, and the pressure deviation change rate ec is obtained through differential operation; S23~According to the combined pressure deviation e and pressure deviation change rate ec, the expert controller of the deviation combination is used to obtain the incremental coefficient of the PID parameter, which is used to adjust the control ratio adjustment coefficient K p , integral adjustment coefficient K i And the differential adjustment coefficient K d , and accordingly, the PID parameters are adjusted online, and finally, the control quantity U under different modes is obtained according to the combined deviation and deviation change rate of the pressure in the transition bag; S3~The electric proportional valve receives the control signal U sent by the PLC controller, changes the opening of the electric proportional valve, and realizes the pressure regulation in the transition package.

2. The method for controlling pressure in a transition bag during production of an ultra-thin amorphous alloy according to claim 1, characterized in that: Establishing the transition bag internal pressure prediction model includes the following steps: S100 ~ data preprocessing, processing abnormal data in the ultra-thin amorphous alloy production process, and performing data normalization processing; S200~determine the topological structure of the BP neural network. The transition bag pressure prediction model constructed by the BP neural network contains only one hidden layer, in which the number of neurons in the input layer is 7, corresponding to the thickness of the ultra-thin amorphous alloy at 5 different positions, the pressure in the transition bag at the last moment, and the temperature in the transition bag; the number of neurons in the output layer is set to 1, which is used to predict the pressure in the transition bag; S300~adaptive PSO optimizes the weights and thresholds of the BP neural network; S400~screening a training sample set, where the training samples of the neural network are offline data with the greatest similarity to the data under the preset working conditions, where the data under the preset working conditions include a preset ultra-thin amorphous alloy thickness, a preset pressure curve in the transition package, and a set temperature in the transition package.

3. The method for controlling pressure in a transition bag during production of an ultra-thin amorphous alloy according to claim 2, characterized in that: The step S100 adopts the following method: S101~Based on the collected data of actual production site of ultra-thin amorphous alloy, the median method of absolute value difference is used to screen out abnormal values ​​in the data to obtain abnormal data; S102~Use the neighboring data average interpolation method for abnormal data points; S103: Normalizing the actual production site data of the ultra-thin amorphous alloy, including the thickness of the ultra-thin amorphous alloy, the pressure in the transition package, and the temperature in the transition package.

4. The method for controlling pressure in a transition bag during production of an ultra-thin amorphous alloy according to claim 3, characterized in that: The step S300 adopts the following method, using the weight and threshold of the BP neural network as the elements contained in the particles in the adaptive PSO algorithm, determining the particle length according to the input, output and number of layers of the BP neural network, and taking the square sum of the error between the predicted value of the transition bag pressure output by the BP neural network and the original value as the fitness function of the adaptive PSO algorithm. The calculation formula of the fitness function is as follows: Where fit is the fitness value; p j is the original value of the pressure in the transition bag; It is the predicted value of the pressure in the transition bag output by the BP neural network; S301~By calculating the fitness value of each particle, the median fitness value is obtained, and the particles in the population are divided into superior zone particles and inferior zone particles according to the median fitness value; the calculation formulas of the median fitness are as follows: Where, X i is the fitness value of the i-th particle; N is the total number of particles; f 0.5 is the median of particle fitness value; Among them, the particles in the superior zone have a fitness value greater than f 0.5 The particles in the difference zone have a fitness value less than f. 0.5 Particles; S302~For particles in the superior zone, which are close to the optimal position of the population and are prone to fall into the local optimum, a self-mutation update mechanism is designed for them, and n is set t It can adaptively decrease with the increase of the number of iterations, which plays a role in balancing the exploration and development of particles in the superior zone. The calculation formula for updating particles in the superior zone is as follows: n t =1-t / t max Where n t To control the parameters of variable step length; rand(0,1) is a random number uniformly distributed between [0,1]; represents the position of the particle in the superior area; t is the current iteration number; t max is the maximum number of iterations; S303~For the difference area, the particle swarm first updates the speed and position of the particles according to the traditional PSO algorithm, and then updates the population in the difference area to its historical optimal position X g1 and its historical suboptimal position X g2 , and make full use of the difference calculation results between the two to perform local search near the historical optimal position of the population to enhance particle diversity. The specific calculation formula is as follows: X g ′1=X g1 +rand(-1,1)×d t ×(X g1 -X g2 ) d t+1 =1-t / t max In the formula, rand(-1,1) is a random number uniformly distributed between [-1,1]; d t is the local scaling factor at the tth generation; t is the current iteration number; t max is the maximum number of iterations; S304~Result X of local search g ′1 adopts the greedy retention strategy, and the calculation formula is as follows: Where fit(x) is the fitness value of x.

5. The method for controlling pressure in a transition bag during production of an ultra-thin amorphous alloy according to claim 1, characterized in that: In step S22, a weight factor a is set to combine the prediction deviation and the measurement deviation, and the calculation formula is as follows: Where, e is the pressure deviation after compensation using the predicted value; e1 is the pressure measurement deviation; e2 is the pressure prediction deviation; a is the weight factor; is the predicted value of pressure in the transition bag; p i is the pressure setting value in the transition bag; p i ' is the measured value of the pressure inside the transition bag.

6. The method for controlling pressure in a transition bag during production of an ultra-thin amorphous alloy according to claim 5, characterized in that: In step S23, the calculation process of the expert controller is as follows: Assuming that the current sampling is the kth time, the current deviation is e(k), and the deviation of the previous sampling time is e(k-1), and the deviation of the previous two sampling times is e(k-2), then the calculation formula of the two deviation increments can be obtained as follows: Δe(k)=e(k)-e(k-1) Δe(k-1)=e(k-1)-e(k-2) Wherein, e(k) is the pressure deviation in the transition bag at time k; e(k-1) is the pressure deviation in the transition bag at time k-1; e(k-2) is the pressure deviation in the transition bag at time k-2; Δe(k) is the pressure deviation increment in the transition bag at time k; Δe(k-1) is the pressure deviation increment in the transition bag at time k-1; 1) When |e(k)|>E max hour, Where U(k) is the control quantity at time k; U max is the maximum output value; U min is the minimum output value; 2) When |e(k)|>E mid hour, U(k)=U(k-1)+k1×(K p Δe(k)+K i e(k)+K d Δe(k-1)) In the formula, U(k-1) is the control quantity at time k-1; k1 is the gain coefficient, k1 is (1,1.5); K p is the proportional adjustment coefficient; K i is the integral adjustment coefficient; K d is the differential adjustment coefficient; 3) When |e(k)| <E mid hour, U(k)=U(k-1)+K p Δe(k)+K i e(k)+K d Δe(k-1) 4) When e(k)Δe(k)<0 and e(k)Δe(k-1)>0 or e(k)=0, U(k)=U(k-1) 5) When e(k)Δe(k)<0 and e(k)Δe(k-1)<0, U(k)=U(k-1)+k2K p e(k),|e(k)|>E mid U(k)=U(k-1)+k3K p e(k),|e(k)|≤E mid In the formula, k2 is the gain coefficient, k2 is (1,5]; k3 is the gain coefficient, k3 is (0,1); 6) When |e(k)|≤E min hour, U(k)=U(k-1)+k4×(K p Δe(k)+K i e(k)) In the formula, k4 is the gain coefficient, and k4 is (1.4, 2.5].

Images