JP7145755B2 - Plating deposition control device and control method - Google Patents

Description

The present invention relates to a coating weight control device and a control method for coating a steel plate with a desired thickness of a hot dip coating bath in a continuous coating plant of a steel line. The present invention relates to a coating weight control method for safely continuing control by controlling the coating weight on the front and back sides of a steel plate to respective target values and minimizing the risk of contact between a nozzle and a steel plate when controlling the coating weight.

Nozzle pressure and nozzle position are operating elements for controlling the amount of plating deposited on the steel sheet. The position of the nozzle is an operating end for changing the distance between the nozzle and the steel plate (hereinafter referred to as the nozzle gap). In general, it is better to operate the nozzle position from the viewpoint of control response and the glossiness of the plated steel sheet. It is not easy to grasp the distance between the steel plate and the steel plate. For this reason, the nozzle position is often controlled manually by the operator. It was necessary to solve the danger of

As a conventional method for controlling the amount of plating deposited, Patent Document 1 discloses that a sensor for detecting the passage position of the steel plate in the nozzle (steel plate pass position) is additionally provided, and the steel plate pass position detected by the sensor is used. Then, an example of controlling the positions of the front nozzle and the back nozzle to appropriate values for the steel plate is shown.

Further, Patent Document 2 discloses a method of providing means for estimating the steel plate path position, and correcting the nozzle gap or issuing an alarm when the estimated distance between the nozzle and the steel plate falls below a certain value. .

Furthermore, in Patent Document 3, a magnetic force generator that can control the steel plate without contact and a displacement gauge that can detect the steel plate path position are provided at the top and bottom of the nozzle, and the steel plate measured by the displacement gauge is controlled to an appropriate position. In addition, a method of controlling the distance between the nozzle and the steel sheet to a value at which a desired coating amount is obtained is shown.

JP 2008-280587 A JP 2009-275266 A JP-A-3-253549

However, in the method of Patent Document 1, since it is necessary to install a steel plate path position detection sensor, the cost of the control system is high, and maintenance and calibration work for the steel plate path position detection sensor are newly required. I had a problem. In addition, since the steel plate pass position detection sensor is usually provided on the upper part of the nozzle, the steel plate pass position detected by the steel plate pass position detection sensor does not correspond to the steel plate pass position of the nozzle part, which may cause plating adhesion. There was a problem that the quantity accuracy was lowered. Furthermore, the detection of the steel plate pass position in the plating plant is technically difficult due to the vibration of the steel plate and warping in the width direction, and there is a problem that it is difficult to detect the steel plate pass position with high accuracy.

The method of Patent Literature 2 includes means for estimating a change in the steel plate pass position when the plate thickness is changed, and further estimating the steel plate pass position by determining a stable control state. However, the steel plate pass position moves due to the operation of the bath rolls (collecting rolls, stabilizing rolls) and changes in the steel plate tension, in addition to the change in plate thickness. Since Patent Document 2 does not consider this point, there is a problem that the estimation accuracy of the steel plate path position decreases during the period from when the in-bath roll operation or tension change occurs until the stable state of control is established. Furthermore, the relationship between the amount of change in thickness and the amount of movement of the steel sheet depends on the thickness and steel type of the steel sheet currently being processed (current steel sheet) and the steel sheet to be processed next time (next steel sheet), the position of the rolls in the bath, and the value of tension. , etc., affect the operating point. By the way, different steel grades have different hardness and yield strength, which affects the amount of movement of the steel plate pass position. Since the method of Patent Document 2 does not consider this point, there is also a problem that the estimation accuracy of the steel plate path position is lowered due to the fact that the influence of the state quantity is not taken into consideration.

The method of Patent Literature 3 requires a large-scale device for restraining the steel plate, so there is a problem that the system becomes expensive.

Therefore, the problem to be solved by the present invention is to accurately estimate the movement of the steel plate pass position without using a special sensor for detecting the steel plate pass position when automatically controlling the nozzle position, and to obtain the estimation result is to control the nozzle position according to As a result, it is possible to prevent the unbalanced amount of plating on the front and back of the steel sheet, improve the accuracy of the amount of plating, and eliminate the risk of contact between the nozzle and the steel sheet, thereby enabling safe continuation of control. be.

In order to solve the above-described problems, in the plating deposition amount control apparatus of the present invention, the continuously sent steel sheet is immersed in a hot dipping bath bath, the hot dipping bath is attached to the steel plate, and the hot dipping bath is lifted out of the bath. Later, gas is blown onto the steel sheet from front nozzles and back nozzles provided on the front and back sides of the steel sheet to remove the excessively adhered hot dipping bath, and the plating plant is controlled to adhere the hot dipping bath to a desired thickness on the steel sheet. In the coating weight control device, the coating weight prediction model that describes the relationship between the plate speed, the nozzle pressure, the distance between the nozzle and the steel plate, and the coating weight on the steel plate, and the steel plate with reference to the coating weight prediction model A control unit that controls at least one of the nozzle pressure and the nozzle position so that the amount of coating adhered to the plate becomes a desired value, and the plate thickness change of the steel plate through the welding point that is the joint of the steel plate. receive determination results from a first determination unit that determines the tension of the steel plate, a second determination unit that determines the tension of the steel plate, and a third determination unit that determines the position of the bath roll that supports the steel plate in the bathtub, a steel plate path movement amount estimating unit for estimating a movement amount of a steel sheet path position, which is a passage position of the steel sheet at a nozzle height, when at least one determination result changes, and outputting the movement amount to the control unit; The portion shifts the nozzle position by the amount of movement of the steel plate path position output from the steel plate path movement amount estimation portion, and the steel plate path movement amount estimation portion calculates the amount of change in plate thickness, the amount of change in tension, and the amount of change in the bath. At least one deviation amount consisting of the movement amount of the roll position is input, and the thickness of the steel sheet, the steel type of the steel sheet, the tension applied to the steel sheet, and the bath It is characterized by being composed of a neural network that takes as input the state quantity consisting of the position of the middle roll and outputs the steel plate path movement amount.

According to the present invention, the steel plate pass movement amount estimating unit is activated when each of factors such as passage of the welding point, change in tension, and in-bath roll operation, which are the timing at which the steel plate pass position moves, occurs, and the steel plate corresponding to the change in the factor occurs. A movement amount of the pass position is calculated, and the nozzle position control unit shifts the front nozzle position and the back nozzle position in accordance with the steel plate pass movement amount. Therefore, since the relative distance between the nozzle and the steel plate can be kept constant, it is possible to prevent the adhesion amounts on the front and back sides of the steel plate from becoming unbalanced due to the movement of the steel plate path. Furthermore, the risk of contact between the nozzle and the steel plate caused by the steel plate approaching either the front or back nozzle can be reduced.

Furthermore, by simultaneously considering various factors that affect the steel plate path movement amount as inputs to the neural network, it is possible to improve the accuracy of estimating the steel plate path movement amount.

In the adjusting neural network, when the input values for the deviation amounts (thickness difference between the current steel sheet and the next steel sheet, tension difference before and after tension change, and position change amount before and after the bath roll operation) are all 0, the steel plate pass movement amount is Estimates of 0 are guaranteed by its structural features. By constructing the steel plate path moving amount estimating unit with an adjusting neural network, it is possible to highly accurately estimate the subtle moving amount of the steel plate path when the deviation amount is small.

It is an explanatory view showing a plating plant. It is an example of instruction information. This is the processing of the welding point passage determination unit. This is the processing of the tension change determination unit. This is the process of the in-bath roll operation determination unit. It is a configuration of a neural network that configures a steel plate path movement amount estimating unit. This is the processing of the learning database construction unit. The relative relationship between the zero point of the nozzle, the position of the front nozzle, the position of the back nozzle, and the steel plate. . 4 is a configuration diagram of a learning database; FIG. This is the processing of the nozzle position control unit. This is the processing of the nozzle pressure control unit. FIG. 4 is an explanatory diagram when a steel plate path movement amount estimating unit is configured by an adjusting neural network; It is an explanatory view showing a plating deposition amount control device. FIG. 4 is an explanatory diagram when a steel plate path movement amount estimating unit is configured with three adjusting neural networks; This is the processing of the steel plate path movement amount estimation result selection unit. BRIEF DESCRIPTION OF THE DRAWINGS It is explanatory drawing which showed the plating adhesion amount control apparatus of this invention. It is the figure which showed the positional relationship in the horizontal cross section of a nozzle and a steel plate. It is a figure explaining the site|part which measures the plating adhesion amount.

Nozzle gap automatic control can be safely introduced in the coating weight control described in the following examples. Compared to the automatic control of only the nozzle pressure, it is possible to improve the accuracy and response of the coating weight control and improve the surface quality of the steel sheet.

FIG. 1 shows an embodiment of the invention. A coating weight control device 100 (details are shown in FIG. 16) controls a coating plant 150 to deposit a desired thickness of a hot dip coating bath on a steel plate (strip) 151 .

First, the plating plant 150 will be explained. A pot (bathtub) 152 of a plating plant 150 stores a hot-dip plating bath, and steel plates 151 connected at welding points 156 are continuously sent. The steel plate 151 is supported by an in-bath roll 160 and controlled to a constant tension value predetermined for each steel plate between itself and a top roll 161 . As the welding point 156 passes, the tension changes from the tension of the currently processed steel plate 151 (current steel plate, preceding material) to the tension of the next processed steel plate 151 (next steel sheet, succeeding material).
The steel plate 151 is immersed in the hot-dip plating bath, and after being lifted up, gas is blown from nozzles 153 consisting of a front nozzle and a back nozzle provided on the front and back sides of the steel plate 151 to remove excessively adhered hot-dip plating bath. By doing so, the amount of deposited plating is controlled to a desired value. The amount of plating adhered to the steel plate 151 is generally determined by the speed of the steel plate 151 (plate speed), the pressure of the gas blown from the nozzle 153, and the distance between the nozzle 153 and the steel plate 151. Considering the vibration of the steel plate 151, the pressure of the front nozzle and the pressure of the back nozzle are usually the same. The adhesion amount on the front and back can be set to the same value.
Here, the passage position at the nozzle height of the steel plate 151 being plated is hereinafter referred to as "steel plate pass position". The distance between the nozzle and the steel plate path position is the nozzle gap, and if the steel plate path position can be estimated, the relative position of the steel plate viewed from the nozzle can be specified. The front nozzle gap can be calculated from the front nozzle position and the steel plate path position, and the back nozzle gap can be calculated from the back nozzle position and the steel plate path position.
Further, by operating the bath rolls 160, the warpage of the steel sheet 151 in the width direction can be changed. When the steel sheet 151 warps, this causes the coating amount of the steel sheet 151 in the width direction to differ in the width direction. However, the position of the bath rolls 160 should be controlled so that the steel sheet 151 does not warp. can correct the phenomenon of different values.
On the other hand, the steel plate pass position changes due to changes in the thickness of the steel plate 151, changes in tension, and roll operations in the bath. When the steel plate path position changes, the steel plate 151 approaches one of the front and back nozzles and moves away from the other. When the target values of the coating amount on the front and back are the same, it is necessary to control the position of the nozzle 153 so that the steel plate 151 is positioned between the front nozzle and the back nozzle even if the steel plate path moves. Different-thickness plating is also conceivable, in which the plating amount on the front and back is intentionally different. In this case, the nozzle 153 must be controlled to a position that takes into account the difference in the plating amount target value on the front and back. In any case, in order to maintain the balance between the front plating amount and the back plating amount, it is necessary to correctly estimate the amount of movement of the steel plate path at the timing when the steel plate path moves, and adjust the position of the nozzle 153 so that the steel path passes. It is necessary to shift by the amount moved.

The relationship between the amount of plating deposited on the steel plate 151, the plate speed, the nozzle pressure, and the nozzle gap (the distance between the nozzle and the steel plate) is represented by Equation 1, for example. In Equation 1, if the values of the surface of the steel plate are input to P and D, the coating weight on the surface of the steel plate is calculated, and if the value of the back surface of the steel plate is input to P and D, the coating weight of the back surface of the steel plate is calculated. can. Furthermore, by inputting the average value of P on the front and back of the steel sheet and the average value of D on the front and back of the steel sheet, the approximate value of the coating weight averaged over the front and back can be calculated.

In this embodiment, Equation 1 is hereinafter referred to as a plating deposit prediction model. In addition to this, nozzle height, steel sheet temperature, and hot dipping bath temperature may be taken into consideration as the coating amount prediction model. The front and rear steel plates are connected by welding at a welding point 156, and the welding point 156 usually corresponds to a switching point of the coating amount target value. The coating amount detector 155 is a device that measures the amount of coating that is actually attached, detects how much coating is attached to the steel sheet 151 on each of the front and back sides of the steel sheet 151, and outputs the detected amount. In the present embodiment, an example will be described in which measured values are output at three points in the width direction of the front and back of the steel plate 151 (total of six points on the front and back). The coating amount detector 155 is attached at a distance of several tens to hundreds of meters from the nozzle 153, and generally moves the steel sheet in the width direction and outputs the value after averaging. For this reason, it usually takes several tens of seconds to two minutes until the amount of deposited plating corresponding to P, V, and D of the nozzle position can be measured.

FIG. 16 shows the configuration of the plating deposition amount control device 100. As shown in FIG. The coating weight control device 100 receives, from the host computer 140, instruction information including the steel plate number, the steel type, the thickness, the width, the target value of the coating weight, etc., for the steel plate 151 to be processed next, and further controls the coating weight. Receive performance information such as the pressure and position of the nozzle 153, the speed of the steel plate 151, and the actual plating amount detected by the plating adhesion amount detector 155 from the plant 150. From these, the plating amount prediction model 104 is referred to, and the target A control unit 101 is provided for calculating a nozzle position and a pressure command for achieving the plating deposition amount. Furthermore, from the performance information taken in from the plating plant 150, the welding point passage determination unit 106 that determines that the welding point 156 has passed the position of the nozzle 153, the tension change determination unit that determines that the tension of the steel plate 151 has changed 107, a bath roll operation determination unit 108 that determines that the bath roll 160 has been operated by the operator, any of these welding point passage determination unit 106, tension change determination unit 107, and bath roll operation determination unit 108 A steel plate path movement amount estimating unit 105 is provided for estimating the movement amount of the steel plate path according to the determination result, and the nozzle position control unit 103 has a function of shifting the position of the nozzle 153 according to the output of the steel plate path movement amount estimating unit 105. It has

In this embodiment, the steel plate path movement amount estimator 105 is composed of a neural network. For this reason, in order to learn the neural network, a learning database constructing unit 110 for retrieving data necessary for learning from performance information taken from the plating plant 150, and a learning for accumulating data output from the learning database constructing unit 110. A learning unit 112 is provided for learning the neural network provided in the steel plate path movement amount estimation unit 105 using the data accumulated in the database 111 for use and the database 111 for learning.

The function of each part will be described in detail below with reference to the drawings. FIG. 2 shows an example of instruction information received by the plating deposition amount control device 100 from the host computer 140. As shown in FIG. The instruction information 201 is composed of basic information such as the steel plate number of the steel plate to be processed next time, steel type, steel thickness, steel plate length, control target values, etc., and is sent prior to the steel plate being processed. In the example of instruction information in FIG. 2, attribute values such as steel plate number, steel type, plate thickness, and plate width; Contains the operating point of the control. Actually, in addition to this, the chemical composition of the steel sheet, the place of delivery, and the information of the next process may be included.

FIG. 3 shows processing executed by the welding point passage determination unit 106. As shown in FIG. The processing is started at the timing when the welding point 156 passes the position of the nozzle 153, and thereafter the processing is repeated at regular intervals (Δt). A tracking value L is initialized in S3-1. In this flow chart, L indicates the distance from the top of the steel sheet to the plated portion. In S3-2, the sheet speed of the steel sheet 151 is read from the plating plant 150. Then, a value obtained by multiplying the plate speed V by the calculation period Δt is added to L, and L is newly set. In S3-3, it is determined whether or not L is greater than the steel plate length L2 taken in from the instruction information 201. If it is not large, the processing of the steel sheet continues, so after Δt has elapsed, the process returns to S3-2, and the processing of S3-2 to S3-3 is repeated. If it is larger, it indicates that the processing of the steel plate has been completed, so in S3-4, the determination result of passage of the welding point is output to the steel plate path movement amount estimating unit 105. FIG. After that, the process returns to S3-1 to start processing the next steel plate.

FIG. 4 shows the processing executed by the tension change determination unit 107. As shown in FIG. In S4-1, the tension of the steel sheet 151 between the bath roll 160 and the top roll 161 is read from the plating plant 150 and compared with the previously read value. If there is no difference in the comparison result, there is no change in tension, so the process returns to S4-1 and repeats the processes from S4-1 to S4-2. If there is a difference in the comparison result, the tension has changed, so in S4-3 the determination result that the tension has changed is output to the steel plate path movement amount estimating unit 105. FIG. After that, the process returns to S4-1 to monitor the next change in tension.

FIG. 5 shows the processing executed by the in-bath roll operation determining unit 108. As shown in FIG. In S5-1, the position of the bath roll 160 is read from the plating plant 150 and compared with the previously read value. Here, at least one of the bath rolls is movable in the horizontal direction, and the position of the bath roll is the horizontal position of the movable roll or the vertical overlap amount of the two bath rolls. is the amount of intermesh.
If there is no difference in the comparison result, the in-bath roll 160 is not operated, so the process returns to S5-1 and repeats the processes of S5-1 to S5-2. If there is a difference in the comparison result, it means that the bath roll 160 has been operated and the position has changed. output the judgment result. Thereafter, the process returns to S5-1 to monitor the next in-bath roll operation.

FIG. 6 shows the configuration of the steel plate path movement amount estimation unit 105. As shown in FIG. In this embodiment, the steel plate path movement amount estimator 105 is composed of a neural network consisting of an input layer 601 , an intermediate layer 602 having one or more layers, and an output layer 603 . Various types of neural networks have been reported, but in this embodiment, a neural network of a type in which a signal propagates in one direction from the input layer 601 to the output layer 603, which is called a multi-layered type, a hierarchical type, a feedforward type, etc. An example consisting of a net is shown. The intermediate layer 602 is composed of n intermediate layers from the first intermediate layer 610 to the n-th intermediate layer 611 , and each neuron 604 transmits and receives signals through a synapse 605 . The output layer consists of one neuron, and the estimated value of the steel plate path movement amount calculated via the input layer 601 and the intermediate layer 602 is output. Each neuron of the input layer 601 receives information necessary for estimating the steel plate path movement amount. The steel plate path moves by a value corresponding to the amount of change in the thickness of the steel plate 151 when the thickness of the steel plate 151 changes as it passes through the welding point. Further, the steel plate path moves by a value corresponding to the amount of change in tension at the timing when the tension of the steel plate 151 changes. Furthermore, the steel plate path moves by a value corresponding to the amount of movement of the position at the timing when the position of the bath roll 160 changes. As described above, an input neuron is provided to take in the plate thickness change amount, the tension change amount, and the bath roll position change amount. The plate thickness change amount, the tension change amount, and the roll position change amount in the bath are referred to as the deviation amount, and are input to the input layer 601 as the deviation amount 620 . In addition, the state quantity when these change affects the relationship between the deviation value and the steel plate path movement amount as the operating point. In the case of this embodiment, the state quantities are the steel type, thickness, width, position of the bath roll 160, height of the nozzle 153 (distance between the bath roll 160 and the nozzle 153), etc. of the steel plate 151, and input Layer 601 has neurons that capture these as state quantities 621 . Additionally, there may be a threshold neuron 612 to which a constant (1 in this example) is input. These signals input in the input layer 601 are sent to the intermediate layer via synapses. A constant called weight is given to each synapse, and a value obtained by multiplying the weight to the synapse is transmitted. For example, a neuron j in the first intermediate layer receives Kj represented by Equation 2 from the input layer.

A neuron j in the first intermediate layer applies nonlinear processing to the received Kj1, converts it to Lj, and outputs it to the next layer. As examples of non-linear processing, the sigmoid function shown in Equation 3, the Relu function shown in Equations 4 and 5, and the like are used.

Such calculations are sequentially performed for each neuron in a layer close to the input layer 601, and finally, the output of the neuron in the output layer 603 becomes the steel plate path movement amount. Neural network calculation methods are introduced in many booklets, such as "Introduction to Mathematics for Understanding Deep Learning" (Technical Reviewer). By constructing the steel plate path movement amount estimating unit 105 with a neural network, it is possible to consider the effects of many parameters that affect the steel plate path movement amount, and to consider the complicated relationship between each parameter and the steel plate path movement amount. Estimation of the steel plate path movement amount can be made highly accurate. In addition, since the relationship between the input parameters and the steel plate path movement amount can be obtained by learning using the data imported from the plating plant 150, there is no need to manually construct the relationship between the input parameters and the steel plate path movement amount. The load for constructing the control device can be reduced.

Next, a learning method of the neural network provided in the steel plate path movement amount estimation unit 105 will be described. FIG. 7 shows the processing of the learning database constructing unit 110. As shown in FIG. The learning database constructing unit 110 constructs a teacher signal database for learning the neural network provided in the steel plate path movement amount estimating unit 105 . From the performance information obtained from the plating plant 150 in S7-1, the information necessary for estimating the amount of steel plate path movement (plating amount on the front and back, nozzle positions on the front and back) is input to the neural network. It is taken in as data to be input to the layer 601 (hereinafter referred to as input data). The steel plate pass position is calculated in S7-2. The steel plate pass position is the distance between the zero point and the steel plate pass position, as shown in FIG. Calculate as Dp. The zero point in FIG. 8 corresponds to the steel plate pass position when the front and back nozzle positions are zero-adjusted. At this time, the front and back nozzles are equidistant across the zero point. That is, the front and back nozzle gaps are the same value. After that, it is assumed that the steel plate is moved to the position shown in FIG. At this time, the nozzle positions output by the coating weight control device 100 as manipulated variables are Dt and Db based on the zero point, but the actual distances between the nozzle and the plate are D't and D'b. Consider first obtaining D't and D'b and then calculating the steel plate path position Dp. D't and D'b can be calculated by Equations 6 and 7, respectively.

Using this result, the steel plate path position Dp can be calculated by Equation (8).

Alternatively, the steel plate path position Dp can be calculated by Equation (9).

In FIG. 8, Dp is defined as positive when it is on the right side of the zero point and negative when it is on the left side. In S7-3, at the timing when Dp changes, the steel plate path movement amount ΔDp is calculated by subtracting the Dp before the change from the Dp after the change. In S7-4, an editing process is performed to set the input data obtained when the steel plate path position Dp is changed and ΔDp, and the result is output to the learning database 111. FIG.

FIG. 9 shows the configuration of the learning database 111. As shown in FIG. The learning database 111 stores a large number of pairs of input data 901 for the neural network and output data 902 corresponding to the steel plate path movement amount. Further, the input data 901 is composed of a deviation amount 903 consisting of a plate thickness change amount, a bath roll position change amount and a tension change amount, and a state quantity 904 such as steel type and plate thickness. No. 1 indicates that the steel plate pass movement amount is 0 when the plate thickness change amount, the roll position change amount in the bath, and the tension change amount are all 0. In No. 2, when the plate thickness of steel plate 151 with steel grade code 5 is reduced by 0.14 mm from 1.55 mm, the position of the rolls in the bath does not change from 10.6 mm, and the amount of tension change is 0, the steel plate pass is reduced by 0.32 mm. In FIG. 8, it has shown that it moved from the right to the left. The actual input data 901 also includes other items, such as the tension value of the steel plate 151 and the steel type of the next steel plate, as state quantities 904 that affect the steel plate path movement amount.

Learning of the neural network included in the steel plate path movement amount estimation unit 105 is performed by the learning unit 112 .
The learning unit 112 sequentially extracts combinations of the input data 901 stored in the learning database 111 and the corresponding output data 902, and applies the input data 901 to the input layer 601 of the neural network provided in the steel plate path movement amount estimation unit 105. in order. Then, the value of the weight of the synapse 605 is updated so that the output value of the output layer 603 that is the calculation result matches the value of the output data 902 .
This updating process is repeated until the difference between the output value of the output layer 603 and the output data 902 is within a certain range. This is a technique called back propagation among neural network learning methods, and can be referred to in many works, such as "Neural Network Information Processing" (Hideki Aso, Sangyo Tosho). Also, a method using an autoencoder is known as a learning method for a deep neural network with multiple intermediate layers. As described above, there are various learning methods for neural networks, and an appropriate method may be used for learning.

FIG. 10 shows the processing of the nozzle position control section 103 provided in the control section 101. As shown in FIG. The nozzle position control unit 103 performs preset control for setting a nozzle position suitable for obtaining the target adhesion amount according to the instruction information 201 regarding the steel plate 151 received from the host computer 140, and determines the steel plate pass position from the steel plate pass movement amount estimation unit 105. The movement information is taken in, and the nozzle shift control that moves the nozzle 153 consisting of the front nozzle and the back nozzle in parallel by the corresponding value. It has three functions: feedback control to detect and change nozzle positions in a direction to equalize them. First, in S10-1, the activation factor is determined, and which of preset control, nozzle shift control, and feedback control is to be executed is determined. All activation factors can be determined from the performance information acquired from the plating plant 150. For example, the preset control is based on the passage of the nozzle position of the welding point 156, and the nozzle shift control is based on the steel plate path position movement information from the steel plate path movement amount estimation unit 105. , and feedback control may be triggered by the detection of a new coating amount from the coating coating amount detector 155 . When the preset control is activated, the nozzle gap Dn of the next steel sheet is read from the instruction information 201 in S10-2. In S10-3, the nozzle positions Dc1 to Dc4 controlled for the current steel sheet are taken in as performance information from the plating plant 150. In this embodiment, a case where a total of four actuators for controlling the nozzle positions are provided, one for each of the front and rear sides and the width direction, will be described as an example. That is, in FIG. 8, the nozzle position on the front side is Dc1, the back side is Dc2, the nozzle position on the front side of the back nozzle is Dc3, and the back side is Dc4. In S10-4, the nozzle positions Dn1 to Dn4 of the next steel sheet are calculated according to Equations 10 to 13 and output to the plating plant 150. See FIG. 17 for each parameter of Equations 10 to 13 (Dc: nozzle position before movement for the current steel sheet, Dn: nozzle position after movement for the next steel sheet).

In the present embodiment, the trigger for the preset control is the timing when the welding point 156 passes the nozzle position. At this time, the welding point 156 may be set to the timing when it passes the bath roll 160, or it may be set to 5 seconds before passing the nozzle position.
On the other hand, when the activation factor is movement of the steel plate path position, nozzle shift control is performed. At S10-5, the current nozzle positions (nozzle positions Dc1 to Dc4 with respect to the current steel plate) are fetched. Further, in S10-6, the steel plate path movement amount ΔDp is taken in from the steel plate path movement amount estimator 105. FIG. In step S10-7, according to equations 14 to 17, each actuator for operating the nozzles is translated by ΔDp, and command values Dn1 to Dn4 for each nozzle position are calculated and output to the plating plant 150.

When a new coating amount actual value is taken in from the plating coating amount detector 155 and the feedback control is started with this as an activation factor, the actual coating amount value is taken in from the plating coating amount detector 155 in S10-8. . In the present embodiment, a case will be described in which a total of 6 points, ie, 3 points at the center and both ends of the steel sheet 151, are detected as the coating amount. Here, the six detection values are defined as follows.
・TL: Amount of adhesion on the left side of the steel plate surface ・TC: Amount of adhesion on the center of the surface of the steel plate ・TR: Amount of adhesion on the right side of the front surface of the steel plate ・BL: Amount of adhesion on the left side of the back surface of the steel plate ・BC: Amount of adhesion on the center of the back surface of the steel plate • BR: Amount of adhesion on the right side of the back surface of the steel plate In S10-9, the unbalance of the amount of adhesion on the front and back sides and in the width direction is calculated. The imbalance is calculated by Equations 18 and 19, for example. See FIG. 18 for the parameters of Equations 18 and 19.

In the present invention, the plating adhesion amount detector 155 measures the plating adhesion amount by a so-called three-point scanning method. That is, when the plating adhesion amount detector 155 moves in the width direction to detect the plating adhesion amount, the adhesion amount is detected by temporarily stopping at three positions on the left side, the center, and the right side. , the measured values for the left, center, and right points in the width direction are output. That is, as described above, a total of 6 measurement values (TL, TC, TR, BL, BC, BR) are output on the front and back.
In addition, usually, both side average (average value of the above 6 points), front average (average value of TL, TC, TR), back average (average value of BL, BC, BR) are also output. In this case, for example, The U value of Equation 18 can also be calculated using the front average and back average.
As for the operation of a general plating amount detector, in addition to the three-point scanning method, when the full scanning method (the plating amount detector 155 continuously moves in the width direction to detect the plating amount) is used. There is also Even in this case, the present invention can be applied as it is by performing calculations using values detected near TL, TC, TR, BL, BC, and BR.

Then, in S10-10, the nozzle position in the direction to eliminate this imbalance is calculated according to Equations 20 to 23, and is output to the plating plant 150.

Since the plate thickness of the steel plate 151 changes as it passes through the welding point 156, it is conceivable that the preset control and the nozzle shift control are started simultaneously. Even in that case, S10-2 to S10-4 and S10-5 to S10-7 should be executed in order and the results should be integrated. Alternatively, as in Equations 24 to 27, it is conceivable to calculate the nozzle position command value by superimposing Equations 18 and 19 and Equations 20 to 23.

In any case, the present invention can be applied as it is.

FIG. 11 shows the processing of the nozzle pressure control section 102 provided in the control section 101. As shown in FIG. The nozzle pressure control unit 102 captures state changes such as preset control for calculating the nozzle pressure that realizes the coating amount target value indicated by the instruction information 201 for the next steel sheet, and the change in the plate speed of the steel sheet 151. Feedforward control that calculates the nozzle pressure correction amount that compensates for the influence on the plating adhesion amount. It has three functions of feedback control that calculates the correction amount of the nozzle pressure for reducing the . At S11-1, the activation factor is determined, and which of preset control, feedforward control, and feedback control is to be executed is determined. All of the activation factors can be determined from the performance information captured from the plating plant 150. For example, the preset control is the passage of the nozzle position of the welding point 156, the feedforward control is the speed change of the steel plate 151, and the feedback control is the coating amount detector 155. The detection of a new adhesion amount from the starting point may be used as the activation factor for each. When the preset control is activated, the current plate speed Vc is read from the plating plant 150 in S11-2. At S11-3, the target coating amount for the next steel sheet is read from the instruction information 201. In S11-4, the nozzle position setting value for the next steel plate is read from the nozzle position control unit. Instead of the nozzle position setting value of the next steel sheet, the nozzle gap of the next steel sheet taken in from the instruction information may be used. In step S11-5, the model for predicting the amount of deposited plating is referred to, and using the values taken in, the preset value of the nozzle pressure is calculated by Equation 28, and is output as the operation amount of the nozzle 153. FIG.

When the feedforward control is started, at S11-6, the plate speed before and after the change is taken in from the plating plant 150. FIG. Then, in S11-7, the nozzle pressure correction amount for compensating for the speed change is calculated according to Equation 29, and the current nozzle pressure is corrected. The coefficient of influence is the ratio of the nozzle pressure and speed required to increase or decrease the coating weight by a unit amount.

When a new coating amount actual value is taken in from the plating coating amount detector 155 and the feedback control is started with this as an activation factor, the actual coating amount value is taken in from the plating coating amount detector 155 in S11-8. . In S11-9, the deviation is calculated from the target adhesion amount Wn taken in from the instruction information 201, and in S11-10, the nozzle pressure that eliminates the deviation is calculated and output to the nozzle 153 as the operation amount. Specifically, the current nozzle pressure is corrected according to the formula 30. The influence coefficient is the amount of change in the nozzle pressure required to change the coating weight by a unit amount.

As described above, by including the nozzle pressure control unit 102 and the nozzle position control unit 103, the control unit 101 can control the plating adhered to the steel plate 151 to a target value and follow the path movement of the steel plate 151 to control the nozzle position. can be controlled, the risk of contact between the nozzle 153 and the steel plate 151 can be eliminated, and the balance of the coating amount on the front and back can be maintained.

In the present embodiment, the change in the speed of the steel plate 151 is shown as an example of the cause for starting the feedforward control of the nozzle pressure control unit 102, but in addition to this, the operator manually corrects the target value of the coating amount. , it is also conceivable that the distance between the steel plate 151 and the nozzle 153 is changed due to a change in the plate thickness of the steel plate 151. Even in this case, feedforward control can be implemented by a similar method. In the present embodiment, an example was shown in which the sum of the coating amount on both sides in the feedback control was controlled by the nozzle pressure, but it is also possible to change the nozzle position (open/close the nozzle) without changing the pressure. . Even in this case, the processing of the steel plate path movement estimator shown in the present embodiment can be applied as it is. Furthermore, in this embodiment, the learning database construction unit 110, the learning database 111, and the learning unit 112 are provided inside the coating weight control device 100, but a separate computer capable of exchanging signals with the coating weight control device 100 is provided. It is also possible to construct a database for learning by periodically fetching performance data from the coating amount control device 100, learn as necessary, and transmit the learning result to the steel plate path movement amount estimation unit 105. can.

Next, as a second embodiment of the present invention, an embodiment in which the steel plate path movement amount is estimated by an adjusting neural network consisting of two neural networks will be described. The configuration and learning method of the adjusting neural network are described in JP-A-7-121206 and JP-A-8-63203, so they are omitted in this embodiment. Only about is described in detail. First, the input is the amount of change in thickness, the amount of change in tension, the amount of deviation in the amount of change in roll position in the bath, the thickness and steel type of the preceding material (current steel sheet) and succeeding material (next steel sheet), the tension, and the position of the roll in the bath. It is separated into state quantities such as FIG. 12 shows a configuration when the adjusting neural network is applied to the steel plate path movement amount estimation unit 105. As shown in FIG. The adjusting neural network 1202 is composed of two neural networks, a normal neural network 1203 and an error calculation neural network 1204. A value obtained by subtracting the output of the error calculation neural network 1204 from the output of the neural network 1203 is Output as path movement amount.
The neural network 1203 and the error calculation neural network 1204 have the same configuration, and the weight values of the synapses 1232 for transmitting and receiving signals between the neurons 1231 are all the same. The neural network 1203 consists of an input layer 1211 , an intermediate layer 1212 and an output layer 1213 , and the error calculation neural network 1204 consists of an input layer 1241 , an intermediate layer 1242 and an output layer 1243 .
The input of the input layer 1211 of the neural network 1203 is the same as the input of the neural network shown in FIG. ing. On the other hand, in the input layer 1241 of the error calculation neural network 1204, 0 is input instead of the deviation amount to the neuron corresponding to the deviation amount (code 1251), and each neuron corresponding to the other state amount 1252 has a neural The value of the state quantity 1222 same as that of the net 1203 is input. It is obvious that the steel plate pass movement amount should be 0 when the plate thickness change amount, tension change amount, and roll position change amount in the bath are all 0. The value obtained by subtracting the output of the error calculation neural network 1204 from the output of the net 1203 is guaranteed to always be 0 due to its structure. As a result, it is possible to avoid the steady-state deviation of the steel plate pass movement amount, and without being affected by the learning error of the neural network, the steel plate pass It is possible to calculate the amount of movement with high accuracy.

In JP-A-7-121206 and JP-A-8-63203, the input layer neuron that inputs 0 to the neural network 1204 for error calculation was single, but in this embodiment, the plate thickness change amount, tension change amount, Since it is necessary to handle three deviations of the roll position variation at the same time, the configuration is expanded to input 0 to three input layer neurons. By doing so, the adjusting neural network 1202 can be applied to the steel plate path movement amount estimating unit 105 of the coating amount control apparatus 100 targeted in this embodiment.

Next, as a third embodiment of the present invention, in FIG. 13, a steel plate path movement amount is estimated by a plurality of adjusting neural networks, and a plurality of obtained steel plate path movement amounts are selected and used for control. An example to do is shown. A steel plate path movement amount estimation unit 1301 in FIG. 13 includes a plurality of adjusting neural networks for estimating the steel plate path movement amount. selects one from a plurality of steel plate path movement amount estimation results calculated by , and outputs the result to the nozzle position control unit 103 provided in the control unit 101 .

FIG. 14 shows the configuration of the steel plate path movement amount estimation unit 1401. As shown in FIG. A steel plate path movement amount estimation unit 1401, which is an example of the steel plate path movement amount estimation unit 1301, includes a first adjusting neural network 1402, a second adjusting neural network 1403, and a third adjusting neural network 1404. . The first adjusting neural network 1402 is composed of a first neural network 1405 and a first error calculation neural network 1406. is output as the steel plate path movement amount. The deviation input to the first neural network 1405 is only the plate thickness variation, and 0 is input to the corresponding input neuron of the first error calculation neural network 1406 . In addition, state quantities similar to those in the second embodiment are input to the first neural network 1405 and the first error calculation neural network 1406 . Similarly, the second adjusting neural network 1403 is composed of a second neural network 1407 and a second error calculating neural network 1408. A value obtained by subtracting the output of 1408 is output as the steel plate path movement amount. The deviation input to the second neural network 1407 is only the change in tension, and 0 is input to the corresponding input neuron of the second error calculation neural network 1408 . In addition, state quantities similar to those in the second embodiment are input to the second neural network 1407 and the second error calculation neural network 1408 . Furthermore, the third adjusting neural network 1404 is composed of a third neural network 1409 and a third error calculating neural network 1410. The output of the third neural network 1409 is used to generate the third error calculating neural network 1410. The value obtained by subtracting the output is output as the steel plate path movement amount. The deviation input to the third neural network 1409 is only the in-bath roll position change amount, and 0 is input to the corresponding input neuron of the third error calculation neural network 1410 . In addition, state quantities similar to those in the second embodiment are input to the third neural network 1409 and the third error calculation neural network 1410 . As described above, the steel plate path movement amount estimator 1301 calculates the steel plate path movement amount corresponding to each of the plate thickness change amount, the tension change amount, and the bath roll position change amount. Output to the steel plate path movement amount estimation result selection unit 1302 .

In FIG. 13, the processing and configuration of the learning database construction unit 110 and the learning database 111 are the same as in the first embodiment, but the learning unit 1303 includes the first adjusting neural network 1402 and the second adjusting neural network 1402 . The sting neural network 1403 and the third adjusting neural network 1404 each have a function of learning individually. For the learning method, for example, the technique disclosed in Japanese Patent Application Laid-Open No. 8-63203 may be used.

FIG. 15 shows processing executed by the steel plate path movement amount estimation result selection unit 1302 . In S15-1, the determination results of the welding point passage determination unit 106, the tension change determination unit 107, and the bath roll operation determination unit 108 are taken in, and which deviation amount has changed is determined as the activation factor. When it is determined that the starting factor is passing through the welding point, the process advances to S15-2, and the steel plate path movement amount calculated by the first adjusting neural network 1402 is output to the control unit 101 as an option. When it is determined that the activation factor is the tension change, the process proceeds to S15-3, and the steel plate path movement amount calculated by the second adjusting neural network 1403 is output to the control unit 101 as an option. Further, when it is determined that the activation factor is the in-bath roll operation, the process proceeds to S15-4, and the steel plate path movement amount calculated by the third adjusting neural network 1404 is output to the control unit 101 as an option. In this manner, an adjusting neural network is provided for each deviation amount, and selected and used depending on the activation factor.

According to this embodiment, by constructing individual adjusting neural networks for each of the deviation amounts, the configuration of each adjusting neural network is simplified. learning error can be reduced. As a result, it is possible to improve the estimation accuracy of the steel plate path movement amount corresponding to each deviation amount.

In the plating deposition amount control device 100 of the first embodiment described above, the welding point passage determination unit 106 that determines the passage of the welding point, the tension change determination unit 107 that determines the tension change, and the operation of the bath roll A bathing roll operation determining unit 108 for determining, a steel plate path movement amount estimating unit 105 that is activated based on the determination result of these determining units and estimates the amount of movement of the steel plate path position at each activation timing, In addition to the nozzle position opening/closing control, the nozzle position control unit 103 has a function of shifting the front nozzle position and the back nozzle position corresponding to the movement amount of the steel plate path position estimated by the steel plate path movement amount estimation unit 105. .
Furthermore, the steel plate path movement amount estimating unit 105 is configured with a neural network, and a learning database construction unit 110 that selects and edits data for learning the neural network from various data taken in from the plating plant 150, edits A learning database 111 for accumulating acquired data and a learning unit 112 for learning a neural network using the data in the learning database 111 are provided.

In the coating amount control device 100 of the second embodiment, the neural network of the steel plate path movement amount estimating unit 105 is configured to output the subtraction result of the two neural networks described in Japanese Patent No. 3040901 (adjusting A neural network 1202) was used, and the thickness difference between the current steel sheet and the next steel sheet, the tension difference before and after the tension change, and the position change amount before and after the in-bath roll operation were simultaneously input as deviation amounts to be input. The conventional adjusting neural network is described in detail in "Configuration and learning method of adjusting neural network for highly accurate model tuning" (The Institute of Electrical Engineers of Japan Transaction D, Vol. 115, No. 4, 1995). .

It can be widely applied to the control of coating weight in steel processing lines.

REFERENCE SIGNS LIST 100 plating adhesion amount control device 101 control unit 102 nozzle pressure control unit 103 nozzle position control unit 104 plating adhesion amount prediction model 105 steel plate path movement amount estimation unit 106 welding point passage determination unit 107 tension change determination unit 108 bath roll operation determination unit 110 learning database construction unit 111 learning database 112 learning unit 140 host computer 150 plating plant 151 steel plate 153 nozzle 155 coating amount detector 156 welding point 1202 adjusting neural network 1203 neural network 1204 error calculation neural network 1302 steel plate path movement Amount estimation result selection unit

Claims (16)

A continuously sent steel sheet is immersed in a bath of a hot-dip coating bath, the hot-dip coating bath is applied to the steel sheet, and after being lifted out of the bath, gas is supplied to the steel sheet from the front nozzle and the back nozzle provided on the front and back of the steel sheet. is sprayed to remove the excessively adhered hot dipping bath to adhere the hot dipping bath of the desired thickness to the steel sheet.
A coating weight prediction model that describes the relationship between the plate speed, nozzle pressure, distance between the nozzle and the steel plate, and the coating weight that adheres to the steel plate;
a control unit that controls at least one of a nozzle pressure and a nozzle position so that the coating amount on the steel sheet becomes a desired value by referring to the coating amount prediction model;
A first determination unit that determines thickness change of the steel plate through a weld point that is a joint of the steel plate , a second determination unit that determines the tension of the steel plate, and a bath that supports the steel plate in the bathtub. Estimate the amount of movement of the steel plate path position, which is the passage position of the steel plate at the nozzle height, when at least one of the determination results changes from the third determination unit that determines the position of the roll. and a steel plate path movement amount estimating unit that outputs to the control unit,
The control unit shifts the nozzle position by the movement amount of the steel plate path position output from the steel plate path movement amount estimation unit,
The steel plate path movement amount estimator receives as input at least one of a deviation amount consisting of a change amount of the plate thickness, a tension change amount, and a movement amount of the roll position in the bath, and further calculates the relationship between the deviation amount and the steel plate path movement amount. It is composed of a neural network that inputs the state quantities that affect the thickness of the steel plate, the steel type of the steel plate, the tension applied to the steel plate, and the position of the roll in the bath, and outputs the steel plate path movement amount. Featured plating deposit control device. a welding point passage determination unit that determines that the welding point has passed through a specific position of the plating plant;
a tension change determination unit that determines that the tension of the steel plate has changed;
At least one or more in-bath roll operation determination units that determine that the in-bath roll has been operated,
The steel plate path movement amount estimating unit is activated according to the determination result of any one of the welding point passage determination unit, the tension change determination unit, and the bath roll operation determination unit, and estimates the amount of movement of the steel plate path position. The plating deposition amount control device according to claim 1. The steel plate path movement amount estimating unit includes the neural network and an error calculation neural network having the same configuration as the neural network and having a neuron to which the deviation amount is input with zero instead of the deviation amount. wherein the adjusting neural network outputs a value obtained by subtracting the output of the error calculation neural network from the output of the neural network as the steel plate path movement amount. The plating deposit control device according to claim 1 or 2. The control unit includes a nozzle position control unit that controls the nozzle position,
The nozzle position control unit translates the positions of the front nozzle and the back nozzle in the change direction of the steel plate pass position by a value corresponding to the amount of movement of the steel plate pass position output by the steel plate pass movement amount estimation unit. ,
The plating deposit control device according to any one of claims 1 to 3, characterized by: From the plating plant, the pressure and position of the front and back nozzles, the plate thickness, tension, moving speed of the steel plate, the actual value of the plating adhesion amount on the front and back sides attached to the steel plate, the position of the roll in the bath, and the deviation The distance between the front nozzle and the front side of the steel sheet is calculated by inputting a signal containing the front side coating amount of the steel sheet, the front nozzle pressure, and the plate speed into the adhesion amount prediction model. The estimated value, the estimated value of the distance between the back nozzle and the back side of the steel plate calculated by inputting the back side coating amount of the steel plate, the back nozzle pressure, and the plate speed into the above-mentioned coating amount prediction model, the position of the front nozzle, and the back nozzle The steel plate pass position is estimated from the position and stored, and a combination of the steel plate pass movement amount, which is the amount of change when the steel plate pass position changes, and the corresponding deviation amount and state quantity, is edited as a pair of data. a learning database construction department that
a learning database that stores and saves pairs of data edited by the learning database construction unit;
Learning means for sequentially presenting pairs of data stored in the learning database to the neural network so that the neural network learns the relationship between the combination of the deviation amount and the state amount and the steel plate path movement amount. The plating deposit control device according to claim 1, claim 2, or claim 4, comprising: The steel plate path movement amount estimating unit includes the neural network and an error calculation neural network having the same configuration as the neural network and having a neuron to which the deviation amount is input with zero instead of the deviation amount. The adjusting neural network outputs a value obtained by subtracting the output of the error calculation neural network from the output of the neural network as a steel plate path movement amount,
By sequentially presenting pairs of data stored in the learning database to the adjusting neural network, the adjusting neural network is provided with a relationship between the combination of the deviation amount and the state amount and the steel plate path movement amount. 6. The plating deposit control apparatus according to claim 5, further comprising learning means for learning. The steel plate path movement amount estimating unit inputs the amount of change in the plate thickness as a deviation amount, further inputs the plate thickness, steel type, tension, and the position of the roll in the bath as state quantities, and outputs the steel plate path movement amount. A second adjusting neural network that inputs the amount of change in tension as a deviation amount, further inputs the plate thickness, steel type, tension, and the position of the roll in the bath as state quantities, and outputs the steel plate pass movement amount. A neural network and the movement amount of the roll position in the bath are input as deviation amounts, and the plate thickness, steel type, tension, and position of the roll in the bath are input as state quantities, and the steel plate path movement amount is output. 7. The coating weight control device according to any one of claims 1 to 6, comprising at least two neural networks. The plating deposition amount control device is
a welding point passage determination unit that determines that the welding point, which is the joint of the steel sheets, has passed through a specific position of the plating plant;
a tension change determination unit that determines that the tension of the steel plate has changed;
At least one or more in-bath roll operation determination units that determine that the in-bath roll has been operated,
The steel plate path movement amount estimating unit is activated according to the determination result of any one of the welding point passage determination unit, the tension change determination unit, and the bath roll operation determination unit, estimates the amount of movement of the steel plate path position,
The control unit includes a nozzle position control unit that controls the nozzle position,
The nozzle position control unit translates the positions of the front nozzle and the back nozzle in the change direction of the steel plate pass position by a value corresponding to the amount of movement of the steel plate pass position output by the steel plate pass movement amount estimation unit,
The plating deposition amount control device is
The determination results of the welding point passage determination unit, the tension change determination unit, and the in-bath roll operation determination unit are used as activation factors, and the steel plate path movement amount output by the first adjusting neural network and the second adjuster are determined. a steel plate path movement amount estimation result selection unit that selects one from the steel plate path movement amount output by the sting neural network and the steel plate path movement amount output by the third adjusting neural network;
The steel plate path movement amount estimation result selection unit selects the steel plate path movement amount output by the first adjusting neural network when the determination result of the welding point passage determination unit is the activation factor, and selects the steel plate path movement amount output by the first adjusting neural network. is the activation factor, the steel plate path movement amount output by the second adjusting neural network is selected, and when the determination result of the in-bath roll operation determination unit is the activation factor, the third adjusting Selecting the steel plate path movement amount output by the neural network and outputting it to the nozzle position control unit,
The nozzle position control unit translates the positions of the front nozzle and the back nozzle in the moving direction of the steel plate path position by a value corresponding to the amount of movement of the steel plate path position output by the steel plate path movement amount estimation unit. ,
The plating deposition amount control device according to claim 7, characterized by: A continuously sent steel sheet is immersed in a bath of a hot-dip coating bath, the hot-dip coating bath is applied to the steel sheet, and after being lifted out of the bath, gas is supplied to the steel sheet from the front nozzle and the back nozzle provided on the front and back of the steel sheet. and removing the excessively adhered hot dipping bath to adhere a hot dipping bath of a desired thickness to the steel plate.
The coating weight control device includes a coating weight prediction model, a control unit, and a steel plate path movement amount estimation unit.
The coating weight prediction model describes the relationship between the plate speed, nozzle pressure, distance between the nozzle and the steel plate, and the coating weight on the steel plate,
The control unit controls at least one of a nozzle pressure and a nozzle position so that the coating weight deposited on the steel sheet becomes a desired value with reference to the coating weight prediction model,
The steel plate path movement amount estimating unit includes a first determination unit that determines a thickness change of the steel plate via a welding point that is a joint of the steel plates , a second determination unit that determines the tension of the steel plate, and the Receive determination results from a third determination unit that determines the position of the bath roll that supports the steel plate in the bathtub, and at the passage position of the steel plate at the nozzle height when at least one determination result changes estimating the movement amount of a certain steel plate path position and outputting it to the control unit;
The control unit shifts the nozzle position by the movement amount of the steel plate path position output from the steel plate path movement amount estimation unit,
The steel plate path movement amount estimator receives as input at least one of a deviation amount consisting of a change amount of the plate thickness, a tension change amount, and a movement amount of the roll position in the bath, and further calculates the relationship between the deviation amount and the steel plate path movement amount. It is composed of a neural network that inputs the state quantities that affect the thickness of the steel plate, the steel type of the steel plate, the tension applied to the steel plate, and the position of the roll in the bath, and outputs the steel plate path movement amount. A plating adhesion amount control method characterized by: The plating deposition amount control device is
a welding point passage determination unit that determines that the welding point has passed through a specific position of the plating plant;
a tension change determination unit that determines that the tension of the steel plate has changed;
At least one or more in-bath roll operation determination units that determine that the in-bath roll has been operated,
The steel plate path movement amount estimating unit is activated according to the determination result of any one of the welding point passage determination unit, the tension change determination unit, and the bath roll operation determination unit, and estimates the amount of movement of the steel plate path position. The plating adhesion amount control method according to claim 9. The steel plate path movement amount estimating unit includes the neural network and an error calculation neural network having the same configuration as the neural network and having a neuron to which the deviation amount is input with zero instead of the deviation amount. wherein the adjusting neural network outputs a value obtained by subtracting the output of the error calculation neural network from the output of the neural network as the steel plate path movement amount. The plating deposit control method according to claim 9 or 10. The control unit includes a nozzle position control unit that controls the nozzle position,
The nozzle position control unit translates the positions of the front nozzle and the back nozzle in the change direction of the steel plate pass position by a value corresponding to the amount of movement of the steel plate pass position output by the steel plate pass movement amount estimation unit. ,
The plating deposit control method according to any one of claims 9 to 11, characterized by: The plating adhesion amount control device includes a learning database construction unit, a learning database, and a learning means,
The learning database construction unit receives from the plating plant the pressure and position of the front nozzle and the back nozzle, the thickness of the steel plate, the tension, the moving speed, the actual value of the coating amount on the front and back of the steel plate, the bath The position of the roll is taken in, and the deviation amount and the state amount are separated and stored, and a table calculated by inputting a signal containing the front side plating adhesion amount of the steel sheet, the surface nozzle pressure, and the plate speed into the adhesion amount prediction model. The estimated value of the distance between the nozzle and the front side of the steel plate, the estimated value of the distance between the back nozzle and the back side of the steel plate calculated by inputting the back side coating amount of the steel plate, the back nozzle pressure, and the plate speed into the above-mentioned coating amount prediction model, and the table The steel plate pass position is estimated from the nozzle position and the back nozzle position and stored, and the steel plate pass movement amount, which is the amount of change when the steel plate pass position changes, the corresponding deviation amount, and the state quantity Edit the combination of as a pair of data,
The learning database stores and saves pairs of data edited by the learning database construction unit,
The learning means sequentially presents pairs of data stored in the learning database to the neural network so that the neural network is provided with a relationship between a combination of the deviation amount and the state amount and the steel plate path movement amount. 13. The plating deposition amount control method according to claim 9, claim 10, or claim 12, wherein the learning is performed. The steel plate path movement amount estimating unit includes the neural network and an error calculation neural network having the same configuration as the neural network and having a neuron to which the deviation amount is input with zero instead of the deviation amount. The adjusting neural network outputs a value obtained by subtracting the output of the error calculation neural network from the output of the neural network as a steel plate path movement amount,
By sequentially presenting pairs of data stored in the learning database to the adjusting neural network, the adjusting neural network is provided with a relationship between the combination of the deviation amount and the state amount and the steel plate path movement amount. 14. The plating adhesion amount control method according to claim 13, further comprising learning means for learning. The steel plate path movement amount estimating unit inputs the amount of change in the plate thickness as a deviation amount, further inputs the plate thickness, steel type, tension, and the position of the roll in the bath as state quantities, and outputs the steel plate path movement amount. A second adjusting neural network that inputs the amount of change in tension as a deviation amount, further inputs the plate thickness, steel type, tension, and the position of the roll in the bath as state quantities, and outputs the steel plate pass movement amount. A neural network and the movement amount of the roll position in the bath are input as deviation amounts, and the plate thickness, steel type, tension, and position of the roll in the bath are input as state quantities, and the steel plate path movement amount is output. 15. The plating deposit control method according to any one of claims 9 to 14, comprising at least two neural networks. The plating deposition amount control device is
a welding point passage determination unit that determines that the welding point, which is the joint of the steel sheets, has passed through a specific position of the plating plant;
a tension change determination unit that determines that the tension of the steel plate has changed;
At least one or more in-bath roll operation determination units that determine that the in-bath roll has been operated,
The steel plate path movement amount estimating unit is activated according to the determination result of any one of the welding point passage determination unit, the tension change determination unit, and the bath roll operation determination unit, estimates the amount of movement of the steel plate path position,
The control unit includes a nozzle position control unit that controls the nozzle position,
The nozzle position control unit translates the positions of the front nozzle and the back nozzle in the change direction of the steel plate pass position by a value corresponding to the amount of movement of the steel plate pass position output by the steel plate pass movement amount estimation unit,
The plating deposition amount control device is
The determination results of the welding point passage determination unit, the tension change determination unit, and the in-bath roll operation determination unit are used as activation factors, and the steel plate path movement amount output by the first adjusting neural network and the second adjuster are determined. a steel plate path movement amount estimation result selection unit that selects one from the steel plate path movement amount output by the sting neural network and the steel plate path movement amount output by the third adjusting neural network;
The steel plate path movement amount estimation result selection unit selects the steel plate path movement amount output by the first adjusting neural network when the determination result of the welding point passage determination unit is the activation factor, and selects the steel plate path movement amount output by the first adjusting neural network. is the activation factor, the steel plate path movement amount output by the second adjusting neural network is selected, and when the determination result of the in-bath roll operation determination unit is the activation factor, the third adjusting Selecting the steel plate path movement amount output by the neural network and outputting it to the nozzle position control unit,
The nozzle position control unit translates the positions of the front nozzle and the back nozzle in the moving direction of the steel plate path position by a value corresponding to the amount of movement of the steel plate path position output by the steel plate path movement amount estimation unit. ,
The plating deposit control method according to claim 15, characterized by:

Images