Method for carrying out width adaptive learning by utilizing dynamic acquisition function

Abstract

The invention relates to a method for carrying out width adaptive learning by utilizing a dynamic acquisition function. The method comprises the following steps: S1, dynamic beating width measurement; S2, dynamic steel throwing width measurement; and S3, width feedforward of an intermediate pass. According to the technical scheme, the precision of width control is improved, data of the width meter is 100% collected from field data collection, and 100% trust is achieved. The redundancy backup mode of width measurement equipment is utilized to ensure that data can be acquired, the accuracy of the data is realized through the design of a dynamic acquisition mode, and meanwhile, the control mode of the model is designed to ensure that the model can efficiently utilize field data to perform width improvement in width setting.

Description

Technical Field

[0001] The present invention relates to a learning method, in particular to a method for performing width adaptive learning by utilizing a dynamic acquisition function, and belongs to the technical field of improving the rough rolling width accuracy of a hot rolling line. Background Art

[0002] There are two width meters between the racks of Meishan Steel's 1422 hot rolling production line, located behind the reversible mills R1 and R2. Due to the water control between the racks of the reversible mill and the influence of the on-site production environment in different seasons, the width meter needs to be installed far away from the rack to ensure that the width meter will not be unable to work due to the influence of water mist on site. However, there is a problem on site that the width meter is far away and cannot collect data in the middle of the strip, resulting in the opposite control of the width control due to dirty data, and serious over-width and narrowness. This technology is to develop a set of on-site data collection and width control methods to address this contradiction on site.

[0003] After checking, "A method for setting the rough rolling width using a width meter between hot rolling stands" mainly introduces how to apply and set the data in the model after the width meter collects the data, and this method can only collect the last width of the stand for setting. This method is aimed at the problem of model feedback learning errors caused by the remote location of the width meter during on-site collection, which leads to the inability to collect correct data due to on-site reasons. It optimizes the collection and transfers the data to the model setting. An important point in the design of this method is to ensure that the strip can receive correct data in all passes of rough rolling and optimize the model setting. The two methods are completely different.

[0004] "A computer model control method for the width of hot-rolled products" is a method for classifying, reading and storing rough rolling width expansion learning values ​​according to steel type, product width, rough rolling mill, width reduction and entrance thickness. The various factors in this classification method are significantly correlated with the width expansion change, and are not affected by the combination of rough rolling passes and product thickness, which significantly narrows the fluctuation range of the rough rolling width expansion learning value, is suitable for the production of products with various incoming slab thicknesses, can effectively improve the product width control accuracy, and reduce the resulting scrap removal rate and cost losses. This method achieves the improvement of width accuracy by accurately and effectively utilizing the on-site width meter data, which is significantly different from the content of the above-mentioned invention patent and paper. Summary of the invention

[0005] The present invention is aimed at the problems existing in the prior art and provides a method for adaptive width learning using a dynamic acquisition function. This technical solution improves the accuracy of width control, starting from on-site data acquisition, and 100% of the data of the width meter is collected and 100% trusted. The redundant backup method of the width measuring equipment is used to ensure that data can be collected, and the accuracy of the data is achieved through the design of a dynamic acquisition method. At the same time, the control method of the model itself is designed to ensure that the model can efficiently use on-site data to increase the width in width setting.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows: a method for performing width adaptive learning using a dynamic acquisition function, the method comprising the following steps:

[0007] S1, dynamic tapping width measurement,

[0008] S2, dynamic steel casting width measurement,

[0009] S3, intermediate pass width feedforward.

[0010] As an improvement of the present invention, S1, dynamic tapping width measurement, is specifically as follows:

[0011] S11, there are two tapping width measuring devices in front of R1, which are located at the descaling box in front of R1 (device 1) and the front of R1 frame (device 2). During on-site control, the tapping device 2 close to the R1 frame is mainly used. The reason is that tapping device 2 is closer to the R1 rolling mill, that is, the entrance device for width control. The width loss caused by its temperature drop will be the smallest, and the width data at the entrance will be more accurate.

[0012] S12. When the following situations occur on site, switch to tapping device 1 for tapping:

[0013] The second tapping device fails and data cannot be collected on site, so the tapping device one is switched;

[0014] Because of the influence of descaling water between racks, flapping device 2 generates a lot of dirty data, so flapping device 1 is switched;

[0015] When producing shorter materials, the strip cannot stop at the tapping position 2 and is easy to rush into the R1 frame, causing equipment and production scrap. At this time, switch to tapping device 1;

[0016] When two tapping devices need to be calibrated, tapping device 1 and tapping device 2 need to be put into use at the same time;

[0017] In summary, there are several ways to combine two beating devices:

[0018] At the same time, patting device 1 and patting device 2 were put into use;

[0019] The flapping device 1 is put into use and the flapping device 2 is not put into use;

[0020] The second flapping device is put into use and the first flapping device is not put into use;

[0021] S13. The model should be designed accordingly for the use and switching mode of the equipment. When flapping 1 and flapping device 2 are used at the same time, the model remembers the values ​​of the two flapping methods and compares the deviation of the two values. When the deviation is within plus or minus 3mm, the width of flapping 2 is used to set the width of the model. When the deviation of the two values ​​exceeds plus or minus 3mm, the original width entry data needs to be introduced to compare the flapping width with the original width data. When the deviation does not exceed plus or minus 5, the width of the model is set with the smaller flapping value according to the control loss on site.

[0022] When the flapping device 1 is put into use and the flapping device 2 is not put into use, the actual value collected by the flapping device 1 is directly used for control;

[0023] When the flapping device 2 is put into use but the flapping device 1 is not put into use, the actual value collected by the flapping device 2 is directly used for control.

[0024] As an improvement of the present invention, S2, dynamic steel casting width measurement, is as follows:

[0025] S21, according to the detector and speed on site, track the tail position TailPos of the strip in the forward direction, and calculate the length Len of the strip in each pass by using the equal volume method.

[0026] S22, calculate the relative position SGPos of the width gauge and the deviation position DeltaPos when the strip leaves the rolling mill,

[0027] S23. To ensure that the middle width of the strip can be detected, it is necessary to ensure that the middle position exceeds the width meter position. At the same time, in order to save the production rhythm of the rolling line, the tail does not need to be detected, so the rolling mill throwing position is dynamically calculated for different strips.

[0028] As an improvement of the present invention, S3, the width feedforward of the intermediate pass is as follows:

[0029] S31. According to the on-site rolling strategy, configure the feedforward learning mode corresponding to the strategy. There are two reversible rolling mills on site, so there are multiple combinations of corresponding feedforward learning modes. When configuring the feedforward learning, since the reverse pass collection function is not turned on, it is only open in the forward direction.

[0030] S32. According to the on-site conditions, configure the feedforward learning confidence of different areas. Since the number and quality of on-site sampling data of width meters in different areas vary greatly, it is necessary to configure different feedforward learning confidences in different areas, ranging from 0 to 1.

[0031] S33, using the final feedforward learning result as the input of the reverse pass, the width model between passes is calculated. The output width of each pass is finally calculated by combining the input of the feedforward learning with the dog bone and horizontal expansion.

[0032] Compared with the prior art, the present invention has the following advantages. This technical solution uploads the accurate width data of the middle part of the strip to the model control by designing a dynamic steel throwing function during width data collection. This not only makes full use of the equipment on site and maximizes the use of the equipment, but also ensures the accuracy and availability of the width data and improves the accuracy of the model setting. By developing the width feedforward learning function corresponding to the model, it can better meet the production needs and control and adjust each width meter separately. The adjustability and adaptive learning functions are more flexible, ensuring the accuracy of the self-learning of the hot-rolled strip width, greatly improving the width control accuracy of hot rolling, and increasing the yield rate of the strip in the downstream process production. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Make a layout of the equipment for the scene;

[0034] Figure 2 It is the distribution diagram of the width between racks;

[0035] Figure 3 It is a dynamic data collection process. DETAILED DESCRIPTION

[0036] In order to deepen the understanding of the present invention, the present embodiment is described in detail below with reference to the accompanying drawings.

[0037] Example 1: See Figure 2 , Figure 3 , a method for performing width adaptive learning using a dynamic acquisition function, the method comprising the following steps:

[0038] In order to improve the accuracy of width control, this method starts with on-site data collection, collects 100% of the data of the width meter, and is 100% trusted. The redundant backup method of the width measuring equipment is used to ensure that data can be collected, and the accuracy of the data is achieved through the design of dynamic collection methods. At the same time, the control method of the model itself is designed to ensure that the model can efficiently use on-site data to improve the width in width setting.

[0039] S1. Dynamic tapping width measurement function

[0040] S2, Dynamic steel casting width measurement function

[0041] S3, intermediate pass width feedforward function

[0042] Among them, S1, dynamic tapping width measurement function:

[0043] The width of the hot-rolled strip remains basically unchanged after it is drawn out of the heating furnace and before the rough rolling R1 mill bites the steel. 1422 Using the two side guide devices on site, the strip can be tapped and measured in front of the descaling box and the R1 rack respectively through redundant backup. There is a descaling device between the two tapping devices. When in use, tapping is performed at different positions of the strip according to the equipment layout diagram. (See the attached figure for the layout diagram of the tapping and width measurement equipment Figure 1 : On-site beating equipment layout diagram)

[0044] The specific implementation of the dynamic tapping width measurement function is as follows:

[0045] S11, the site mainly uses the second beating device. The reason is that the second beating device is closer to the R1 rolling mill, that is, the entrance device of width control, and the width loss caused by its temperature drop will be the smallest, and the entrance width data is more accurate;

[0046] S12. When the following situations occur on site, switch to tapping device 1 for tapping:

[0047] The second tapping device fails and data cannot be collected on site, so the tapping device one is switched;

[0048] Because of the influence of descaling water between racks, flapping device 2 generates a lot of dirty data, so flapping device 1 is switched;

[0049] When producing shorter materials, the strip cannot stop at the second beating device and is prone to rushing into the R1 frame, causing equipment and production scrap. At this time, switch to the first beating device;

[0050] When two tapping devices need to be calibrated, tapping device 1 and tapping device 2 need to be put into use at the same time;

[0051] In summary, there are several ways to combine two beating devices:

[0052] Put the beating device 1 and beating device 2 into use at the same time; put the beating device 1 into use + beating device 2 is not put into use; put the beating device 2 into use + beating device 1 is not put into use;

[0053] S13. The model should be designed accordingly according to the commissioning and switching methods of the equipment.

[0054] When flapping device 1 and flapping device 2 are put into use at the same time, the model remembers the values ​​of the two flapping methods, and compares the deviation of the two values. When the deviation is within plus or minus 3mm, the width of the model is set using the width of flapping device 2. When the deviation of the two values ​​exceeds plus or minus 3mm, it is necessary to introduce the original width entry data, compare the flapping width with the original width data, and within the range of plus or minus 5, according to the control loss on site, the width of the model is set with the smaller flapping value.

[0055] When the flapping device 1 is put into use and the flapping device 2 is not put into use, the actual value collected by the flapping device 1 is directly used for control;

[0056] When the second beating device is in use and the first beating device is not in use, the actual value collected by the second beating device is directly used for control; S2, dynamic steel casting width measurement function

[0057] The two rough rolling mills now basically meet the function of reversible rolling. The width meter between the stands can maximize the use of resources during the production of the reversible rolling mill, so the width meter between the stands plays a key role in width control during the production process. It is understood that the width meter between the stands of other domestic steel mills can only be used in the calculation of the width of the last pass of each stand, and the use efficiency is not high. For width control, the thickness becomes thinner in the last pass, and the rolling force of the vertical roll limits the correction range of the width, which greatly affects the efficiency of width control.

[0058] One of the important features of this method is to make full use of the width meter and use it in the calculation of the passes involving all vertical rollers, which improves the efficiency by 50% compared with the previous method.

[0059] During hot rolling production, the water between the racks will produce mist as the ambient temperature changes, which affects the authenticity of the width gauge sampling data. Therefore, in order to avoid the influence of mist, the width gauge is slightly moved back. (See attached figure Figure 2 Width between racks (distribution diagram)

[0060] Although this equipment distribution can avoid the influence of fog, the width sampling length is too short and the reliability is not high. When short materials are produced, the collected data is even 0. In order to solve this contradiction, a dynamic steel throwing model of the width meter is designed.

[0061] When controlling the strip width, the strip will lose width at the head and tail in the length direction, so it is necessary to avoid this area when collecting the width, and try to collect the width in the middle of the strip. (See the attached figure for the distance between the strip and the width gauge.) Figure 3 :Dynamic data collection process)

[0062] S21, according to the detector and speed on site, track the tail position TailPos of the strip in the forward direction, and calculate the length Len of the strip in each pass by using the equal volume method

[0063] S22, calculate the relative position of the width gauge SGPos and the deviation position DeltaPos when the strip leaves the rolling mill

[0064] S23. To ensure that the middle width of the strip can be detected, it is necessary to ensure that the middle position exceeds the width meter position. At the same time, in order to save the production rhythm of the rolling line, the tail does not need to be detected, so the rolling mill throwing position is dynamically calculated for different strips.

[0065] S3, intermediate pass width feedforward function

[0066] Since the hot rolling rough rolling area adopts reversible rolling, the model needs to be frequently calculated during the rolling between passes. This algorithm is a function of the model's own error correction. At present, most of the input data for this calculation in hot rolling is the data pre-calculated by the previous model, which has little effect on the width improvement on site. This method makes full use of the data from the pass after the preliminary data collection, screening, cleaning, etc., and substitutes the data into the calculation between passes through the model feedforward learning method to achieve the purpose of correction.

[0067] S31. According to the on-site rolling strategy, configure the feedforward learning mode corresponding to the strategy. There are two reversible rolling mills on site, so there are multiple combinations of corresponding feedforward learning modes. When configuring the feedforward learning, since the reverse pass collection function is not enabled, it is only open in the forward direction.

[0068] S32. Configure the confidence of feedforward learning in different areas according to the on-site conditions. Since the quantity and quality of the on-site sampling data of the width meter in different areas vary greatly, it is necessary to configure different confidence of feedforward learning in different areas, ranging from 0 to 1.

[0069] S33, using the final feedforward learning result as the input of the reverse pass, the width model between passes is calculated. The output width of each pass is finally calculated by combining the input of the feedforward learning with the dog bone and horizontal expansion.

[0070] Embodiment 2:

[0071] 1422 hot rolling width double tapping function, one tapping is placed before R1 descaling, and the other is placed before R1 rolling mill. The two tapping devices can be switched at will or used at the same time. Generally, for short materials, tapping is used before R1 descaling to prevent inaccurate strip positioning from causing only the strip head to be measured or the strip to rush into the rack. The control process is as follows:

[0072] 1. The width of each exit calculated by the model initially

[0073] ps obj Model calculated outlet width 1 e1 1101.9 1 r1 1147 2 r1 1157.4 2 e1 1157.4 3 e1 1100.8 3 r1 1142.3 4 e2 1110.9 4 r2 1136.6 5 r2 1141.3 5 e2 1141.3 6 e2 1139 6 r2 1143.1

[0074] 2. Width deviation of each pass after dynamic measurement on site;

[0075]

[0076] It can be seen from the maximum deviation and minimum deviation that the data deviation after dynamic width measurement and cleaning and screening is reasonable and can truly reflect the actual measured width on site.

[0077] 3. After feedback calculation, the entrance width of each pass is set;

[0078] ps obj Entry width calculated from width meter feedback 1 e1 1180.6 1 r1 1102.3 2 r1 1146.1 2 e1 1157.5 3 e1 1157.4 3 r1 1101.1 4 e2 1136.9 4 r2 1110.9 5 r2 1136.3 5 e2 1141.2 6 e2 1140.9 6 r2 1138.8 .

[0079] It should be noted that the above embodiments are not intended to limit the protection scope of the present invention, and equivalent changes or substitutions made on the basis of the above technical solutions all fall within the protection scope of the claims of the present invention.

Claims

1. A method for performing width adaptive learning using a dynamic acquisition function, characterized in that: The method comprises the following steps: S1, dynamic tapping width measurement, S2, dynamic steel casting width measurement, S3, intermediate pass width feedforward.

2. The method for performing width adaptive learning using a dynamic acquisition function according to claim 1, characterized in that: S1. Dynamic tapping width measurement, details are as follows: S11, there are two flapping width measuring devices in front of R1, which are located at the descaling box in front of R1 (device one) and the front of R1 frame (device two). During on-site control, flapping device two close to the R1 frame is mainly used because flapping device two is closer to R1 rolling mill, i.e. the entrance device for width control, and the width loss caused by its temperature drop will be the smallest, so the width data at the entrance will be more accurate. S12. When the following situations occur on site, switch to tapping device 1 for tapping: The second tapping device fails and data cannot be collected on site, so the tapping device one is switched; Because of the influence of descaling water between racks, flapping device 2 generates a lot of dirty data, so flapping device 1 is switched; When producing shorter materials, the strip cannot stop at the tapping position 2 and is easy to rush into the R1 frame, causing equipment and production scrap. At this time, switch to tapping device 1; When two tapping devices need to be calibrated, tapping device 1 and tapping device 2 need to be put into use at the same time; In summary, there are several ways to combine two beating devices: At the same time, patting device 1 and patting device 2 were put into use; The flapping device 1 is put into use and the flapping device 2 is not put into use; The second flapping device is put into use and the first flapping device is not put into use; S13. The model should be designed accordingly for the use and switching mode of the equipment. When flapping 1 and flapping device 2 are used at the same time, the model remembers the values ​​of the two flapping methods and compares the deviation of the two values. When the deviation is within plus or minus 3mm, the width of flapping 2 is used to set the width of the model. When the deviation of the two values ​​exceeds plus or minus 3mm, the original width entry data needs to be introduced to compare the flapping width with the original width data. When the deviation does not exceed plus or minus 5, the width of the model is set with the smaller flapping value according to the control loss on site. When the flapping device 1 is put into use and the flapping device 2 is not put into use, the actual value collected by the flapping device 1 is directly used for control; When the flapping device 2 is put into use but the flapping device 1 is not put into use, the actual value collected by the flapping device 2 is directly used for control.

3. The method for performing width adaptive learning using a dynamic acquisition function according to claim 2, characterized in that: S2, dynamic steel casting width measurement, details are as follows: S21, according to the detector and speed on site, track the tail position TailPos of the strip in the forward direction, and calculate the length Len of the strip in each pass by using the equal volume method. S22, calculate the relative position SGPos of the width gauge and the deviation position DeltaPos when the strip leaves the rolling mill, S23. To ensure that the middle width of the strip can be detected, it is necessary to ensure that the middle position exceeds the width meter position. At the same time, in order to save the production rhythm of the rolling line, the tail does not need to be detected, so the rolling mill throwing position is dynamically calculated for different strips.

4. The method for performing width adaptive learning using a dynamic acquisition function according to claim 3, characterized in that: S3, the width feedforward of the intermediate pass is as follows: S31. According to the on-site rolling strategy, configure the feedforward learning mode corresponding to the strategy. S32. According to the on-site conditions, configure the confidence of feedforward learning in different areas. Since the quantity and quality of the on-site sampling data of the width meter in different areas vary greatly, it is necessary to configure different confidence of feedforward learning in different areas, ranging from 0 to 1. S33. Based on the final feedforward learning result, as the input of the reverse pass, the width model calculation between passes is performed, and the output width of each pass is finally calculated by combining the input of the feedforward learning through dog bones and horizontal expansion.

Images