KR20240082882A - Systrm and method for classifying defects in steel plate using weight value - Google Patents

Abstract

The present invention is to provide a steel sheet defect classification system and method that uses different types of defect classifiers using weights. The steel sheet defect classification system according to an embodiment of the present invention includes a defect detection unit that detects defects in the steel sheet. , a plurality of defect classifiers with different classification methods, each of the plurality of defect classifiers being a defect classification unit for classifying defects of the steel sheet detected by the defect detection unit, and classification results of each of the plurality of defect classifiers of the defect classification unit. The plurality of defect classifiers may include a weighting unit that classifies the final defects of the steel sheet by assigning a weight according to the accuracy of the classification results of each of the plurality of defect classifiers. In the steel sheet defect classification method according to an embodiment of the present invention, the defect detection unit classifies the final defect of the steel sheet. Detecting defects, a defect classification unit comprising a plurality of defect classifiers with different classification methods, each of the plurality of defect classifiers classifying defects of the steel sheet detected by the defect detection unit, a weighting unit classifying the defects It may include classifying the final defect of the steel sheet by assigning a weight to the classification result of each of the plurality of defect classifiers of the classification unit according to the accuracy of the classification result of each of the plurality of defect classifiers.

Description

Steel plate defect classification system and method using weight {SYSTRM AND METHOD FOR CLASSIFYING DEFECTS IN STEEL PLATE USING WEIGHT VALUE}

The present invention relates to a system and method for classifying steel plate defects using weights.

In general, the hot rolling process involves reheating and extracting materials (slabs) supplied from a continuous casting machine to a temperature appropriate for rolling in a heating furnace device, then performing width rolling and thickness rolling in a rough rolling mill, and performing as desired by the consumer in a finishing (finishing) rolling mill. After performing the final thickness rolling into a thick product, the rolled steel sheet is wound in the form of a coil in a winder. Immediately after winding, the tail part is partially inspected to inspect for defects. After winding, the high-temperature coil is cooled and then the coil is removed. It is processed by unwrapping and inspecting the steel plate for defects.

Conventionally, defects are detected and classified by acquiring an image of a steel plate. There is a type of defect classifier that extracts defect characteristics and classifies defects according to the characteristics, or a defect classifier that detects defects and then gives the defect a name in advance. When a similar defect is detected through learning, a different type of defect classifier that assigns a defect name to the defect has been used independently depending on the process or steel sheet.

In addition, in the case of the above-described type of defect classifier, several types of fault classifiers are used in groups to increase the defect classification rate, and in the case of other types of fault classifiers, several types of fault classifiers are similarly used in a grouping method.

The above-described conventional method has the disadvantage of having a low classification rate when detecting and classifying defects because each defect classifier has its own unique characteristics, which results in the occurrence of lot-type mass defects or the occurrence of defects in advance. There are problems that make it difficult to take action.

Republic of Korea Patent Publication No. 10-2018-0123775

According to one embodiment of the present invention, a steel plate defect classification system and method using different types of defect classifiers using weights are provided.

In order to solve the problems of the present invention described above, a steel sheet defect classification system according to an embodiment of the present invention includes a defect detection unit that detects defects in a steel sheet, a plurality of defect classifiers with different classification methods, and the plurality of defects. Each classifier is a defect classification unit that classifies defects in the steel sheet detected by the defect detection unit, and a weight is given to the classification results of each of the plurality of defect classifiers of the defect classification unit according to the accuracy of the classification results of each of the plurality of defect classifiers. It may include a weighting unit that classifies final defects of the steel sheet.

A steel sheet defect classification method according to an embodiment of the present invention includes the step of a defect detection unit detecting defects in a steel sheet, the defect classification unit having a plurality of defect classifiers with different classification methods, and each of the plurality of defect classifiers is provided by the defect detection unit. Classifying defects of a steel sheet detected by a weighting unit, wherein a weighting unit assigns a weight to the classification results of each of the plurality of defect classifiers of the defect classification unit according to the accuracy of the classification results of each of the plurality of defect classifiers to determine the final defect of the steel sheet. It may include a classification step.

According to an embodiment of the present invention, the defect classification rate is improved, thereby improving the quality of the steel sheet.

1 is a schematic configuration diagram of a steel plate defect classification system according to an embodiment of the present invention.
Figure 2 is a schematic flowchart of a method for classifying defects in a steel sheet according to an embodiment of the present invention.
3 and 4 are schematic flow charts showing the defect classification operation of each classifier with different classification methods in the defect classification unit of the steel sheet defect classification system according to an embodiment of the present invention.
Figure 5 is a diagram for explaining the operation of the weighting unit of the steel sheet defect classification system according to an embodiment of the present invention.
Figure 6 is a schematic flowchart showing a weight setting method of the weight setting unit of the weighting unit of the steel sheet defect classification system according to an embodiment of the present invention.
Figure 7 is a schematic flowchart showing a method of storing the average value of defect names for each classifier of the determination unit of the weighting unit of the steel sheet defect classification system according to an embodiment of the present invention.
8 is a diagram illustrating an example computing environment in which a steel plate defect classification system according to an embodiment of the present invention may be implemented.

Hereinafter, with reference to the attached drawings, preferred embodiments will be described in detail so that those skilled in the art can easily practice the present invention.

Figure 1 is a schematic configuration diagram of a steel sheet defect classification system according to an embodiment of the present invention, and Figure 2 is a schematic flow chart of a steel sheet defect classification method according to an embodiment of the present invention.

Referring to FIGS. 1 and 2 , the steel sheet defect classification system 100 according to an embodiment of the present invention may include a defect detection unit 110, a defect classification unit 120, and a weighting unit 130.

The defect detection unit 110 may detect defects in the steel sheet based on images from an optical system, such as a camera that acquires an image of the steel sheet or a light that irradiates light to the steel sheet. The defect detection unit 110 may be at least a part of a steel sheet surface defect detector (SDD). In this case, the defect detection unit 110 may detect surface defects of the steel sheet by acquiring a surface image of the steel sheet, but is limited to this. It doesn't work. For example, images of internal defects of the steel sheet S or surface defects and internal defects of the steel sheet S may be obtained by irradiating ultrasonic waves or electromagnetic waves to the steel sheet (Sa).

The defect classification unit 120 may be provided with a plurality of defect classifiers 121 and 122 having different classification methods, and each of the plurality of defect classifiers 121 and 122 classifies defects in the steel sheet detected by the defect detection unit. I can do it (Sb). Although FIG. 1 shows first and second classifiers 1221 and 122 as a plurality of defect classifiers, this is for illustrative purposes only, and the present invention is not limited thereto, and the configuration and number of the plurality of defect classifiers may vary.

3 and 4 are schematic flow charts showing the defect classification operation of each classifier with different classification methods in the defect classification unit of the steel sheet defect classification system according to an embodiment of the present invention.

First, referring to FIG. 3 together with FIG. 1 , the first defect classifier 121 may classify the type of defect based on the physical characteristic value extracted from the defect detected by the defect detection unit 110. When a defect is detected (Sa1), the first defect classifier 121 extracts physical feature values from the detected defect (Sb1), and conditions 1 to condition n (where n is a natural number) based on the extracted feature values. By setting , you can divide the range of each feature value into specific sections and decide how many conditions to divide it into (Sc1).

For example, if the detected defect is a linear defect among surface defects, physical characteristic values such as width, size, and length of the defect can be extracted, and the ratio of defect width to length calculated to determine how to divide it into several conditions.

Defects can be classified (Sd1) and assigned a defect name (Se1) based on feature values that meet the above-mentioned standard conditions.

Next, referring to FIG. 4 together with FIG. 1, the second defect classifier 122 assigns a defect name to the defect Sa2 previously detected by the defect detection unit 110 to classify the defect type (Sb2). , the defects detected by the detection unit 110 are classified (Se2) using the learned classification model (Sd2), which is trained according to the collected defect group (Sc2), and learned until the operation is completed (Sf2). The defects (Sg2, Sh2) detected by the detection unit 110 can be classified (Sd2) using the classified classification model.

Referring again to FIGS. 1 and 2, the weighting unit 130 classifies each of the plurality of defect classifiers 121 and 122 to the classification results of each of the plurality of defect classifiers 121 and 122 of the defect classification unit 120. The final defect of the steel sheet can be classified (Sd) by assigning a weight set according to the result accuracy (Sc).

Figure 5 is a diagram for explaining the operation of the weighting unit of the steel sheet defect classification system according to an embodiment of the present invention.

Referring to FIG. 5 along with FIG. 1 , the weighting unit 130 may include a weight setting unit 131, a weight calculating unit 132, and a determination unit 133.

The weight setting unit 131 may set a weight according to the classification results of each of the plurality of defect classifiers 121 and 122 of the defect classification unit 120, and the accuracy of the classification results of each of the plurality of defect classifiers 121 and 122. You can set the weight according to. As shown in Figure 5, looking at the table, it can be seen that generally, when each defect is classified, a probability value for being classified as each defect is given. The table shown is an example of the results of defect A being classified by two classifiers, the first defect classifier 121 (e.g., selection tree classifier) and the second defect classifier 122 (e.g., deep It can be seen that the learning classifier) produces different results due to each feature. This shows that, depending on the specific defect, there are classifiers that can classify well, and there are classifiers that are not good at classifying. Although only the first and second defect classifiers are illustrated in Figure 5, the plurality of defect classifiers may include two or more defect classifiers, and the two or more defect classifiers may have different classification methods, or may be the same or A number of defect classifiers with similar classification methods may be included in a group, but are not limited to this.

Figure 6 is a schematic flowchart showing a weight setting method of the weight setting unit of the weighting unit of the steel sheet defect classification system according to an embodiment of the present invention.

Referring to FIG. 6 along with FIGS. 1 and 5 , the weight setting unit 131 of the weighting unit 130 of the steel sheet defect classification system 100 according to an embodiment of the present invention uses each defect classifier 121 and 122. ), the weight can be set according to the accuracy of the defect classification result.

First, the weight setting unit 131 receives the defect classification results of the plurality of defect classifiers 121 and 122 of the defect classification unit 120 (S11) and calculates the defect classification results of each of the plurality of defect classifiers 121 and 122. The accuracy can be analyzed (S12). That is, the weight setting unit 131 analyzes the probability that each of the plurality of defect classifiers 121 and 122 correctly classifies defect A as A defect, and determines the probability that the first defect classifier 121 classifies defect A as A defect. After setting (probability 1A) and the probability (probability 2A) that the second defect classifier 122 classified A defect as A defect (S13), the first weight (W1) regarding the defect result of the first defect classifier 121 ) and a second weight (W2) regarding the defect results of the second defect classifier 122 can be set (S14, S15). Here, the weight setting unit 131 may set the sum of the first weight (W1) and the second weight (W2) to '1' or '100%'. In the weight setting operation of the weight setting unit 131 described above, the first and second weights (W1, W2) of the first and second defect classifiers (121, 122) are exemplified, but are not limited thereto, and for example, defect When the classification unit 12 includes first to Nth classifiers (where N is a natural number of 2 or more), the accuracy of the defect classification results of each of the first to Nth classifiers is analyzed to classify the defects of each defect classifier. Depending on the accuracy of the result, the first to Nth weights (where N is a natural number of 2 or more) may be set, and the sum of the first to Nth weights may be set to '1' or '100%'.

To describe in more detail the first and second weights (W1, W2) of the first and second defect classifiers 121 and 122 in the weight setting operation of the weight setting unit 131 described above, the weight setting unit ( 131) First, the first weight W1 is divided into the probability that the first defect classifier 121 classified A defect as A defect (probability 1A) and the second defect classifier 122 classified A defect as A defect. After setting according to probability (probability 2A) (S14) (W1=(probability 1A/100)/{(probability 1A+probability 2A)/100}), the second weight (W2) is used to classify A defect as A defect. It can be set according to the probability (probability 1A) and the probability (probability 2A) that the second defect classifier 122 classified defect A as defect A or the first weight (W1) (S15) (W2 = (probability 2A / 100 )/{(Probability 1A+Probability 2A)/100} or W2=1.0-W1).

The weight calculation unit 132 multiplies the result of the first defect classifier 121 by the first weight (W1) for a specific defect, defect A, and multiplies the second defect classifier 122 by the second weight (W2). The results are added together with the same classification name.

In this way, you can get the following results:

A defect 85% x W1 + A defect 3% x W2 One)

B defect 10% x W1 + B defect 75% x W2 2)

C-defect 3.0% x W1 + C-defect 20% x W2 3)

D Defect 1.5% x W1 + D Defect 2.0% x W2 4)

EDefect 1.5% x W1 + EDefect 0.5% x W2 5)

Based on the W1 and W2 values derived in this way, the defect name with the highest probability from the final result is assigned as the final defect name of defect A.

For example, if W1 = 0.8 and W2 = 1-0.8 = 0.2, the above-mentioned equations 1) to 5) can be calculated as follows.

Probability of A defect = 0.85*0.8 + 0.03*0.2 = 0.686 = 68.6 % One')

Probability of B defect = 0.1*0.8 + 0.75*0.2 = 0.23 = 23% 2')

C Probability of defect = 0.03*0.8 + 0.2*0.2 = 0.064 = 5.4 % 3')

D Probability of defect = 0.015*0.8 + 0.02*0.2 = 0.016 = 1.6 % 4')

Probability of E defect = 0.015*0.8 + 0.005*0.2 = 0.013 = 1.3 % 5')

The above-described weights may assign different weights to each defect name, or may assign weights to all defect names at once. However, in the present invention, different weights are assigned to each defect classifier for each defect. This is because the ability of each defect classifier to properly classify a specific defect is different.

That is, in equations 1) to 5), the first defect classifier 121 has the highest probability of classifying a specific defect as an A defect at 85%, and conversely, the second defect classifier 122 has a probability of classifying a specific defect as an A defect. It can be seen that it is low at 3%.

So far, statistically, each defect classifier must have an indicator of its classification performance for a specific defect. In the present invention, there is an average value of the classification results of each defect classifier for a specific defect, and a weight can be assigned to the specific defect based on this.

Figure 7 is a schematic flowchart showing a method of storing the average value of defect names for each classifier of the determination unit of the weighting unit of the steel sheet defect classification system according to an embodiment of the present invention.

Referring to FIG. 7 together with FIG. 1, first, a defect classifier (m) and a defect name (n) to be classified are set (S21), and the images collected for each defect name (S22) are classified by each defect classifier (S21). A defect name classified according to S23) can be assigned (S24, S25, S26), and the average value of the defect name for each defect classifier can be stored (S27). Saving the above-mentioned defect name average value can be repeated according to the number of defect name types set (S28), and when the average defect name value of one defect classifier is completed, the storage of the average defect name value of the next defect classifier is repeated and the set defect classifier It can be repeated as many times as necessary (S29, S30, S31). Here, the case where the first and second defect classifiers 121 and 122 are set as an example is used.

Based on the average value for each defect classifier for a specific defect derived as described above, weights can be assigned as follows.

In general, each classifier can show the top five defect names for a specific defect, starting from the most probable. In the present invention, based on the first probability, a weight is assigned to each defect classifier for that specific defect based on the following equation.

Let A be the sum of the top first and second probabilities of the first defect classifier 121 for a specific defect, and let B be the sum of the top first and second probabilities of the second defect classifier 122 for that specific defect. If so, the difference between A and B is called C, and the first and second weights (W1, W2) can be determined according to the range of the C value as follows.

C ≥ 80% -> W1 = 0.8, W2 = 1 - 0.8 = 0.2 6)

C ≥ 60% -> W1 = 0.6, W2 = 1 - 0.6 = 0.4 7)

C ≥ 40% -> W1 = 0.4, W2 = 1 - 0.4 = 0.6 8)

C ≥ 20% -> W1 = 0.2, W2 = 1 - 0.2 = 0.8 9)

According to the above-mentioned formula, the determination unit 133 can output the final defect classification result.

8 is a diagram illustrating an example computing environment in which a steel plate defect classification system according to an embodiment of the present invention may be implemented.

8, an example system 1000 is shown that includes a computing device 1100 configured to implement one or more embodiments described above. For example, computing device 1100 may include a personal computer, server computer, handheld or laptop device, mobile device (mobile phone, PDA, media player, etc.), multiprocessor system, consumer electronics, minicomputer, mainframe computer, Distributed computing environments including any of the above-described systems or devices, etc. are included, but are not limited thereto.

Computing device 1100 may include at least one processing unit 1110 and memory 1120. Here, the processing unit 1110 may include, for example, a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, an application specific integrated circuit (ASIC), or a field programmable gate array (Field Programmable Gate Arrays). FPGA), etc., and may have multiple cores. Memory 1120 may be volatile memory (eg, RAM, etc.), non-volatile memory (eg, ROM, flash memory, etc.), or a combination thereof.

Additionally, computing device 1100 may include additional storage 1130. Storage 1130 includes, but is not limited to, magnetic storage, optical storage, etc. The storage 1130 may store computer-readable instructions for implementing one or more embodiments disclosed in this specification, and other computer-readable instructions for implementing an operating system, application program, etc. may also be stored. Computer-readable instructions stored in storage 1130 may be loaded into memory 1120 for execution by processing unit 1110.

Computing device 1100 may also include input device(s) 1140 and output device(s) 1150. Here, the input device(s) 1140 may include, for example, a keyboard, mouse, pen, voice input device, touch input device, infrared camera, video input device, or any other input device, etc. Additionally, output device(s) 1150 may include, for example, one or more displays, speakers, printers, or any other output devices. Additionally, the computing device 1100 may use an input device or output device provided in another computing device as the input device(s) 1140 or the output device(s) 1150.

Additionally, computing device 1100 may include communication connection(s) 1160 that enable communication with another device (e.g., computing device 1300) via network 1200. Here, communication connection(s) 1160 may include a modem, network interface card (NIC), integrated network interface, radio frequency transmitter/receiver, infrared port, USB connection, or other device for connecting computing device 1100 to another computing device. May contain interfaces. Additionally, communication connection(s) 1160 may include a wired connection or a wireless connection.

Each component of the computing device 1100 described above may be connected by various interconnections such as buses (e.g., peripheral component interconnect (PCI), USB, firmware (IEEE 1394), optical bus structure, etc.) and may be interconnected by a network.

As used herein, terms such as “defect detection unit,” “defect classification unit,” “weighting unit,” and the like generally refer to computer-related entities that are hardware, a combination of hardware and software, software, or software in execution. For example, components such as "fault detection unit", "fault classification unit", "weighting unit", etc. are processes running on a processor, a processor, an object, an executable, a thread of execution, a program, and/or a computer. It may be, but it is not limited to this. For example, both the application running on the controller and the controller can be components. One or more components may reside within a process and/or thread of execution, and a component may be localized on one computer or distributed between two or more computers.

As described above, according to the present invention, two defect classifiers with different advantages are used to utilize the strengths of each defect classifier to increase the defect classification rate and thereby preemptively determine the cause of the defect. Not only can it identify and take action, but it can also prevent the occurrence of mass defects in lots.

Additionally, when producing steel sheets used in the rolling process, they pass through numerous rollers. If a defect occurs in a specific roller and is transferred to the steel sheet, periodic defects occur. At this time, if the defect classification rate is low, not only does periodic inspection not work well, but it also becomes impossible to know which roller the defect occurred in.

Accordingly, according to the present invention, the surface quality of the product can be improved by effectively extracting periodic defects so that it can be known which roller needs to be replaced or cleaned.

Accordingly, the expected effects related to steel sheet quality of the steel sheet defect classification system according to the present invention are as follows.

　 ㅇ Improved classification rate

　 　 　- Before improvement: 85 ~ 90%　

　 　 　- After improvement: 91 ~ 95%

　　 ㅇ Quality improvement due to improved classification rate

　 　 　- Improvement after improvement

　 　 　 . Improved agreement rate: The rate at which the judgment results of the inspector and the steel plate defect classification system match.

. Reduce nonconformity rate

. Reduction in claims

. Increased productivity

The present invention described above is not limited by the above-described embodiments and the accompanying drawings, but is limited by the scope of the patent claims described later, and the configuration of the present invention can be varied without departing from the technical spirit of the present invention. Those skilled in the art can easily see that changes and modifications can be made.

100: Steel plate defect classification system
110: defect detection unit
120: Defect classification unit
121: First defect classifier
122: Second defect classifier
121, 122: Multiple fault classifiers
130: weighting unit
131: Weight setting unit
132: Weight calculation unit
133: Judgment unit

Claims (14)

A defect detection unit that detects defects in the steel sheet;
a defect classification unit comprising a plurality of defect classifiers using different classification methods, each of the plurality of defect classifiers classifying defects of the steel sheet detected by the defect detection unit; and
A weighting unit that classifies the final defect of the steel sheet by assigning a weight to the classification results of each of the plurality of defect classifiers of the defect classification unit according to the accuracy of the classification results of each of the plurality of defect classifiers.
Steel plate defect classification system using weights including.
According to paragraph 1,
Multiple defect classifiers
a first defect classifier that classifies types of defects based on physical feature values extracted from defects detected by the defect detection unit; and
A second defect classifier that learns defects collected in advance and classifies the types of defects detected by the defect detection unit.
Steel plate defect classification system using weights including.
According to paragraph 1,
The weighting unit
a weight setting unit that sets a weight based on classification results of each of the plurality of defect classifiers;
a weight calculation unit that assigns the weight set by the weight setting unit to classification results of each of the plurality of defect classifiers; and
A determination unit that determines the final defect type according to the classification results of each of the plurality of defect classifiers processed by the weight calculation unit.
Steel plate defect classification system using weights including.
According to paragraph 3,
A steel sheet defect classification system using weights, wherein the weight setting unit sets weights according to the accuracy of classification results of each of the plurality of defect classifiers.
According to clause 4,
A steel sheet defect classification system using weights, wherein the weight setting unit sets weights according to the accuracy of each classification result of each of the plurality of defect classifiers.
According to clause 5,
The weight setting unit sets the total weight given to the same classified defect types among the classification results of each of the plurality of defect classifiers to either '1' or '100%'. A steel sheet defect classification system using weights.
According to paragraph 3,
A steel sheet defect classification system using weights, wherein the determination unit determines the classification result with the highest probability among the classification results of each of the plurality of defect classifiers processed by the weight calculation unit as the final classification result.
A defect detection unit detecting defects in the steel sheet;
A defect classification unit comprising a plurality of defect classifiers with different classification methods, each of the plurality of defect classifiers classifying defects of the steel sheet detected by the defect detection unit; and
Classifying the final defect of the steel sheet by a weighting unit assigning a weight to the classification results of each of the plurality of defect classifiers of the defect classification unit according to the accuracy of the classification results of each of the plurality of defect classifiers.
Steel plate defect classification method using weights including.
According to clause 8,
The steps to classify defects in steel plates are:
Classifying the type of defect by a first defect classifier of the plurality of defect classifiers based on physical feature values extracted from the defect detected by the defect detection unit; and
A second defect classifier of the plurality of defect classifiers classifies the type of defect detected by the defect detection unit by learning defects collected in advance.
Steel plate defect classification method using weights including.
According to clause 8,
The step of classifying the final defects of the steel plate is
A weight setting unit of the weighting unit setting a weight based on classification results of each of the plurality of defect classifiers;
assigning the weight set by the weight setting unit to the classification result of each of the plurality of defect classifiers; and
A determination unit of the weighting unit determining a final defect type according to classification results of each of the plurality of defect classifiers processed by the weight calculation unit.
Steel plate defect classification method using weights including.
According to clause 10,
In the step of setting the weight, the weight setting unit sets the weight according to the accuracy of the classification result of each of the plurality of defect classifiers.
According to clause 11,
In the step of setting the weight, the weight setting unit sets the weight according to the accuracy of each classification result of each of the plurality of defect classifiers.
According to clause 12,
The step of setting the weight uses a weight in which the weight setting unit sets the total weight given to the types of defects that are equally classified among the classification results of each of the plurality of defect classifiers to either '1' or '100%'. Steel plate defect classification method.
According to clause 10,
In the step of determining the final defect type, the determination unit determines the classification result with the highest probability among the classification results of each of the plurality of defect classifiers processed by the weight calculation unit as the final classification result. A steel sheet defect classification method using weights.

Images