Method and system for identifying longitudinal crack defects of continuous casting slab based on crystallizer temperature

Abstract

The application discloses a method for identifying longitudinal crack defects of continuous casting slabs based on a crystallizer temperature, which comprises the following steps: arranging a plurality of rows of thermocouples below a molten steel liquid level position of the crystallizer along a continuous casting pulling speed direction, wherein each row of thermocouples comprises a plurality of thermocouples; collecting temperature data of each thermocouple in m time point windows in real time; calculating average values of the temperature data of each row of thermocouples in each time point window, and obtaining m temperature data average values of each row of thermocouples; performing abnormality detection on the m temperature data average values of each row of thermocouples by using an abnormality detection algorithm, and obtaining the number n of abnormality detection points of each row of thermocouples i ; comparing the number of abnormality detection points of each row of thermocouples along the continuous casting pulling speed direction, and if the number of abnormality detection points of the next row is not less than that of the previous row, it is judged that longitudinal crack defects are generated in the slab, otherwise, it is judged that the slab is normal.

Description

TECHNICAL FIELD

[0001] The present application relates to a slab quality online prediction method, in particular to a slab longitudinal crack online prediction method. BACKGROUND

[0002] It is known that the hot charging and hot sending of continuous casting billets and the continuous rolling technology can greatly reduce the equipment investment and production cost and improve the product competitiveness, which requires that the production line produces defect-free billets, i.e. the surface quality and internal quality of the billets basically meet the requirements of direct rolling without cleaning.

[0003] However, the current continuous casting technology cannot completely eliminate the production of defective billets, so it is often necessary to detect and judge the defective billets, and timely online prediction and detection of billet quality to timely sort off-line or take measures for defective billets. Such a method for detecting and judging the defects of the slab is the key to ensure the production of defect-free billets, which is of great significance to ensure the continuity of production, improve product quality and reduce production cost.

[0004] At present, in the existing continuous casting billet quality inspection method, the traditional cold continuous casting billet quality inspection method has been difficult to meet the demand, and the hot online inspection method has become a necessary means to control the quality of the billet. Among them, the hot online inspection method can include two different detection means, which are divided into online detection judgment based on physical means and continuous casting billet quality prediction judgment based on model.

[0005] However, it should be noted that since the continuous casting process is a high-temperature, rapid and fully enclosed production, the online detection judgment method based on physical means (infrared method, eddy current method and flaw detection method) cannot meet the needs of continuous casting and rolling in the hot online inspection method; this method needs to rely on the defect detection device working in high-temperature harsh environment for online detection and judgment, which is technically complex and expensive, requires a lot of daily maintenance work and mainly detects the surface quality of the billet.

[0006] Therefore, under normal circumstances, the hot online inspection of continuous casting billets can only rely on the continuous casting billet quality prediction judgment system based on the model.

[0007] Among the surface cracks of the continuous casting slab, the longitudinal crack is the most common crack defect, and the surface longitudinal crack usually appears in the middle of the wide surface of the continuous casting slab and is parallel to the casting direction. For the slab crystallizer, the heat flux density should be uniform throughout the width of the crystallizer, especially in the meniscus region, otherwise uneven heat transfer will cause the very thin billet shell to be irregularly separated from the crystallizer wall, and the heat recovery in the separation area will increase the thermal stress, which may cause the continuous casting slab to have a surface longitudinal crack.

[0008] In order to predict the longitudinal cracks on the surface of the continuous casting slab, some researchers have conducted in-depth research and selected to monitor the temperature change of the mold to predict the surface longitudinal cracks. However, the above prediction method still has a deficiency. In the actual production process, the factors affecting the temperature change of the mold are not only longitudinal cracks. The performance of the casting powder, the uniformity of its inflow distribution, other defects on the surface of the casting slab and the like can all affect the temperature value of the embedded thermocouple. Only the temperature deviation in the same column cannot separate the longitudinal cracks from other factors that cause temperature deviation in the same column. This will cause false reporting of longitudinal crack defects, resulting in low prediction accuracy.

[0009] Therefore, in order to overcome this defect, a Chinese patent document with publication number CN1428216A and publication date of July 9, 2003, entitled "A continuous casting slab longitudinal crack prediction method" discloses a continuous casting slab longitudinal crack prediction method. It not only requires the temperature of the adjacent two thermocouples in the same column to start to decrease, but also further requires the continuous decreasing time of the next column to be not less than the continuous decreasing time of the previous row. That is, the temperature decreasing time of the next row is not less than the temperature continuous decreasing time of the previous row, which means that the crack is expanding in the longitudinal direction, so as to be identified as a longitudinal crack.

[0010] In theory, compared with other patents, the above patent technical solution can more accurately distinguish longitudinal cracks from other factors causing temperature drop and reduce false reporting. However, this patent has two deficiencies: (1) It is limited to the condition of stable casting speed, and the fluctuation of casting speed is also one of the reasons for the abnormal temperature; (2) It is stipulated that the temperature drop rate > 3℃ / s is abnormal, but according to different cooling conditions and different steel grades, the temperature drop rate reaching a certain value is a stable longitudinal crack threshold. The above two factors are easy to cause the operator to miss the longitudinal crack when using the continuous casting slab longitudinal crack prediction method.

[0011] In view of the defects and deficiencies in the prior art, the inventors have conducted in-depth research in order to obtain a new and easy-to-implement method and system for identifying longitudinal crack defects of continuous casting slabs.

[0012] In recent years, with the rapid development of information technology, especially the successful application of big data and AI technology in the Internet and medical industries, these information technologies have also begun to be applied in the industry. At present, K-neighbor, SVM and isolated forest are common machine learning anomaly detection algorithms, so these algorithms can be used to detect abnormal points (compared to other points) in real time to obtain the abnormal features of the detected features. These machine learning algorithms based on python language have mature frameworks, and the environment for building the framework can quickly identify abnormal features by calling these anomaly detection algorithms, which is very convenient and fast.

[0013] Based on this, the application expects to obtain a new and easy-to-implement method and system for identifying longitudinal crack defects of continuous casting slabs, which can obtain abnormal features of the crystallizer temperature in real time based on a machine learning anomaly algorithm, and identify longitudinal crack defects of the slab according to the characteristics of the abnormal features.

[0014] The method and system can fully consider the uniqueness of longitudinal crack defects of continuous casting slabs, can effectively avoid false positives and false negatives, and improve the longitudinal crack prediction accuracy, are simple to calculate, easy to implement and transplant, and have very good promotion prospects and application value. SUMMARY

[0015] One of the purposes of the application is to provide a method for identifying longitudinal crack defects of continuous casting slabs based on the temperature of the crystallizer, which fully considers the uniqueness of longitudinal crack defects of continuous casting slabs, can effectively avoid false positives and false negatives, and improve the longitudinal crack prediction accuracy, is simple to calculate, easy to implement and transplant, and has very good promotion prospects and application value.

[0016] In order to achieve the above-mentioned purpose, the application provides a method for identifying longitudinal crack defects of continuous casting slabs based on the temperature of the crystallizer, which includes the following steps:

[0017] A plurality of rows of thermal couples are arranged along the continuous casting speed direction below the molten steel liquid level position of the crystallizer, wherein each row of thermal couples includes a plurality of thermal couples;

[0018] The temperature data of each thermal couple at m time point windows are collected in real time;

[0019] The average value of the temperature data of each row of thermal couples at each time point window is calculated, and each row of thermal couples obtains m temperature data average values;

[0020] An anomaly detection algorithm is used to detect anomalies of the m temperature data average values of each row of thermal couples, and the number n of anomaly detection points of each row of thermal couples is obtained i ;

[0021] The number of anomaly detection points of each row of thermal couples is compared along the continuous casting speed direction, and if the number of anomaly detection points of the next row is not less than that of the previous row, it is judged that the slab has longitudinal crack defects, otherwise, it is judged that the slab is normal.

[0022] In the present application, the inventors have found that, after the longitudinal crack source is formed on the initial shell of the meniscus, an air gap is formed between the slag film and the shell, which increases the interfacial thermal resistance and thus causes abnormal temperature reduction rate in the corresponding area of the crystallizer; as the casting proceeds downward, the air gap caused by the longitudinal crack is located on the shell and thus moves downward gradually during the casting process, which inevitably affects the crystallizer area passed by. Therefore, the abnormal timing of the temperature of the thermocouples arranged in at least two rows along the casting speed direction below the molten steel liquid level position of the crystallizer is one of the necessary conditions for judging whether the longitudinal crack source is generated.

[0023] On the other hand, it should be noted that the longitudinal crack extends in the casting direction with time, otherwise disappears, so the abnormal time of the temperature of the next row of thermocouples in the timing should be no less than the abnormal time of the temperature of the previous row of thermocouples in the timing, which is the second necessary condition for the generation of the longitudinal crack and is also an important feature for distinguishing the longitudinal crack from other factors causing the thermal resistance to increase.

[0024] It should be noted that the above-mentioned "the abnormal time of the temperature of the next row of thermocouples in the timing should be no less than the abnormal time of the temperature of the previous row of thermocouples in the timing" can also be equivalent to the above-mentioned technical solution: "the number of abnormal detection points of the next row is no less than the number of abnormal detection points of the previous row".

[0025] Therefore, based on the two points, the present application can judge the longitudinal crack of the continuous casting slab, which has the advantage of high prediction accuracy. When the system implementing the above method predicts the longitudinal crack and feeds back the L1 system collecting instrument and other device information, preventive measures such as reducing the casting speed and the specific water quantity of secondary cooling can be taken in the subsequent process to slow down the expansion of the longitudinal crack of the continuous casting slab, so as to achieve the purpose of reducing the longitudinal crack defect of the continuous casting slab.

[0026] Further, in the method for identifying the longitudinal crack defect of the continuous casting slab based on the temperature of the crystallizer, the method further comprises the step of outputting an alarm signal when it is considered that the slab has a longitudinal crack defect.

[0027] Further, in the method for identifying the longitudinal crack defect of the continuous casting slab based on the temperature of the crystallizer, the abnormal detection algorithm is a K-neighbor algorithm.

[0028] Further, in the method for identifying the longitudinal crack defect of the continuous casting slab based on the temperature of the crystallizer, the abnormal detection algorithm is a SVM algorithm.

[0029] Further, in the method for identifying the longitudinal crack defect of the continuous casting slab based on the temperature of the crystallizer, the abnormal detection algorithm is an isolated forest algorithm.

[0030] It should be noted that in the method for identifying longitudinal crack defects of continuous casting slabs based on the temperature of the crystallizer, the anomaly detection algorithm can be K-neighbor algorithm, SVM algorithm or isolated forest algorithm. The above algorithms are all anomaly detection algorithms commonly used in the prior art, and will not be described in detail here.

[0031] Correspondingly, another object of the present application is to provide a system for identifying longitudinal crack defects of continuous casting slabs based on the temperature of the crystallizer, which can be used to implement the above-mentioned method of the present application, has wide applicability and has very broad application prospects.

[0032] In order to achieve the above-mentioned object, the present application provides the above-mentioned system for identifying longitudinal crack defects of continuous casting slabs based on the temperature of the crystallizer, which comprises the steps of:

[0033] a plurality of rows of heat couples arranged below the molten steel liquid level position of the crystallizer and arranged along the continuous casting pulling speed direction, wherein each row of heat couples comprises a plurality of heat couples;

[0034] a data acquisition device for collecting temperature data of each heat couple in m time point windows in real time;

[0035] a control device arranged to perform the following steps:

[0036] calculating the average temperature data of each row of heat couples at each time point window based on the temperature data of each heat couple in m time point windows, so that each row of heat couples obtains m average temperature data;

[0037] using an anomaly detection algorithm to perform anomaly detection on the m average temperature data of each row of heat couples, and obtaining the number n of anomaly detection points of each row of heat couples i

[0038] comparing the number of anomaly detection points of each row of heat couples along the continuous casting pulling speed direction, and if the number of anomaly detection points of the next row is not less than that of the previous row, outputting a judgment that the slab has longitudinal crack defects, otherwise, outputting a judgment that the slab is normal.

[0039] Further, in the system for identifying longitudinal crack defects of continuous casting slabs based on the temperature of the crystallizer, an alarm device is further included, which is connected with the control device, wherein when the control device outputs the judgment that the slab has longitudinal crack defects, the alarm device outputs an alarm signal.

[0040] Further, in the system for identifying longitudinal crack defects of continuous casting slabs based on the temperature of the crystallizer, the anomaly detection algorithm is K-neighbor algorithm.

[0041] Further, in the system for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the present application, the anomaly detection algorithm is an SVM algorithm.

[0042] Further, in the system for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the present application, the anomaly detection algorithm is an isolation forest algorithm.

[0043] Compared with the prior art, the method and system for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the present application have the following advantages and beneficial effects:

[0044] In the present application, the method for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the present application is based on a machine learning anomaly algorithm, which can obtain mold temperature anomaly features in real time and identify longitudinal crack defects according to the characteristics of the anomaly features.

[0045] The method for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the present application can improve the accuracy of longitudinal crack prediction: on the one hand, by identifying and comparing the anomaly probabilities of different mold temperatures, longitudinal crack judgment is performed, longitudinal cracks are distinguished from other factors causing temperature anomalies, and the false positive rate is reduced; on the other hand, the judgment of anomaly features is not limited to a constant rate, nor is it limited to the use of conventional variance or temperature drop rate, which can effectively reduce the false negative rate of longitudinal crack prediction.

[0046] The method fully considers the uniqueness of longitudinal crack defects of continuous casting slabs, can effectively avoid false positives and false negatives, and improve the accuracy of longitudinal crack prediction. With the help of the python machine learning library, the calculation is simple, easy to implement and transplant, and has very good popularization prospect and application value.

[0047] Correspondingly, the system for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the present application can be used to implement the above-mentioned method, and also has the above-mentioned advantages and beneficial effects. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The thermocouple positions of the mold of the system for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the present application in one embodiment are schematically shown.

[0049] Figure 2 The implementation flowchart of the method for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the present application in one embodiment is schematically shown. DETAILED DESCRIPTION

[0050] The method and system for identifying longitudinal cracks in continuously cast slabs based on crystallizer temperature, as described in this invention, will be further explained and described below with reference to specific embodiments and accompanying drawings. However, this explanation and description do not constitute an undue limitation on the technical solution of this invention.

[0051] In this invention, the system for identifying longitudinal crack defects in continuously cast slabs based on crystallizer temperature can be used to implement the method for identifying longitudinal crack defects in continuously cast slabs based on crystallizer temperature described in this invention. It has wide applicability and a very broad application prospect.

[0052] Figure 1 The diagram schematically illustrates the thermocouple position in the crystallizer of one embodiment of the system for identifying longitudinal crack defects in continuously cast slabs based on crystallizer temperature, as described in this invention.

[0053] In this invention, the system for identifying longitudinal crack defects in continuously cast slabs based on crystallizer temperature may include: several rows of thermocouples, a data acquisition device, and a control device. The data acquisition device can collect temperature data from each thermocouple over m time points in real time; the control device can be used to implement the steps of the method for identifying longitudinal crack defects in continuously cast slabs based on crystallizer temperature as described in this invention.

[0054] like Figure 1 As shown, in this embodiment, the system of the present invention may include two rows of thermocouples, namely a first row of thermocouples 3 and a second row of thermocouples 2. The first row of thermocouples 3 and the second row of thermocouples 2 are both arranged below the molten steel surface of the crystallizer 1, and the first row of thermocouples 3 and the second row of thermocouples 2 are arranged along the continuous casting speed direction (i.e., the direction shown by the arrow in the figure), wherein each row of thermocouples includes 8 thermocouples.

[0055] Figure 2 The diagram illustrates the implementation process of the method for identifying longitudinal crack defects in continuously cast slabs based on crystallizer temperature according to one embodiment of the present invention.

[0056] In this invention, the method for identifying longitudinal crack defects in continuously cast slabs based on crystallizer temperature can be implemented using the above-described system. The data acquisition device in the system described in this invention can collect temperature data of each thermocouple at 200 (m) time points in real time and perform data analysis.

[0057] Accordingly, the control device in the system can implement the method steps of the present invention for identifying longitudinal crack defects in continuously cast slabs based on crystallizer temperature, which includes the following steps:

[0058] Step 1: Read in the temperature information of two rows of thermocouples from 200 time points in real time.

[0059] Step 2: Calculate the average value of the temperature data of each row of thermocouples at each time point window based on the temperature data of each thermocouple at 200 time point windows, so that each row of thermocouples obtains 200 average values of temperature data, and the calculation formula is as follows:

[0060] B1(i) = (A 11 + A 12 + A 13 + A 14 + A 15 + A 16 + A 17 + A 18 ) / 8

[0061] B2(i) = (A 21 + A 22 + A 23 + A 24 + A 25 + A 26 + A 27 + A 28 ) / 8

[0062] wherein B1(i) represents the average value of the temperature of the first row of thermocouples 3 at the i-th time node, A 11 , A 12 … A 18 respectively represent the value of the single thermocouple of the first row at the i-th time node; B2(i) represents the average value of the temperature of the second row of thermocouples 2 at the i-th time node, A 21 , A 22 … A 28 respectively represent the value of the single thermocouple of the second row at the i-th time node.

[0063] Step 3: Use an anomaly detection algorithm to perform anomaly detection on the 200 average values of temperature data of each row of thermocouples, and obtain the number of anomaly detection points n1, n2 of the two rows of thermocouples.

[0064] It should be noted that in the above step 3, the anomaly detection algorithm can be selected from K-neighbor algorithm, SVM algorithm or isolated forest algorithm. The above algorithms are all common anomaly detection algorithms in the prior art, and will not be described in detail here.

[0065] Step 4: Compare the number of anomaly detection points of each row of thermocouples along the continuous casting speed direction, if the number of anomaly detection points of the next row is not less than that of the previous row, i.e. the number of anomaly detection points of the second row of thermocouples n2 ≥ the number of anomaly detection points of the first row of thermocouples n1, output the judgment that the slab produces longitudinal crack defects, and output an alarm signal, otherwise, output the judgment that the slab is normal.

[0066] Step 5: update the time, that is, the time window can be moved downward by 20 time points to form a new time window slab, and steps 1-4 are performed again on the updated time window to predict whether a crack defect occurs in the new time window slab.

[0067] As can be seen, in the embodiment, the system for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the application can detect real-time abnormalities of two rows of thermocouples in the mold of the slab through the abnormality detection algorithm of machine learning, obtain the number of abnormal detection points (n1, n2) of each row of thermocouples on the slab in a certain time window, and when the number of abnormal detection points n2 of the second row of thermocouples is greater than or equal to the number of abnormal detection points n1 of the first row of thermocouples, it is predicted that the continuous casting slab has a longitudinal crack. The method fully considers the uniqueness of the longitudinal crack, avoids false positives and false negatives, and can effectively improve the accuracy of longitudinal crack prediction. The calculation is simple and easy to implement and promote.

[0068] Of course, in some other embodiments, when the system is provided with a plurality of rows of thermocouples below the molten steel level position of the mold, and each row of thermocouples includes a plurality of thermocouples, the control device in the system can calculate the average temperature data of each row of thermocouples in each time point window based on the temperature data of each thermocouple in m time point windows collected in real time by the data acquisition device, so that each row of thermocouples obtains m average temperature data.

[0069] The control device can further use an abnormality detection algorithm to detect the m average temperature data of each row of thermocouples to obtain the number n of abnormal detection points of each row of thermocouples. i Then, the number of abnormal detection points of each row of thermocouples can be compared in the continuous casting speed direction. If the number of abnormal detection points of the next row is not less than that of the previous row, a judgment that the slab has a longitudinal crack defect is output, and an alarm signal is output. Otherwise, a judgment that the slab is normal is output.

[0070] In summary, the method for identifying longitudinal crack defects of continuous casting slabs based on mold temperature according to the application fully considers the uniqueness of longitudinal crack defects relative to other factors that cause temperature rate to decrease, and improves the accuracy of longitudinal crack prediction. When the system implementing the above method has a longitudinal crack prediction and feeds back the L1 system collecting instrument and other device information, preventive measures such as reducing the casting speed and the specific water quantity of secondary cooling can be taken in the subsequent process to slow down the expansion of the longitudinal crack of the continuous casting slab, so as to achieve the purpose of reducing the longitudinal crack defects of the continuous casting slab.

[0071] The method fully considers the uniqueness of the longitudinal crack of the continuous casting slab, can effectively avoid false positives and false negatives, and improves the longitudinal crack prediction accuracy. With the help of the python machine learning library, the calculation is simple, easy to realize and transplant, and has very good popularization prospect and application value.

[0072] It should be noted that the combination of various technical features in the case is not limited to the combination of the claims in the case or the combination of the specific embodiments described in the case. All technical features described in the case can be freely combined or combined in any way, unless contradictory to each other.

[0073] It should also be noted that the above-mentioned embodiments are only specific embodiments of the present application. Obviously, the present application is not limited to the above-mentioned embodiments, and similar changes or modifications made by those skilled in the art from the content disclosed by the present application are directly derived or easily conceived, and all should belong to the protection scope of the present application.

Claims

1. A method for identifying a longitudinal crack defect of a continuously cast slab based on a temperature of a mold, characterized by, The method comprises the steps of: a plurality of rows of thermocouples are arranged below the molten steel level position of the crystallizer and along the continuous casting speed direction, wherein each row of thermocouples comprises a plurality of thermocouples; temperature data of each thermocouple at m time point windows are collected in real time; average values of temperature data of each row of thermocouples at each time point window are calculated, and each row of thermocouples obtains m average values of temperature data; Adopting the anomaly detection algorithm to each row of thermocouple m temperature data average value carries out the anomaly detection, obtains each row of thermocouple anomaly detection point quantity n i ; the number of abnormal detection points of each row of thermocouples is compared along the continuous casting speed direction, and if the number of abnormal detection points of the next row is not less than that of the previous row, it is judged that the slab has longitudinal crack defects, otherwise, it is judged that the slab is normal.

2. The method for identifying a longitudinal crack defect of a continuous casting slab based on a mold temperature according to claim 1, wherein The method further comprises the step of outputting an alarm signal when it is considered that the slab has longitudinal crack defects.

3. The method for identifying a longitudinal crack defect of a continuous casting slab based on a mold temperature according to claim 1, wherein The abnormal detection algorithm is a K-nearest neighbor algorithm.

4. The method for identifying a longitudinal crack defect of a continuous casting slab based on a mold temperature according to claim 1, wherein The abnormal detection algorithm is a SVM algorithm.

5. The method for identifying a longitudinal crack defect of a continuous casting slab based on a mold temperature according to claim 1, wherein The abnormal detection algorithm is an isolated forest algorithm.

6. A system for identifying longitudinal crack defects of a continuously cast slab based on a temperature of a mold, characterized by, The method comprises: a plurality of rows of thermocouples are arranged below the molten steel level position of the crystallizer and along the continuous casting speed direction, wherein each row of thermocouples comprises a plurality of thermocouples; a data acquisition device that collects temperature data of each thermocouple at m time point windows in real time; a control device configured to perform the following steps: average values of temperature data of each row of thermocouples at each time point window are calculated based on the temperature data of each thermocouple at m time point windows, so that each row of thermocouples obtains m average values of temperature data; Adopting the anomaly detection algorithm to each row of thermocouple m temperature data average value carries out the anomaly detection, obtains each row of thermocouple anomaly detection point quantity n i ; the number of abnormal detection points of each row of thermocouples is compared along the continuous casting speed direction, and if the number of abnormal detection points of the next row is not less than that of the previous row, it is judged that the slab has longitudinal crack defects, otherwise, it is judged that the slab is normal.

7. The system for identifying longitudinal crack defects of a continuous casting slab based on a mold temperature according to claim 6, wherein The method further comprises a warning device connected to the control device, wherein when the control device outputs the judgment that the slab has longitudinal crack defects, the warning device outputs an alarm signal.

8. The system for identifying longitudinal crack defects of a continuous casting slab based on a mold temperature according to claim 6, wherein The abnormal detection algorithm is a K-nearest neighbor algorithm.

9. The system for identifying longitudinal crack defects of a continuous casting slab based on a mold temperature according to claim 6, wherein The abnormal detection algorithm is a SVM algorithm.

10. The system for identifying longitudinal crack defects of a continuous casting slab based on a mold temperature according to claim 6, wherein The abnormal detection algorithm is an isolated forest algorithm.

Images