A method for on-line detection and identification of cracks defects in continuous casting slabs

Abstract

This invention discloses an online method for detecting and identifying cracks in continuously cast slabs, comprising the following steps: real-time collection of instrument detection information of the slab in the ladle, tundish, crystallizer, primary cooling, secondary cooling and air cooling zones at m time points to construct a dataset of dimension m×n, where n is the number of instrument detection information collected at each time point; for each time point, performing: (1) based on the dataset, using an anomaly detection algorithm to obtain the anomaly value I of each time point. i (2) Obtain the anomaly probability P of the slab at that time point: P = p / m, where p is the outlier value I. i (3) Collect steel grade information in real time and obtain the abnormal threshold S of the steel grade based on the steel grade information; (4) Compare the abnormal probability P of the steel grade with the abnormal threshold S. If P≥S, output the judgment that the production abnormality probability of the steel grade will lead to slab crack defects; otherwise, output the judgment that the slab is normal.

Description

Technical Field

[0001] This invention relates to a method for inspecting and identifying the quality of slabs, and more particularly to a method for detecting and identifying crack defects in slabs. Background Technology

[0002] As is well known, hot charging and hot delivery of continuously cast billets and continuous rolling technology can significantly reduce equipment investment and production costs, and improve product competitiveness. It requires that the production line produces defect-free billets, that is, the surface quality and internal quality of the billets can meet the requirements of direct rolling without cleaning.

[0003] However, current continuous casting technology cannot completely eliminate the generation of defective slabs. Therefore, it is often necessary to detect and judge defective slabs, and to predict and monitor slab quality online in a timely manner during production, so as to sort out defective slabs or take appropriate measures. This method of detecting and judging slab defects is the key to ensuring defect-free slab production technology, and it is of great significance for ensuring production continuity, improving product quality, and reducing production costs.

[0004] Currently, traditional cold-state continuous casting billet quality inspection methods are no longer sufficient to meet the requirements, and hot-state online inspection methods have become an essential means of billet quality control. Hot-state online inspection methods can be divided into physical-based online detection and judgment and model-based continuous casting billet quality prediction and judgment systems.

[0005] However, due to the high-temperature, rapid, and fully enclosed nature of continuous casting, physical-based online inspection methods (infrared spectroscopy, eddy current testing, and flaw detection) are insufficient for the needs of continuous casting and rolling. These methods require defect detection devices operating in harsh, high-temperature environments, which are technically complex, expensive, and require extensive daily maintenance, primarily focusing on inspecting the surface quality of the cast billet. Therefore, under normal circumstances, online hot inspection of continuously cast billets relies on model-based quality prediction and assessment systems.

[0006] In recent years, with the rapid development of information technology, especially the successful application of big data and AI technologies in industries such as the internet and healthcare, these technologies have begun to be applied in industry. Currently, some researchers have begun to apply machine learning / deep learning methods to the online monitoring of longitudinal cracks in cast billets.

[0007] For example, Chinese patent document CN102937784A, published on February 20, 2013, entitled "Control Method for Online Prediction of Slab Quality Based on Artificial Neural Network", discloses a control method for online prediction of slab quality based on artificial neural network. The method first learns and trains a three-layer BP neural network through training samples to obtain a well-trained and validated network model; secondly, it uses the well-trained and validated BP neural network model to control the online prediction of cracks in the middle of continuously cast slabs.

[0008] For example, three Chinese patent documents, namely: CN111618265A, published on September 4, 2020, entitled "An Online Detection Method for Longitudinal Cracks in Continuous Casting Billets Based on K-Nearest Neighbor Classification"; CN111666710A, published on September 15, 2020, entitled "A Method for Predicting Longitudinal Cracks in Continuous Casting Billets Using Logistic Regression Classification"; and CN111680448A, published on September 18, 2020, entitled "A Method for Predicting Longitudinal Cracks in Continuous Casting Billets Based on Support Vector Machine (SVM) Classification", respectively disclose methods for predicting longitudinal cracks in continuous casting billets using different machine learning classification algorithms (K-Nearest Neighbor, Logistic Regression, and Support Vector Machine (SVM)).

[0009] These three different methods can obtain temperature samples and a sample library by splicing the temperature change rate of longitudinal cracks and the temperature of the thermocouples in the same row under normal operating conditions. The sample library and the preprocessed results of the online real-time detection of the temperature of the thermocouples in the same row can be classified by the classification algorithm, thereby enabling the identification and prediction of longitudinal cracks in the continuously cast billet.

[0010] Therefore, it is evident that the publicly available patents applying machine learning algorithms all require offline or online learning based on sample libraries for slab quality classification. Their underlying logic for adaptability is that the samples can represent the population, meaning the patterns in the samples can be extended to online data. However, in actual continuous casting production, the condition of the continuous casting equipment changes over time, making it difficult to use historical samples that match the equipment's condition at that time to represent the quality of the equipment at that current state.

[0011] Based on this, and addressing the shortcomings and deficiencies of existing technologies, this invention aims to provide a novel method for online detection and identification of cracks in continuously cast slabs. Unlike the aforementioned published literature, the online detection and identification method for cracks in continuously cast slabs proposed in this invention is an anomaly detection method that does not require sample learning. This method uses an anomaly detection algorithm to detect anomalies in the slab in real time, obtaining the anomaly probability on the slab within a certain time window to characterize the fluctuation of production parameters. Furthermore, it compares the anomaly probability with an anomaly threshold. When the anomaly probability exceeds the anomaly threshold, it predicts that cracks will occur in that segment of the slab, thus identifying and predicting cracks in continuously cast slabs.

[0012] The online detection and identification method for cracks in continuously cast slabs described in this invention is relatively simple and widely applicable. It does not require the addition of additional detection equipment and can use existing detection instruments to predict the presence of cracks in continuously cast slabs in a timely manner, so as to sort them off the production line or take measures. It is of great significance for ensuring the continuity of production, improving product quality and reducing production costs, and has a very broad application prospect.

[0013] This online method for detecting and identifying cracks in continuously cast slabs does not require learning from historical samples, effectively preventing errors in identifying abnormal points due to differences in equipment conditions and improving the accuracy of slab defect prediction. Summary of the Invention

[0014] One of the objectives of this invention is to provide a method for online detection and identification of cracks in continuously cast slabs. This method is relatively simple and widely applicable. It does not require the addition of extra detection equipment and can use existing detection instruments to predict the presence of cracks in continuously cast slabs in a timely manner. This is of great significance for ensuring the continuity of production, improving product quality, and reducing production costs, and has a very broad application prospect.

[0015] The online detection and identification method for cracks in continuously cast slabs described in this invention can detect and determine the presence of cracks in continuously cast slabs. It can predict and detect the quality of cast slabs in a timely manner during production, so that continuously cast slabs with cracks can be sorted and removed from the production line in a timely manner or measures can be taken to ensure the production of defect-free cast slabs, thereby effectively improving the hot charging and hot delivery rate.

[0016] To achieve the above objectives, this invention proposes a method for online detection and identification of cracks in continuously cast slabs, comprising the following steps:

[0017] The instrument detection information of slabs in the ladle, tundish, crystallizer, first cooling, second cooling and air cooling zones is collected in real time at m time points to construct a dataset with dimension m×n, where n is the number of instrument detection information collected at each time point window;

[0018] For each time window, the following steps are performed:

[0019] (1) Based on the dataset, an anomaly detection algorithm is used to obtain the outlier value I for each time window. i (x1,x2,...,x n ); where outlier I i (x1,x2,...,x n This corresponds to whether time point window i is considered normal or abnormal in the dataset;

[0020] (2) Obtain the anomaly probability P of the slab at that time point:

[0021] P = p / m, where p is the outlier I. i The number of anomalies;

[0022] (3) Collect steel grade information in real time and obtain the abnormal threshold S of the steel grade based on the steel grade information;

[0023] (4) Compare the abnormal probability P of the steel grade with the abnormal threshold S. If P≥S, output the judgment that the abnormal probability of the steel grade will lead to slab crack defects; otherwise, output the judgment that the slab is normal.

[0024] The online detection and identification method for cracks in continuously cast slabs proposed in this invention is an anomaly detection method that does not require sample learning. This method can perform real-time anomaly detection on slabs. Specifically, it uses an anomaly detection algorithm to identify anomalies in parameters of the continuous casting production process on the slab within a certain time window based on "self-comparison" rather than "historical data." This allows for the acquisition of anomaly probabilities on the slab within a specific time window, which characterize the fluctuation of production parameters. Furthermore, the anomaly probabilities are compared with a pre-set anomaly threshold. When the anomaly probability exceeds the threshold, cracks are predicted to occur in that segment of the slab, thus identifying and predicting cracks in the continuously cast slab.

[0025] Therefore, the online detection and identification method for cracks in continuously cast slabs described in this invention does not require learning from historical samples, which can effectively prevent errors in identifying abnormal points due to different equipment conditions and improve the accuracy of slab defect prediction.

[0026] Furthermore, in the method for online detection and identification of crack defects in continuously cast slabs described in this invention, step (3) further includes: when outputting the judgment that the probability of a production abnormality occurring in this steel grade will lead to crack defects in the slab, an alarm signal is output at the same time.

[0027] Furthermore, in the method for online detection and identification of crack defects in continuously cast slabs described in this invention, in step (1), the anomaly detection algorithm is one of the following: K-nearest neighbor algorithm, SVM algorithm, or isolated forest algorithm.

[0028] Furthermore, in the method for online detection and identification of crack defects in continuously cast slabs described in this invention, the instrument detection information comes from the L1 computer system.

[0029] Furthermore, in the method for online detection and identification of crack defects in continuously cast slabs described in this invention, the steel grade information comes from the L2 computer system.

[0030] Furthermore, in the method for online detection and identification of crack defects in continuously cast slabs described in this invention, the steel grade information includes steel grade markings and steel grade composition information.

[0031] Furthermore, in the method for online detection and identification of crack defects in continuously cast slabs described in this invention, the instrument detection information includes at least two of the following: tundish weight, tundish liquid level, thermocouple temperature of each region on the crystallizer, crystallizer heat flux density, slab casting speed, crystallizer cooling water temperature, and outlet pressure.

[0032] Furthermore, in the method for online detection and identification of crack defects in continuously cast slabs described in this invention, the instrument detection information also includes argon pressure and stopper rod position.

[0033] Furthermore, in the method for online detection and identification of crack defects in continuously cast slabs described in this invention, the abnormal threshold S has a value range of (0,1).

[0034] Compared with existing technologies, the online detection and identification method for cracks in continuously cast slabs described in this invention has the following advantages and beneficial effects:

[0035] Unlike existing technologies, the crack prediction proposed in this invention is based on an anomaly detection method that does not require sample learning. This online detection and identification method for cracks in continuously cast slabs can perform real-time anomaly detection on the slab. Specifically, it uses an anomaly detection algorithm to identify anomalies in parameters of the continuous casting production process on the slab within a certain time window based on "self-comparison" rather than "historical data." This allows the determination of the anomaly probability of the slab in production within a certain time window. When the anomaly probability exceeds a set anomaly threshold, it is predicted that a crack will occur in that section of the slab.

[0036] The method for online detection and identification of cracks in continuously cast slabs described in this invention provides the following two beneficial effects:

[0037] (1) The method described in this invention does not require learning from historical data, thus avoiding the adverse effects of historical equipment status on the prediction effect. In the actual continuous casting production process, the status of continuous casting equipment changes over time, and it is difficult to use historical samples that match the equipment status at that time to represent the quality of subsequent equipment status. Therefore, learning from historical data to judge the production anomalies at that time is very likely to affect the prediction effect.

[0038] (2) The method for online detection and identification of cracks in continuous casting slabs described in this invention is based on the abnormal probability criterion obtained by the abnormal detection algorithm, which conforms to the process logic of crack generation in continuous casting slabs. The essential reason for crack generation in continuous casting slabs is that the production data has changed (abnormal) within a certain period of time. During the solidification process, the temperature field, solidification field, and solute field change, resulting in internal stress exceeding the fracture strength. Therefore, by identifying the abnormal probability of the slab within a certain period of time, the degree of change of process parameters in the continuous casting slab during production is essentially characterized. This abnormal probability determines whether cracks are generated during continuous casting, which conforms to the process logic of crack generation in continuous casting slabs.

[0039] In summary, the online detection and identification method for cracks in continuously cast slabs described in this invention is relatively simple and widely applicable. It does not require additional detection equipment and can predict the presence of cracks in continuously cast slabs in a timely manner using existing detection instruments. It is of great significance for ensuring production continuity, improving product quality, and reducing production costs, and has a very broad application prospect.

[0040] This online method for detecting and identifying cracks in continuously cast slabs does not require learning from historical samples, effectively preventing errors in identifying abnormal points due to differences in equipment conditions and improving the accuracy of slab defect prediction.

[0041] The online detection and identification method for cracks in continuously cast slabs described in this invention can detect and determine the presence of cracks in continuously cast slabs. It can predict and detect the quality of cast slabs in a timely manner during production, so that continuously cast slabs with cracks can be sorted and removed from the production line in a timely manner or measures can be taken to ensure the production of defect-free cast slabs, thereby effectively improving the hot charging and hot delivery rate. Attached Figure Description

[0042] Figure 1 This is a schematic diagram illustrating the implementation process of the online detection and identification method for cracks in continuously cast slabs according to one embodiment of the present invention. Detailed Implementation

[0043] The method for online detection and identification of cracks in continuously cast slabs according to the present 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 the present invention.

[0044] Figure 1 This is a schematic diagram illustrating the implementation process of the online detection and identification method for cracks in continuously cast slabs according to one embodiment of the present invention.

[0045] like Figure 1 As shown, in Figure 1 The embodiment shown employs the online detection and identification method for cracks in continuously cast slabs as described in this invention:

[0046] Step 1: Collect instrument detection information of slabs at 200 time points in real time, including temperature, heat flux density, and flow rate of 32 instruments in the ladle, tundish, crystallizer, primary cooling zone, secondary cooling zone, and air cooling zone. This information is then transmitted to the model system (i.e., to implement the online detection and identification method for continuous casting slab cracks described in this case) to construct a dataset with a dimension of 200×32.

[0047] It should be noted that, in this embodiment, the instrument detection information comes from the L1 computer system, which may include at least two of the following: tundish weight, tundish liquid level, thermocouple temperature of each region on the crystallizer, crystallizer heat flux density, slab drawing speed, crystallizer cooling water temperature, outlet pressure, argon pressure, and stopper position.

[0048] Step 2: For each time window, the model system applies anomaly detection algorithms to perform anomaly detection on the constructed 200×32 dimension dataset, calculating the outlier values ​​I at 200 time points. i ,(x1,x2,...,x 32 ).

[0049] Among them, outlier I i (x1,x2,...,x 32 This corresponds to whether time point window i in the dataset is considered normal or abnormal. When I... i When I = 1, it can be represented as an anomaly. i When the value is -1, it can be considered normal.

[0050] Furthermore, it can be based on outlier I i Calculate the probability P of anomalies in the 200 time-point window slabs by counting the number of times p = 1: P = p / 200, where p can be represented as I i The number of outliers is represented by the probability P of this outlier, which characterizes the degree of change in the slab's production during the continuous casting process at that time point.

[0051] In the above technical solution of the present invention, the anomaly detection algorithm can be selected from one of the K-nearest neighbor algorithm, SVM algorithm, and isolated forest algorithm; the above anomaly detection algorithms are all known in the prior art. The feature of the present invention is not the anomaly detection algorithm itself, but how it is applied. The anomaly probability P is calculated based on the anomaly detection algorithm.

[0052] Step 3: Collect steel grade information from the L2 computer system at each point in time. The steel grade information may include the steel grade mark and steel grade composition information.

[0053] Step 4: Collect steel grade information in real time and obtain the abnormal threshold S for that steel grade based on the steel grade information. In Step 4 above, the abnormal threshold S can be obtained based on the defect index of the steel grade. The value range of the abnormal threshold S can be controlled between (0, 1) and stored in the database. Its value is maintained by process personnel based on process experience.

[0054] Step 5: Compare the abnormal probability P of the steel grade with the abnormal threshold S. If P≥S, output the judgment that the production abnormality of the steel grade will lead to slab crack defects, and output an alarm signal. Otherwise, output the judgment that the slab is normal.

[0055] Step 6: Update the time, that is, the time window can be moved down by 20 time points to form a new time window slab. Steps 1-5 above are executed again for the updated time window, so that it can be predicted whether crack defects will occur in the slab in the new time window.

[0056] In summary, it can be seen that, in order to address the shortcomings of existing technologies, this method uses an anomaly detection algorithm to detect anomalies in slabs in real time. It can obtain the anomaly probability P on the slab within a certain time window to characterize the degree of fluctuation of production parameters. Furthermore, it compares the anomaly probability P with the anomaly threshold S. When the anomaly probability P exceeds the anomaly threshold S, it predicts that cracks will occur in the slab within that window, and simultaneously outputs an alarm signal.

[0057] The online detection and identification method for cracks in continuously cast slabs described in this invention is relatively simple and widely applicable. It does not require the addition of additional detection equipment and can use existing detection instruments to predict the presence of cracks in continuously cast slabs in a timely manner, so as to sort them off the production line or take measures. It is of great significance for ensuring the continuity of production, improving product quality and reducing production costs, and has a very broad application prospect.

[0058] This online method for detecting and identifying cracks in continuously cast slabs does not require learning from historical samples, effectively preventing errors in identifying abnormal points due to differences in equipment conditions and improving the accuracy of slab defect prediction.

[0059] It should be noted that the combination of the technical features in this case is not limited to the combination methods described in the claims of this case or the combination methods described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way, unless they contradict each other.

[0060] It should also be noted that the embodiments listed above are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and similar changes or modifications made thereto are those that can be directly derived or easily conceived by those skilled in the art from the content disclosed in the present invention, and should all fall within the protection scope of the present invention.

Claims

1. A method for online detection and identification of crack defects in continuously cast slabs, characterized in that, Including the following steps: The instrument detection information of the slab in the ladle, tundish, crystallizer, primary cooling, secondary cooling and air cooling zones is collected in real time at m time points to construct a dataset of dimension m×n, where n is the number of instrument detection information collected at each time point window. The instrument detection information includes at least two of the following: tundish weight, tundish liquid level, thermocouple temperature of each area on the crystallizer, heat flux density of the crystallizer, slab drawing speed, cooling water temperature of the crystallizer, and outlet pressure. For each time window, the following steps are performed: (1) Based on the dataset, an anomaly detection algorithm is used to obtain the outlier values ​​for each time window. I i (x1, x2,..., x n ); among which outliers I i (x1, x2, ..., x n Corresponding to the time point window i Is it considered normal or abnormal in the dataset? (2) Obtain the anomaly probability of the slab at that time point. P : P = p / m, where p is an outlier I i The number of anomalies; (3) Collect steel grade information in real time, and obtain the abnormal threshold S of the steel grade based on the steel grade information; (4) Compare the abnormal probability P of the steel grade with the abnormal threshold S. If P≥S, output the judgment that the abnormal probability of the steel grade will lead to slab crack defects; otherwise, output the judgment that the slab is normal.

2. The method for online detection and identification of crack defects in continuously cast slabs as described in claim 1, characterized in that, Step (3) further includes: when outputting the judgment that the probability of a production abnormality occurring in the steel grade will lead to slab crack defects, an alarm signal is output at the same time.

3. The method for online detection and identification of crack defects in continuously cast slabs as described in claim 1, characterized in that, In step (1), the anomaly detection algorithm is one of the following: K-nearest neighbor algorithm, SVM algorithm, or isolated forest algorithm.

4. The method for online detection and identification of crack defects in continuously cast slabs as described in claim 1, characterized in that, The instrument detection information comes from the L1 computer system.

5. The method for online detection and identification of crack defects in continuously cast slabs as described in claim 1, characterized in that, The steel grade information comes from the L2 computer system.

6. The method for online detection and identification of crack defects in continuously cast slabs as described in claim 1, characterized in that, The steel grade information includes the steel grade designation and steel grade composition information.

7. The method for online detection and identification of crack defects in continuously cast slabs as described in claim 1, characterized in that, The instrument detection information also includes argon pressure and stopper position.

8. The method for online detection and identification of crack defects in continuously cast slabs as described in claim 1, characterized in that, The abnormal threshold S has a value range of (0, 1).

Images