A hot continuous rolling strip steel thickness abnormality root cause tracing method based on representation learning

Abstract

The present disclosure provides a hot continuous rolling strip thickness abnormality root cause tracing method based on representation learning, comprising: when a to-be-detected strip exists thickness abnormality, obtaining thickness control related integrated information of the to-be-detected strip, the thickness control related integrated information of the to-be-detected strip comprising: multi-source process data of the to-be-detected strip and multi-source process data of a previous strip adjacent to the to-be-detected strip; performing dimension reduction feature extraction on the thickness control related integrated information of the to-be-detected strip to obtain a feature vector with a preset fixed dimension corresponding to the to-be-detected strip; according to the similarity between the feature vector of the to-be-detected strip and the feature vectors contained in each historical strip abnormal thickness data recorded in a preset strip abnormal thickness database, screening part of the historical strip abnormal thickness data from the preset strip abnormal thickness database as candidate cases; and determining a root cause tracing result of the thickness abnormality of the to-be-detected strip according to the abnormal root cause labels of the candidate cases.

Description

Technical Field

[0001] This disclosure relates to the field of hot-rolled strip steel product quality diagnosis technology, and in particular to a method for tracing the root causes of thickness anomalies in hot-rolled strip steel based on representation learning, electronic equipment, and computer-readable media. Background Technology

[0002] In the hot-rolled strip steel production process, strip thickness control is one of the key indicators affecting product quality, and its control accuracy directly relates to the performance of the finished product and the stability of subsequent processes. The setting and control of strip thickness mainly relies on the synergistic effect of various process parameters, such as rolling force model, roll gap setting, temperature model, and mill stiffness, and is achieved through dynamic adjustments between each stand to ensure overall thickness accuracy. In actual production, the thickness control model is typically set based on the process conditions at the strip head (such as entry thickness, temperature, rolling force, etc.), and the model coefficients are continuously corrected through a self-learning mechanism to improve control accuracy.

[0003] However, existing methods for diagnosing thickness control quality in hot-rolled strip steel have limitations. Many methods for diagnosing abnormal strip thickness rely on single-parameter threshold judgments or local rule analysis, failing to fully utilize the correlation information between multi-dimensional process data and lacking the ability to comprehensively model historical cases and various failure modes. Furthermore, when these key parameters are missing, traditional expert systems may fail to make a judgment on a particular branch, leading to failed source tracing. For example, existing methods often struggle to accurately distinguish and locate the root cause of different types of thickness faults, such as inaccurate temperature models, deviations in rolling force model coefficients, and abnormal mill stiffness coefficients. This is especially true when data dimensionality is high and category distribution is uneven, further reducing diagnostic accuracy and robustness.

[0004] Therefore, there is an urgent need in this field for a technical solution that can comprehensively utilize multi-source high-dimensional process data, fully consider the influence relationship between adjacent strips, and have the ability to perform root cause analysis on abnormal strip thickness, so as to improve the level of intelligence and production stability of hot continuous rolling thickness control. Summary of the Invention

[0005] This disclosure aims to solve at least one of the technical problems existing in the prior art, and proposes a method, electronic equipment and medium for tracing the root causes of abnormal thickness in hot-rolled strip steel.

[0006] Firstly, this disclosure provides a method for tracing the root causes of abnormal thickness in hot-rolled strip steel, including:

[0007] When the strip steel to be inspected has an abnormal thickness, the thickness control related integrated information of the strip steel to be inspected is obtained. The thickness control related integrated information of the strip steel to be inspected includes: multi-source process data of the strip steel to be inspected and multi-source process data of the previous strip steel adjacent to the strip steel to be inspected.

[0008] The thickness control-related integrated information of the strip steel to be tested is subjected to dimensionality reduction feature extraction to obtain the feature vector with a preset fixed dimension corresponding to the strip steel to be tested;

[0009] Based on the similarity between the feature vector of the strip to be detected and the feature vector contained in each historical strip abnormal thickness data recorded in the preset strip abnormal thickness database, some historical strip abnormal thickness data are selected from the preset strip abnormal thickness database as candidate cases. The historical strip abnormal thickness data includes: feature vectors and abnormal root cause labels of historical strips with thickness abnormalities.

[0010] Based on the abnormal root cause labels of the candidate cases, the root cause tracing results of the abnormal strip thickness to be detected are determined.

[0011] In some embodiments, the step of performing dimensionality reduction feature extraction on the thickness control-related integrated information to obtain a feature vector with a preset fixed dimension corresponding to the strip steel to be detected includes:

[0012] The thickness control-related integrated information is input into a pre-trained multilayer perceptron, which acts as a feature extraction network to map the high-dimensional thickness control-related integrated information into a low-dimensional feature vector with a preset fixed dimension.

[0013] In some embodiments, prior to the step of inputting the thickness control-related integrated information into a pre-trained multilayer perceptron, the method further includes: training the multilayer perceptron, specifically including:

[0014] Multiple strip steel samples with abnormal thickness and their corresponding root cause labels are obtained. The strip steel samples with abnormal thickness include: multi-source process data of the strip steel with abnormal thickness and multi-source process data of the previous strip steel adjacent to the strip steel with abnormal thickness.

[0015] The multilayer perceptron is trained using steel strip samples with abnormal thickness and their corresponding abnormal root cause labels. The multilayer perceptron includes: an input splicing layer, a dual-channel shared encoder, a feature fusion layer, and a feature transformation output layer arranged sequentially.

[0016] The input splicing layer is configured to splice the multi-source process data x0 of the previous strip adjacent to the strip with the thickness abnormality and its configured mask m0, and then send it to one input of the dual-channel shared encoder; and is further configured to splice the multi-source process data x1 of the strip with the thickness abnormality and its configured mask m1, and then send it to the other input of the dual-channel shared encoder.

[0017] The dual-channel shared encoder is configured to perform dimensionality reduction feature extraction on the two input data (x0, m0) and (x1, m1) respectively, to obtain the intermediate feature vector V0 corresponding to the data (x0, m0) and the intermediate feature vector V1 corresponding to the data (x1, m1);

[0018] The feature fusion layer is configured to fuse the intermediate feature vector V0 and the intermediate feature vector V1 to obtain a fused intermediate feature vector.

[0019] The feature transformation output layer is configured to perform dimensionality reduction and feature extraction on the fused intermediate feature vector again to obtain the feature vector with a preset fixed dimension corresponding to the thickness abnormal strip steel sample.

[0020] In some embodiments, cosine similarity is used as the similarity between two vectors, or the similarity between two vectors is determined based on the Euclidean distance between the vectors.

[0021] In some embodiments, the step of selecting a portion of historical abnormal strip thickness data from the preset abnormal strip thickness database as candidate cases includes:

[0022] From the preset abnormal strip thickness database, the top N historical abnormal strip thickness data with the highest similarity between the feature vector contained therein and the feature vector of the strip to be detected are selected as candidate cases, where N is a preset positive integer.

[0023] In some embodiments, the root cause tracing result of the detected strip thickness anomaly is determined based on the anomaly root cause labels of the candidate cases:

[0024] The abnormal root cause labels of all the candidate cases are classified and statistically analyzed, and the root cause tracing results of the abnormal strip thickness to be detected are determined based on the statistical results; the statistical results include: the candidate cases corresponding to each type of abnormal root cause label.

[0025] In some embodiments, the step of determining the root cause of the abnormal thickness of the strip to be detected based on statistical results includes:

[0026] Perform label statistical voting: determine the cumulative voting score of each type of abnormal root cause label included in the statistical results; wherein, the cumulative voting score of the abnormal root cause label is equal to the sum of the single-case voting scores of all the candidate cases corresponding to the abnormal root cause label, and the single-case voting score of the candidate case is equal to the similarity between the feature vector corresponding to the candidate case and the feature vector of the strip to be detected multiplied by the sample weight configured for the candidate case;

[0027] The anomaly root cause label corresponding to the highest cumulative voting score is taken as the root cause tracing result of the anomaly in the thickness of the strip to be detected.

[0028] In some embodiments, after using the anomaly root cause label corresponding to the highest cumulative voting score as the root cause tracing result of the detected strip thickness anomaly, the method further includes:

[0029] Check whether the concentration of the abnormal root cause labels of all candidate cases is greater than or equal to a preset threshold, and whether the maximum cumulative vote score is greater than or equal to a preset score threshold;

[0030] When the detected label distribution level is less than the preset level threshold, or when the detected maximum cumulative vote score is less than the preset score threshold, a prompt message is generated and output to indicate that the root cause tracing results need to be manually reviewed.

[0031] Secondly, this disclosure provides an electronic device, including:

[0032] One or more processors;

[0033] Memory, used to store one or more programs;

[0034] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as provided in the first aspect.

[0035] Thirdly, this disclosure also provides a computer-readable medium on which a computer program is stored, which, when executed by a processor, implements the steps of the method provided in the first aspect.

[0036] The technical solution disclosed herein, when tracing the root cause of abnormal thickness in the strip under test, not only incorporates multi-source process data of the strip under test, but also integrates multi-source process data of the preceding strip that has a successor relationship with the strip under test, which can greatly increase the accuracy of tracing the root cause of thickness abnormalities. Furthermore, the technical solution disclosed herein can also summarize the root causes of failures by combining historical candidate cases, thus possessing good interpretability of the results. Attached Figure Description

[0037] Figure 1A flowchart illustrating a method for tracing the root causes of abnormal thickness in hot-rolled strip steel, provided in an embodiment of this disclosure;

[0038] Figure 2 This is a flowchart of an optional implementation method of step S4 in the embodiments of this disclosure;

[0039] Figure 3 This is a schematic diagram of the feature extraction network used in step S2 of this embodiment.

[0040] Figure 4 A flowchart illustrating another method for tracing the root causes of abnormal thickness in hot-rolled strip steel provided in this embodiment of the present disclosure;

[0041] Figure 5 A confusion matrix heatmap obtained by comparing the traceability results of strip steel on the test set with the actual labels;

[0042] Figure 6 This is a t-SNE visualization diagram of the feature vector set in the strip abnormal thickness database of this disclosure;

[0043] Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0044] To enable those skilled in the art to better understand the technical solutions of this disclosure, exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of this disclosure to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0045] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0046] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0047] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the said feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded.

[0048] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined herein.

[0049] To address at least one of the technical problems existing in the aforementioned related technologies, this disclosure provides a method for tracing the root causes of thickness anomalies in hot-rolled strip steel based on representation learning, an electronic device, and a computer-readable medium. Figure 1 This is a flowchart illustrating a method for tracing the root causes of thickness anomalies in hot-rolled strip steel based on representation learning, provided in an embodiment of this disclosure. Figure 1 As shown, this root cause tracing method includes:

[0050] Step S1: When there is a thickness abnormality in the strip to be inspected, obtain the thickness control related integrated information of the strip to be inspected. The thickness control related integrated information of the strip to be inspected includes: multi-source process data of the strip to be inspected and multi-source process data of the previous strip adjacent to the strip to be inspected.

[0051] In practical applications, the thickness deviation of the strip (actual thickness of the strip - target thickness of the strip) can be monitored in real time. When the thickness deviation of the strip to be tested is within a preset reasonable range, it is determined that the thickness of the strip to be tested is normal; when the thickness deviation of the strip to be tested exceeds the preset reasonable range, it is determined that the thickness of the strip to be tested is abnormal. At this time, step S1 can be executed.

[0052] Unlike related technologies that only acquire process data of the strip under test when performing root cause analysis on strips with thickness anomalies, the technical solution of this disclosure acquires not only multi-source process data of the strip under test but also multi-source process data of the preceding strip adjacent to it. This is because the inventors discovered that the thickness anomaly of the strip under test is not only directly affected by the process data during its production process but also indirectly affected by the production process of the preceding strip (e.g., cumulative process errors). Based on this, this disclosure integrates the multi-source process data of the preceding strip adjacent to the strip under test as contextual information of the strip under test with the multi-source process data of the strip under test to characterize the inheritance and influence relationships between adjacent strips. This improves the accuracy of the root cause analysis results for thickness anomalies.

[0053] In this disclosure, the multi-source process data for strip steel refers to multiple process characteristic data related to strip steel thickness during the strip steel production process. The multi-source process data for strip steel includes, but is not limited to, at least one of the following: strip steel grade, strip steel entry thickness, rolling force learning coefficient, temperature learning coefficient, pre-calculated relative deviation of rolling force, actual rolling force, strip steel temperature deviation, roll gap deviation, mill stiffness coefficient, target strip steel thickness, stand speed, cooling water parameters, rolling interval, strip steel width, thickness deviation of the previous strip steel, and actual exit thickness, etc.

[0054] In the actual process of collecting multi-source process data of strip steel, the collected multi-source process data can be preprocessed, such as filling missing values ​​and removing outliers, to ensure the integrity and compliance of the integrated information related to thickness control.

[0055] Step S2: Perform dimensionality reduction feature extraction on the integrated information related to thickness control of the strip to be inspected to obtain the feature vector with a preset fixed dimension corresponding to the strip to be inspected.

[0056] In step S2, a feature extraction network can be used to perform dimensionality reduction feature extraction on the integrated information related to thickness control of the strip to be detected; that is, on the one hand, feature extraction processing is performed on the integrated information related to thickness control, and on the other hand, the goal of dimensionality reduction is achieved. The feature extraction network in this disclosure can adopt any network structure that has both feature extraction and dimensionality reduction functions, and this disclosure does not limit it.

[0057] In step S2, the extracted feature vectors are limited to having a preset fixed dimension, which is also to facilitate the subsequent similarity calculation of different feature vectors with the same dimension.

[0058] Step S3: Based on the similarity between the feature vector of the strip to be detected and the feature vector contained in each historical strip abnormal thickness data recorded in the preset strip abnormal thickness database, select some historical strip abnormal thickness data from the preset strip abnormal thickness database as candidate cases.

[0059] The preset strip thickness abnormality database contains multiple historical strip thickness abnormality data. Each historical strip thickness abnormality data includes: the feature vector of the historical strip with thickness abnormality (with preset fixed dimensions) and the abnormality root cause label (pre-labeled).

[0060] Among them, the feature vectors in the historical strip steel abnormal thickness data can be compared with the feature vectors of the strip steel to be detected (to calculate similarity), thereby filtering out historical strip steels with similar production conditions (also indicating similar thickness abnormalities) as candidate cases; and the abnormal root cause labels in the historical strip steel abnormal thickness data (i.e., candidate cases) with the same or similar production conditions can provide a reference for subsequently determining the root cause of the thickness abnormality of the strip steel to be detected.

[0061] The types, number, and corresponding fields of the abnormal root cause labels can be pre-designed according to actual needs. For example, abnormal root cause labels can be "inaccurate temperature model," "deviation in rolling force model coefficients," and "abnormal mill stiffness coefficients," to characterize different types of thickness fault root causes.

[0062] In addition, in practical applications, other related information can be added to each historical strip thickness anomaly data point as needed. For example, the unique sample identifier, strip number, multi-source process data, generation time, and sample batch number can be added to each historical strip thickness anomaly data point.

[0063] In addition, an index can be built for a preset strip thickness anomaly database to facilitate retrieval of the database. For example, a field index can be built for at least some process fields (e.g., thickness grade, temperature grade) in multi-source process data, and a vector index can be built for feature vectors; among them, the vector index can be built using an approximate nearest neighbor retrieval structure to support fast searching of historical similar samples.

[0064] Through the similarity calculation and screening steps in step S3, candidate cases that are similar to the production conditions (similar thickness anomalies) of the strip steel to be tested can be identified.

[0065] In some embodiments, cosine similarity is used as the similarity between two vectors, or the similarity between two vectors is determined based on the Euclidean distance between the vectors (the smaller the Euclidean distance, the greater the determined similarity).

[0066] Step S4: Based on the abnormal root cause labels of the candidate cases, determine the root cause tracing results of the abnormal strip thickness to be detected.

[0067] Since the candidate cases selected in step S3 are the same as or similar to the production conditions (thickness anomalies) of the strip steel to be tested, the anomaly root cause labels of these candidate cases can provide a valid reference for determining the root cause of the thickness anomaly of the strip steel to be tested.

[0068] In this disclosure, no limitation is made on the algorithm for determining the root cause of the abnormal strip thickness based on the abnormal root cause labels of candidate cases. For example, the abnormal root cause label with the highest frequency among all candidate cases can be used as the root cause of the abnormal strip thickness, or the root cause of the abnormal strip thickness can be determined based on statistical voting (this scheme will be described in detail later).

[0069] The technical solution disclosed herein, when tracing the root cause of abnormal thickness in the strip under test, not only incorporates multi-source process data of the strip under test, but also integrates multi-source process data of the preceding strip that has a successor relationship with the strip under test, which can greatly increase the accuracy of tracing the root cause of thickness abnormalities. Furthermore, the technical solution disclosed herein can also summarize the root causes of failures by combining historical candidate cases, thus possessing good interpretability of the results.

[0070] In this disclosure, the purpose of step S3 is to select historical strip steel with the same or similar production conditions as the strip steel to be inspected from the strip steel abnormal thickness database. This can be achieved in various specific ways.

[0071] In some embodiments, historical strip abnormal thickness data with a similarity greater than a preset similarity threshold between the feature vector contained in the preset strip abnormal thickness database can be selected as candidate cases (at this time, the number of selected candidate cases will change dynamically according to the actual situation).

[0072] In other embodiments, the step of selecting some historical strip abnormal thickness data from the preset strip abnormal thickness database as candidate cases in step S3 includes: selecting the top N historical strip abnormal thickness data with the highest similarity between the feature vector contained in the preset strip abnormal thickness database and the feature vector of the strip to be detected, as candidate cases, where N is a preset positive integer (at this time, the number of candidate cases selected is fixed at N).

[0073] Figure 2 This is a flowchart of an optional implementation method of step S4 in an embodiment of this disclosure. For example... Figure 2 As shown, in some embodiments, step S4 specifically involves: classifying and statistically analyzing the anomaly root cause labels of all candidate cases, and determining the root cause tracing result of the anomaly in the strip thickness to be detected based on the statistical results. The statistical results include: the candidate cases corresponding to each type of anomaly root cause label. Step S4 includes:

[0074] Step S401: Perform label statistical voting: Determine the cumulative voting score of each type of abnormal root cause label included in the statistical results.

[0075] The cumulative voting score of the abnormal root cause label is equal to the sum of the individual voting scores of all candidate cases corresponding to the abnormal root cause label. The individual voting score of a candidate case is equal to the similarity between the feature vector corresponding to the candidate case and the feature vector of the strip to be detected multiplied by the sample weight configured for the candidate case.

[0076] In this disclosure, a corresponding sample weight can be assigned to each historical strip steel anomaly thickness data point in the strip steel anomaly thickness database, either manually or automatically using an algorithm. This is because the number of historical strip steel anomaly thickness data points corresponding to different category labels in the strip steel anomaly thickness database varies (the distribution of the number of historical strip steel anomaly thickness data points for different category labels is uneven). When selecting candidate cases in step S3, category labels with a larger number of corresponding historical strip steel anomaly thickness data points are more likely to be selected, which may affect the accuracy of the final traceability result. Therefore, a relatively small sample weight can be configured for historical strip steel anomaly thickness data points with category labels that have a larger number of data points, and a relatively large sample weight can be configured for historical strip steel anomaly thickness data points with category labels that have a smaller number of data points.

[0077] Of course, the sample weights of historical strip thickness anomaly data corresponding to each category of anomaly can be adjusted according to the importance of different anomaly root causes. For example, the sample weights of historical strip thickness anomaly data with important anomaly root cause labels can be appropriately increased to ensure that important anomaly root causes can be detected in a timely manner.

[0078] Step S402: Take the abnormal root cause label corresponding to the largest cumulative voting score as the root cause tracing result of the abnormal strip thickness to be detected.

[0079] Unlike traditional methods of counting votes based on frequency, this disclosure first determines the individual vote score of a candidate case by multiplying its similarity with its sample weight. Then, it calculates the sum of the individual vote scores for multiple candidate cases based on their label categories to obtain the cumulative vote score for each label category. Finally, it selects the anomaly root cause label corresponding to the highest cumulative vote score as the root cause tracing result for the detected strip thickness anomaly. This improves the accuracy and robustness of the root cause tracing results.

[0080] In some embodiments, the method further includes the following after step S402:

[0081] Step S403: Detect whether the concentration of the abnormal root cause labels of all candidate cases is greater than or equal to a preset threshold, and whether the maximum cumulative vote score is greater than or equal to a preset score threshold.

[0082] The degree of concentration of the abnormal root cause labels for all candidate cases can be determined based on the number of candidate cases corresponding to each type of abnormal root cause label. The degree of concentration of label distribution represents the concentration of abnormal root cause labels for all candidate cases; a higher value indicates a more concentrated label distribution.

[0083] Generally, the degree of concentration of label distribution can be characterized in a variety of different ways. For example, the degree of concentration of label distribution can be represented by the normalized entropy (first determine the information entropy of these labels and then normalize it), or by the concentration ratio (directly calculate the sum of the proportions of the top M dominant categories of labels, where M is a preset positive integer, for example, M takes the value of 1, 2, or 3).

[0084] When the detected label distribution degree is greater than or equal to the preset degree threshold, it indicates that the label distribution of the abnormal root cause labels of all candidate cases is concentrated; otherwise, it indicates that the label distribution of the abnormal root cause labels of all candidate cases is dispersed.

[0085] When the detected maximum cumulative vote score is greater than or equal to the preset score threshold, it indicates that the confidence level corresponding to the maximum cumulative vote score is high; otherwise, it indicates that the confidence level corresponding to the maximum cumulative vote score is low.

[0086] In step S403, if the detected label distribution degree is greater than or equal to a preset degree threshold and the maximum cumulative vote score is greater than or equal to a preset score threshold, it indicates that the confidence level of the root cause tracing result obtained in step S402 is high, and no manual review is required. However, if the detected label distribution degree is less than the preset degree threshold, or the detected maximum cumulative vote score is less than the preset score threshold, it indicates that the confidence level of the root cause tracing result obtained in step S402 is low, and manual review is required. In this case, step S404 is executed.

[0087] Step S404: Generate and output the prompt message.

[0088] The message indicates that the confidence level of the root cause tracing results is low, suggesting that manual review is needed to reduce the risk of misjudgment.

[0089] Figure 3 This is a schematic diagram of the feature extraction network used in step S2 of this embodiment. Figure 3 As shown, in some embodiments, step S2 specifically includes: inputting thickness control-related integrated information into a pre-trained multilayer perceptron (MLP), whereby the MLP acts as a feature extraction network and maps the high-dimensional thickness control-related integrated information into a low-dimensional feature vector (also known as an embedding vector) with a preset fixed dimension.

[0090] Figure 4 This is a flowchart illustrating another method for tracing the root causes of thickness anomalies in hot-rolled strip steel based on representation learning, provided in an embodiment of this disclosure. Figure 4 As shown, step S0 is included before step S2. Figure 4 The example illustrates the case where step S0 is executed before step S1. In this disclosure, it is only necessary to ensure that step S0 is executed before step S2. Step S0 will be described in detail below.

[0091] Step S0: Train the multilayer perceptron.

[0092] In some embodiments, step S0 specifically includes:

[0093] Step S01: Obtain multiple strip steel samples with abnormal thickness and their corresponding root cause labels. The strip steel samples with abnormal thickness include: multi-source process data of the strip steel with abnormal thickness and multi-source process data of the previous strip steel adjacent to the strip steel with abnormal thickness.

[0094] The thickness of the strip produced on the production line can be monitored in advance. When an abnormal thickness is detected (when the actual thickness of the strip deviates from the target thickness beyond a preset reasonable range), the strip with the abnormal thickness can be used as a training sample. Multi-source process data of the strip with the abnormal thickness and the preceding strip adjacent to it can be collected. For a description of the multi-source process data, please refer to the previous content, which will not be repeated here.

[0095] Step S02: Train the multilayer perceptron using steel strip samples with thickness anomalies and their corresponding root cause labels.

[0096] The multilayer perceptron includes, in sequence, an input splicing layer, a dual-channel shared encoder, a feature fusion layer, and a feature transformation output layer.

[0097] The input splicing layer is configured to splice the multi-source process data x0 of the previous strip adjacent to the strip with abnormal thickness and its configured mask m0, and then send it to one input of the dual-channel shared encoder; and is also configured to splice the multi-source process data x1 of the strip with abnormal thickness and its configured mask m1, and then send it to the other input of the dual-channel shared encoder.

[0098] In this disclosure, a mask is used to mark the validity of data (it can distinguish between real data and invalid padding data in multi-source process parameters, and make up for the problems caused by sensor packet loss, signal abnormality and data missing in industrial field). A feature masking mechanism is introduced during the training process to simulate data missing caused by sensor failure or data transmission error in actual production process, so as to enhance the robustness of the network in the case of partial feature missing.

[0099] The dual-channel shared encoder is configured to perform dimensionality reduction feature extraction on the two input data (x0, m0) and (x1, m1) respectively, to obtain the intermediate feature vector V0 corresponding to the data (x0, m0) and the intermediate feature vector V1 corresponding to the data (x1, m1).

[0100] In some embodiments, a residual structure is introduced into the dual-channel shared encoder. The accompanying drawings exemplarily illustrate a residual structure comprising two residual blocks, Residual Block 1 and Residual Block 2. The number of residual blocks in this disclosure can also be one, three, or more; this disclosure does not impose any limitation on this. In this disclosure, the introduction of the residual structure ensures that the dual-channel shared encoder can stably deepen the network layers and learn more complex front and rear strip working conditions, while preventing deep training degradation. Simultaneously, it preserves the original effective features of x0 / m0 and x1 / m1, making the encoded V0 and V1 representations more accurate.

[0101] The feature fusion layer is configured to fuse intermediate feature vectors V0 and V1 to obtain a fused intermediate feature vector.

[0102] The feature transformation output layer is configured to perform dimensionality reduction and feature extraction again on the fused intermediate feature vector to obtain the feature vector e with a preset fixed dimension corresponding to the strip steel sample with thickness anomalies. i During training, the feature transformation output layer is also configured to calculate the detection score based on this feature vector, indicating its category to which the corresponding "abnormal root cause label" belongs. i This is for subsequent loss calculation and network parameter tuning (which will be described in detail later with specific examples). Among them, the aforementioned e... i Let o represent the feature vector corresponding to the i-th sample. i This represents the detection result score corresponding to the i-th sample.

[0103] After training, each strip steel sample can output a corresponding fixed-dimensional embedding vector and a category score to facilitate subsequent retrieval and fast inference.

[0104] In some embodiments, step S0 further includes: step S03, constructing a strip abnormal thickness database.

[0105] After training, the strip steel samples with thickness anomalies are used to form corresponding structured historical strip steel anomaly thickness data (including at least the feature vector and root cause label of the strip steel with the anomaly, and may also include other field information, such as unique sample identifier, strip steel number, original process field, production time, sample source batch, etc.), and written to the strip steel anomaly thickness database. Additionally, indexes can be created in the strip steel anomaly thickness database as needed.

[0106] A specific implementation example of the technical solution disclosed herein:

[0107] The following example illustrates the process using strip steel production data collected over six consecutive months from a 1780mm hot strip mill. The original sample consisted of 25,036 records, including 22,571 normal records and 2,465 fault records. The multi-source process fields comprised 48 dimensions (i.e., the multi-source process data was 48-dimensional). The root cause labels configured for thickness anomalies included types such as "caused by global thickness learning," "inaccurate temperature model," "inconsistency between settings and actual control results," "inaccurate coefficients in the rolling force model itself," "self-learning interference between rolling force models," "inaccurate mill stiffness coefficient," and "roll gap - gauge change."

[0108] Database preparation and network training phases (1) to (6):

[0109] (1) Extract production data related to strip thickness from the process database of the hot strip rolling production line. The data includes strip grade, strip entry thickness, rolling force learning coefficient, temperature learning coefficient, pre-calculated rolling force relative deviation, actual rolling force, strip temperature deviation, roll gap deviation, mill stiffness coefficient, strip target thickness, stand speed, cooling water, rolling interval, strip width, thickness deviation of the previous strip, actual strip thickness at the exit, and corresponding fault labels (this field is temporarily "empty" during the original data collection).

[0110] (2) The original data of strip thickness is deduplicated and obvious abnormal records are removed; missing values ​​are filled by mean filling, median filling or interpolation according to the adjacent time period of the process.

[0111] (3) To determine whether the strip thickness is abnormal, the standard is: whether the thickness deviation between the actual thickness of the strip and the target thickness of the strip is within a preset reasonable range (e.g., -20μm to 20μm). When the thickness deviation exceeds the preset reasonable range, the strip thickness is considered abnormal. Extract this data and the previous data and add them to the abnormal dataset. When constructing strip training samples, associate and splice the current strip sample with its previous strip sample, and use the data of the previous strip as the context information input of the current sample to describe the inheritance relationship and influence relationship between adjacent strips; at the same time, the abnormal root cause label of the sample also needs to be labeled.

[0112] (4) The labeled samples are divided into training and testing sets, allocated in a 7:3 ratio, ensuring that the proportion of each fault category remains consistent in both sets. A multilayer perceptron is used as the encoder, and residual connections are introduced into the network to mitigate the degradation problem during deep network training. The neural network structure is shown in the attached figure. Figure 2 As shown, the network outputs a score for each category.i After the Softmax layer, the probability that a sample belongs to class c is:

[0113]

[0114] After training, the original high-dimensional features of the strip steel are compressed into a 16-dimensional low-dimensional vector representation, so that each strip steel sample corresponds to a fixed-length embedding vector (i.e., features are concatenated), which facilitates subsequent retrieval and fast inference, such as... Figure 3 e i As shown.

[0115] The loss function of this network is a combination of weighted cross-entropy loss and supervised contrastive loss.

[0116]

[0117] Where L represents the total loss, Represents cross-entropy loss, For supervised comparative loss, This is the weighting coefficient, which is set to 0.5 in this embodiment.

[0118] The calculation formula is:

[0119]

[0120] Where B is the training batch size. The weight of the current category. Given the embedding vector e i The probability that the sample belongs to class c is calculated. Supervised contrastive learning is used for optimization, resulting in samples of the same type of fault being closer together in the vector space, and samples of different fault categories having larger intervals.

[0121] The calculation formula is:

[0122]

[0123] in, This represents the set of anchor samples that contains at least one positive sample in the current batch.

[0124] The contrast loss for a single sample is calculated using the following formula:

[0125]

[0126] This is the set of positive samples for the anchor point. To define the similarity between two samples i and j, , This is the temperature coefficient, and in this embodiment, it is taken as 0.07; The similarity result obtained by subtracting the maximum value of the row from each sample is used to prevent overflow in exponential operations.

[0127] Figure 5 This is a confusion matrix heatmap obtained by comparing the traceability results of strip steel on the test set with the actual labels. For example... Figure 5 As shown, after training is completed on the training set, the network can be tested using the test set. Specifically, fault tracing is performed on all strip steel in the test set and compared with the real labels to obtain the confusion matrix of the experimental results.

[0128] (5) After training, each sample vector output by the network is saved as a structured file, such as CSV format, while retaining information such as strip grade, fault label, and original process field index, as input data for the subsequent strip abnormal thickness database. After this step, a set of low-dimensional feature vectors suitable for similarity calculation can be obtained. Figure 6 This is a t-SNE visualization diagram of the feature vector set within the strip anomaly thickness database of this disclosure. (Example:) Figure 6 As shown, t-SNE is a dimensionality reduction method used to map high-dimensional data to a low-dimensional space in order to visualize the similarity relationship between samples. In this disclosure, the visualization of all feature vectors in the strip steel abnormal thickness database after training and t-SNE dimensionality reduction shows that the inter-class distance of the samples is large, but the spatial distance of samples of the same class is relatively close.

[0129] (6) Write the 16-dimensional sample vectors output by the neural network and their associated information into a vector database to construct a strip steel abnormal thickness database. Each record in the database includes at least: a unique sample identifier, strip steel number, 16-dimensional vector representation, fault label, original process field summary, generation time, and sample source batch. The vector index can be established using an approximate nearest neighbor retrieval structure to support rapid searching of historical similar samples. In a specific implementation, all strip steel samples formed from the past 6 months of data can be stored in the database, and auxiliary retrieval tags can be established according to fault type, specification category, thickness grade, or temperature grade. This preserves both the efficiency of vector retrieval and the constraints of the process scenario.

[0130] On-site diagnostic steps (7) to (9)

[0131] (7) When a strip steel to be inspected is found on site, first, perform the same data preprocessing and feature construction as in steps (1)-(3) on the strip steel, and input the integrated data of the strip steel to be inspected with thickness anomalies into the trained dimensionality reduction feature extraction network to obtain its corresponding 16-dimensional feature vector. Then, send the feature vector into the strip steel anomaly thickness database for similarity retrieval, and take the Top-N most similar historical samples as candidate cases. The similarity measurement method can be cosine similarity or Euclidean distance.

[0132] As a specific example, a strip steel processing line has 7 stands. Stand 1 is the first stand, stand 6 is the second to last stand, and the thickness of the strip steel processed by the last stand, 7, is the final thickness. For any strip steel to be inspected, its 48-dimensional multi-source process data is concatenated with the 48-dimensional multi-source process data of the previous strip steel to obtain integrated data in a 2×48 matrix. The integrated data of the strip steel to be inspected after concatenation is shown in Table 1 below:

[0133] Table 1. Integrated Data Sheet of the Strip Steel to be Inspected

[0134]

[0135]

[0136]

[0137] Table 2 below is a table showing the meaning of each field name in the above 48-dimensional multi-source process data.

[0138] Table 2. Meaning of Field Names in Multi-Source Process Data

[0139]

[0140]

[0141] The integrated 2×48-dimensional data in Table 1 is input into the trained dimensionality reduction feature extraction network to obtain a 16-dimensional feature vector.

[0142] A=[4.2664,-2.5001,-1.4077,-3.5678,3.5191,-3.6066,-3.3620,3.3570 ,-2.6243,-0.1437,0.8696,0.4405,-0.3515,-0.8101,-3.8663,-2.7025]

[0143] The cosine similarity between the feature vector A of the strip to be detected and the feature vectors of each historical strip anomaly thickness data in the strip anomaly thickness database is calculated sequentially. The top N historical strip anomaly thickness data with the largest values ​​are returned. In this embodiment, K is set to 5. The feature vectors of the 5 candidate cases are as follows:

[0144] V1=[4.0101,-1.5153,-2.0582,-2.8927,4.1284,-3.5873,-2.4194,2.2024, -4.5698,0.0892,0.3332,0.9624,-0.1726,0.5595,-4.1818,-2.6557]

[0145] V2=[4.5773,-0.2961,-2.2306,-3.2218,2.4519,-3.7583,-2.5108,1.3273, -4.3102,-0.4931,-0.0199,0.8799,-0.7638,0.7157,-4.2762,-1.4976]

[0146] V3=[3.2501,-3.9578,1.5239,-2.3527,5.1820,-1.4663,-3.1853,3.2842, -2.4606,0.1109,1.1406,0.1580,0.2036,-0.5039,-3.7935,-2.6223]

[0147] V4=[5.5317,0.1578,-2.7029,-3.4416,3.1249,-4.0920,-4.2439,1.6736, -4.2833,1.4338,-1.0279,0.3837,0.2645,0.7698,-4.9113,-1.7681]

[0148] V5=[4.8410,-0.1687,-3.7446,-3.9792,3.6688,-3.5008,-3.3492,1.8861, -4.1014,1.4871,-0.1753,0.5116,0.1923,0.5011,-3.9466,-0.6821]

[0149] The cosine similarity between the feature vectors of these 5 candidate cases and the feature vector A of the strip to be detected are 0.9538, 0.9127, 0.9081, 0.8992, and 0.8973, respectively; the root cause label for these 5 candidate cases is "inaccurate temperature model".

[0150] (8) Based on the search results of (7), statistical voting is performed on the fault labels of the Top-5 samples: when most of the neighboring samples belong to the same category, the root cause of the abnormal thickness of the strip to be tested is also the fault, that is, the root cause of the strip to be tested is "inaccurate temperature model".

[0151] (9) The system can output root cause analysis results; it can also output candidate root causes (a set of abnormal root cause labels for all candidate cases), the number of candidate cases in the strip thickness abnormality database, and other data as needed, so that on-site process personnel can perform parameter correction, network updates, and subsequent production optimization. For root cause analysis results with low confidence, the system will also provide feedback prompts to indicate that the root cause analysis results need to be manually reviewed in order to avoid misjudgment affecting production.

[0152] This disclosed technical solution transforms high-dimensional, multi-source, and strongly coupled hot-rolled strip steel process data into low-dimensional vector representations, and utilizes a strip steel abnormal thickness database for similar sample retrieval and fault diagnosis. This effectively improves the accuracy, retrieval efficiency, and interpretability of thickness fault identification. Furthermore, this disclosed technical solution can also summarize the root causes of faults by combining historical cases, providing a basis for adjusting on-site thickness control parameters, and exhibits good engineering application effects.

[0153] Through experimental comparison, compared with traditional thickness defect tracing systems based on expert experience, this invention avoids the situation where tracing is impossible when data is missing. In the absence of missing data, the tracing accuracy reaches 90.67%, compared to 90.27% for the expert experience system; with a 10% missing data rate, the tracing accuracy is 84.17%; and with a 20% missing data rate, the tracing accuracy still remains at 75.83%.

[0154] Based on the same inventive concept, this disclosure also provides an electronic device. Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of this disclosure. Figure 7 As shown, this disclosure provides an electronic device including: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement any of the representation learning-based root cause tracing methods for hot-rolled strip thickness anomalies in the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory, configured to enable information interaction between the processor and the memory.

[0155] The processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read / write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus).

[0156] In some embodiments, the processor 101, memory 102, and I / O interface 103 are interconnected via bus 104, and thus connected to other components of the computing device.

[0157] In some embodiments, the one or more processors 101 include a field-programmable gate array.

[0158] This disclosure also provides a computer-readable medium. The computer-readable medium stores a computer program, which, when executed by a processor, implements the steps in any of the representation learning-based root cause tracing methods for hot-rolled strip thickness anomalies described above. The computer-readable storage medium may be volatile or non-volatile.

[0159] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code. When the computer-readable code is run in the processor of an electronic device, the processor in the electronic device executes the above-described method for tracing the root cause of thickness anomalies in hot-rolled strip steel based on representation learning.

[0160] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0161] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable program instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0162] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0163] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0164] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0165] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0166] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0167] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0168] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0169] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in connection with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in connection with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. A method for tracing the root causes of thickness anomalies in hot-rolled strip steel based on representation learning, characterized in that, include: When the strip steel to be inspected has an abnormal thickness, the thickness control related integrated information of the strip steel to be inspected is obtained. The thickness control related integrated information of the strip steel to be inspected includes: multi-source process data of the strip steel to be inspected and multi-source process data of the previous strip steel adjacent to the strip steel to be inspected. The thickness control-related integrated information of the strip steel to be tested is subjected to dimensionality reduction feature extraction to obtain the feature vector with a preset fixed dimension corresponding to the strip steel to be tested; Based on the similarity between the feature vector of the strip to be detected and the feature vector contained in each historical strip abnormal thickness data recorded in the preset strip abnormal thickness database, some historical strip abnormal thickness data are selected from the preset strip abnormal thickness database as candidate cases. The historical strip abnormal thickness data includes: feature vectors and abnormal root cause labels of historical strips with thickness abnormalities. Based on the abnormal root cause labels of the candidate cases, the root cause tracing results of the abnormal strip thickness to be detected are determined.

2. The method according to claim 1, characterized in that, The steps of performing dimensionality reduction feature extraction on the integrated information related to thickness control to obtain the feature vector with a preset fixed dimension corresponding to the strip to be detected include: The thickness control-related integrated information is input into a pre-trained multilayer perceptron, which acts as a feature extraction network to map the high-dimensional thickness control-related integrated information into a low-dimensional feature vector with a preset fixed dimension.

3. The method according to claim 2, characterized in that, Before the step of inputting the thickness control-related integrated information into the pre-trained multilayer perceptron, the method further includes: training the multilayer perceptron, specifically including: Multiple strip steel samples with abnormal thickness and their corresponding root cause labels are obtained. The strip steel samples with abnormal thickness include: multi-source process data of the strip steel with abnormal thickness and multi-source process data of the previous strip steel adjacent to the strip steel with abnormal thickness. The multilayer perceptron is trained using steel strip samples with abnormal thickness and their corresponding abnormal root cause labels. The multilayer perceptron includes: an input splicing layer, a dual-channel shared encoder, a feature fusion layer, and a feature transformation output layer arranged sequentially. The input splicing layer is configured to splice the multi-source process data x0 of the previous strip adjacent to the strip with the thickness abnormality and its configured mask m0, and then send it to one input of the dual-channel shared encoder; and is further configured to splice the multi-source process data x1 of the strip with the thickness abnormality and its configured mask m1, and then send it to the other input of the dual-channel shared encoder. The dual-channel shared encoder is configured to perform dimensionality reduction feature extraction on the two input data (x0, m0) and (x1, m1) respectively, to obtain the intermediate feature vector V0 corresponding to the data (x0, m0) and the intermediate feature vector V1 corresponding to the data (x1, m1); The feature fusion layer is configured to fuse the intermediate feature vector V0 and the intermediate feature vector V1 to obtain a fused intermediate feature vector. The feature transformation output layer is configured to perform dimensionality reduction and feature extraction on the fused intermediate feature vector again to obtain the feature vector with a preset fixed dimension corresponding to the thickness abnormal strip steel sample.

4. The method according to claim 1, characterized in that, Cosine similarity can be used as the similarity between two vectors, or the similarity between two vectors can be determined based on the Euclidean distance between the vectors.

5. The method according to claim 1, characterized in that, The step of selecting a portion of historical abnormal strip thickness data from the preset abnormal strip thickness database as candidate cases includes: From the preset abnormal strip thickness database, the top N historical abnormal strip thickness data with the highest similarity between the feature vector contained therein and the feature vector of the strip to be detected are selected as candidate cases, where N is a preset positive integer.

6. The method according to claim 1, characterized in that, Based on the anomaly root cause labels of the candidate cases, the root cause tracing results of the detected strip thickness anomaly were determined: The abnormal root cause labels of all the candidate cases are classified and statistically analyzed, and the root cause tracing results of the abnormal strip thickness to be detected are determined based on the statistical results. The statistical results include: candidate cases corresponding to each type of abnormal root cause label.

7. The method according to claim 6, characterized in that, The steps for determining the root cause of the abnormal thickness of the strip steel to be tested based on statistical results include: Perform label statistical voting: determine the cumulative voting score of each type of abnormal root cause label included in the statistical results; wherein, the cumulative voting score of the abnormal root cause label is equal to the sum of the single-case voting scores of all the candidate cases corresponding to the abnormal root cause label, and the single-case voting score of the candidate case is equal to the similarity between the feature vector corresponding to the candidate case and the feature vector of the strip to be detected multiplied by the sample weight configured for the candidate case; The anomaly root cause label corresponding to the highest cumulative voting score is taken as the root cause tracing result of the anomaly in the thickness of the strip to be detected.

8. The method according to claim 7, characterized in that, After using the anomaly root cause label corresponding to the highest cumulative voting score as the root cause tracing result of the strip thickness anomaly to be detected, it also includes: Check whether the concentration of the abnormal root cause labels of all candidate cases is greater than or equal to a preset threshold, and whether the maximum cumulative vote score is greater than or equal to a preset score threshold; When the detected label distribution level is less than the preset level threshold, or when the detected maximum cumulative vote score is less than the preset score threshold, a prompt message is generated and output to indicate that the root cause tracing results need to be manually reviewed.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 8.

10. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images