A method for evaluating abnormal production state of a back light aluminized zinc plated steel sheet

Abstract

The application discloses a kind of backlight source aluminium-zinc plated steel plate production state abnormality evaluation methods, it is related to backlight source aluminium-zinc plated steel plate production state abnormality evaluation technical field.The method includes: obtaining the real-time magnetic field response signal of backlight source aluminium-zinc plated steel plate at multiple sampling time on continuous hot galvanizing aluminium production line;Perform multi-order differential geometry feature extraction operation, obtain the differential geometry characteristic parameter of each sampling time;Dynamic topological feature map of aluminium-zinc plated steel plate production state is constructed on time axis;Apply homotopy mapping transformation, generate the homotopy invariant characteristic quantity describing production state stability;According to the deviation degree of homotopy invariant characteristic quantity relative to preset standard state baseline, calculate the abnormality evaluation index of current production state.The method can realize accurate evaluation to production state abnormality.

Description

Technical Field

[0001] This invention relates to the field of abnormal production status assessment technology for aluminum-zinc coated steel sheets for backlights, specifically a method for assessing abnormal production status of aluminum-zinc coated steel sheets for backlights. Background Technology

[0002] In the continuous hot-dip galvanizing process of aluminum-zinc coated steel sheets for backlighting, real-time monitoring and anomaly assessment of the steel sheet's production status are necessary to avoid quality problems such as uneven coating thickness and poor adhesion caused by fluctuations in process parameters. Existing technologies typically employ time-domain statistical characteristic analysis of single process parameters or magnetic field response signals to assess the production status. For example, by calculating statistical quantities such as the mean, variance, or peak value of the magnetic field signal and comparing them with preset thresholds, it is determined whether the current state deviates from the normal range. This monitoring method, relying on manually set thresholds and simple statistical quantities, struggles to capture the dynamic geometric structure information reflecting subtle changes in the production status contained within the magnetic field signal when dealing with complex production processes like continuous hot-dip galvanizing, which involve multi-physics coupling and time-varying nonlinearity. When early minor anomalies occur in the production status, the statistical characteristics of the magnetic field signal often have not yet shown significant changes, resulting in a significant time lag in anomaly detection and a tendency to miss transient and intermittent anomalies. Meanwhile, most existing assessment methods are based on independent analysis of signals from individual sampling points, failing to fully explore the correlation and evolution patterns between states at different times during the production process, resulting in insufficient overall understanding of the stability of the entire production state. Therefore, how to extract higher-order and more sensitive geometric deformation features from magnetic field response signals, and how to construct methods that can characterize the continuous evolution stability of the production state and quantify its degree of deviation, are urgent problems to be solved. Summary of the Invention

[0003] This invention provides a method for assessing anomalies in the production status of backlight-coated aluminum-zinc steel sheets. This method extracts multi-order differential geometric features from real-time magnetic field response signals and constructs a dynamic topological feature map based on this. Then, it obtains homotopy-invariant feature quantities through homotopy mapping transformation to calculate the anomaly assessment index, thereby achieving highly sensitive detection of minor anomalies in the production status and objective quantitative assessment of the stability of the status evolution.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a method for assessing abnormal production status of backlight aluminum-zinc coated steel sheets, the method comprising:

[0005] Real-time magnetic field response signals of the backlight aluminized zinc-coated steel sheet at multiple sampling moments on a continuous hot-dip galvanized aluminum production line are acquired. Multi-order differential geometric feature extraction is performed on the real-time magnetic field response signals to obtain differential geometric feature parameters at each sampling moment. Based on the evolution trajectory of the differential geometric feature parameters on the time axis, a dynamic topological feature map of the aluminized zinc-coated steel sheet's production state is constructed. A homotopy mapping transformation is applied to the dynamic topological feature map to generate homotopy-invariant feature quantities describing the stability of the production state. Based on the degree of deviation of the homotopy-invariant feature quantities relative to a preset standard state baseline, an anomaly assessment index of the current production state of the backlight aluminized zinc-coated steel sheet is calculated.

[0006] This method applies an alternating excitation magnetic field to the surface of an aluminized zinc-coated steel sheet and collects the induced magnetic field response signal. It transforms the subtle changes in the magnetic field response into quantifiable geometric features, which can sensitively capture the minute fluctuations in coating uniformity, alloy layer formation quality, and substrate surface condition during the production process, thereby achieving high-precision assessment of abnormal production conditions.

[0007] As a preferred embodiment of the present invention, when performing multi-order differential geometric feature extraction on the real-time magnetic field response signal, the first-order numerical derivative of the real-time magnetic field response signal at each sampling moment is performed to obtain the first-order magnetic field rate of change component, and then the second-order numerical derivative is performed to obtain the second-order magnetic field acceleration component. Subsequently, the signal curvature feature value in phase space at that sampling moment is calculated, and the third-order numerical derivative of the signal curvature feature value is performed to obtain the signal torsion feature value. Finally, the first-order magnetic field rate of change component, the second-order magnetic field acceleration component, the signal curvature feature value, and the signal torsion feature value are combined and encoded to generate the differential geometric feature parameter at that sampling moment. Preferably, the order of the multi-order derivative is at least fourth-order, and higher-order curvature rates of change of the signal are further calculated based on the fourth-order derivative as supplementary components of the differential geometric feature parameter. This multi-order differential geometric feature extraction method extracts multi-dimensional information such as rate of change, acceleration, degree of bending and degree of torsion from the magnetic field response signal layer by layer. It effectively separates normal fluctuations and abnormal precursors that were originally difficult to distinguish in the differential geometric space, making the subsequent anomaly discrimination more sensitive and specific.

[0008] As a further preferred embodiment of the present invention, when constructing the dynamic topological feature graph, the differential geometric feature parameter at each sampling time is treated as a node. Nodes are connected in the order of sampling times to form an initial feature evolution path. The Hausdorff distance between two adjacent sampling times is calculated as the weight value of the edge connecting these two nodes. A node color label is assigned based on the sign of the signal torsion feature value in each node, with a positive sign corresponding to the first color label and a negative sign corresponding to the second color label. Finally, an initial weighted colorized graph is constructed based on the color labels of all nodes and the weight values ​​of all edges to obtain the dynamic topological feature graph of the production state of the aluminized zinc steel sheet. Preferably, the Hausdorff distance is calculated using the Chebyshev distance metric to determine the absolute value of the maximum component difference between two adjacent sampling times. This dynamic topological feature graph maps the continuous evolution process of the production state into a visually distinguishable graph structure. The edge weights characterize the drastic degree of state change, and the node colors intuitively indicate the change in the direction of magnetic field response torsion, facilitating subsequent identification of the essential transformation of the production state through the calculation of topological invariants.

[0009] In another preferred embodiment of the present invention, when applying a homotopy mapping transformation to a dynamic topological feature graph, a continuous homotopy mapping parameter is constructed with an initial value of zero and a final value of one. When the homotopy mapping parameter is zero, the dynamic topological feature graph is mapped to a complete graph. The edge weight between each pair of different nodes in the complete graph is taken as the shortest path distance between that pair of different nodes in the dynamic topological feature graph. During the continuous change of the homotopy mapping parameter from zero to one, the edge weights of the dynamic topological feature graph are continuously deformed so that when the homotopy mapping parameter is one, the original edge weight values ​​are restored. The set of critical homotopy parameter values ​​at which the color labels of all nodes in the dynamic topological feature graph change during this continuous change is recorded, and the statistical distribution characteristics of this set are calculated to obtain the homotopy-invariant feature quantity. Preferably, the homotopy mapping transformation uses a linear deformation function, and the relationship between the edge weight and the homotopy mapping parameter is a convex combination between the original weight and the weight of the complete graph. The homotopy mapping transformation process simulates the process of gradually restoring the topological graph from a fully connected state to the original state. The critical parameter value at which the node color label changes reveals the key transformation point in the topological structure. The homotopy invariant feature quantity extracted from this process provides a global description of the stability of the production state that is not affected by local noise.

[0010] In calculating the anomaly assessment index, as a preferred embodiment of the present invention, a standard state baseline is extracted from historical normal production state data. This baseline includes a standard homotopy parameter probability distribution function and a standard node color transfer entropy. The relative entropy value between the actual homotopy parameter probability distribution function and the standard homotopy parameter probability distribution function in the current homotopy invariant feature is calculated as a first distribution offset, and the absolute value of the difference between the current actual node color transfer entropy and the standard node color transfer entropy is calculated as a second entropy offset. The first distribution offset and the second entropy offset are then input into a preset weighted aggregation function, and a weighted sum is calculated using a first preset weight and a second preset weight. This weighted sum is used as the anomaly assessment index. This assessment method based on dual offset weighted aggregation takes into account both the global offset degree of the homotopy parameter distribution and the disordered changes in node color transfer, enabling the anomaly assessment index to comprehensively reflect the degree to which the current production state deviates from the normal pattern from both statistical distribution and transfer pattern dimensions.

[0011] Preferably, before inputting the first distribution offset and the second entropy offset into the weighted aggregation function, real-time thickness uniformity data collected by the radiation detector on the aluminized zinc steel sheet production line is acquired. The thickness fluctuation variance is calculated based on this data, and the values ​​of the first and second preset weights are dynamically adjusted according to the thickness fluctuation variance. Specifically, the larger the thickness fluctuation variance, the smaller the value of the first preset weight and the larger the value of the second preset weight. By introducing online detection data on thickness uniformity, which is directly related to the final product quality, adaptive adjustment of the evaluation index weights is achieved, making the evaluation results more closely related to the actual physical quality of the product and improving the engineering practicality of the evaluation.

[0012] As a complete technical solution of the present invention, it also includes a process for determining the production status of the backlight aluminum-zinc coated steel sheet based on the anomaly assessment index: the calculated anomaly assessment index is compared with preset mild anomaly thresholds and severe anomaly thresholds. When the anomaly assessment index is less than the mild anomaly threshold, it is determined to be in a normal state; when the anomaly assessment index is greater than or equal to the mild anomaly threshold but less than the severe anomaly threshold, it is determined to be in a mild anomaly state; when the anomaly assessment index is greater than or equal to the severe anomaly threshold, it is determined to be in a severe anomaly state. When it is determined to be in a mild anomaly state, the sampling time corresponding to the critical homotopy parameter value where the node color label changes in the dynamic topology feature graph is located, and this sampling time is marked as the anomaly start time point. This hierarchical judgment mechanism not only clearly distinguishes the severity level of the anomaly, but also accurately traces the start time of the anomaly at the mild anomaly stage, providing production line operators with a valuable early warning window and accurate investigation basis.

[0013] Preferably, when locating the anomaly initiation time point, the smallest critical homotopy parameter value in the set of critical homotopy parameter values ​​is extracted as the earliest change threshold. In the dynamic topology feature map, the target node whose critical homotopy parameter value equals this earliest change threshold when its node color label changes is searched. The sampling time corresponding to the target node is obtained as the candidate anomaly initiation time. The change trend of the differential geometric feature parameters within a preset time window before the candidate anomaly initiation time is traced. If the change trend shows a monotonically increasing or monotonically decreasing regular change, the candidate anomaly initiation time is determined as the final anomaly initiation time point. This tracing and verification mechanism can effectively eliminate misjudgments caused by occasional disturbances, ensuring that the marked anomaly initiation time point has real physical meaning.

[0014] As another preferred aspect of the invention, a preprocessing procedure is performed before acquiring the real-time magnetic field response signal: a fixed-frequency alternating excitation magnetic field is applied to the surface of the backlight-emitting aluminum-zinc coated steel plate. An array of magnetic field sensors is used to collect the induced magnetic field intensity on the steel plate surface at the annealing furnace outlet, zinc pot inlet, and air knife scraping point of the continuous hot-dip galvanized aluminum production line. The collected induced magnetic field intensities are time-synchronized and aligned according to the sampling time to obtain the original magnetic field intensity vector. Then, an adaptive filtering operation is performed on the original magnetic field intensity vector to eliminate power frequency interference and environmental noise, ultimately obtaining the real-time magnetic field response signal. Multi-location synchronous acquisition and adaptive filtering preprocessing ensure the spatiotemporal consistency and signal-to-noise ratio of the magnetic field response signal, providing a high-quality data foundation for subsequent differential geometric feature extraction.

[0015] Furthermore, after obtaining the anomaly assessment index, a production status feedback adjustment step is executed: based on the magnitude of the anomaly assessment index, a corresponding parameter adjustment command is matched from the preset control strategy library, and the parameter adjustment command is sent to the air knife pressure controller and zinc layer thickness feedback control system of the continuous hot-dip galvanized aluminum production line; when the production status corresponding to the anomaly assessment index is a mild anomaly, a pressure fine-tuning command is sent to the air knife pressure controller, and a sampling frequency increase command is sent to the zinc layer thickness feedback control system; when the production status corresponding to the anomaly assessment index is a severe anomaly, a shutdown alarm command is sent to the main controller of the continuous hot-dip galvanized aluminum production line, and the set of critical homotopy parameter values ​​where the color labels of all nodes in the dynamic topology feature graph change is packaged into an anomaly diagnosis log for storage. This hierarchical feedback adjustment mechanism realizes a closed-loop linkage from the assessment result to the production line control. Mild anomalies can be suppressed by slight corrections to process parameters and increased monitoring density, while severe anomalies trigger shutdown protection and complete anomaly information recording in a timely manner, balancing production efficiency and quality safety.

[0016] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0017] By performing multi-order differential geometric feature extraction on real-time magnetic field response signals, the algorithm moves beyond the time-domain statistics of the signal and delves into its geometric structure to obtain differential geometric feature parameters, including first-order magnetic field change rate components, second-order magnetic field change acceleration components, signal curvature eigenvalues, and signal torsion eigenvalues. These parameters can comprehensively describe the local morphology of the magnetic field signal in phase space from multiple dimensions such as change rate, curvature, and torsion. When initial minor disturbances occur in the production state, although the signal amplitude statistics may not have shifted significantly, the curvature and torsion characteristics of its internal geometry have already changed sensitively, allowing abnormal information to be captured earlier and more comprehensively. By constructing a dynamic topological feature map from the evolution trajectory of differential geometric feature parameters on the time axis and applying a homotopy mapping transformation to generate homotopy-invariant feature quantities, a quantitative description of the continuous evolution stability of the production state is achieved. This process concatenates isolated state feature points at each sampling time into a whole topological structure and uses homotopy mapping to track the critical parameters of node color label changes during continuous deformation. The homotopy-invariant characteristic quantity generated thus records the characteristic information of the dynamic topology of the production state remaining unchanged under continuous deformation, rather than the instantaneous state value at a certain moment. The anomaly assessment index calculated based on this characteristic quantity reflects the degree of deviation in the stability of the entire production process, and can effectively identify those abnormal patterns that appear normal in single-point statistical values ​​but whose overall evolutionary trend has potentially become unstable. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a flowchart of the method for assessing abnormal production status of aluminum-zinc coated steel sheets for backlights;

[0020] Figure 2 This is a flowchart of the multi-order differential geometric feature extraction operation;

[0021] Figure 3 This is a flowchart of the dynamic topology feature map construction process for the production status of aluminized zinc-coated steel sheets;

[0022] Figure 4 This is a schematic diagram of the process of generating homotopy-invariant eigenvalues ​​by homotopy mapping transformation;

[0023] Figure 5 This is a flowchart of the anomaly assessment index generation process based on hybrid offset weighted calculation;

[0024] Figure 6This is a flowchart for determining and adjusting the production status of aluminum-zinc coated steel sheets for backlighting. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] See Figure 1 This invention provides a method for assessing the abnormal production status of a backlight galvanized steel sheet, comprising: acquiring real-time magnetic field response signals of the backlight galvanized steel sheet at multiple sampling times on a continuous hot-dip galvanized aluminum production line; performing multi-order differential geometric feature extraction on the real-time magnetic field response signals to obtain differential geometric feature parameters at each sampling time; constructing a dynamic topological feature map of the production status of the galvanized steel sheet based on the evolution trajectory of the differential geometric feature parameters on the time axis; applying a homotopy mapping transformation to the dynamic topological feature map to generate a homotopy-invariant feature quantity describing the stability of the production status; and calculating an abnormal assessment index of the current production status of the backlight galvanized steel sheet based on the degree of deviation of the homotopy-invariant feature quantity relative to a preset standard state baseline.

[0027] Example 1:

[0028] In specific implementation, please refer to Figure 2 The real-time magnetic field response signal is subjected to multi-order differential geometric feature extraction to obtain the differential geometric feature parameters at each sampling time, which includes the following process.

[0029] For the real-time magnetic field response signal of the backlight aluminum-zinc coated steel sheet acquired at each sampling moment on the continuous hot-dip galvanizing aluminum production line, the first-order numerical derivative of the real-time magnetic field response signal is performed to obtain the first-order magnetic field change rate component at each sampling moment. The first-order numerical derivative adopts the three-point central difference method. At each sampling time, the amplitude of the real-time magnetic field response signal is denoted as... The time interval between adjacent sampling times is First-order magnetic field rate of change component The following calculation formula is used to obtain: At the initial sampling time of the real-time magnetic field response signal sequence, the first-order numerical derivative is calculated using the forward difference method. At the last sampling time, the first-order numerical derivative is calculated using the backward difference method. The second-order numerical derivative is then performed on the first-order magnetic field rate of change component sequence to obtain the second-order magnetic field acceleration component at the sampling time. This second-order numerical derivative directly affects the amplitude of the real-time magnetic field response signal, using the three-point center difference formula: At the sequence endpoints, the second-order numerical derivative uses the same one-sided difference processing logic as the first-order numerical derivative. This is based on the first-order magnetic field rate of change components at the sampling time. and the acceleration component of the second-order magnetic field change Calculate the signal curvature eigenvalues ​​of the real-time magnetic field response signal in phase space at the sampling time. The expression for the signal curvature eigenvalue is:

[0030]

[0031] in, Indicates the sampling time The first-order magnetic field rate of change component Indicates the sampling time The second-order magnetic field change acceleration component, This indicates the absolute value operation. The constant 1 in the denominator is a smoothing constant used to prevent singular values ​​from appearing when the first-order magnetic field rate of change component is zero. The denominator as a whole... This is the normalization factor. The signal curvature eigenvalue reflects the degree of local curvature of the real-time magnetic field response signal in the two-dimensional phase space trajectory composed of the signal value and the first-order magnetic field rate of change component.

[0032] After obtaining the signal curvature eigenvalue sequence at all sampling times, the third-order numerical derivative of the signal curvature eigenvalue sequence is performed to obtain the signal torsion eigenvalue at each sampling time. The third-order numerical derivative is achieved by successively performing the first-order numerical derivative: First, apply the three-point central difference to the sequence of signal curvature eigenvalues ​​to obtain the sequence of first-order derivatives of the signal curvature eigenvalues. : The second step is to process the sequence of first derivatives. By applying the three-point central difference again, the second derivative sequence of the signal curvature eigenvalues ​​is obtained. : The third step is to process the second derivative sequence. The third application of the three-point central difference yields the third derivative sequence of the signal curvature eigenvalues, which is the signal torsion eigenvalue at the sampling time. : At the boundary sampling moments of the signal curvature eigenvalue sequence, each of the above differential operations is replaced by the corresponding forward or backward differential. The first-order magnetic field rate of change component obtained at each sampling moment is then... Second-order magnetic field change acceleration component Signal curvature eigenvalues and signal torsion characteristic value Combinatorial encoding is performed to generate differential geometric feature parameters at each sampling time. The combinational encoding method involves concatenating the four scalar feature values ​​into a four-dimensional feature vector, denoted as... Each sampling time generates a four-dimensional differential geometric feature parameter.

[0033] The multi-order differential geometric feature extraction operation has a minimum order of fourth, and further calculates the higher-order curvature change rate of the signal based on the fourth derivative as a supplementary component of the differential geometric feature parameters. In implementation, a fourth-order numerical derivative is performed on the real-time magnetic field response signal to obtain the fourth-order magnetic field change rate component. The specific process of fourth-order numerical differentiation is as follows: First, the third-order magnetic field rate of change component is obtained by successively differentiating the real-time magnetic field response signal. Then, by performing the first-order numerical derivative on the third-order magnetic field rate of change component, the fourth-order magnetic field rate of change component is obtained. Using the fourth-order magnetic field rate of change component Noise suppression is applied to the calculation of the first derivative of the signal curvature eigenvalues. Using the fourth-order magnetic field rate of change component as a constraint term, the overshoot caused by high-frequency noise in the first derivative of the signal curvature eigenvalues ​​is corrected, thus obtaining the higher-order curvature rate of change of the signal. Higher-order rate of change of curvature As a supplementary component to the differential geometric characteristic parameters. When obtaining the rate of change of higher-order curvature. Subsequently, the differential geometric feature parameters at the sampling time are expanded from four dimensions to five dimensions, and the feature vector formed by the combined encoding is: .

[0034] Example 2:

[0035] In specific implementation, please refer to Figure 3 Based on the evolution trajectory of differential geometric feature parameters on the time axis, a dynamic topological feature map of the production state of aluminized zinc steel sheet is constructed, which includes the following process.

[0036] The differential geometric feature parameter at each sampling time is treated as a node, and the nodes are connected in chronological order of sampling time to form the initial feature evolution path. This is applied to data collected from a continuous hot-dip galvanized aluminum production line. At each sampling time, we obtain There are several differential geometric feature parameter vectors, each corresponding to a sampling time. Each differential geometric feature parameter vector is mapped to a graph structure. There are 10 nodes, and the node number is marked as the index of the sampling time. , The value range is 1 to Starting from the node at the first sampling time, connect the nodes at any two adjacent sampling times with an undirected edge. With nodes An edge is generated between them, and a total of [number] edges are generated. The set of edges constitutes the initial feature evolution path.

[0037] The Hausdorff distance between the differential geometric feature parameters at two adjacent sampling times is calculated and used as the weight of the edge connecting these two nodes. The Hausdorff distance is calculated using the Chebyshev distance metric to determine the absolute value of the maximum component difference between the differential geometric feature parameters at two adjacent sampling times. Let the th... The differential geometric feature parameters at each sampling time are vectors. ,vector Include One portion, The value of is determined by the dimension of the differential geometric characteristic parameter, the first Each component is denoted as ,in The component number is the component sequence number. The value range is 1 to Let the differential geometric characteristic parameter at the next adjacent sampling time be a vector. , its first Each component is denoted as Hausdorff distance between differential geometric feature parameters at two adjacent sampling times. Calculate using the following formula:

[0038]

[0039] in, Indicates the first Differential geometric feature parameter vector at each sampling time point In the The values ​​of each component Indicates the first Differential geometric feature parameter vector at each sampling time point In the The values ​​of each component This indicates the absolute value operation. This indicates traversing all component indices. The absolute value of the difference is calculated, and the maximum value is taken. The calculated Hausdorff distance is used as the connection node. With nodes The weight values ​​of the edges.

[0040] Based on the sign of the signal torsion characteristic value in the differential geometric characteristic parameters of each node, a node color label is assigned to that node, where a positive sign corresponds to the first color label and a negative sign corresponds to the second color label. For the ... Differential geometric feature parameter vector at each sampling time point The signal torsion eigenvalues ​​are vectors. A component at a fixed position in the signal is used as the criterion for judgment: when the signal torsion characteristic value is greater than or equal to zero, the node... The node color label is set to the first color label; when the signal torsion characteristic value is less than zero, the node... The node color label is set to the second color label. The first color label is displayed as a preset color, and the second color label is displayed as a different preset color. In the visualization, the first color label is red and the second color label is blue.

[0041] Based on the node color labels and edge weights of all nodes, an initial weighted colorized graph is constructed as the dynamic topological feature graph representing the production status of aluminized zinc-coated steel sheets. The node set of the initial weighted colorized graph is as follows: The sampling time corresponding to each sampling time There are nodes, and the set of edges connects nodes at adjacent sampling times. Each edge carries a corresponding Hausdorff distance weight, and each node carries a corresponding node color label. The initial weighted colorized graph is stored in a data structure consisting of an adjacency list and a node attribute list. The adjacency list records the neighboring nodes and their corresponding edge weights for each node, and the node attribute list records the node color label for each node.

[0042] Example 3:

[0043] In specific implementation, please refer to Figure 4 Applying a homotopy mapping transformation to the dynamic topological feature graph generates homotopy-invariant feature quantities that describe the stability of the production state, which includes the following process.

[0044] Construct a continuous homotopy mapping parameter, initializing it to zero and terminating it to one. The homotopy mapping parameter is denoted as . , For continuous variables in the real number field, In the closed interval The value is taken from the inner value. When the homotopy mapping parameter... When the value is 0, it corresponds to the initial state of the homotopy mapping transformation; when the homotopy mapping parameter... A value of 1 corresponds to the termination state of the homotopy mapping transformation. Homotopy mapping parameters. During the change from 0 to 1, with a preset step size Discrete sampling is performed to obtain a set of discrete parameter values. ,in, , This represents the total number of discrete sampling steps. Preset step size. Set to 0.01.

[0045] In homotopy mapping parameters When the value is zero, the dynamic topological feature graph is mapped to a complete graph, where the edge weight between each pair of distinct nodes in the complete graph is the shortest path distance between that pair of distinct nodes in the dynamic topological feature graph. The dynamic topological feature graph is a weighted colored graph containing... Each node and the nodes connecting adjacent sampling times. Each edge carries a corresponding Hausdorff distance weight. The weights of all edges in the dynamic topological feature graph are obtained, and an adjacency matrix is ​​constructed. Adjacency matrix The Line number Column elements Represents a node With nodes The edge weight between nodes, if the node With nodes If there is no direct edge between them, then The value can be positive infinity. In the adjacency matrix... Based on this, the Floyd shortest path algorithm is used to calculate the shortest path distance between any two nodes in the dynamic topological feature graph. For any pair of nodes... ,node With nodes The shortest path distance in the dynamic topology feature graph is denoted as . , Traverse all intermediate nodes using the Floyd algorithm. calculate Obtain. Nodes in the complete graph. With nodes The edge weight values ​​between them are set to The complete graph contains There are nodes, and there is an edge between each pair of different nodes. The edge weight is the shortest path distance between the corresponding node pair in the dynamic topology feature graph.

[0046] In homotopy mapping parameters During the continuous transformation from zero to one, the edge weights of the dynamic topological feature graph undergo continuous deformation, thus changing the homotopy mapping parameters. This method restores the original edge weights to a dynamic topological feature map. The homotopy mapping transformation uses a linear deformation function, and the edge weights vary with the homotopy mapping parameters. The relationship between the changes is a convex combination of the original weights and the weights of the complete graph. Let the nodes be... With nodes The original edge weights in the dynamic topological feature graph are... The edge weights in the complete graph are . During deformation, the corresponding homotopy mapping parameter edge weight values Determine according to the following formula:

[0047]

[0048] in, For nodes in the dynamic topology feature graph With nodes The original edge weights between nodes, if the nodes With nodes If there are no direct edges in the dynamic topological feature graph, then Values ; For nodes in the complete graph With nodes The edge weight between nodes With nodes The shortest path distance in the dynamic topology feature graph; For homotopy mapping parameters, ; These are the interpolation coefficients for the weights of the complete graph; These are the interpolation coefficients for the original weights. When hour, The edge weights take values ​​that are exactly the same as the full graph weights; when hour, The edge weights are restored to their original values ​​in the dynamic topological feature map; when When the edge weight is taken as a convex combination of the complete graph weight and the original edge weight, the edge weight is taken as a combination of the complete graph weight and the original edge weight.

[0049] In homotopy mapping parameters Each discrete sample value At this point, the weighted graph under the current parameters is reconstructed using the deformed edge weights, and all nodes in this weighted graph are reassigned color labels. The node color label assignment method is as follows: keeping the node color label assignment rules unchanged in the dynamic topology feature graph, the assignment is based on the positive or negative sign of the signal torsion feature value in the differential geometric feature parameter corresponding to the node; a positive sign corresponds to the first color label, and a negative sign corresponds to the second color label. Since the continuous deformation of the edge weights does not change the differential geometric feature parameter of the node itself, the node color label remains consistent across the homotopy mapping parameters. The changes are only affected by the original attributes of the nodes.

[0050] Recorded in homotopy mapping parameters During continuous change, the condition for all node color labels in the dynamic topological feature map to change is: under the homotopy mapping parameter At a certain critical value, the sign of the signal torsion eigenvalue of a node in the reconstructed weighted graph flips, causing a change in the node's color label. This sign flip is not directly caused by the deformation of the homotopy mapping edge weights, but rather by changes in the homotopy mapping parameters. When the value changes to a specific value, combined with changes in the differential geometric characteristic parameters of other nodes, the signal torsion characteristic value of that node is recalculated and its sign changes. When a node is in a homotopy mapping parameter... Node color label and homotopy mapping parameters When the node color labels are different, the homotopy mapping parameters will be changed. This is recorded as a critical homotopy parameter value. Iterate through all... For each node, collect the critical homotopy parameter values ​​corresponding to the node color label switching, forming a critical homotopy parameter value set. ,in In order to be in The total number of critical homotopy parameter values ​​that change across all node color labels during the entire process from 0 to 1.

[0051] For the set of critical homotopy parameter values Statistical distribution characteristics are calculated to obtain homotopy-invariant characteristics. The calculation of statistical distribution characteristics includes: calculating the sample mean of the set of critical homotopy parameter values. and sample standard deviation ; Hotopy mapping parameters The range of values Evenly divided into Statistical intervals, The value is 20, and the width of each statistical interval is... Count the number of critical homotopy parameter values ​​falling within each statistical interval, construct a frequency distribution histogram, and obtain the probability density estimation vector of the critical homotopy parameter. ,in, Indicates the first The frequency of the critical homotopy parameter value within a statistical interval; calculate the cumulative distribution function sample value vector of the critical homotopy parameter value. ,in For the front The sum of frequencies across statistical intervals. Homotopic invariant characteristics are derived from the sample mean. Sample standard deviation and probability density estimation vector Together, they constitute a homotopic invariant characteristic quantity that describes the stability of the production state of aluminized zinc steel sheets.

[0052] Example 4:

[0053] In specific implementation, please refer to Figure 5Based on the degree of deviation of the homotopy invariant characteristic quantity from the preset standard state baseline, the abnormal assessment index of the current production state of the backlight aluminum-zinc coated steel sheet is calculated, which includes the following process.

[0054] Standard baselines are extracted from historical normal production data. This data originates from real-time magnetic field response signal records collected during multiple production batches of backlight aluminum-zinc coated steel sheets operating continuously within the normal process parameter range on a continuous hot-dip galvanized aluminum production line. For each normal production batch's historical real-time magnetic field response signal records, multi-order differential geometric feature extraction, dynamic topological feature map construction, and homotopy mapping transformation operations are repeatedly performed to obtain the homotopy-invariant feature quantities corresponding to each normal production batch. The critical homotopy parameter values ​​from all normal production batches are aggregated to form a standard critical homotopy parameter set; statistical distribution calculations are then performed on this standard critical homotopy parameter set to obtain the standard homotopy parameter probability distribution function. Standard homotopy parameter probability distribution function Represented in the form of a probability density estimation vector, denoted as ,in, To count the total number of intervals, The value is 20, and the width of each statistical interval is 0.05. Indicates the first The frequency of the standard critical homotopy parameter value within a statistical interval. Extract the node color label switching events recorded in the homotopy-invariant features of all normal production batches, and statistically analyze the homotopy mapping parameters throughout the entire process. Calculate the number of node pairs whose node color labels change during the transition from 0 to 1, and calculate the standard node color transition entropy. Standard node color transfer entropy The calculation method is as follows: Iterate through all normal production batches and count the total number of events where the node color label changes from the first color label to the second color label. The total number of events in which a node's color label changes from the second color label to the first color label. The transition probability matrix is , Standard node color transfer entropy ,when or When the value is 0, the corresponding logarithmic term takes the value 0. The standard state baseline contains the standard homotopy parameter probability distribution function. and standard node color transfer entropy .

[0055] Calculate the relative entropy between the actual probability distribution function of the homotopy parameter in the current homotopy-invariant eigenvalue and the standard homotopy parameter probability distribution function, and use this as the first distribution offset. The actual probability distribution function of the homotopy parameter in the current homotopy-invariant eigenvalue is denoted as... , The probability density estimation vector is derived from the homotopy-invariant eigenvalues ​​of the current production process. Given: First distribution offset The calculation uses the relative entropy formula:

[0056]

[0057] in, Describes the actual probability distribution function of the homotopy parameter. relative to the standard homotopy parameter probability distribution function The first distribution offset; This represents the total number of intervals in the statistics. The value is 20; The actual probability distribution function of the homotopy parameter is represented at the th... Probability values ​​within a statistical interval; The standard homotopy parameter probability distribution function is represented in the th case. Probability values ​​within a statistical interval; Represents logarithmic operations to base 2; when When the value is 0, the entire product term takes the value of 0; when Greater than 0 and When the value is 0, the product term takes the value of 0. At this point, the first distribution offset is directly set to a preset maximum constant of 100.

[0058] Calculate the absolute value of the difference between the actual node color transfer entropy and the standard node color transfer entropy in the current homotopy-invariant features, and use this as the second entropy offset. Extract the actual node color transfer entropy from the current homotopy-invariant features. Actual node color transition entropy The calculation method is the same as the standard node color transfer entropy. Consistency: Count the number of times the first color label changes to the second color label from all node color label switching events recorded in the homotopy mapping transformation of the dynamic topology feature graph in the current production process. And the number of times the second color label switches to the first color label. Calculate the transition probability and Actual node color transition entropy ,when or When it is 0, the corresponding logarithmic term takes the value of 0. Second entropy offset. The calculation method is as follows , This indicates the absolute value operation.

[0059] The first distribution offset and the second entropy offset are input into a preset weighted aggregation function. The weighted aggregation function calculates a weighted sum based on the first preset weight of the first distribution offset and the second preset weight of the second entropy offset. This weighted sum is used as the anomaly assessment index. The weighted aggregation function is denoted as... The abnormal assessment index is denoted as The calculation formula is ,in The first preset weight has a value range of [value range missing]. ; This is the second preset weight; This is the offset of the first distribution; This is the offset of the second entropy.

[0060] Before inputting the first distribution offset and the second entropy offset into the preset weighted aggregation function, the following processing procedure is performed: real-time thickness uniformity data collected by a radiation detector on the aluminized zinc steel sheet production line is acquired. The radiation detector is installed after the air knife scraping point on the continuous hot-dip galvanized aluminum production line. By emitting X-rays or gamma rays that penetrate the surface of the aluminized zinc steel sheet, the intensity attenuation of the penetrating rays is detected to invert the coating thickness distribution on the surface of the aluminized zinc steel sheet in real time, obtaining a set of thickness sample values ​​along the width direction of the steel sheet. The real-time thickness uniformity data consists of multiple thickness sample values.

[0061] The thickness fluctuation variance is calculated based on real-time thickness uniformity data. Let the real-time thickness uniformity data include... The thickness sample value, the first Each thickness sample value is denoted as , The value range is 1 to The arithmetic mean of the thickness sample values ​​is denoted as thickness fluctuation variance The calculation formula is The thickness fluctuation variance reflects the uniformity of the coating thickness on the surface of the aluminized zinc-coated steel sheet at the current moment. A larger value indicates more drastic fluctuations in thickness.

[0062] The values ​​of the first and second preset weights are dynamically adjusted based on the thickness fluctuation variance. The larger the thickness fluctuation variance, the smaller the value of the first preset weight and the larger the value of the second preset weight. A baseline value for the thickness fluctuation variance is set. , The value is the average of the thickness fluctuation variance of multiple consecutive batches under historical normal production conditions. Less than or equal to At that time, the first preset weight The value is 0.7, the second preset weight. The value is 0.3; when Greater than And less than or equal to 2 times At that time, the first preset weight The value is 0.5, the second preset weight. The value is 0.5; when More than twice At that time, the first preset weight The value is 0.3, the second preset weight. The value is set to 0.7. By adjusting the weights based on the thickness fluctuation variance, the contribution of node color transfer entropy shift to the anomaly assessment index is increased when coating thickness fluctuations are aggravated, while distribution shift is used as the main contributing factor to the anomaly assessment index when coating thickness fluctuations are normal. Final Anomaly Assessment Index .

[0063] Example 5:

[0064] In specific implementation, please refer to Figure 6 The production status of the backlight aluminum-zinc coated steel sheet is determined based on the anomaly assessment index, which includes the following process.

[0065] The calculated anomaly assessment index is compared with preset thresholds for mild and severe anomalies. The threshold for mild anomalies is denoted as... The threshold for severe abnormalities is denoted as , and The value of is determined through statistical analysis of historical production data. A series of anomaly assessment index samples were collected during the normal production period of the backlight aluminum-zinc coated steel sheet on the continuous hot-dip galvanized aluminum production line, and the mean of the anomaly assessment index samples was calculated. and standard deviation Set a threshold for mild abnormalities Set a threshold for severe abnormalities .

[0066] When the anomaly assessment index is less than the mild anomaly threshold At this point, the current production status is determined to be normal. Under normal conditions, the coating quality of the aluminum-zinc coated steel sheet surface of the backlight is stable, the process parameters of the continuous hot-dip galvanized aluminum production line are within the controlled range, and the anomaly assessment index remains within the historical normal fluctuation range.

[0067] When the anomaly assessment index is greater than or equal to the mild anomaly threshold And less than the severe abnormality threshold At this point, the current production status is determined to be a slightly abnormal state. Under this slightly abnormal state, the coating quality of the aluminum-zinc coated steel sheet surface of the backlight shows a tolerable deviation trend, and a certain process parameter or magnetic field response characteristic of the continuous hot-dip galvanized aluminum production line deviates but has not yet reached the level requiring immediate shutdown.

[0068] When the anomaly assessment index is greater than or equal to the severe anomaly threshold At this point, the current production status is determined to be a severely abnormal state. Under this severely abnormal state, the coating quality on the surface of the aluminum-zinc coated steel sheet for the backlight exhibits significant deterioration, and the process parameters of the continuous hot-dip galvanized aluminum production line deviate severely from the normal range, posing a risk of producing a batch of unqualified products.

[0069] When a mild anomaly is identified, the sampling time corresponding to the critical homotopy parameter value where the node color label changes in the dynamic topology feature map is located, and this sampling time is marked as the anomaly initiation time. The specific implementation method for locating the sampling time corresponding to the critical homotopy parameter value where the node color label changes in the dynamic topology feature map is as follows: It involves using the set of critical homotopy parameter values ​​recorded during the generation process of the homotopy-invariant feature quantity. The critical homotopy parameter value with the smallest numerical value is extracted and taken as the earliest change critical value, which is denoted as . In the dynamic topology feature graph, traverse all nodes and find the critical homotopy parameter value corresponding to the earliest change critical value when the node's color label changes. The target node. There may be one or more target nodes. When multiple target nodes exist, the one with the smallest node index number is selected as the unique target node. The sampling time corresponding to the target node is obtained; the index of the sampling time corresponding to the target node is... Sampling time As the starting point of the candidate anomaly.

[0070] Tracing the changing trend of differential geometric feature parameters within a preset time window prior to the initial time of the candidate anomaly. The length of the preset time window is set to... Each sampling time, The value is 10. Extracting from the sampling time... up to the sampling time This continuous Differential geometric feature parameters at each sampling time. Extracting the signal curvature feature value sequence from the differential geometric feature parameters. Monotonicity detection is performed on the signal curvature eigenvalue sequence. The monotonicity detection method is as follows: calculate the first-order difference sequence of the signal curvature eigenvalue sequence. ,in, from Received When all first-order differences When all are greater than zero, the signal curvature eigenvalue sequence is determined to exhibit a monotonically increasing regular change; when all first-order differences are greater than zero, the signal curvature eigenvalue sequence exhibits a monotonically increasing regular change. When all values ​​are less than zero, the signal curvature feature value sequence is determined to exhibit a monotonically decreasing trend. If the signal curvature feature value sequence exhibits a monotonically increasing or monotonically decreasing trend, then the candidate anomaly's starting time is selected. The final anomaly initiation time point is determined. If the trend of the signal curvature characteristic value sequence does not satisfy the regularity of monotonically increasing or monotonically decreasing changes, the current candidate anomaly initiation time is discarded, and the set of critical homotopy parameter values ​​is selected. Remove from Then, the earliest change threshold is extracted again, and the steps of locating the target node and tracing the change trend are repeated until the abnormal starting time point that satisfies the monotonicity condition or the set of critical homotopy parameter values ​​is determined. The positioning process terminates when the set becomes empty.

[0071] Before obtaining the real-time magnetic field response signal of the backlight aluminum-zinc coated steel sheet at multiple sampling times on the continuous hot-dip galvanized aluminum production line, the following preprocessing procedure is also performed.

[0072] An alternating excitation magnetic field of fixed frequency is applied to the surface of the aluminum-zinc coated steel sheet used as a backlight. This alternating excitation magnetic field is generated by an excitation coil installed above the annealing furnace outlet of the continuous hot-dip galvanized aluminum production line. A fixed frequency is applied to the excitation coil. sinusoidal alternating current, The value is set to 1000 Hz, the excitation current amplitude is set to 5 amperes, the number of turns of the excitation coil is 200, and the generated alternating excitation magnetic field is incident perpendicularly on the surface of the aluminum-zinc coated steel plate of the backlight.

[0073] An array-type magnetic field sensor was used to collect the induced magnetic field intensity on the steel plate surface at the annealing furnace exit, zinc pot inlet, and air knife scraping point of a continuous hot-dip galvanized aluminum production line. The array-type magnetic field sensor contains multiple magnetic field sensing units, each a Hall effect sensor with a sensitivity of 1.3 mV / Gauss, arranged uniformly along the width of the steel plate, with a spacing of 20 mm between adjacent units, for a total of 16 units. At the annealing furnace exit, the array-type magnetic field sensor was installed 500 mm downstream of the excitation coil to collect the induced magnetic field intensity generated by the backlight-coated aluminum-zinc steel plate after annealing. At the zinc pot inlet, the array-type magnetic field sensor was installed 300 mm above the zinc pot surface to collect the induced magnetic field intensity of the backlight-coated aluminum-zinc steel plate before entering the zinc pot. At the air knife scraping point, the array-type magnetic field sensor was installed 100 mm downstream of the air knife nozzle to collect the induced magnetic field intensity of the backlight-coated aluminum-zinc steel plate after air knife scraping. Each of the three collection points yielded one output signal from the array-type magnetic field sensor.

[0074] The acquired induced magnetic field strengths are time-synchronized and aligned according to the sampling time to obtain the original magnetic field strength vector for each sampling time. The array-type magnetic field sensors at the three sampling locations are triggered by a unified external clock source, with a sampling frequency set to 1000 Hz, and the time interval between adjacent sampling times... The time is 1 millisecond. At each sampling time, the array magnetic field sensor at the annealing furnace outlet outputs 16 magnetic field strength values, the array magnetic field sensor at the zinc pot inlet outputs 16 magnetic field strength values, and the array magnetic field sensor at the air knife scraping point outputs 16 magnetic field strength values. The three sets of magnetic field strength values ​​are concatenated into an original magnetic field strength vector containing 48 elements.

[0075] An adaptive filtering operation is performed on the original magnetic field strength vector to eliminate power frequency interference and environmental noise, obtaining the real-time magnetic field response signal. The adaptive filtering operation uses a minimum mean square error (MMS) adaptive filter. The reference input signal of this MMS adaptive filter is a 50 Hz sinusoidal signal extracted from the power frequency power supply by a phase-locked loop (PLL) and its 90-degree phase-shifted signal. The filter order is set to 32, and the step size factor is set to 0.005. The MMS adaptive filter outputs a power frequency interference estimate. Subtracting the power frequency interference estimate from the original magnetic field strength vector yields the first-stage filtered signal. The first-stage filtered signal is further subjected to wavelet threshold denoising to eliminate environmental noise. The wavelet basis function is the Daubechiesdb8 wavelet, the decomposition level is set to 5 levels, and the threshold is selected using an adaptive threshold based on Stein's unbiased risk estimation. The signal after wavelet threshold denoising is the real-time magnetic field response signal, which corresponds to a 48-dimensional signal vector at each sampling time.

[0076] After obtaining the anomaly assessment index, the following production status feedback adjustment steps are also performed.

[0077] Based on the magnitude of the anomaly assessment index, corresponding parameter adjustment commands are matched from a pre-defined control strategy library. The pre-defined control strategy library is a lookup table structure; the index of the lookup table is the interval in which the anomaly assessment index falls, and the output of the lookup table is the corresponding parameter adjustment command. The anomaly assessment index interval is divided as follows: if the anomaly assessment index is less than... For the first interval, the anomaly assessment index is greater than or equal to and less than For the second interval, the anomaly assessment index is greater than or equal to This corresponds to the third interval. The parameter adjustment instructions for the first interval are no-operation instructions, indicating that no adjustment is needed; the parameter adjustment instructions for the second interval are the set of instructions for minor anomalies; and the parameter adjustment instructions for the third interval are the set of instructions for handling severe anomalies.

[0078] The parameter adjustment commands are sent to the air knife pressure controller and zinc layer thickness feedback control system of the continuous hot-dip galvanized aluminum production line. The air knife pressure controller is a programmable logic controller that receives analog control signals and controls the zinc layer thickness on the surface of the aluminum-zinc coated steel sheet by adjusting the pressure of the compressed air ejected from the air knife nozzle. The zinc layer thickness feedback control system is a closed-loop control system that adjusts the pressure setpoint of the air knife pressure controller by receiving the deviation between the feedback signal from the coating thickness gauge and the set target value.

[0079] When the production status corresponding to the anomaly assessment index is a minor anomaly, a pressure fine-tuning command is sent to the air knife pressure controller. The pressure fine-tuning command increases the pressure setpoint of the air knife pressure controller based on the current pressure value. The offset, The value is 0.05 MPa; at the same time, a sampling frequency increase command is sent to the zinc layer thickness feedback control system. The sampling frequency increase command increases the sampling frequency of the control loop of the zinc layer thickness feedback control system from the current 200 Hz to 400 Hz, so as to enhance the response speed to coating thickness fluctuations.

[0080] When the production status corresponding to the anomaly assessment index is a severe anomaly, a shutdown alarm command is sent to the main controller of the continuous hot-dip galvanized aluminum production line. The shutdown alarm command triggers the main controller to execute a preset emergency stop procedure, stopping the continuous conveying of the steel strip and shutting down the zinc pot heating device; at the same time, the set of critical homotopy parameter values ​​where the color labels of all nodes in the dynamic topology feature graph change is also recorded. The abnormal diagnostic logs are packaged and stored as binary files on the industrial computer hard drive of the production line. The filenames include timestamp information, and the storage path is the preset diagnostic log directory.

[0081] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for assessing abnormal production status of aluminum-zinc coated steel sheets for backlights, characterized in that, The method includes: Acquire real-time magnetic field response signals of the backlight aluminum-zinc coated steel sheet at multiple sampling moments on a continuous hot-dip galvanized aluminum production line; Perform multi-order differential geometric feature extraction on the real-time magnetic field response signal to obtain the differential geometric feature parameters at each sampling time; Based on the evolution trajectory of the differential geometric feature parameters on the time axis, a dynamic topological feature map of the production state of aluminized zinc steel sheet is constructed. Apply a homotopy mapping transformation to the dynamic topology feature graph to generate homotopy-invariant feature quantities that describe the stability of the production state; Based on the degree of deviation of the homotopy invariant characteristic quantity relative to the preset standard state baseline, the abnormality assessment index of the current production state of the backlight aluminum-zinc coated steel sheet is calculated.

2. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 1, characterized in that, The process of performing multi-order differential geometric feature extraction on the real-time magnetic field response signal to obtain differential geometric feature parameters at each sampling time specifically includes: The first-order numerical derivative of the real-time magnetic field response signal at each sampling time is obtained to obtain the first-order magnetic field change rate component at that sampling time. The second-order numerical derivative of the first-order magnetic field rate of change component is then performed to obtain the second-order magnetic field acceleration component at the sampling time. Based on the first-order magnetic field change rate component and the second-order magnetic field change acceleration component, calculate the signal curvature characteristic value of the real-time magnetic field response signal in phase space at the sampling moment. The signal torsion characteristic value at the sampling time is obtained by taking the third-order numerical derivative of the signal curvature characteristic value. The first-order magnetic field rate of change component, the second-order magnetic field acceleration component, the signal curvature characteristic value, and the signal torsion characteristic value are combined and encoded to generate the differential geometric characteristic parameter at the sampling moment.

3. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 2, characterized in that, The order of the multi-order derivative includes at least the fourth order, and the higher-order curvature change rate of the signal is further calculated based on the fourth-order derivative as a supplementary component of the differential geometric feature parameter.

4. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 2, characterized in that, Based on the evolution trajectory of the differential geometric feature parameters on the time axis, a dynamic topological feature map of the production state of aluminized zinc-coated steel sheet is constructed, specifically including: The differential geometric feature parameter at each sampling time is treated as a node, and the nodes are connected in the order of sampling time to form the initial feature evolution path. Calculate the Hausdorff distance between the differential geometric feature parameters at two adjacent sampling times, and use it as the weight value of the edge connecting these two nodes; Based on the sign of the signal torsion characteristic value in the differential geometric characteristic parameter of each node, a node color label is assigned to the node, where a positive sign corresponds to the first color label and a negative sign corresponds to the second color label; Based on the node color labels of all nodes and the weight values ​​of all edges, an initial weighted coloring graph is constructed as a dynamic topological feature graph of the production status of aluminized zinc steel sheets.

5. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 4, characterized in that, The Hausdorff distance is calculated using the Chebyshev distance metric to determine the absolute value of the maximum component difference between the differential geometric feature parameters at two adjacent sampling times.

6. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 4, characterized in that, Applying a homotopy mapping transformation to the dynamic topological feature graph generates homotopy-invariant feature quantities describing the stability of the production state, specifically including: Construct continuous homotopy mapping parameters, such that the initial value of the homotopy mapping parameters is zero and the final value is one; When the homotopy mapping parameter is zero, the dynamic topology feature graph is mapped to a complete graph, where the edge weight between each pair of different nodes in the complete graph is the shortest path distance between the pair of different nodes in the dynamic topology feature graph. During the process of the homotopy mapping parameter continuously changing from zero to one, the edge weights of the dynamic topological feature graph are continuously deformed so that the dynamic topological feature graph with the homotopy mapping parameter of one is restored to the original edge weight values. Records the set of critical homotopy parameter values ​​at which the color labels of all nodes in the dynamic topological feature map change during the continuous variation of the homotopy mapping parameters. The statistical distribution characteristics of the critical homotopy parameter value set are calculated to obtain the homotopy invariant characteristic quantity.

7. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 6, characterized in that, The homotopy mapping transformation uses a linear deformation function, where the relationship between the edge weights and the homotopy mapping parameters is a convex combination between the original weights and the weights of the complete graph.

8. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 6, characterized in that, Based on the degree of deviation of the homotopy invariant characteristic quantity relative to the preset standard state baseline, an anomaly assessment index for the current production state of the backlight aluminum-zinc coated steel sheet is calculated, specifically including: Extract a standard state baseline from historical normal production data. This standard state baseline includes a standard homotopy parameter probability distribution function and a standard node color transfer entropy. Calculate the relative entropy between the actual probability distribution function of the homotopy parameter and the standard homotopy parameter probability distribution function in the current homotopy invariant feature quantity, and use it as the first distribution offset; Calculate the absolute value of the difference between the actual node color transfer entropy and the standard node color transfer entropy in the current homotopy invariant features, and use it as the second entropy offset; The first distribution offset and the second entropy offset are input into a preset weighted aggregation function. The weighted aggregation function calculates a weighted sum based on the first preset weight of the first distribution offset and the second preset weight of the second entropy offset. The weighted sum is used as the anomaly evaluation index.

9. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 8, characterized in that, Before inputting the first distribution offset and the second entropy offset into the preset weighted aggregation function, the following processing procedure is also performed: Obtain real-time thickness uniformity data collected by radiation detectors on the aluminized zinc steel sheet production line; The thickness fluctuation variance is calculated based on the real-time thickness uniformity data. The values ​​of the first preset weight and the second preset weight are dynamically adjusted according to the thickness fluctuation variance. The larger the thickness fluctuation variance, the smaller the value of the first preset weight and the larger the value of the second preset weight.

10. The method for assessing abnormal production status of aluminized zinc-coated steel sheet for backlighting according to claim 1, characterized in that, The method further includes: The production status of the aluminized zinc-plated steel sheet for the backlight is determined based on the anomaly assessment index, specifically including: The calculated anomaly assessment index is compared with the preset thresholds for mild and severe anomalies. When the anomaly assessment index is less than the mild anomaly threshold, the current production status is determined to be normal. When the anomaly assessment index is greater than or equal to the mild anomaly threshold and less than the severe anomaly threshold, the current production status is determined to be a mild anomaly. When the anomaly assessment index is greater than or equal to the severe anomaly threshold, the current production status is determined to be a severe anomaly. When a mild anomaly is identified, the sampling time corresponding to the critical homotopy parameter value at which the node color label changes in the dynamic topology feature map is located, and this sampling time is marked as the anomaly start time point.

Images