A pressure mapping method based on an array of pressure sensors

Abstract

The application discloses a pressure mapping method based on a pressure sensor array, which comprises the following steps: performing gradient direction analysis on an original pressure matrix output by the pressure sensor array in real time, matching a material motion direction, and generating a direction-sensitive pressure feature matrix; obtaining the direction-sensitive pressure feature matrix generated at different collection time points within a preset time window, obtaining a trajectory sequence, generating a pressure distribution kinematics feature vector sequence, inputting the pressure evolution state prediction model, and outputting evolution state information; generating a per-sensor unit single-point-global pressure evolution feature vector, inputting the per-sensor unit single-point-global pressure evolution feature vector into a pressure mapping abnormality identification and evaluation model, and outputting a local risk value of the sensor unit; and generating a risk thermodynamic map according to the local risk value of each sensor unit in the pressure sensor array, and identifying a pressure mapping abnormal area. Thus, accurate monitoring, analysis and abnormal mapping of pressure distribution in the material motion process are realized.

Description

Technical Field

[0001] This invention relates to the field of sensor technology, and in particular to a pressure mapping method based on a pressure sensor array. Background Technology

[0002] In industrial production, many scenarios require monitoring the pressure distribution and motion of materials on equipment. For example, the movement of fabrics or metal strips on tensioners, guide rollers, grippers, or conveyor belts directly reflects the stress and motion of the material. Accurately acquiring pressure distribution information and analyzing its dynamic changes is crucial for ensuring product quality, stable equipment operation, and precise control.

[0003] Traditional pressure monitoring methods often employ single pressure sensors, acquiring only localized pressure values ​​and failing to comprehensively reflect pressure distribution. This is particularly true for capturing the dynamic changes and directional characteristics of pressure during material movement. Furthermore, when faced with complex operating conditions and interference factors such as equipment vibration, traditional methods struggle to accurately distinguish between real pressure signals and noise interference, leading to measurement errors and hindering reliable data for equipment control. Moreover, current technologies lack comprehensive analysis of pressure distribution and material motion states, making it difficult to predict abnormal movement trends in advance and failing to meet the high-precision, high-reliability, and intelligent control requirements of modern industrial production.

[0004] Therefore, a pressure mapping method is needed that can comprehensively and accurately monitor pressure distribution and analyze its dynamic changes, providing a reliable basis for operating condition monitoring. Summary of the Invention

[0005] This application provides a pressure mapping method based on a pressure sensor array, which enables accurate monitoring, analysis, and anomaly mapping of pressure distribution during material movement.

[0006] This application provides a pressure mapping method based on a pressure sensor array, including: S101 generates a direction-sensitive pressure feature matrix by performing gradient direction analysis on the original pressure matrix output in real time from the preset pressure sensor array and matching it with the corresponding material movement direction. S102: Within a preset time window, obtain the direction-sensitive pressure feature matrix generated at different acquisition time points. By continuously tracking the spatial displacement of the pressure center, obtain the trajectory sequence, generate the pressure distribution kinematic feature vector sequence, input it into the pre-trained pressure evolution state prediction model, and output the evolution state information. S103, based on the direction-sensitive pressure feature matrix of continuously acquired time points within the time window, obtains the spatial gradient amplitude characteristics of each sensor unit; S104: Based on each sensor unit, generate a single-point-global pressure evolution feature vector, input it into the pre-trained sensor unit pressure mapping anomaly identification and evaluation model, and output the local risk value of the sensor unit. S105 generates a risk heat map based on the local risk value of each sensor unit in the pressure sensor array, and identifies abnormal areas of pressure mapping based on the risk heat map.

[0007] Therefore, S101 specifically includes: S201, Obtain the original M×N dimensional pressure matrix output by the pressure sensor array at the current acquisition time point. M and N represent the number of rows and columns of the pressure sensor array, respectively, and the actual motion direction vector of the corresponding material is obtained through a preset encoder; S202, perform gradient direction analysis on the original pressure matrix, calculate the gradient direction angle, and convert it into a biangular vector. ; S203, based on the biangular vector and the material motion direction vector, calculates the pressure-motion direction matching factor at the corresponding sensor unit location. ; S204, based on a preset direction-sensitive enhancement algorithm, enhances the original pressure matrix according to the pressure-motion direction matching factor to generate a direction-sensitive pressure feature matrix. .

[0008] Preferably, the preset direction sensitivity enhancement algorithm is set as follows:

[0009] in, For sensor unit The directional pressure characteristic value at that location, The preset directional gain coefficient, This is the pressure-motion direction matching factor corresponding to the location of the sensor unit. For sensor unit The original pressure value at the location.

[0010] Preferably, S102 specifically includes: S301, locates the pressure center coordinates corresponding to each acquisition time point within the preset time window, forming a trajectory sequence. , Let T be the k-th sampling time point within the time window, and T be the total number of sampling time points within the time window. The coordinates of the kth pressure center; S302, Generate a sequence of kinematic feature vectors for pressure distribution based on the trajectory sequence. Each kinematic feature vector includes the instantaneous velocity value, instantaneous acceleration value, and direction change of the pressure center at the corresponding acquisition time point. S303 inputs the kinematic feature vector sequence of pressure distribution into the pre-trained pressure evolution state prediction model and outputs evolution state information, including state labels and predicted trajectory offset vectors.

[0011] Preferably, the pressure center coordinates at each sampling time point are obtained as follows:

[0012] in, For sensor unit The directional pressure characteristic value at that location, This represents the cumulative vertical pressure value in column j. This represents the cumulative value of the lateral pressure in the i-th row. The sum of global pressure. Let j be the column-space weight value of the j-th column. Let be the row-space weight value of the i-th row.

[0013] Preferably, the column-direction spatial weight value and the row-direction spatial weight value are determined based on the pressure-motion direction matching factor, specifically: The column-space weight value of column j is determined as follows: The pressure-motion direction matching factor of all sensor units in column j is normalized and then averaged to obtain the pressure-motion direction matching factor of column j. ; Match the pressure-motion direction factor in column j. As the column-space weight value of the j-th column.

[0014] The row-space weight value of the i-th row is determined as follows: The pressure-motion direction matching factor of all sensor units in the i-th row is obtained by normalizing the factor and then averaging the results. ; Match the pressure-motion direction factor in the i-th row As the row-space weight value of the i-th row.

[0015] Preferably, the pre-trained stress evolution state prediction model is obtained as follows: C1. Collect trajectory sequences and pressure distribution kinematic feature vector sequences within a large number of historical event windows of the pressure sensor array. Label each historical pressure distribution kinematic feature vector sequence, and set the label content as the actual motion state label and the predicted trajectory offset vector as the actual evolution state information. C2. Use all the labeled historical pressure distribution kinematic feature vector sequences as the training sample set, and use the training sample set to train the pre-selected neural network structure to continuously optimize the model parameters and obtain the pressure evolution state prediction model.

[0016] Preferably, the single-point-global pressure evolution feature vector of the sensor unit includes the position coordinates of the sensor unit in the pressure sensor array, spatial gradient amplitude characteristics, and evolution state information.

[0017] Preferably, S105 specifically includes: The local risk values ​​of all sensor units in the pressure sensor array are mapped according to the array position to obtain the risk heat map of the pressure sensor array within the time window. All sensor units with local risk values ​​greater than the preset risk threshold are identified as target heat points. Risk dominance domain analysis is performed on the target heat points, and the risk dominance domain is mapped into the pressure sensor array to lock the abnormal areas in the corresponding directional sensitive feature matrix of the pressure sensor array.

[0018] Preferably, the risk-dominant domain analysis includes: clustering based on the distance values ​​between target heat points, and taking the cluster with the most target heat points as the risk-dominant domain.

[0019] One or more technical solutions provided in this application have at least the following technical effects or advantages: By analyzing the gradient direction and matching it with the material motion direction based on the raw pressure matrix output in real time from a pressure sensor array, a direction-sensitive pressure feature matrix is ​​generated, eliminating measurement bias. A kinematic feature vector sequence is generated by tracking the spatial displacement of the pressure center, and evolution state information is output using a pre-trained model to quantify the dynamic evolution law. The spatial gradient amplitude characteristics of sensor units are obtained to generate a single-point-to-global pressure evolution feature vector and assess local risk values. A risk heatmap is generated to identify abnormal regions. Combined with information on abnormal regions and evolution state, equipment control decision commands are generated and converted into an equivalent tension physical model. This system achieves comprehensive and accurate monitoring of pressure distribution and analysis of its dynamic changes, effectively eliminating interference and providing a reliable basis for equipment control. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of the pressure mapping method based on a pressure sensor array according to an embodiment of the present invention. Detailed Implementation

[0021] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0022] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] Example 1: Figure 1 This is a schematic flowchart of a pressure mapping method based on a pressure sensor array according to an embodiment of the present invention.

[0025] like Figure 1 As shown, a pressure mapping method based on a pressure sensor array includes the following steps: S101 generates a direction-sensitive pressure feature matrix by performing gradient direction analysis on the original pressure matrix output in real time from the preset pressure sensor array and matching it with the corresponding material movement direction.

[0026] Specifically, the raw pressure matrix output by the pressure sensor array is first acquired. This matrix consists of M rows and N columns of pressure values. Gradient direction analysis is then performed on the raw pressure matrix, which involves calculating the pressure change rate (i.e., pressure gradient component) at each sensor unit position in the horizontal and vertical directions to obtain the pressure gradient direction angle at each point. To avoid ambiguity in direction judgment, the angle is converted into a biangular vector form (containing cosine and sine components). The biangular vector is then matched with the material movement direction detected externally. Based on the matching degree, the raw pressure values ​​are enhanced to generate a direction-sensitive pressure feature matrix.

[0027] The material movement direction is determined by an external sensor (encoder or vision system), which indicates the direction of movement of the detected material. Based on preset parameters according to the material's friction characteristics, the influence of the direction matching on the pressure enhancement is adjusted.

[0028] It should be noted that the pressure sensor array is mounted on the contact surface of the associated equipment (e.g., tensioner: guide roller / gripper / conveyor belt pressure surface) according to the actual scenario, and is in direct contact with the material being tested (e.g., fabric / metal strip). The pressure sensor array is arranged along the preset movement direction of the material (e.g., 8×16 units, spacing 0.5-2mm).

[0029] In some embodiments, step S101 specifically includes: S201, Obtain the original M×N dimensional pressure matrix output by the pressure sensor array at the current acquisition time point. M and N represent the number of rows and columns of the pressure sensor array, respectively. The actual motion direction vector of the corresponding material is obtained through a preset encoder. , is used to represent the components of a material in the horizontal and vertical directions.

[0030] S202, perform gradient direction analysis on the original pressure matrix, calculate the gradient direction angle, and convert it into a biangular vector. .

[0031] Specifically, gradient direction analysis is performed on the original pressure matrix, including: A1. Based on the rate of change of pressure values ​​at the position (i,j) of each sensor unit in the horizontal and vertical directions, calculate the gradient component in the x-direction. and gradient components in the y direction This reflects the changing trends of the pressure field in the horizontal and vertical directions.

[0032] Wherein, the gradient component in the x-direction and gradient components in the y direction The results are calculated using the following formulas:

[0033] For sensor unit Gradient component in the x-direction For sensor unit gradient component in the y-direction For sensor unit The original pressure value at the location, For sensor unit The original pressure value at the location, The distance (mm) between the two sensor units in the x-direction. For sensor unit The original pressure value at the location, For sensor unit The original pressure value at the location, y is the distance (mm) between the two sensor units in the y direction.

[0034] It should be noted that if the sensor unit At the boundary of the sensor array, in this case, when in the boundary row:

[0035] When in a boundary column:

[0036] A2. Calculate the pressure gradient direction angle at the corresponding sensor unit location based on the x-direction gradient components and the y-direction gradient components. This indicates the dominant direction of pressure change at each location.

[0037] Wherein, the pressure gradient direction angle Calculated using the following formula:

[0038] For sensor unit The pressure gradient direction angle at that location, It is the arctangent function in the four quadrants.

[0039] A3. Adjust the pressure gradient direction angle Convert to biangular vector To avoid the ambiguity of 180° direction determination in special mathematical expressions, cosine is used to strengthen the orthogonal direction characteristics, and sine is used to eliminate the interference of abrupt changes in direction, thus solving the problem of discontinuous direction in traditional angle representation.

[0040] S203, based on the biangular vector and the material motion direction vector, calculates the pressure-motion direction matching factor at the corresponding sensor unit location. .

[0041] Specifically, the pressure-motion direction matching factor is calculated using the following formula:

[0042] in, Represents the vector dot product. Represents the magnitude of the vector. The range is [-1, 1]. The larger the value, the higher the directional consistency, which quantifies the degree of matching between the pressure gradient and the material motion.

[0043] S204, based on a preset direction-sensitive enhancement algorithm, enhances the original pressure matrix according to the pressure-motion direction matching factor to generate a direction-sensitive pressure feature matrix. .

[0044] Specifically, the preset direction sensitivity enhancement algorithm is set as follows:

[0045] in, For sensor unit The directional pressure characteristic value at that location, The preset directional gain coefficient (ranging from 0.2 to 0.8) is pre-set by experts based on the material's friction characteristics to adjust the degree of influence of directional matching on pressure enhancement. For example, for metallic materials... The value is 0.5 (high friction), corresponding to polymer materials. The value is 0.3 (low friction). Enhancement is applied only to positive matching, where the pressure gradient direction is in the same direction as the material movement (e.g., the pressure leading edge during strip advancement), reflecting the true tension load and enhancing the reliability of the signal. For example, in metal rolling, enhancing the high-pressure zone at the leading edge can improve thickness control accuracy. Reverse matching refers to pressure gradients in the opposite direction to the movement (e.g., pressure fluctuations caused by vibration, which are usually automatically classified as equipment vibration characteristics). If enhancement amplifies noise, it can lead to misjudgment; therefore, excessive processing should be avoided to prevent trajectory distortion.

[0046] The directional pressure characteristic values ​​of all sensor units in the pressure sensor array are used to form a directional pressure characteristic matrix according to the array position. .

[0047] Therefore, the directional measurement deviation caused by material motion is eliminated, the pressure signal characteristics in the main motion direction are enhanced, and pressure characteristic values ​​that can reflect the sensitivity of different directions are obtained, thus eliminating the pressure distribution distortion caused by material motion.

[0048] S102: Within a preset time window, the direction-sensitive pressure feature matrix generated at different acquisition time points is obtained. By continuously tracking the spatial displacement of the pressure center, a trajectory sequence is obtained, and a pressure distribution kinematic feature vector sequence is generated. This sequence is then input into the pre-trained pressure evolution state prediction model, and the evolution state information, including state labels and predicted trajectory offset vectors, is output.

[0049] Specifically, based on the direction-sensitive pressure feature matrix, the position of the "center of gravity" of the entire pressure field, i.e. the pressure center, is calculated. The pressure center is continuously recorded at multiple time points to form a trajectory sequence. The movement speed at adjacent time points and the rate of change of speed over time (acceleration) are analyzed through the trajectory sequence. The change in angle of movement direction is also calculated to extract the kinematic features of the pressure distribution. Using a pre-trained pressure evolution state prediction model, the current pressure distribution motion state label (stable operation, lateral offset, accelerated sliding, high pressure gradient, etc.) and the trajectory offset vector of the prediction window are determined.

[0050] Therefore, by analyzing the trajectory of the pressure center, the static pressure distribution is transformed into dynamic motion characteristics, and the dynamic evolution law of the pressure distribution is quantified. Its output provides a kinematic basis for the mapping of pressure in subsequent anomaly judgment and control decision-making, which helps to provide early warning of abnormal motion trends. It is a key link connecting pressure distribution characteristics with actual equipment control.

[0051] In some embodiments, step S102 specifically includes: S301, locates the pressure center coordinates corresponding to each acquisition time point within the preset time window, forming a trajectory sequence. , Let T be the k-th sampling time point within the time window, and T be the total number of sampling time points within the time window. Let be the coordinates of the kth pressure center.

[0052] Specifically, the pressure center coordinates at each data collection time point are obtained as follows:

[0053] in, For sensor unit The directional pressure characteristic value at that location, This represents the cumulative vertical pressure value in column j. This represents the cumulative value of the lateral pressure in the i-th row. The sum of global pressure. Let j be the column-space weight value of the j-th column. This is the row-space weight value for the i-th row, which is set according to the actual application scenario.

[0054] Specifically, the column-direction spatial weight values ​​and row-direction spatial weight values ​​are determined based on the pressure-motion direction matching factor (since the reverse matching mentioned above is generally caused by vibration, the vibration source can be reflected based on the pressure-motion direction matching factor. The smaller the pressure-motion direction matching factor, the smaller the vibration source at the corresponding position. Therefore, a smaller weight value is assigned to this pressure value, thereby suppressing vibration interference in the determination of the pressure center). Specifically: The column-space weight value of column j is determined as follows: The pressure-motion direction matching factor of all sensor units in column j is normalized and then averaged to obtain the pressure-motion direction matching factor of column j. ; The pressure-motion direction matching factor in column j As the column-space weight value of the j-th column.

[0055] The row-space weight value of the i-th row is determined as follows: The pressure-motion direction matching factor of all sensor units in the i-th row is obtained by normalizing the factor and then averaging the results. ; The pressure-motion direction matching factor in the i-th row As the row-space weight value of the i-th row.

[0056] Therefore, the trajectory sequence is used to record the spatial evolution path of the pressure center, providing a temporal basis for kinematic analysis.

[0057] S302 generates a sequence of kinematic feature vectors for pressure distribution based on the trajectory sequence. Each kinematic feature vector includes the instantaneous velocity value, instantaneous acceleration value, and directional change of the pressure center at the corresponding acquisition time point.

[0058] The pressure distribution kinematic feature vector sequence consists of T pressure distribution kinematic feature vectors, arranged in the order of the acquisition time points, and corresponding one-to-one with the pressure center coordinates in the trajectory sequence.

[0059] The instantaneous velocity value is calculated using the following formula:

[0060] in, Let be the instantaneous velocity value of the pressure center change at the k-th sampling time point. Let k be the coordinates of the pressure center. Let k-1 be the coordinates of the pressure center. For the k-th sampling time point within the time window, This represents the (k-1)th sampling time point within the time window.

[0061] The instantaneous acceleration value is calculated using the following formula:

[0062] in, Let be the instantaneous acceleration value of the pressure center change at the k-th sampling time point. Let be the instantaneous velocity value of the pressure center change at the k-th sampling time point. This represents the instantaneous velocity value of the pressure center change at the (k-1)th sampling time point.

[0063] The change in direction is calculated using the following formula:

[0064] in, This represents the change in the direction of the pressure center at the k-th sampling time point. The meanings of other parameters are as described above, and will not be repeated here.

[0065] S303 inputs the kinematic feature vector sequence of pressure distribution into the pre-trained pressure evolution state prediction model and outputs evolution state information, including state labels and predicted trajectory offset vectors.

[0066] Specifically, the pre-trained stress evolution state prediction model is obtained as follows: C1. Collect trajectory sequences and pressure distribution kinematic feature vector sequences within a large number of historical event windows of the pressure sensor array. Label each historical pressure distribution kinematic feature vector sequence. The label content is set as the actual motion state label (based on the actual scenario, the coded value label can be used. The larger the coded value, the greater the degree of abnormal working condition of the motion state) and the predicted trajectory offset vector.

[0067] Specifically, for each historical pressure distribution kinematic feature vector sequence, the actual motion state label is determined as follows: it is judged based on a predefined kinematic feature vector threshold (set according to the actual application scenario and expert experience, used to measure the motion state of the pressure center within the current time window), or it can be directly labeled by experts (based on the actual working condition of the pressure sensor array mapping tensioner surface or the corresponding material within the corresponding historical time window).

[0068] The predicted trajectory offset vector is set as follows: obtain the trajectory sequence in the next historical time window corresponding to the historical time window, compare the offset vectors of the mean pressure center coordinates corresponding to the two trajectory sequences respectively, including the offset distance value and offset direction angle value of the two mean pressure center coordinates.

[0069] It should be noted that the collected historical trajectory sequences need to cover a large number of normal operating condition trajectory sequences and abnormal operating condition trajectory sequences in order to improve the generalization of the model training.

[0070] C2. Use all the labeled historical pressure distribution kinematic feature vector sequences as the training sample set, and use the training sample set to train the pre-selected neural network structure to continuously optimize the model parameters and obtain the pressure evolution state prediction model.

[0071] S103, based on the directional pressure feature matrix of continuously acquired time points within the time window, obtains the spatial gradient amplitude characteristics of each sensor unit, including the average spatial gradient amplitude and the spatial gradient amplitude fluctuation factor.

[0072] Specifically, the spatial gradient magnitude characteristics of each sensor unit are obtained as follows: D1. The spatial gradient magnitude of the sensor unit at each acquisition time point is calculated using the following formula:

[0073] in, Indicates sensor unit Spatial gradient magnitude at the corresponding acquisition time point, For sensor unit With sensor unit Horizontal spacing value, For sensor unit With sensor unit The vertical spacing value. Therefore, the spatial gradient magnitude is used to quantify the intensity of local pressure abrupt changes.

[0074] D2. Based on the spatial gradient amplitudes of all acquisition time points within the time window, take the average and standard deviation to obtain the average spatial gradient amplitude and the spatial gradient amplitude fluctuation factor.

[0075] S104: Based on each sensor unit, generate a single-point-global pressure evolution feature vector, input it into the pre-trained sensor unit pressure mapping anomaly identification and evaluation model, and output the local risk value of the sensor unit.

[0076] Specifically, the single-point-global pressure evolution feature vector of the sensor unit includes the sensor unit's position coordinates in the pressure sensor array (which can be row and column coordinates, or position coordinates obtained from a coordinate system established with the center of the pressure sensor array as the reference), spatial gradient magnitude features, and evolution state information. The pre-trained sensor unit pressure mapping anomaly identification and evaluation model is obtained as follows: E1. Collect the single-point-global pressure evolution feature vectors of sensor units in the pressure sensor array within a large number of historical time windows, and label each historical single-point-global pressure evolution feature vector with the label content set as the degree of pressure mapping anomaly of the corresponding sensor unit.

[0077] It should be noted that the pressure mapping anomaly level of the sensor unit is set between 0 and 1. The larger the value, the greater the anomaly level of the pressure value mapping at the corresponding time point of the sensor unit. It is scored by experts or by preset rules.

[0078] For example, since pressure sensor arrays apply a standard directional pressure feature matrix in different application scenarios or environments (such as the contact surface scenario of the mounting tensioner and the properties of the tested material), an anomaly degree value can be obtained by comparing the standard pressure value of the corresponding sensor unit in the standard directional pressure feature matrix with the pressure value to be evaluated.

[0079] It should be noted that physical simulations can also be used to generate a partial training sample set, but this invention will not elaborate on this.

[0080] E2. Use all the labeled historical single-point-global pressure evolution feature vectors as the training sample set, and use the training sample set to train the pre-selected neural network structure to continuously optimize the model parameters and obtain the sensor unit pressure mapping anomaly identification and evaluation model.

[0081] S105 generates a risk heat map based on the local risk value of each sensor unit in the pressure sensor array, and identifies abnormal areas of pressure mapping based on the risk heat map.

[0082] Specifically, the local risk values ​​of all sensor units in the pressure sensor array are mapped according to the array position to obtain a risk heat map of the pressure sensor array within a time window. All sensor units with local risk values ​​greater than a preset risk threshold (e.g., set to 0.6) are identified as target heat points. Risk dominance domain analysis is performed on the target heat points, and the risk dominance domain is mapped onto the pressure sensor array to lock the abnormal regions in the corresponding direction sensitive feature matrix of the pressure sensor array.

[0083] The risk-dominant domain analysis includes: clustering based on the distance values ​​(Euclidean distance values ​​between location coordinates) between target heat points, and taking the cluster with the most target heat points as the risk-dominant domain.

[0084] Furthermore, the identification of abnormal pressure mapping areas needs to be promptly synchronized to the administrator platform for data marking and maintenance.

[0085] S106 generates equipment control decision commands through abnormal region and evolution state information, and converts the abnormal region of the direction-sensitive pressure characteristic matrix into a physical model of equivalent tension to reflect the tension control state of the material movement process in the associated equipment.

[0086] Specifically, the equipment control decision commands are pre-set according to the actual scenario. For example: F1. If the proportion of abnormal areas in the pressure sensor array is greater than a preset proportion threshold (predefined based on actual conditions and expert experience): When the evolution status information meets the first preset condition, control the working parameters of the associated device to the preset state and send a prompt to the management platform; Otherwise, maintain the current operating parameters of the associated device and send a notification to the management platform.

[0087] F2. If the proportion of the abnormal area in the pressure sensor array is less than or equal to the preset proportion threshold: When the evolution status information meets the second preset condition, control the working parameters of the associated device to the preset state and send a prompt to the management platform; Otherwise, maintain the current operating parameters of the associated device.

[0088] It should be noted that the threshold for meeting the first preset condition must be lower than the threshold for meeting the second preset condition. This can be understood as the requirement of the first preset condition to measure the degree of anomaly in the evolved state information being lower than that of the second preset condition. For example, in the evolved state information, the degree of anomaly corresponding to the state label and the offset degree of the predicted trajectory offset vector are compared with preset anomaly thresholds. The anomaly threshold determined by the first preset condition must be lower than the anomaly threshold determined by the second preset condition. This invention will not elaborate further on this. It should be noted that if the degree of anomaly of any object in the evolved state information is greater than the anomaly threshold, it is defined as meeting the preset condition. Furthermore, the degree of anomaly corresponding to the state label is predefined by the administrator, i.e., different state labels correspond to different degree of anomaly values ​​and anomaly thresholds (used to define the degree of anomaly). The anomaly threshold of the predicted trajectory offset vector is also predefined by the administrator, i.e., different offset degrees correspond to different degree of anomalies and anomaly thresholds.

[0089] As an example, when the proportion of abnormal areas is greater than the preset proportion threshold, if the evolution status information is: the status label is stable operation and the predicted trajectory offset is less than the abnormal threshold (indicating that the degree of offset is small), the current working parameters of the corresponding associated equipment can be maintained; if the evolution status information is: the status label is accelerated sliding and the predicted trajectory offset is greater than the abnormal threshold (indicating that the degree of offset is large), the tension roller (main drive frequency converter) of the associated equipment is controlled to slow down.

[0090] Specifically, the physical model of the direction-sensitive pressure characteristic matrix is ​​converted into an equivalent tension, and set as follows:

[0091] in, For equivalent tension, For sensor unit The directional pressure characteristic value at that location, This is an abnormal area. The effective sensing area of ​​the sensor unit (set at the factory when the pressure sensor array is shipped). The elastic modulus of the material being tested (preset based on material properties, e.g., steel: 200 GPa, rubber: 0.01 GPa). The contact arc length between the tested material and the associated equipment is measured in real time using an encoder.

[0092] Furthermore, the equivalent tension converted from the abnormal region of the direction-sensitive pressure characteristic matrix is ​​compared with the pre-set theoretical tension, and the relevant equipment is controlled through the management platform. As an example: in the cold-rolled steel strip production line, the equivalent tension corresponding to the abnormal region is detected to be 12375N, while the target theoretical tension is 12000N, triggering a control command: the hydraulic system is pressurized by 3%.

[0093] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: By performing gradient direction analysis on the raw pressure matrix output in real time from a preset pressure sensor array and matching it with the corresponding material motion direction, a direction-sensitive pressure feature matrix is ​​generated. This eliminates directional measurement bias caused by material motion, enhances the pressure signal characteristics in the main motion direction, and obtains pressure feature values ​​reflecting different directional sensitivities. It also eliminates pressure distribution distortion caused by material motion, making pressure measurement results more accurate and truly reflecting the force situation of the material during motion. This solves the problem that traditional single pressure sensors cannot comprehensively reflect pressure distribution and directional characteristics, as well as measurement bias caused by material motion.

[0094] Within a preset time window, direction-sensitive pressure feature matrices are obtained at different acquisition time points. By continuously tracking the spatial displacement of the pressure center, a trajectory sequence is obtained, generating a kinematic feature vector sequence of pressure distribution. This sequence is input into a pre-trained pressure evolution state prediction model, outputting evolution state information. This transforms static pressure distribution into dynamic motion features, quantifying the dynamic evolution law of pressure distribution and providing kinematic basis for subsequent anomaly judgment and control decisions, thus facilitating early warning of abnormal movement trends. This solves the problem of traditional methods lacking analysis of dynamic changes in pressure distribution and failing to provide early warning of abnormal movement trends.

[0095] Based on the direction-sensitive pressure feature matrix acquired continuously within a time window, the spatial gradient amplitude characteristics of each sensor unit are obtained, including the average spatial gradient amplitude and the spatial gradient amplitude fluctuation factor. The spatial gradient amplitude is used to quantify the intensity of local pressure abrupt changes, while the average spatial gradient amplitude and the spatial gradient amplitude fluctuation factor reflect the spatial and temporal variations in pressure, providing important feature information for subsequent pressure mapping anomaly identification. This provides local pressure abrupt change characteristics for a more comprehensive analysis of pressure distribution, compensating for the shortcomings of analyzing only the overall pressure distribution.

[0096] For each sensor unit, a single-point-to-global pressure evolution feature vector is generated and input into a pre-trained sensor unit pressure mapping anomaly identification and evaluation model, outputting the local risk value of that sensor unit. By comprehensively considering the sensor unit's location, spatial gradient magnitude features, and evolution state information, the degree of pressure mapping anomaly of the sensor unit is accurately assessed. This solves the problem that traditional methods cannot accurately assess the pressure mapping anomaly of a single sensor unit.

[0097] A risk heatmap is generated based on the local risk value of each sensor unit in the pressure sensor array, and abnormal pressure mapping areas are identified based on the risk heatmap. This provides a visual representation of the risk status of each region in the pressure sensor array, and accurately pinpoints abnormal areas through risk-dominant domain analysis. This solves the problem that traditional methods cannot visually display and accurately identify abnormal pressure mapping areas.

[0098] By analyzing information on abnormal regions and their evolution, this method generates equipment control decision commands. It transforms abnormal regions in the direction-sensitive pressure feature matrix into a physical model of equivalent tension, generating reasonable equipment control decision commands based on actual conditions. This conversion of pressure characteristics into an equivalent tension physical model provides a more intuitive basis for equipment control. Specifically, different decision commands are generated based on the proportion of abnormal regions and their evolution status information. A specific formula is used to convert the direction-sensitive pressure feature matrix into an equivalent tension physical model. This solves the problem that traditional methods cannot generate effective equipment control decision commands based on pressure distribution and intuitively reflect the tension control status.

[0099] In summary, this method generates a direction-sensitive pressure feature matrix by analyzing the gradient direction and matching it with the material motion direction based on the raw pressure matrix output in real time from the pressure sensor array, thus eliminating measurement bias. It also tracks the spatial displacement of the pressure center to generate a kinematic feature vector sequence, uses a pre-trained model to output evolution state information, and quantifies the dynamic evolution law. Furthermore, it acquires the spatial gradient amplitude characteristics of sensor units, generates a single-point-to-global pressure evolution feature vector, and assesses local risk values. A risk heatmap is generated to identify abnormal regions. Finally, it combines the abnormal regions and evolution state information to generate equipment control decision commands and converts them into an equivalent tension physical model. This method achieves comprehensive and accurate monitoring of pressure distribution and analysis of its dynamic changes, effectively eliminates interference, provides a reliable basis for equipment control, and solves the technical problems of traditional methods failing to comprehensively reflect pressure distribution, analyze dynamic changes, accurately assess anomalies, and generate effective control commands.

[0100] It enables precise monitoring, analysis, and anomaly identification of pressure distribution during material movement, providing strong support for real-time control and optimization of equipment in industrial production, and improving production efficiency and product quality.

[0101] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A pressure mapping method based on a pressure sensor array, characterized in that, include: S101, by performing gradient direction analysis on the raw pressure matrix output in real time from the preset pressure sensor array and matching it with the corresponding material movement direction, a direction-sensitive pressure feature matrix is ​​generated: obtaining the M×N dimensional raw pressure matrix output by the pressure sensor array at the current acquisition time point. M and N represent the number of rows and columns of the pressure sensor array, respectively. The actual motion direction vector of the corresponding material is obtained through a preset encoder. Gradient direction analysis is performed on the original pressure matrix to calculate the gradient direction angle and convert it into a bi-angle vector. ; Based on the biangular vector and the material motion direction vector, the pressure-motion direction matching factor at the corresponding sensor unit location is calculated. Based on a preset direction-sensitive enhancement algorithm, the original pressure matrix is ​​enhanced according to the pressure-motion direction matching factor to generate a direction-sensitive pressure feature matrix. ; S102: Within a preset time window, obtain the direction-sensitive pressure feature matrix generated at different acquisition time points. By continuously tracking the spatial displacement of the pressure center, obtain the trajectory sequence, generate the pressure distribution kinematic feature vector sequence, input it into the pre-trained pressure evolution state prediction model, and output the evolution state information. S103, based on the direction-sensitive pressure feature matrix of continuously acquired time points within the time window, obtains the spatial gradient amplitude characteristics of each sensor unit; S104: Based on each sensor unit, generate a single-point-global pressure evolution feature vector, input it into the pre-trained sensor unit pressure mapping anomaly identification and evaluation model, and output the local risk value of the sensor unit. S105 generates a risk heat map based on the local risk value of each sensor unit in the pressure sensor array, and identifies abnormal areas of pressure mapping based on the risk heat map.

2. The pressure mapping method based on a pressure sensor array as described in claim 1, characterized in that, The preset direction sensitivity enhancement algorithm is set as follows: ， in, For sensor unit The directional pressure characteristic value at that location, The preset directional gain coefficient, This is the pressure-motion direction matching factor corresponding to the location of the sensor unit. For sensor unit The original pressure value at the location.

3. The pressure mapping method based on a pressure sensor array as described in claim 2, characterized in that, S102 specifically includes: S301, locates the pressure center coordinates corresponding to each acquisition time point within the preset time window, forming a trajectory sequence. , Let T be the k-th sampling time point within the time window, and T be the total number of sampling time points within the time window. The coordinates of the kth pressure center; S302, Generate a sequence of kinematic feature vectors for pressure distribution based on the trajectory sequence. Each kinematic feature vector includes the instantaneous velocity value, instantaneous acceleration value, and direction change of the pressure center at the corresponding acquisition time point. S303 inputs the kinematic feature vector sequence of pressure distribution into the pre-trained pressure evolution state prediction model and outputs evolution state information, including state labels and predicted trajectory offset vectors.

4. The pressure mapping method based on a pressure sensor array as described in claim 3, characterized in that, The method for obtaining the pressure center coordinates at each acquisition time point is as follows: ， ， in, For sensor unit The directional pressure characteristic value at that location, This represents the cumulative vertical pressure value in column j. This represents the cumulative value of the lateral pressure in the i-th row. The sum of global pressure. Let j be the column-space weight value of the j-th column. Let be the row-space weight value of the i-th row.

5. The pressure mapping method based on a pressure sensor array as described in claim 4, characterized in that, The column-direction spatial weight values ​​and row-direction spatial weight values ​​are determined based on the pressure-motion direction matching factor, specifically: The column-space weight value of column j is determined as follows: The pressure-motion direction matching factor of all sensor units in column j is normalized and then averaged to obtain the pressure-motion direction matching factor of column j. ; The pressure-motion direction matching factor in column j As the column-space weight value of the j-th column, The row-space weight value of the i-th row is determined as follows: The pressure-motion direction matching factor of all sensor units in the i-th row is obtained by normalizing the factor and then averaging the results. ; Match the pressure-motion direction factor in the i-th row As the row-space weight value of the i-th row.

6. The pressure mapping method based on a pressure sensor array as described in claim 4, characterized in that, The pre-trained stress evolution state prediction model is obtained as follows: C1. Collect trajectory sequences and pressure distribution kinematic feature vector sequences within a large number of historical event windows of the pressure sensor array. Label each historical pressure distribution kinematic feature vector sequence, and set the label content as the actual motion state label and the predicted trajectory offset vector as the actual evolution state information. C2. Use all the labeled historical pressure distribution kinematic feature vector sequences as the training sample set, and use the training sample set to train the pre-selected neural network structure to continuously optimize the model parameters and obtain the pressure evolution state prediction model.

7. The pressure mapping method based on a pressure sensor array as described in claim 6, characterized in that, The single-point-global pressure evolution feature vector of the sensor unit includes the position coordinates of the sensor unit in the pressure sensor array, spatial gradient amplitude characteristics, and evolution state information.

8. The pressure mapping method based on a pressure sensor array as described in claim 7, characterized in that, S105 specifically includes: The local risk values ​​of all sensor units in the pressure sensor array are mapped according to the array position to obtain the risk heat map of the pressure sensor array within the time window. All sensor units with local risk values ​​greater than the preset risk threshold are identified as target heat points. Risk dominance domain analysis is performed on the target heat points, and the risk dominance domain is mapped into the pressure sensor array to lock the abnormal areas in the corresponding directional sensitive feature matrix of the pressure sensor array.

9. The pressure mapping method based on a pressure sensor array as described in claim 8, characterized in that, The risk-dominant domain analysis includes: clustering based on the distance values ​​between target heat points, and taking the cluster with the most target heat points as the risk-dominant domain.

Images