A lead electrolytic cell pole plate short circuit grading early warning method and device

Abstract

The application provides a lead electrolytic cell pole plate short circuit grading early warning method and device. It includes continuous acquisition of infrared image sequence, establishment of pole plate arrangement topology model and obtaining of each pole plate area sequence, extraction of temperature mean value, local temperature gradient mean value and hot spot area proportion, calculation of adjacent pole plate difference feature and thermal field trend feature; the fused time sequence feature sequence is input into a short circuit time sequence recognition model to obtain state probability of each short circuit risk state, and a rule risk value is determined according to the deviation degree of the early warning index relative to the normal benchmark, and the short circuit early warning level is determined in combination with the state probability, the rule risk value and the grading threshold value. The application can continuously identify, locate and grade early warning of the short circuit early thermal anomaly of the single pole plate, reduce false positives and false negatives, provide grading basis for inspection, pole plate adjustment, sediment cleaning or tank maintenance, thereby reducing invalid current and power loss, reducing pole plate burning and product quality fluctuation, and reducing the risk of unplanned tank shutdown.

Description

Technical Field

[0001] This invention belongs to the field of electrolytic metallurgical equipment condition monitoring and industrial infrared intelligent detection technology, specifically relating to a method and device for graded early warning of short circuits in lead electrolytic cell plates. Background Technology

[0002] Lead electrolysis production typically involves alternating cathode and anode plates along a predetermined direction within an electrolytic cell containing electrolyte, operating continuously under high cell current conditions. With the development of infrared thermal imaging, industrial vision, and online monitoring technologies, utilizing the increased local current density and Joule heat accumulation caused by electrode short circuits to identify anomalies has become an important technological direction for replacing some manual inspections and contact temperature measurements.

[0003] In actual production, the electrodes are exposed to strong acid, high humidity, acid mist, precipitate growth, and electrolyte disturbance for extended periods. Conductive paths may form due to installation misalignment, plate deformation, cathode deposit overlap, anode mud or other precipitate accumulation, and localized reduction in plate spacing. Electrode short circuits increase in ineffective current, reduce current efficiency, and increase energy loss, potentially leading to electrode burnout, cathode product quality fluctuations, tank shutdowns for maintenance, and production interruptions. Early thermal anomalies in the short circuit phase are usually not significant high-temperature points, but rather manifest as a slow rise in the temperature of a single electrode, a gradual increase in the local temperature gradient, a continuous expansion of the hotspot area, and a gradual widening of the thermal field difference between the hotspot and adjacent electrodes.

[0004] Existing detection methods include manual touch or temperature measurement inspection, calculating the average temperature of a single frame or the temperature difference between the electrode and the electrolyte based on a pre-calibrated electrode area and applying a threshold, identifying anomalies by segmenting the electrolytic cell image into blocks and using local feature values, and determining the electrode state using target detection, instance segmentation, clustering, or ordinary time-series classification models. These methods can provide a certain level of detection capability when a short circuit has already caused a significant temperature rise.

[0005] However, fixed-area or single-frame threshold methods struggle to consistently correspond to the same electrode under conditions of camera pose changes, perspective scale changes, acid fog obstruction, thermal reflection, and short-term image blurring, easily leading to electrode area mismatch or number jumps. Relying solely on absolute temperature or maximum temperature cannot adequately characterize the small but progressively increasing thermal anomalies in the early stages of a short circuit, and synchronous temperature rises across the entire tank caused by changes in ambient temperature, tank current, or overall operating conditions can easily result in false alarms. When relying solely on a time-series model, alarm results lack verifiable physical indicators by on-site personnel; relying solely on fixed rules is prone to missed or false alarms due to normal baseline drift under different tank types, production cycles, and loads. Existing methods also struggle to stably distinguish between short circuit initiation, minor short circuits, and severe short circuits according to the degree of risk evolution. Summary of the Invention

[0006] The technical problem to be solved by the present invention is that the existing short circuit monitoring of lead electrolytic cell plates is difficult to stably locate and continuously track a single plate under complex working conditions. Furthermore, it is not timely in identifying weak and continuously evolving local thermal anomalies in the early stage of short circuits, and is easily affected by the temperature rise of the entire cell and the normal reference drift, resulting in false alarms and missed alarms, making it difficult to achieve reliable graded early warning.

[0007] In a first aspect, the present invention provides a graded early warning method for short circuits in lead electrolytic cell plates, the method comprising the following steps: Multiple infrared images are continuously acquired according to a preset sampling period to obtain an infrared image sequence covering at least one electrode group in the lead electrolytic cell; the electrode group includes cathode plates and anode plates arranged alternately along the electrode arrangement direction; Based on the structural parameters of the lead electrolytic cell and the electrode parameters of the electrode group, an electrode arrangement topology model is established to determine the corresponding region of each electrode in each infrared image sequence, thereby obtaining the electrode region sequence corresponding to each electrode. The electrode arrangement topology model is used to characterize the arrangement order, polarity, and adjacency relationship of each electrode. For any target electrode in the electrode group, the single-plate thermal field features of the target electrode are extracted based on the temperature values ​​of the corresponding regions in each infrared image; the single-plate thermal field features include the average temperature, the average local temperature gradient, and the proportion of hot spot area. Calculate the difference in single-plate thermal field characteristics between the target plate and its adjacent plates to obtain the differential characteristics of adjacent plates; adjacent plates are determined based on their adjacency relationship. The thermal field characteristics of the target electrode are fitted with a trend over time to obtain the thermal field trend characteristics of the target electrode; the thermal field trend characteristics characterize the trend of the thermal field of the target electrode changing over time. The thermal field characteristics of a single plate, the differential characteristics of adjacent plates, and the thermal field trend characteristics are fused to obtain the temporal feature sequence of the target plate. The temporal feature sequence is then input into a pre-trained short-circuit temporal identification model to obtain the state probability of the target plate belonging to each preset short-circuit risk state. The short-circuit temporal identification model is a multi-class temporal model trained using historical temporal feature sequences labeled with different preset short-circuit risk states, which is used to output the state probability of each preset short-circuit risk state. Early warning indicators are determined from time-series feature sequences, and rule-based risk values ​​are determined based on the degree of deviation of the early warning indicators from preset normal benchmarks. The early warning indicators include at least one of the following: the average temperature of the target electrode, the average local temperature gradient, the proportion of hot spot area, the difference in average temperature between the target electrode and adjacent electrodes, and the slope of the average temperature change. The short-circuit warning level of the target plate is determined based on the comparison results of the state probability, rule risk value and preset classification threshold.

[0008] Optionally, based on the cell structure parameters of the lead electrolytic cell and the electrode parameters of the electrode group, an electrode arrangement topology model is established to determine the corresponding region of each electrode in the infrared image sequence, including: The effective area of ​​the tank is determined based on the tank structure parameters. Based on the number of plates, the spacing between plates, the arrangement direction of plates, and the polarity of each plate, a candidate search area corresponding to each plate is generated within the effective area of ​​the tank. In each candidate search area, the plate region is divided, and the plate regions are assigned plate numbers according to the plate arrangement order. Based on the matching cost between the polariton region in the current infrared image and the historical trajectory prediction region, the polariton regions at adjacent sampling times are correlated; the expression for calculating the matching cost is:

[0009] in, Indicates time The The current electrode region and the first The matching cost between historical polar plate trajectories; Indicates center distance Non-negative weights Indicates the first The center coordinates of the current electrode region Indicates the first The trajectory of the historical plate at any moment The predicted center coordinates; Indicates the degree of regional overlap Non-negative weights Indicates the first The current electrode region at time The area mask, Indicates the first The trajectory of the historical plate at any moment The predicted region mask, This indicates the calculation of the intersection-union ratio; Indicates area change Non-negative weights; Indicates the first The area of ​​the current electrode region Indicates the first The predicted area of ​​the historical polar plate trajectory. Indicates a preset positive number. This indicates taking the maximum value; The historical electrode trajectory corresponding to the current electrode region is determined according to the principle of minimizing matching cost, so as to keep the electrode number of the same electrode consistent in the electrode region sequence.

[0010] Optionally, the average temperature, the average local temperature gradient, and the proportion of hot spot area are determined according to the following formulas:

[0011]

[0012]

[0013]

[0014] in, Indicates the first At any time, the block plate average temperature Indicates the first At any time, the block plate The mean of the local temperature gradient, Indicates the first At any time, the block plate The percentage of hot spot area; Indicates the first At any time, the block plate The set of pixels in the region; Indicates the number of pixels contained in the region's pixel set; Represents pixel coordinates; Indicates time Pixel coordinates in infrared images Temperature value; This represents the temperature gradient along the horizontal direction. This represents the temperature gradient in the vertical direction. This is an indicator function that takes the value of 1 when the condition inside the parentheses is true and takes the value of 0 when the condition inside the parentheses is false. Indicates the first Hot spot determination threshold for bulk electrode plates; Indicates the first The baseline average temperature of the bulk electrode under historical normal operating conditions; Indicates the first The baseline standard deviation of the temperature of the bulk electrode under historical normal operating conditions; This represents the safety margin coefficient used to determine the hot spot detection threshold.

[0015] Optionally, the difference in single-plate thermal field characteristics between the target plate and its adjacent plates is calculated to obtain the differential characteristics of adjacent plates, including: For each plate's thermal field characteristics among the average temperature, average local temperature gradient, and hot spot area percentage, the difference between the target plate's plate thermal field characteristics and the average value of the corresponding plate thermal field characteristics of adjacent plates is calculated to obtain the adjacent plate differential characteristics; the calculation expression for the adjacent plate differential characteristics is as follows:

[0016] in, Indicates the first At any time, the block plate The average temperature, the average local temperature gradient, or the percentage of hot spot area; Indicates the first The difference between adjacent plates corresponding to the thermal field characteristics of a single plate; This indicates that the topological model of the electrode arrangement determines the relationship with the first... The set of adjacent plates of a block electrode; Indicates the number of plates in the plate set; In the set of plates, the first... At any time, the block plate The corresponding single-plate thermal field characteristics.

[0017] Optionally, the calculation expression for the thermal trend characteristics is as follows:

[0018] in, Indicates the first The single-plate thermal field characteristics corresponding to the block plate at time The slope of the change; Indicates the cutoff time. The set of sampling moments within a time window; Represents the first in the set of sampling times. Each sampling time; This represents the average value of all sampling times in the set of sampling times; Indicates the first At the sampling time of the block plate Corresponding single-plate thermal field characteristics; Indicates the first time within the time window The average value of the single-plate thermal field characteristics corresponding to the block plate.

[0019] Optionally, the thermal field characteristics of a single plate, the differential characteristics of adjacent plates, and the thermal field trend characteristics are fused to obtain the temporal characteristic sequence of the target plate, including: The single-plate thermal field characteristics, adjacent plate difference characteristics, and thermal field trend characteristics at each sampling time are scaled and spliced ​​to obtain the fused feature vector corresponding to each sampling time. Multiple fused feature vectors are arranged in the order of sampling time to obtain a temporal feature sequence.

[0020] Optionally, the short-circuit temporal identification model includes a Long Short-Term Memory (LSTM) network, a fully connected layer, and a Softmax output layer. The LTM network extracts temporal information about the evolution of short-circuit risk states over time from the temporal feature sequence. The fully connected layer outputs a discrimination score for each preset short-circuit risk state based on the temporal information. The Softmax output layer converts the discrimination score into a state probability. The expression for calculating the state probability is:

[0021] in, Indicates the first At any time, the block plate Belongs to the The state probability of a preset short-circuit risk state; Indicates that the fully connected layer is for the first The first of the block plates The discrimination score of the preset short-circuit risk state output; This indicates the total number of preset short-circuit risk states; A category index representing the preset short-circuit risk status; Represents the natural exponential function; Indicates that the fully connected layer is for the first The first of the block plates The discrimination score of the preset short-circuit risk state output; The preset short-circuit risk states include normal state, minor short-circuit state, and severe short-circuit state.

[0022] Optionally, early warning indicators are determined from the time-series feature sequence, and rule-based risk values ​​are determined based on the degree of deviation of the early warning indicators from a preset normal benchmark, including: Based on the baseline mean and baseline standard deviation of each early warning indicator under historical normal operating conditions, the deviation of each early warning indicator from the preset normal baseline is standardized to obtain the standardized deviation. The positive deviations of each standardized deviation level are weighted and summed to obtain the rule risk value; the formula for calculating the rule risk value is:

[0023]

[0024] in, Indicates the first At any time, the block plate The rule risk value; This indicates the number of early warning indicators used to determine the risk value of the rule; Indicates the index of early warning indicators; Indicates the first The non-negative risk weights of each early warning indicator are equal to one; Indicates the first The degree of standardized deviation of each early warning indicator from the preset normal benchmark; Indicates the first At any time, the block plate The One early warning indicator; Indicates the first The baseline average of each early warning indicator under historical normal operating conditions; Indicates the first The baseline standard deviation of each early warning indicator under historical normal operating conditions.

[0025] Optionally, based on the comparison results of the state probability, rule risk value, and preset classification threshold, the short-circuit warning level of the target plate is determined, including: Set the first, second, and third level thresholds to increase sequentially; When the rule risk value is not lower than the first level threshold and the sum of the probability of a minor short circuit state and the probability of a severe short circuit state is not lower than the first probability threshold, and this condition is maintained for a first preset duration, the target plate is determined to be in a first-level warning state. When the rule risk value is not lower than the second level threshold and the sum of the probability of a minor short circuit state and the probability of a severe short circuit state is not lower than the second probability threshold, and this condition is maintained for the second preset duration, the target plate is determined to be in a level two warning state. When the rule risk value is not lower than the third level threshold and the probability of a severe short circuit is not lower than the third probability threshold, and the third preset duration is maintained continuously, the target plate is determined to be in a level three warning state. When two or more short-circuit warning levels are met simultaneously, the short-circuit warning level with the highest level is determined as the short-circuit warning level of the target plate.

[0026] In a second aspect, the present invention provides a graded early warning device for short circuits on lead electrolytic cell plates, comprising: The image acquisition module is used to continuously acquire multiple infrared images according to a preset sampling period to obtain an infrared image sequence covering at least one electrode group in the lead electrolytic cell; the electrode group includes cathode plates and anode plates arranged alternately along the electrode arrangement direction; The electrode region determination module is used to establish an electrode arrangement topology model based on the cell structure parameters of the lead electrolytic cell and the electrode parameters of the electrode group to determine the corresponding region of each electrode in each infrared image of the infrared image sequence, and obtain the electrode region sequence corresponding to each electrode. The electrode arrangement topology model is used to characterize the arrangement order, polarity and adjacency relationship of each electrode. The feature extraction module is used to extract single-plate thermal field features, including the average temperature, the average local temperature gradient, and the hot spot area ratio, based on the temperature value of the target plate in the corresponding region of each infrared image for any target plate in the plate group. It calculates the difference between the single-plate thermal field features of the target plate and the adjacent plates determined based on the adjacency relationship to obtain the adjacent plate differential features, and performs trend fitting on the single-plate thermal field features of the target plate in the time dimension to obtain the thermal field trend features. The state recognition module is used to fuse the thermal field characteristics of a single plate, the differential characteristics of adjacent plates, and the thermal field trend characteristics to obtain the time-series feature sequence of the target plate. The time-series feature sequence is then input into a pre-trained short-circuit time-series recognition model to obtain the state probability of the target plate belonging to each preset short-circuit risk state. The short-circuit time-series recognition model is a multi-class time-series model trained using historical time-series feature sequences labeled with different preset short-circuit risk states, which is used to output the state probability of each preset short-circuit risk state. The rule risk determination module is used to determine early warning indicators from the time-series feature sequence and to determine the rule risk value based on the degree of deviation of the early warning indicators from the preset normal benchmark. The early warning indicators include at least one of the following: the average temperature of the target electrode, the average local temperature gradient, the proportion of hot spot area, the difference between the average temperature of the target electrode and the adjacent electrode, and the slope of the change of the average temperature. The graded early warning module is used to determine the short circuit warning level of the target plate based on the comparison results of the state probability, rule risk value and preset graded threshold.

[0027] The present invention has at least the following beneficial effects: By continuously acquiring infrared images and establishing a topological model of electrode arrangement based on tank structure parameters and electrode parameters to characterize the arrangement order, polarity, and adjacency relationship, the thermal field data of the same electrode can be continuously correlated in time. The average temperature, the average local temperature gradient, and the proportion of hot spot area are used to describe the overall temperature rise, thermal field non-uniformity, and local heat accumulation, respectively. The differential of adjacent electrodes is used to suppress the influence of synchronous temperature rise across the entire tank caused by changes in ambient temperature and tank current. The thermal field trend reflects the continuous evolution of weak anomalies. Furthermore, the state probability output by the time-series identification model and the rule risk value of the early warning index relative to the normal benchmark are integrated, and the early warning level is determined by combining the graded threshold. This improves the accuracy of identifying and locating single-block short-circuit hazards before obvious high temperature forms, reduces false alarms and missed alarms, and provides a graded basis for inspection, electrode adjustment, sediment cleaning, or tank shutdown maintenance. This reduces ineffective current and power loss, mitigates electrode burn-out and product quality fluctuations, and reduces the risk of unplanned tank shutdowns. Attached Figure Description

[0028] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0029] Figure 1 This is a flowchart of a graded early warning method for short circuits in lead electrolytic cell plates according to one embodiment of this application; Figure 2 This is a structural diagram of a lead electrolytic cell electrode plate short-circuit classification and early warning device according to one embodiment of this application. Detailed Implementation

[0030] The technical solution of the present invention will now be described in detail and completely with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] In the description of this invention, it should be noted that the terms "upper", "lower", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0032] Example 1 This embodiment uses an in-service lead electrolytic cell in a lead electrolysis production line as an example to illustrate the graded early warning method for short circuits in lead electrolytic cells provided by this invention. In this embodiment, 49 anode plates and 48 cathode plates are alternately arranged along the length of the electrolytic cell, with a center-to-center spacing of approximately 100 mm. During production, the cell current varies with the process load. During the growth process, cathode lead deposits may form protrusions towards adjacent anode plates. Simultaneously, electrode plate installation misalignment and surface deformation can further reduce the local spacing. In the early stages of short circuit formation, only a slow temperature rise and small-scale thermal inhomogeneity appear on the surface of the target electrode plate. Acid mist scattering, radiation from adjacent electrode plates, changes in ambient temperature, and fluctuations in cell current will cause synchronous changes in the thermal field of the entire cell. If a single-frame maximum temperature or fixed absolute threshold alarm is used, it is easy to miss the early intervention opportunity or misjudge normal operating condition changes as short circuits.

[0033] like Figure 1 As shown, the lead electrolytic cell electrode plate short circuit classification early warning method provided by the present invention includes steps 11 to 18.

[0034] Step 11: Collect multiple infrared images continuously according to a preset sampling period to obtain an infrared image sequence covering at least one electrode group in the lead electrolytic cell; the electrode group includes cathode plates and anode plates arranged alternately along the electrode arrangement direction.

[0035] In one feasible implementation, a fixed infrared thermal imaging camera can be installed approximately 4.5 meters above the side of the electrolytic cell, ensuring that the camera's field of view covers the upper edge of all 97 electrode plates and the plate surface area exposed above the electrolyte. The camera outputs a radiation temperature measurement image with a resolution of 640×512 pixels, with a temperature measurement range set from 0℃ to 150℃, and images are acquired according to a preset sampling period of 10 seconds. The on-site industrial computer synchronizes the infrared images with data on cell current, cell voltage, ambient temperature, and acid mist concentration via timestamps. The above quantities and parameters are used to illustrate the implementation of this scheme and do not constitute a limitation on the number of electrode plates, camera type, or sampling period.

[0036] In practice, the infrared thermal imaging camera outputs an infrared image containing pixel temperature values ​​and acquisition timestamps at each sampling moment. The industrial computer uses 40 consecutive frames of images as a sliding time window. Each time a new frame is acquired in an adjacent window, the window slides forward one frame, thus obtaining an infrared image sequence covering approximately 390 seconds of operation. For production phases where short-circuit evolution is rapid, the sampling period can be shortened to 2 to 5 seconds; for stable operation phases, the sampling period can be extended to 15 to 30 seconds. When adjusting the sampling period, the number of frames or the duration of the time window are adjusted accordingly to cover the evolution of the thermal anomaly to be identified.

[0037] In practice, infrared thermal imaging cameras output radiation temperature measurement data corresponding to each pixel. The original infrared image may contain inconsistencies in detector response, dead pixels, fixed pattern noise, and zero-point drift caused by long-term camera operation. To address this, the detector output values ​​can be converted into temperature values ​​first, based on the infrared thermal imaging camera's factory calibration parameters or on-site blackbody calibration results. Then, dead pixels can be repaired based on adjacent pixels and historical frame data, and offset correction can be performed using stable background areas, blackbody reference areas, or background statistics from historical normal operating conditions.

[0038] In this embodiment, the corrected infrared image is represented as follows: .in, Indicates time raw infrared image Mid-pixel coordinates Pixel value at that location, This represents the corrected pixel value. Represents the radiation calibration function. This indicates the result of the dead pixel repair. This indicates the background correction result.

[0039] In one feasible implementation method, .in, This represents the average background brightness determined from multiple frames of infrared images taken when there are no personnel, no moving heat sources, and the electrolytic cell is in a stable operating state. This represents the standard brightness reference value, which can be determined based on the data bit width of the infrared image and the brightness distribution under normal operating conditions, for example, between 100 and 120. By subtracting the average background brightness at the corresponding position from the current pixel value and then superimposing the standard brightness reference value, the global deviation caused by fixed background heat sources and camera zero-point drift can be reduced.

[0040] For isolated bad pixels, abnormal response points, or short-time impulse noise points, weighted repair can be performed based on the spatial distance and grayscale similarity of pixels in the neighborhood of the pixel to be processed. The result of bad pixel repair can be expressed as:

[0041] in, Indicates the radius of the filter kernel. This represents the normalization coefficient that normalizes the sum of the weights of each neighborhood. Gaussian function representing spatial distance This represents a Gaussian function indicating gray-level similarity. Neighboring pixels that are spatially closer and have values ​​closer to the center pixel have a greater repair weight. For fixed defective pixels already registered in the defective pixel table of the infrared thermal imaging camera, the original pixel value is directly replaced using the above repair results; for unregistered defective pixels, replacement can be performed when the deviation of the current pixel value from the neighborhood median exceeds a preset defect threshold. This allows for the preservation of the local temperature boundary and hotspot contour of the electrode while reducing the impact of defective pixels and non-uniform responses on the calculation results of the average temperature, average local temperature gradient, and hotspot area ratio.

[0042] After completing radiation calibration and non-uniformity correction, acid fog interference suppression and scene normalization are performed on the corrected infrared image. Acid fog in the lead electrolysis workshop absorbs and scatters infrared radiation from the electrode surface. Furthermore, non-electrode heat sources such as pipes, tank edges, and workers may cause thermal reflection, resulting in different apparent brightness of the same electrode under different acid fog concentrations or ambient temperatures. To reduce this interference, this embodiment employs infrared defogging based on an atmospheric scattering model to compensate for the radiation attenuation caused by acid fog. The defogging image can be represented as follows:

[0043] in, Indicates the first time after defogging Pixel coordinates in a frame of infrared image Pixel value at that location, Indicates atmospheric light composition. This represents the fog concentration correction factor. Represents pixel coordinates The corresponding fog concentration distance. Indicates acid mist transmittance, and according to Sure, This represents the preset minimum transmittance, which is greater than 0 and less than 1. Represents the natural exponential function. Atmospheric light composition. The fog concentration correction coefficient can be estimated based on a stable background region in the image that is far from the electrode and has minimal temperature variation. The concentration can be determined based on measurements from an acid mist sensor, or by the degree of contrast attenuation of the current infrared image relative to a historically clear image; mist concentration distance. The minimum transmittance can be determined based on the spatial distance between the infrared thermal imaging camera and the corresponding tank surface, as well as the acid mist distribution. Using the aforementioned minimum transmittance limit can avoid... Too small a value leads to unstable compensation results; as Increase The corresponding reduction will compensate for the radiation attenuation when the acid mist concentration or propagation distance increases.

[0044] After defogging, the edges of the tank, fixed components in the workshop, or areas that remain stable over a long period in the infrared image are used as background reference areas. The global temperature offset of the current frame relative to historical normal reference frames is calculated, and this global temperature offset is removed from the current frame to ensure that the background temperature benchmark remains consistent across different sampling times. Simultaneously, areas outside the electrode area, such as those containing personnel, mobile devices, steam plumes, and high-temperature pipes, are shielded. For localized low-contrast areas caused by short-term acid mist obstruction, compensation can be made by combining the corresponding areas from the previous sampling time. Through acid mist interference suppression and scene normalization, the synchronous rise and fall of the entire image caused by changes in acid mist concentration, ambient temperature, and non-electrode heat sources can be reduced, avoiding misjudging global thermal field changes as local anomalies of a single electrode.

[0045] It should be noted that local contrast enhancement and other processing aimed at improving the visibility of the electrode boundary are mainly used for electrode region segmentation. When calculating the mean electrode temperature, the mean local temperature gradient, and the proportion of hot spot area, a temperature matrix that has been calibrated by radiation, compensated for acid mist, and has the same coordinate relationship as the segmentation result is used, thereby avoiding the direct use of grayscale values ​​used only for display enhancement as the actual electrode temperature value.

[0046] When an infrared thermal imaging camera is mounted on a pan-tilt unit or when the camera is viewed from an oblique angle, the electrodes at different locations in the image will exhibit problems such as near-large and far-small appearance, and scale changes with imaging distance. To ensure that the electrodes at different locations have a uniform spatial scale, this embodiment further performs image unfolding on the infrared image based on the reference plane of the electrolytic cell surface. Specifically, the plane containing the surface of the electrolytic cell, the upper edge of the electrodes, or the plane corresponding to the main area to be detected on the electrodes is determined as the reference plane; when the surface of the cell and the upper edge of the electrodes cannot be approximated by the same plane, the imaging area is divided into multiple sub-regions, and piecewise homography transformation is performed on each sub-region based on the corresponding reference plane.

[0047] The target area can be formed by a plane For example, ground points Compared with the original image pixels The projection relationship can be expressed as:

[0048]

[0049] in, Indicates the scaling factor. The homography matrix represents the distance from the reference plane to the original image plane. Represents the camera intrinsic parameter matrix. and The camera attitude rotation matrix corresponds to the two direction vectors of the slot reference plane. This is the translation vector determined by the camera's mounting position. (Camera intrinsic parameter matrix) Calibration can be performed using a calibration plate before installation; for fixed cameras, , as well as The image remains unchanged after installation; for gimbal-mounted cameras, the camera attitude rotation matrix is ​​updated based on the horizontal and pitch angles of the gimbal at the current sampling time to obtain the homography matrix at the corresponding sampling time. .

[0050] Define the pixel coordinates in the unfolded image as The actual distance on the reference plane corresponding to a single pixel in the unfolded image is The relationship between the pixel coordinates of the unfolded image and the coordinates of the reference plane is as follows:

[0051] in, The matrix representing the mapping from the base plane coordinates to the unfolded diagram coordinates. This represents the inverse projection matrix from the original image plane to the reference plane. For each pixel position in the unfolded image... Based on the above transformation relationship, calculate its corresponding sampling position in the original infrared image. The temperature values ​​at corresponding locations were resampled using bilinear interpolation to obtain the unfolded infrared image of the ground. Pixels mapped outside the effective field of view of the original image were marked as invalid pixels and not included in subsequent electrode region segmentation and thermal field feature calculations.

[0052] By applying the same unfolded region boundary and ground resolution to each frame in the infrared image sequence, and updating the corresponding homography matrix according to the gimbal attitude at each sampling time, the same physical electrode is mapped to the same or adjacent unfolded map positions at different sampling times. This reduces electrode size variations and positional drift caused by camera tilt imaging and gimbal attitude changes, ensuring that the electrode spacing, arrangement direction, and effective slot area remain consistent across continuous infrared images. This provides a stable spatial reference for subsequently establishing an electrode arrangement topology model based on slot structure parameters and electrode parameters, generating candidate search regions, and maintaining continuous and consistent electrode numbering.

[0053] The aforementioned calibration, acid mist suppression, quality control, and image unfolding processing make the pixel temperature and spatial scale at different sampling times comparable, reducing the interference of acid mist, camera drift, and shooting angle changes on the subsequent thermal field characteristics of a single plate.

[0054] Step 12: Based on the cell structure parameters of the lead electrolytic cell and the electrode parameters of the electrode group, establish an electrode arrangement topology model to determine the corresponding region of each electrode in the infrared image sequence, and obtain the electrode region sequence corresponding to each electrode.

[0055] The electrode arrangement topology model in this embodiment is used to characterize the arrangement order, polarity, and adjacency relationship of each electrode.

[0056] Specifically, the inner wall boundary, effective length and width of the tank surface are used as structural parameters of the tank, while the number of pole plates, the center-to-center spacing of the pole plates, the pole plate arrangement direction, the exposed height of the pole plates, the pole plate width, and the anode and cathode polarities are used as pole plate parameters. The pole plate arrangement topology model includes a set of nodes and a set of edges. Each node corresponds to one pole plate, and the node attributes include at least the pole plate number, polarity, nominal center position, and candidate search region. For two pole plates that are adjacent to each other in the arrangement direction, an edge is established between the corresponding nodes to record the adjacency relationship. When numbering from the head to the tail of the tank, the pole plates from the 1st to the 97th are numbered sequentially. Except for the pole plates at both ends, the set of adjacent pole plates of the i-th pole plate includes the... Block and the first The target electrode has a block electrode plate, and the adjacent electrode plates have opposite polarities to the target electrode plate.

[0057] Based on the tank structure parameters, the effective area of ​​the tank is determined in the unfolded image. Candidate search regions, corresponding one-to-one with each electrode, are generated, centered on the nominal center position and with dimensions equal to preset multiples of the electrode width and spacing. These candidate search regions are distributed sequentially along the electrode arrangement direction, and their centers can be dynamically corrected based on the current tank boundary, reference centerline, and the trajectory position from the previous moment. The candidate search regions limit the segmentation range, preventing adjacent electrodes from being merged into the same target under conditions of acid mist, reflection, or local blurring.

[0058] Within each candidate search region, an adaptive thresholding method is first used to obtain the foreground based on local temperature or grayscale distribution. Then, the pole region is filtered by combining the approximate straightness of the pole edge, the region aspect ratio, the connected area, the vertical continuity, and the distance from the center of the candidate region. Holes and breaks are repaired using closing operations, and isolated small regions are removed using opening operations. When multiple connected regions exist within the same candidate search region, the region with the lowest combined shape and position cost is selected as the current pole region, and pole numbers are assigned according to their order. Instance segmentation networks, edge detection, or template matching can also replace the above local thresholding segmentation, as long as the output is a region mask that corresponds one-to-one with each pole.

[0059] To maintain consistent numbering of the same electrode in consecutive images, electrode regions at adjacent sampling times are associated based on the matching cost between the electrode region in the current infrared image and the historical trajectory prediction region; the expression for calculating the matching cost is:

[0060] in, Indicates time The The current electrode region and the first The matching cost between historical polar plate trajectories; Indicates center distance Non-negative weights Indicates the first The center coordinates of the current electrode region Indicates the first The trajectory of the historical plate at any moment The predicted center coordinates; Indicates the degree of regional overlap Non-negative weights Indicates the first The current electrode region at time The area mask, Indicates the first The trajectory of the historical plate at any moment The predicted region mask, This indicates the calculation of the intersection-union ratio; Indicates area change Non-negative weights; Indicates the first The area of ​​the current electrode region Indicates the first The predicted area of ​​the historical polar plate trajectory. Indicates a preset positive number. This indicates taking the maximum value.

[0061] In this feasible implementation, the center distance weight is taken. Regional overlap weight Area change weight Preset positive number A 1-pixel area is used. The historical trajectory prediction region is predicted by Kalman filtering based on the position and velocity at the previous moment. Infeasible matches are first eliminated if the center distance exceeds 0.6 times the width of the corresponding candidate search region or the intersection-union ratio is less than 0.05. Then, the Hungarian algorithm is used to determine the historical electrode trajectory corresponding to the current electrode region according to the principle of minimizing matching cost. The weights and thresholds can be calibrated using normal operating samples based on the camera viewpoint and electrode size.

[0062] If a plate is not segmented within three consecutive sampling times due to acid mist obscuring or local distortion, a temporary region is generated based on its predicted position from its historical trajectory, the spacing between adjacent plates, and their arrangement order. Once the target reappears, if its matching cost with the temporary region meets the association condition, the original plate number is restored. If the plate is not recovered after a preset loss period, it is marked as data unavailable, and an image quality warning is output; however, this does not directly trigger a short-circuit warning.

[0063] Through the above topological constraints, local segmentation and temporal correlation, the electrode region sequence of each electrode in each infrared image can be obtained, and the arrangement order, polarity and adjacency relationship are kept stable, thereby avoiding the replacement of the target electrode temperature sequence by adjacent electrode data due to region mismatch.

[0064] Step 13: For any target electrode in the electrode group, extract the single-plate thermal field characteristics of the target electrode based on the temperature value of the target electrode in the corresponding region of each infrared image.

[0065] In this embodiment, the thermal field characteristics of the single plate include the average temperature, the average local temperature gradient, and the proportion of hot spot area.

[0066] Specifically, the electrode area mask is etched inward by two pixels to remove boundary pixels that are easily affected by electrolyte reflection and radiation from adjacent electrodes; abnormal pixels that are outside the camera's effective temperature measurement range, are marked as bad pixels, or have a temperature deviation from the median temperature of the area exceeding 6 times the median absolute deviation are removed, and the remaining pixels constitute the second... At any time, the block plate The set of pixels in the region. The average temperature, average local temperature gradient, and hot spot area ratio are determined according to the following formulas:

[0067]

[0068]

[0069]

[0070] in, Indicates the first At any time, the block plate average temperature Indicates the first At any time, the block plate The mean of the local temperature gradient, Indicates the first At any time, the block plate The percentage of hot spot area; Indicates the first At any time, the block plate The set of pixels in the region; Indicates the number of pixels contained in the region's pixel set; Represents pixel coordinates; Indicates time Pixel coordinates in infrared images Temperature value; This represents the temperature gradient along the horizontal direction. This represents the temperature gradient in the vertical direction. This is an indicator function that takes the value of 1 when the condition inside the parentheses is true and takes the value of 0 when the condition inside the parentheses is false. Indicates the first Hot spot determination threshold for bulk electrode plates; Indicates the first The baseline average temperature of the bulk electrode under historical normal operating conditions; Indicates the first The baseline standard deviation of the temperature of the bulk electrode under historical normal operating conditions; This represents the safety margin coefficient used to determine the hot spot detection threshold.

[0071] The horizontal and vertical temperature gradients can be calculated using the Sobel operator or by calculating the difference between adjacent pixels. The hot spot determination threshold is determined by the mean and standard deviation of the temperature reference for each electrode under historical normal operating conditions. In this embodiment, no less than 1000 normal samples are collected within the same tank current range and ambient temperature range, with a safety margin coefficient λ of 3. When the standard deviation of the temperature reference is less than 0.1℃, 0.1℃ is used as the standard deviation for calculation to avoid threshold distortion caused by excessively small fluctuations in normal samples.

[0072] For example, the effective area of ​​the 24th cathode plate at a certain sampling time contains 8300 pixels. The calculated average temperature is 39.8℃, the average local temperature gradient is 0.74℃ / pixel, and the hot spot area ratio is 0.062. Its historical normal temperature baseline average is 38.1℃, and the temperature baseline standard deviation is 0.45℃. When the value is 3, the hot spot detection threshold is 39.45℃. The above three characteristics respectively reflect the overall temperature level of the electrode, the intensity of internal thermal field changes, and the proportion of high-temperature regions exceeding the normal safety margin.

[0073] Simultaneously extracting the thermal field characteristics of three types of single boards can avoid being overly sensitive to noise when judging solely by the highest temperature at a single point, and enable quantifiable characterization of the overall temperature rise, local thermal field inhomogeneity, and heat accumulation caused by short circuits.

[0074] Step 14: Calculate the difference in single-plate thermal field characteristics between the target plate and its adjacent plates to obtain the differential characteristics of adjacent plates. Adjacent plates are determined based on their adjacency relationship.

[0075] Specifically, for each plate's thermal field characteristic among the average temperature, average local temperature gradient, and hot spot area ratio, the difference between the target plate's thermal field characteristic and the average value of the corresponding thermal field characteristics of the adjacent plates is calculated to obtain the adjacent plate differential characteristics; its calculation expression is:

[0076] in, Indicates the first At any time, the block plate The average temperature, the average local temperature gradient, or the percentage of hot spot area; Indicates the first The difference between adjacent plates corresponding to the thermal field characteristics of a single plate; This indicates that the topological model of the electrode arrangement determines the relationship with the first... The set of adjacent plates of a block electrode; Indicates the number of plates in the plate set; In the set of plates, the first... At any time, the block plate The corresponding single-plate thermal field characteristics.

[0077] In practical implementation, the following calculation formula can be used:

[0078]

[0079]

[0080] The first one was obtained respectively At any time, the block plate Difference between adjacent plates corresponding to the average temperature , No. At any time, the block plate Difference between adjacent plates corresponding to the mean of the local temperature gradient , No. At any time, the block plate The difference between adjacent plates corresponding to the hot spot area ratio .

[0081] For a target electrode located inside an electrode group, the set of adjacent electrodes typically includes the preceding and following electrodes in the arrangement direction; for a target electrode located at the end of an electrode group, the set of adjacent electrodes may only include the innermost electrode. If one of the adjacent electrodes is marked as unavailable at the current time, the other valid adjacent electrode is used for calculation; if both are unavailable, the differential features of the adjacent electrodes are not updated at this time.

[0082] For example, the average temperature of the 24th cathode plate is 39.8℃, while the average temperatures of its adjacent 23rd and 25th anode plates are 38.2℃ and 38.4℃, respectively. Therefore, the temperature difference between adjacent plates is 1.5℃. The difference in the average local temperature gradient and the difference in hot spot area are calculated in the same way. When changes in ambient temperature or tank current cause the entire tank to heat up synchronously, the absolute temperatures of the target plate and its adjacent plates usually change simultaneously with a small difference. However, when localized conduction occurs in the target plate, the difference will continuously increase.

[0083] Therefore, the differential characteristics of adjacent plates, using the local reference provided by the regular alternating arrangement of plates in the electrolytic cell, can suppress the influence of the common temperature rise of the entire cell on the anomaly judgment and locate the warning target to a specific plate.

[0084] Step 15: Perform trend fitting on the single-plate thermal field characteristics of the target electrode in the time dimension to obtain the thermal field trend characteristics of the target electrode. The thermal field trend characteristics characterize the trend of the thermal field of the target electrode changing over time.

[0085] This embodiment performs trend fitting on the mean temperature, mean local temperature gradient, and hot spot area ratio within a sliding time window containing 40 sampling moments. To reduce the impact of outliers caused by occasional acid mist obstruction or thermal reflection on the fitting results, outliers are first identified using the median and median absolute deviation within the window. These outliers are then replaced with linear interpolations from adjacent valid time moments, and the slope of change is calculated using the least squares method. The expression for calculating the thermal field trend characteristics is as follows:

[0086] in, Indicates the first The single-plate thermal field characteristics corresponding to the block plate at time The slope of the change; Indicates the cutoff time. The set of sampling moments within a time window; Represents the first in the set of sampling times. Each sampling time; This represents the average value of all sampling times in the set of sampling times; Indicates the first At the sampling time of the block plate Corresponding single-plate thermal field characteristics; Indicates the first time within the time window The average value of the single-plate thermal field characteristics corresponding to the block plate.

[0087] In the formula, when the time variable is in minutes, the unit of the slope of the mean temperature change is ℃ / minute, the unit of the slope of the mean local temperature gradient change is ℃ / (pixel·minute), and the unit of the slope of the hot spot area percentage change is 1 / minute. The time window can be 30 to 50 frames, or it can be set to a fixed duration covering 3 to 10 minutes depending on the sampling period.

[0088] For example, the average temperature of the 24th cathode plate gradually increased from 38.6℃ to 39.8℃ in about 6.5 minutes, and the fitted slope of the average temperature change was 0.18℃ / minute; the average temperature of its adjacent plates remained basically stable within the same time period. Even though 39.8℃ has not yet reached the traditional high temperature alarm threshold, the continuously positive slope of change, which is significantly higher than the normal fluctuation range, still indicates that the thermal anomaly of this plate is developing.

[0089] Therefore, the thermal trend feature can identify the process in which the single frame feature has not yet obviously exceeded the limit but the anomaly continues to increase, so that the warning time point is moved forward to the short circuit initiation stage.

[0090] Step 16: The thermal field characteristics of the single plate, the differential characteristics of adjacent plates, and the thermal field trend characteristics are fused to obtain the time series feature sequence of the target plate. The time series feature sequence is then input into the pre-trained short-circuit time series recognition model to obtain the state probability of the target plate belonging to each preset short-circuit risk state.

[0091] In one feasible implementation, at each sampling time, the mean temperature, mean local temperature gradient, hot spot area ratio, three adjacent plate differences, and three slopes are Z-score unified according to the mean and standard deviation obtained from the training set, and then concatenated in a fixed order to form a 9-dimensional fusion feature vector. The 40 fusion feature vectors within the current time window are arranged according to the sampling time order to obtain a 40×9-dimensional temporal feature sequence.

[0092] In another feasible implementation, without changing the above core features, the synchronously collected cell voltage, cell current, ambient temperature and acid mist concentration can be scaled and then used as additional operating condition features to be spliced ​​into the fused feature vector, or used to select the normal benchmark for the corresponding operating condition range.

[0093] In this embodiment, the short-circuit timing identification model is trained using historical timing feature sequences labeled with different preset short-circuit risk states. The historical samples include at least three categories: normal state, minor short-circuit state, and severe short-circuit state. The state labels are jointly determined based on on-site inspection records, electrode adjustment or cleaning records, abnormal cell current records, and subsequent cell shutdown inspection results. To avoid data leakage caused by adjacent windows of the same short-circuit process simultaneously entering the training and test sets, the training, validation, and test sets are divided according to the electrolytic cell number and production time period. This embodiment uses a 7:2:1 ratio.

[0094] The short-circuit timing identification model in this embodiment includes a two-layer Long Short-Term Memory (LSTM) network, a fully connected layer, and a Softmax output layer. The first LSM network has 64 hidden units, and the second layer has 32 hidden units. A Dropout with a ratio of 0.2 is applied between the two layers. The hidden state at the last sampling time of the second layer is input to the fully connected layer, which outputs three discrimination scores: normal state, slight short-circuit state, and severe short-circuit state. The Softmax output layer converts the discrimination scores into state probabilities, and the expression for calculating the state probabilities is as follows:

[0095] in, Indicates the first At any time, the block plate Belongs to the The state probability of a preset short-circuit risk state; Indicates that the fully connected layer is for the first The first of the block plates The discrimination score of the preset short-circuit risk state output; This indicates the total number of preset short-circuit risk states; A category index representing the preset short-circuit risk status; Represents the natural exponential function; preset short-circuit risk states include normal state, minor short-circuit state, and severe short-circuit state; Indicates that the fully connected layer is for the first The first of the block plates The discrimination score of the output of a preset short-circuit risk state.

[0096] Cross-entropy loss is used during training:

[0097]

[0098]

[0099] in, Indicates the number of training samples; Indicates the total number of categories; Indicates sample Corresponding category The tag value; This represents the class probability output by the model; Indicates that the fully connected layer is for the first The discrimination score output for a preset short-circuit risk state is the raw score that has not been normalized by Softmax. Indicates that the fully connected layer is for the first The discrimination score of the preset short-circuit risk state output. For category index, values ​​range from 1 to... ; Indicates the discrimination score The natural index value, Indicates the discrimination score The natural index value; Indicates sample The corresponding comprehensive feature vector; Indicates the sequence length.

[0100] The Adam optimizer is used, with an initial learning rate of 0.001, a batch size of 32, and a maximum training duration of 100 epochs. Training is stopped early if the validation set loss does not decrease for 10 consecutive epochs. When the number of class samples is imbalanced, non-negative weights inversely proportional to the class frequency can be added to the cross-entropy.

[0101] During online execution, the temporal feature sequence of each electrode in the current time window is input into the trained and parameter-fixed model to obtain three state probabilities. For example, the probabilities of the normal state, slight short circuit state, and severe short circuit state of the 24th cathode plate at a certain moment are 0.18, 0.61, and 0.21, respectively. The state probabilities not only give the most likely state at the current moment, but also retain the risk distribution of the evolution from a slight short circuit to a severe short circuit, providing input for subsequent rule verification.

[0102] By integrating the thermal field of a single plate, the difference between adjacent plates, and the time trend, and using a time series model to identify state evolution within a continuous window, it is possible to avoid directly identifying isolated noise frames as short circuits, while improving the ability to identify various early thermal anomaly combination patterns.

[0103] Step 17: Determine the early warning indicators from the time-series feature sequence, and determine the rule risk value based on the degree of deviation of the early warning indicators from the preset normal benchmark.

[0104] The early warning indicators include at least one of the following: the average temperature of the target electrode, the average local temperature gradient, the proportion of hot spot area, the difference in average temperature between the target electrode and adjacent electrodes, and the slope of the average temperature change.

[0105] Specifically, this embodiment selects the above five indicators to jointly calculate the rule risk value. Normal baselines are statistically analyzed according to plate number, tank current range, and ambient temperature range. The baseline mean and standard deviation of each indicator are obtained from historical samples confirmed to be free of short circuits on-site, and are updated daily using the most recent normal data throughout the production cycle. However, plate data that has triggered warnings and low-quality image data are not included in the update. This maintains the baseline's adaptability to changes in operating conditions while avoiding contamination of the normal baseline by abnormal samples.

[0106] Based on the baseline mean and standard deviation of each early warning indicator under historical normal operating conditions, the deviation of each early warning indicator from the preset normal baseline is standardized to obtain the standardized deviation. The positive deviations of each standardized deviation are then weighted and summed to obtain the rule-based risk value, the calculation expression of which is as follows:

[0107]

[0108] in, Indicates the first At any time, the block plate The rule risk value; This indicates the number of early warning indicators used to determine the risk value of the rule; Indicates the index of early warning indicators; Indicates the first The non-negative risk weights of each early warning indicator are equal to one; Indicates the first The degree of standardized deviation of each early warning indicator from the preset normal benchmark; Indicates the first At any time, the block plate The One early warning indicator; Indicates the first The baseline average of each early warning indicator under historical normal operating conditions; Indicates the first The baseline standard deviation of each early warning indicator under historical normal operating conditions.

[0109] In this embodiment, the risk weights corresponding to the mean temperature, mean local temperature gradient, hot spot area ratio, mean temperature difference, and mean temperature change slope are respectively set to 0.25, 0.15, 0.20, 0.25, and 0.15, with each weight being non-negative and summing to 1. The weights can be jointly calibrated based on the discriminative power of historical fault samples, field process experience, or the false alarm and false negative rates of the validation set, without directly replacing rule-based risk with model training loss.

[0110] For example, if the standardized deviations of the five early warning indicators for the target electrode are 1.3, 0.8, 1.1, 1.5, and 1.2 respectively, all of which are positive deviations, then the rule risk value is 0.25×1.3+0.15×0.8+0.20×1.1+0.25×1.5+0.15×1.2=1.22. If any indicator is lower than the normal benchmark, its positive deviation is counted as zero, thus preventing negative fluctuations within the normal range from offsetting abnormal positive deviations of other indicators.

[0111] The rule-based risk value is based on temperature, gradient, hot spot, adjacent differences and trend indicators that can be verified by on-site personnel. It can independently verify the output of the time series model and avoid generating unexplainable alarms based solely on the probability of the model category.

[0112] Step 18: Determine the short-circuit warning level of the target plate based on the comparison results of the state probability, rule risk value and preset classification threshold.

[0113] The first, second, and third level thresholds are set to increase sequentially. In this embodiment, the thresholds are calibrated based on the high quantile of the risk value of normal sample rules, the distribution of minor and severe short-circuit samples, and the acceptable false alarm rate on-site. The first, second, and third level thresholds are set to 1.0, 1.8, and 2.8, respectively; the first, second, and third probability thresholds are set to 0.55, 0.70, and 0.80, respectively; and the first, second, and third preset durations are set to 60 seconds, 40 seconds, and 20 seconds, respectively. The shorter third preset duration allows severe short circuits to be triggered quickly under high confidence conditions. These values ​​can be recalibrated based on historical data from different production lines.

[0114] When the rule risk value is not lower than the first level threshold, and the sum of the probability of a minor short circuit and the probability of a severe short circuit is not lower than the first probability threshold, and this condition is maintained for a first preset duration, the target electrode is determined to be in a first-level warning state. The first-level warning corresponds to the short circuit initiation stage. The industrial computer outputs the electrode number, current indicators, state probability, and trend curve, prompting on-site personnel to conduct fixed-point inspections and check the electrode spacing, electrode surface deformation, and deposit growth.

[0115] When the rule risk value is not lower than the second level threshold, and the sum of the probabilities of a minor short circuit and a severe short circuit is not lower than the second probability threshold, and this condition is maintained for a second preset duration, the target electrode is determined to be in a level two warning state. The level two warning corresponds to a minor short circuit or the stage where conduction is about to occur. In addition to outputting an audible and visual alarm, it also prompts users to adjust the electrode position, clean overlapping deposits, or reduce the operating current of the corresponding tank, and records the handling time.

[0116] When the rule risk value is not lower than the third-level threshold, and the probability of a severe short circuit is not lower than the third probability threshold, and this condition is maintained for a continuous third preset duration, the target plate is determined to be in a level three warning state. A level three warning corresponds to a severe short circuit or a forming short circuit stage. The system sends a high-priority alarm to the centralized control terminal, prompting immediate isolation of the relevant circuits, reduction of the tank current, or scheduling of tank shutdown for maintenance to prevent further development of localized overheating.

[0117] When the conditions for determining two or more short-circuit warning levels are met simultaneously, the short-circuit warning level with the highest level is determined as the short-circuit warning level for the target plate. To prevent the level from frequently switching near the threshold, a reset threshold lower than the trigger threshold can be set when the warning is lifted, and it is required that the warning be lifted level by level after being continuously below the reset threshold for 5 minutes; this reset logic does not change the trigger conditions of the warning level.

[0118] Taking the 24th cathode plate as an example, its initial rule risk value is 1.22, and the sum of the probabilities of a minor short circuit and a severe short circuit is 0.82. A Level 1 warning is triggered after 60 consecutive seconds of meeting the Level 1 condition. Subsequently, as the sediment continues to grow, its rule risk value rises to 1.95, and the sum of the probabilities of the two short circuit states reaches 0.88. After 40 consecutive seconds, it is upgraded to a Level 2 warning. If no action is taken and the rule risk value rises to 3.10 and the probability of a severe short circuit rises to 0.86, it is upgraded to a Level 3 warning after 20 consecutive seconds. Completing electrode plate adjustment and sediment removal on-site at the Level 1 or Level 2 stage can eliminate potential conductivity risks before significant high temperatures and forming short circuits occur, thereby reducing ineffective current, energy loss, electrode plate burnout, and unplanned tank shutdowns.

[0119] Example 2 Based on the lead electrolytic cell electrode plate short-circuit graded early warning method provided in Example 1, this example also provides a lead electrolytic cell electrode plate short-circuit graded early warning device. For example... Figure 2 As shown, the device includes an image acquisition module 201, an electrode area determination module 202, a feature extraction module 203, a state recognition module 204, a rule-based risk determination module 205, and a graded early warning module 206. Each module can be implemented as a software function module executed by a processor in an industrial computer, or it can be implemented separately by a processor, a programmable logic device, or a dedicated circuit.

[0120] The image acquisition module 201 is used to continuously acquire multiple infrared images according to a preset sampling period to obtain an infrared image sequence covering at least one electrode group in the lead electrolytic cell; the electrode group includes cathode plates and anode plates arranged alternately along the electrode arrangement direction. This module can also perform timestamp generation, radiation calibration, bad pixel repair, acid mist effect suppression, image quality evaluation, and auxiliary operating condition data synchronization.

[0121] The electrode region determination module 202 is used to establish an electrode arrangement topology model based on the cell structure parameters of the lead electrolytic cell and the electrode parameters of the electrode group to determine the corresponding region of each electrode in each infrared image of the infrared image sequence, thereby obtaining the electrode region sequence corresponding to each electrode. The electrode arrangement topology model is used to characterize the arrangement order, polarity, and adjacency relationship of each electrode. This module performs candidate search region generation, electrode region segmentation, numbering, trajectory prediction, and adjacent frame association, and outputs a region mask with stable electrode numbers to the feature extraction module 203.

[0122] The feature extraction module 203 is used to extract single-plate thermal field features, including the average temperature, the average local temperature gradient, and the hot spot area ratio, for any target plate in the plate group based on the temperature value of the target plate in the corresponding region of each infrared image. It calculates the difference between the single-plate thermal field features of the target plate and the adjacent plates determined based on the adjacency relationship to obtain the differential features of the adjacent plates, and performs trend fitting on the single-plate thermal field features of the target plate in the time dimension to obtain the thermal field trend features.

[0123] The state recognition module 204 is used to fuse the thermal field characteristics of the single plate, the differential characteristics of adjacent plates, and the thermal field trend characteristics to obtain the time series feature sequence of the target plate. The time series feature sequence is input into the pre-trained short-circuit time series recognition model to obtain the state probability of the target plate belonging to each preset short-circuit risk state. The short-circuit time series recognition model is a multi-class time series model trained using historical time series feature sequences labeled with different preset short-circuit risk states, which is used to output the state probability of each preset short-circuit risk state.

[0124] The rule risk determination module 205 is used to determine early warning indicators from the time-series feature sequence and determine the rule risk value based on the degree of deviation of the early warning indicators from the preset normal benchmark. The early warning indicators include at least one of the following: the average temperature of the target electrode, the average local temperature gradient, the proportion of hot spot area, the difference in average temperature between the target electrode and adjacent electrodes, and the slope of the change in the average temperature.

[0125] The graded early warning module 206 is used to determine the short circuit warning level of the target electrode plate based on the comparison results of the state probability, rule risk value and preset graded threshold, and send the electrode plate number, warning level, state probability, rule risk value, trigger index and duration to the on-site audible and visual alarm, host computer or production control system.

[0126] During device operation, the infrared image sequence output by the image acquisition module 201 is transmitted to the electrode region determination module 202. The electrode region determination module 202 outputs the sequence and topological relationship of each electrode region. The feature extraction module 203 forms various thermal field features based on the above data. The state recognition module 204 and the rule risk determination module 205 output the state probability and rule risk value, respectively. The graded early warning module 206 performs a joint judgment on the two. Thus, the modules are sequentially connected in the data flow and jointly realize the continuous positioning, early identification, and graded handling of specific electrodes.

[0127] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. Their specific functions and technical effects can be found in the method embodiments section, and will not be repeated here. Those skilled in the art will understand that, for the sake of convenience and brevity, the division of the above-mentioned functional units and modules is only used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0128] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of protection of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of one or more embodiments of this application as described above, which are not provided in detail for the sake of brevity.

[0129] One or more embodiments in this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments in this application should be included within the protection scope of this application.

Claims

1. A graded early warning method for short circuits in lead electrolytic cell plates, characterized in that, include: Multiple infrared images are continuously acquired according to a preset sampling period to obtain an infrared image sequence covering at least one electrode group in the lead electrolytic cell. The electrode assembly includes cathode plates and anode plates arranged alternately along the electrode plate arrangement direction; Based on the cell structure parameters of the lead electrolytic cell and the electrode parameters of the electrode group, an electrode arrangement topology model is established to determine the corresponding region of each electrode in the infrared image sequence, thereby obtaining the electrode region sequence corresponding to each electrode. The electrode arrangement topology model is used to characterize the arrangement order, polarity, and adjacency relationship of each electrode. For any target electrode in the electrode group, the single-plate thermal field features of the target electrode are extracted based on the temperature values ​​of the target electrode in the corresponding region of each infrared image; the single-plate thermal field features include the average temperature, the average local temperature gradient, and the proportion of hot spot area. Calculate the difference in single-plate thermal field characteristics between the target plate and its adjacent plates to obtain the differential characteristics of adjacent plates; the adjacent plates are determined based on the adjacent relationship. The thermal field characteristics of the target electrode plate are fitted with a trend in the time dimension to obtain the thermal field trend characteristics of the target electrode plate. The thermal field trend characteristics characterize the trend of the target electrode's thermal field changing over time; The thermal field characteristics of the single plate, the differential characteristics of the adjacent plates, and the thermal field trend characteristics are fused to obtain the temporal feature sequence of the target plate. The temporal feature sequence is then input into a pre-trained short-circuit temporal identification model to obtain the state probability of the target plate belonging to each preset short-circuit risk state. The short-circuit temporal identification model is a multi-class temporal model trained using historical temporal feature sequences labeled with different preset short-circuit risk states, which is used to output the state probability of each preset short-circuit risk state. Early warning indicators are determined from the time-series feature sequence, and rule risk values ​​are determined based on the degree of deviation of the early warning indicators from the preset normal benchmark; the early warning indicators include at least one of the following: the average temperature of the target electrode, the average local temperature gradient, the proportion of hot spot area, the difference in average temperature between the target electrode and adjacent electrodes, and the slope of the change of the average temperature. The short-circuit warning level of the target plate is determined based on the comparison results of the state probability, the rule risk value, and the preset classification threshold.

2. The method for graded early warning of short circuits in lead electrolytic cell plates according to claim 1, characterized in that, The step of establishing an electrode arrangement topology model based on the cell structure parameters of the lead electrolytic cell and the electrode parameters of the electrode group to determine the corresponding region of each electrode in the infrared image sequence includes: The effective area of ​​the tank is determined based on the tank structure parameters. Based on the number of plates, the spacing between plates, the arrangement direction of plates, and the polarity of each plate, a candidate search area corresponding to each plate is generated within the effective area of ​​the tank. In each candidate search area, the plate region is divided, and the plate regions are assigned plate numbers according to the plate arrangement order. Based on the matching cost between the electrode region in the current infrared image and the historical trajectory prediction region, the electrode regions at adjacent sampling times are correlated; the calculation expression for the matching cost is: in, Indicates time The The current electrode region and the first The matching cost between historical polar plate trajectories; Indicates center distance Non-negative weights Indicates the first The center coordinates of the current electrode region Indicates the first The trajectory of the historical plate at any moment The predicted center coordinates; Indicates the degree of regional overlap Non-negative weights Indicates the first The current electrode region at time The area mask, Indicates the first The trajectory of the historical plate at any moment The predicted region mask, This indicates the calculation of the intersection-union ratio; Indicates area change Non-negative weights; Indicates the first The area of ​​the current electrode region Indicates the first The predicted area of ​​the historical polar plate trajectory. Indicates a preset positive number. This indicates taking the maximum value; The historical electrode trajectory corresponding to the current electrode region is determined according to the principle of minimizing the matching cost, so as to keep the electrode number of the same electrode consistent in the electrode region sequence.

3. The method for graded early warning of short circuits in lead electrolytic cell plates according to claim 2, characterized in that, The average temperature, the average local temperature gradient, and the proportion of hot spot area are determined according to the following formulas: in, Indicates the first At any time, the block plate average temperature Indicates the first At any time, the block plate The mean of the local temperature gradient, Indicates the first At any time, the block plate The percentage of hot spot area; Indicates the first At any time, the block plate The set of pixels in the region; This indicates the number of pixels contained in the region pixel set; Represents pixel coordinates; Indicates time Pixel coordinates in infrared images Temperature value; This represents the temperature gradient in the horizontal direction of the stated temperature value; This represents the temperature gradient of the temperature value in the vertical direction; This is an indicator function that takes the value of 1 when the condition inside the parentheses is true and takes the value of 0 when the condition inside the parentheses is false. Indicates the first Hot spot determination threshold for bulk electrode plates; Indicates the first The baseline average temperature of the bulk electrode under historical normal operating conditions; Indicates the first The baseline standard deviation of the temperature of the bulk electrode under historical normal operating conditions; This represents the safety margin coefficient used to determine the hot spot detection threshold.

4. The graded early warning method for short circuits in lead electrolytic cell plates according to claim 3, characterized in that, The calculation of the difference in single-plate thermal field characteristics between the target plate and its adjacent plates to obtain the differential characteristics of adjacent plates includes: For each plate thermal field characteristic among the average temperature, average local temperature gradient, and hot spot area ratio, the difference between the target plate's plate thermal field characteristic and the average value of the corresponding plate thermal field characteristics of adjacent plates is calculated to obtain the adjacent plate differential characteristics; the calculation expression for the adjacent plate differential characteristics is as follows: in, Indicates the first At any time, the block plate The average temperature, the average local temperature gradient, or the percentage of hot spot area; Indicates the first The difference between adjacent plates corresponding to the thermal field characteristics of a single plate; This indicates that the topology determined based on the electrode arrangement model is related to the first... The set of adjacent plates of a block electrode; This indicates the number of electrodes in the electrode set; Indicates the first in the set of electrodes At any time, the block plate The corresponding single-plate thermal field characteristics.

5. The method for graded early warning of short circuits in lead electrolytic cell plates according to claim 4, characterized in that, The calculation expression for the thermal trend characteristics is as follows: in, Indicates the first The thermal field characteristics of a single plate corresponding to a block plate at time 1 The slope of the change; Indicates the cutoff time. The set of sampling moments within the time window; Represents the first in the set of sampling times. Each sampling time; This represents the average value of each sampling time in the set of sampling times; Indicates the first At the sampling time of the block plate Corresponding single-plate thermal field characteristics; Indicates the first time within the time window The average value of the single-plate thermal field characteristics corresponding to the block plate.

6. The method for graded early warning of short circuits in lead electrolytic cell plates according to claim 5, characterized in that, The process of fusing the thermal field characteristics of the single plate, the differential characteristics of the adjacent plates, and the thermal field trend characteristics to obtain the time-series characteristic sequence of the target plate includes: The single-plate thermal field features, the adjacent plate differential features, and the thermal field trend features at each sampling time are scaled and spliced ​​to obtain the fused feature vector corresponding to each sampling time. Multiple fused feature vectors are arranged in the order of sampling time to obtain the temporal feature sequence.

7. The method for graded early warning of short circuits in lead electrolytic cell plates according to claim 1, characterized in that, The short-circuit temporal identification model includes a Long Short-Term Memory (LSTM) network, a fully connected layer, and a Softmax output layer. The LSM network extracts temporal information about the evolution of short-circuit risk states over time from the temporal feature sequence. The fully connected layer outputs a discrimination score for each preset short-circuit risk state based on the temporal information. The Softmax output layer converts the discrimination score into a state probability. The expression for calculating the state probability is: in, Indicates the first At any time, the block plate Belongs to the The state probability of a preset short-circuit risk state; This indicates that the fully connected layer is for the first... The first of the block plates The discrimination score of the preset short-circuit risk state output; This represents the total number of the preset short-circuit risk states; A category index representing the preset short-circuit risk state; Represents the natural exponential function; This indicates that the fully connected layer is for the first... The first of the block plates The discrimination score of the preset short-circuit risk state output; The preset short-circuit risk states include normal state, minor short-circuit state, and severe short-circuit state.

8. The method for graded early warning of short circuits in lead electrolytic cell plates according to claim 7, characterized in that, Determining early warning indicators from the time-series feature sequence, and determining rule-based risk values ​​based on the degree of deviation of the early warning indicators from a preset normal benchmark, includes: Based on the baseline mean and baseline standard deviation of each early warning indicator under historical normal operating conditions, the deviation of each early warning indicator from the preset normal baseline is standardized to obtain the standardized deviation. The positive deviations of each standardized deviation level are weighted and summed to obtain the rule risk value; the calculation expression for the rule risk value is: in, Indicates the first At any time, the block plate The rule risk value; This indicates the number of early warning indicators used to determine the risk value of the rule; Indicates the index of early warning indicators; Indicates the first The non-negative risk weights of each early warning indicator are equal to one; Indicates the first The degree of standardized deviation of each early warning indicator relative to the preset normal benchmark; Indicates the first At any time, the block plate The One early warning indicator; Indicates the first The baseline average of each early warning indicator under historical normal operating conditions; Indicates the first The baseline standard deviation of each early warning indicator under historical normal operating conditions.

9. The method for graded early warning of short circuits in lead electrolytic cell plates according to claim 8, characterized in that, The step of determining the short-circuit warning level of the target plate based on the comparison result of the state probability, the rule risk value, and the preset classification threshold includes: Set the first, second, and third level thresholds to increase sequentially; When the rule risk value is not lower than the first classification threshold and the sum of the probability of a minor short circuit state and the probability of a severe short circuit state is not lower than the first probability threshold, and this condition is maintained for a first preset duration, the target electrode is determined to be in a first-level warning state. When the rule risk value is not lower than the second classification threshold and the sum of the probability of a minor short circuit state and the probability of a severe short circuit state is not lower than the second probability threshold, and this condition is maintained for a second preset duration, the target electrode is determined to be in a level two warning state. When the rule risk value is not lower than the third classification threshold and the probability of a severe short circuit is not lower than the third probability threshold, and is maintained for a third preset duration, the target plate is determined to be in a level three warning state. When two or more short-circuit warning levels are simultaneously met, the short-circuit warning level with the highest level is determined as the short-circuit warning level of the target plate.

10. A graded early warning device for short circuits on lead electrolytic cell plates, characterized in that, include: The image acquisition module is used to continuously acquire multiple infrared images according to a preset sampling period to obtain an infrared image sequence covering at least one electrode group in the lead electrolytic cell; the electrode group includes cathode plates and anode plates arranged alternately along the electrode arrangement direction; The electrode region determination module is used to establish an electrode arrangement topology model based on the cell structure parameters of the lead electrolytic cell and the electrode parameters of the electrode group to determine the corresponding region of each electrode in the infrared image sequence of the infrared image sequence, thereby obtaining the electrode region sequence corresponding to each electrode; the electrode arrangement topology model is used to characterize the arrangement order, polarity and adjacency relationship of each electrode. The feature extraction module is used to extract single-plate thermal field features, including the average temperature, the average local temperature gradient, and the hot spot area ratio, for any target plate in the plate group based on the temperature value of the target plate in the corresponding region of each infrared image. It calculates the difference in single-plate thermal field features between the target plate and the adjacent plates determined based on the adjacent relationship to obtain the adjacent plate difference features, and performs trend fitting on the single-plate thermal field features of the target plate in the time dimension to obtain the thermal field trend features. The state recognition module is used to fuse the thermal field characteristics of the single plate, the differential characteristics of the adjacent plates, and the thermal field trend characteristics to obtain the time-series feature sequence of the target plate. The time-series feature sequence is then input into a pre-trained short-circuit time-series recognition model to obtain the state probability of the target plate belonging to each preset short-circuit risk state. The short-circuit time-series recognition model is a multi-class time-series model trained using historical time-series feature sequences labeled with different preset short-circuit risk states, and is used to output the state probability of each preset short-circuit risk state. The rule risk determination module is used to determine early warning indicators from the time-series feature sequence and determine the rule risk value based on the degree of deviation of the early warning indicators from the preset normal benchmark; the early warning indicators include at least one of the following: the average temperature of the target electrode, the average local temperature gradient, the proportion of hot spot area, the difference in average temperature between the target electrode and adjacent electrodes, and the slope of the change of the average temperature. The graded early warning module is used to determine the short circuit warning level of the target plate based on the comparison result of the state probability, the rule risk value and the preset graded threshold.

Images