Intelligent temperature control system for glass brick annealing furnace

Abstract

The present application relates to the technical field of glass brick annealing, and discloses a glass brick annealing furnace intelligent temperature control system, comprising: a collection module for acquiring multi-source data in the annealing process and generating feeding and discharging image sequences; an analysis module for regularly analyzing the feeding image sequences, establishing a reference template when the arrangement is normal and switching to a fixed time sampling detection mode, and calculating equivalent heat load and generating risk dimension markers through a detection model when the arrangement is abnormal; a control module for generating heating element feedforward compensation power and cooling fan target speed instructions according to the equivalent heat load, combining PID feedback and step limit output control instructions; the present application quickly acquires load information when the arrangement is normal, quantifies the heat load of each brick when the arrangement is abnormal, calculates the feedforward compensation amount under the condition of frequent load changes, avoids the fluctuation of furnace temperature caused by load estimation deviation, and thus ensures the stress elimination effect of glass brick annealing.

Description

Technical Field

[0001] This invention relates to the technical field of glass brick annealing, specifically to an intelligent temperature control system for a glass brick annealing furnace. Background Technology

[0002] The intelligent temperature control system of glass annealing furnaces generally adopts a closed-loop control scheme based on PID algorithm. The system collects temperature data in real time through thermocouples distributed in each temperature zone of the furnace, compares the measured temperature with the process set temperature, calculates the deviation by the PID controller, and adjusts the power of the heating element through solid-state relays, thereby realizing closed-loop regulation of the furnace temperature. At the same time, the furnace body is divided into multiple independent temperature control zones along the length direction. Each zone is equipped with an independent temperature instrument and heating circuit to realize segmented control of the annealing process temperature curve. The hot air in the furnace is forced to circulate through the circulating fan to improve the temperature uniformity in the furnace. When glass bricks of different specifications and thicknesses are sequentially introduced into the furnace, the differences in their mass and heat capacity cause changes in the heat absorption load entering the furnace per unit time. Existing PID control systems only initiate power adjustment after temperature deviation occurs. However, the thermal inertia of the annealing furnace means that the effect of heating power adjustment takes a long time to be reflected in the furnace temperature. This lag causes periodic fluctuations in furnace temperature under conditions of frequent load changes. The temperature coupling effect between different temperature zones intensifies the expansion of the transverse and longitudinal temperature differences, thereby affecting the stress relief effect of the glass bricks. Summary of the Invention

[0003] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an intelligent temperature control system for glass brick annealing furnaces, which has the advantage of multi-source data fusion and solves the problem of large temperature fluctuations and lag in adjustment caused by frequent load disturbances during continuous production, thus affecting the annealing quality of glass bricks.

[0004] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: an intelligent temperature control system for a glass brick annealing furnace, comprising: The acquisition module is used to acquire image data, speed data, temperature data and process parameter data during the annealing process, and align the multi-source data according to the time window to generate feed image sequence and discharge image sequence; The analysis module is used to perform regularity analysis on the feed image sequence. When the arrangement is determined to be normal, a baseline template is established and the fixed-time sampling detection mode is switched. When the arrangement is determined to be abnormal, the bounding box and specification category of each glass brick are output through the detection model, the equivalent heat load is calculated and risk dimension markers are generated. The control module is used to generate feedforward compensation power for heating elements in each temperature zone and target speed commands for cooling fans based on equivalent heat load, and to combine PID feedback to adjust power and step limit output control commands. The verification module is used to verify the directional defects of glass bricks marked with risk dimensions in the output image sequence. It performs attribution analysis by comparing the defect incidence rates of risk bricks and normal bricks, and feeds the results back to the control module.

[0005] Preferably, the feed image sequence undergoes regularity analysis, specifically including: Extract the geometric center coordinates of each glass brick from several consecutively acquired images to obtain a point set. If any of the following conditions are met, an abnormal arrangement is determined and the detection model is triggered to analyze the image frame: The number of points in the point set is less than the preset minimum effective point threshold; The range of the nearest neighbor distances of all points in the point set is greater than a preset range threshold or the standard deviation is greater than a preset standard deviation threshold. The concentration of nearest neighbor distances between points in the point set is lower than the preset concentration threshold.

[0006] Preferably, the calculation process of the nearest neighbor distance between points in the point set specifically includes: When the number of points in the point set is greater than or equal to the preset minimum effective point threshold, the distance between each point and its nearest neighbor is calculated to obtain a distance set. A histogram is constructed based on the distance set, and the range and standard deviation of the distance set are calculated.

[0007] Preferably, the process of calculating the concentration of nearest neighbor distances for each point in the point set specifically includes: Find the interval with the highest frequency in the histogram, take the center value of the interval as the candidate standard interval, count the number of distances corresponding to the candidate standard interval, divide the number of distances corresponding to the interval by the total number of distances in the distance set, and obtain the concentration of the candidate standard interval.

[0008] Preferably, the specific process by which the detection model outputs the bounding box and specification category of each glass brick includes: The geometric center coordinates of the point set are used as the candidate region anchor points of the detection model. The backbone network of the detection model performs layer-by-layer convolution downsampling on the input image to extract multi-scale feature maps. The detection head only activates the detection channel at the feature map position corresponding to the geometric center coordinates. Based on each geometric center, multiple prior boxes of different sizes and aspect ratios are preset around it. The dimensions of the prior frame are based on the physical width and length corresponding to the glass brick specifications obtained from the process parameter data, combined with the camera calibration coefficient, and two sets of width and height values ​​are preset in the horizontal and vertical directions respectively.

[0009] Preferably, the specific process of estimating the equivalent heat load and generating risk dimension markers includes: Based on the bounding box of each glass brick output by the detection model, the horizontal and vertical spacing between adjacent bounding boxes, as well as the physical area of ​​each brick, are calculated to obtain the total area of ​​the bricks and the total area of ​​the spacing in the current frame. The heat absorption of all glass bricks in the current frame is summed, divided by the time window corresponding to the frame, and then multiplied by the ratio of the total area of ​​the bricks to the total area of ​​the bricks plus the total area of ​​the spacing to obtain the equivalent heat load. The heat absorption of each brick is calculated based on its physical area, thickness, density and specific heat capacity corresponding to its specification category. The equivalent heat load expression is as follows: , where P thermal Represents the equivalent heat load; the numerator ∑Q i Represents the sum of heat absorbed by all bricks in the current frame; Δt represents the time window; the ratio part This is the spacing correction coefficient, reflecting the coverage of glass bricks on the conveyor belt. The larger the value, the denser the glass bricks are arranged and the greater the feed rate per unit time. This results in a higher equivalent heat load, and consequently, a higher compensation power is output to the heating elements in each temperature zone in advance, overcoming the lag caused by the thermal inertia of the annealing furnace; S brick S represents the sum of the physical areas of all bricks in the current frame. gap This represents the total area of ​​the spacing between all adjacent bricks in the current frame; Read the distance between each brick and its adjacent bricks, mark bricks with a distance lower than the preset safe distance threshold as spacing risk bricks, and mark bricks with a distance greater than the preset edge distance threshold as edge location risk bricks based on the distance between the horizontal coordinate of the center of each brick's bounding box and the center line of the conveyor belt. The preset edge distance threshold is determined based on the total width of the conveyor belt and the width of the glass brick.

[0010] Preferably, the calculation process for the feedforward compensation power of the heating element in each temperature zone includes: The acquisition module is also used to generate a position tracking sequence. The position tracking sequence is generated by calculating the time when the glass brick arrives at each temperature zone corresponding to each frame image based on the pre-calibrated physical distance from the feed inlet to the center of each temperature zone and the conveyor belt linear speed measured in real time by the rotary encoder. The equivalent heat load is distributed according to the preset heat distribution coefficients of each temperature zone to obtain the initial feedforward compensation power of each temperature zone. The heat distribution coefficients are determined according to the annealing process curve and the thermophysical properties of the glass bricks. The sum of the heat distribution coefficients of each temperature zone is one. The travel events of the current batch of glass bricks arriving at each temperature zone are read from the position tracking sequence and used as delay events for the feedforward signal of each temperature zone. The time when the batch of glass bricks is photographed at the feed inlet is taken as the timing start point. A delay event equal to the travel time is set for each temperature zone. The feedforward compensation power is not output during the delay event and is superimposed on the output of the PID controller of that temperature zone after the delay ends.

[0011] Preferably, the calculation process for the target speed command of the cooling fan includes: Based on the preset calibration relationship between cooling fan speed and heat dissipation, the equivalent heat load of the current batch is converted into the target speed of the cooling fan. The calibration relationship is established by measuring the heat carried away from the cooling zone by the fan per unit time at different speeds. The cooling fan speed is adjusted based on the cooling rate required by the process. The cooling rate setpoint is extracted from the process parameter data, specifying the upper and lower limits of the cooling rate of a batch of glass bricks in the cooling zone. The travel time of the current batch of glass bricks to the entrance of the cooling zone is read from the position tracking sequence and used as the delay time of the feedforward signal of the cooling fan. The timing starts from the moment when the industrial camera at the feed inlet takes the picture of the batch of glass bricks. The fan speed remains unchanged during the delay time. After the delay ends, the target speed command is output to the fan controller to ensure that the cooling air volume is adjusted synchronously when the batch of bricks enters the cooling zone.

[0012] Preferably, the specific process of combining PID feedback to adjust power and step limit output control commands includes: The acquisition module is also used to generate temperature sequences, which are generated by aligning the furnace temperatures collected in real time by thermocouples in each temperature zone according to time windows. The temperature sequence is read, and the PID controller generates feedback regulation power based on the temperature deviation between the actual temperature and the process set temperature extracted from the process parameter data. Before the feedforward control is put into operation, each temperature zone operates in pure PID control mode. The maximum and minimum values ​​of the PID output power during the period of minimum temperature fluctuation are extracted, and half of the difference between the two is taken as the maximum single-step adjustment of the feedforward compensation power of the temperature zone. The feedforward compensation power after step limitation is added to the feedback regulation power to form the heating power command of the temperature zone.

[0013] Preferably, the specific process for targeted defect verification of the abnormal glass bricks marked as risk dimensions includes: For glass bricks marked with risk dimensions, conduct targeted inspections to check for defects in specific areas based on their risk types. If the defect rate of risky bricks is higher than that of normal bricks in several consecutive batches recently, the attribution results will be fed back to the control module to correct the heat distribution coefficient of the corresponding temperature zone. If the defect rate increases after adjustment, an alarm will be triggered.

[0014] (III) Beneficial Effects Compared with the prior art, the present invention provides an intelligent temperature control system for glass brick annealing furnaces, which has the following beneficial effects: 1. This invention analyzes the regularity of the feed image sequence. When the glass bricks are arranged normally on the conveyor belt, a reference template is established and the sampling detection mode is switched to a fixed time. When the arrangement is abnormal, the detection model outputs the boundary box and specification type of each glass brick and calculates the equivalent heat load. This allows for graded processing of different arrangement states, enabling rapid acquisition of load information when the arrangement is normal and accurate quantification of the heat load of each brick when the arrangement is abnormal. This ensures accurate calculation of feedforward compensation under conditions of frequent load changes, avoiding the problem of furnace temperature fluctuations caused by load estimation deviations, which in turn affect the stress relief effect of glass brick annealing.

[0015] 2. This invention distributes the equivalent heat load according to the heat distribution coefficient of each temperature zone and sets a delayed trigger time in combination with the position tracking sequence. This allows the feedforward compensation power to be superimposed on the output of the PID controller when the glass enters the corresponding temperature zone. At the same time, the upper limit of the step is determined based on the steady-state fluctuation range extracted from each temperature zone in pure PID control mode to limit the feedforward power. This achieves synchronization between the feedforward compensation and the arrival of the bricks in time and matching with the thermal inertia of the furnace in amplitude. It avoids interference with the existing bricks in the furnace due to preheating and temperature overshoot caused by power step, thereby ensuring the furnace temperature is stable during the annealing process of the glass bricks and ensuring uniform and sufficient stress relief.

[0016] 3. This invention verifies the directional defects of glass bricks marked with risk dimensions in the output image sequence, uses the comparison of defect incidence rates between risk bricks and normal bricks to determine the cause analysis, and feeds the cause analysis results back to the control module to correct the heat distribution coefficients of each temperature zone, thereby achieving closed-loop correction of the output quality to the feedforward model and ensuring the stability of the annealing quality of glass bricks. Attached Figure Description

[0017] Figure 1 This is a system diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] As attached Figure 1 As shown, this embodiment provides an intelligent temperature control system for a glass brick annealing furnace, including a data acquisition module, an analysis module, a control module, and a verification module; The acquisition module communicates with various detection components deployed on-site to obtain image data, speed data, temperature data, and process parameter data during the annealing process. The detection components include industrial cameras, rotary encoders, thermocouples, and photoelectric sensors. At least two industrial cameras are provided, one installed above the annealing furnace's inlet and the other above its outlet, used to acquire image data from the inlet and outlet conveyor belts. The rotary encoder is installed at the end of the annealing furnace's drive roller to obtain the conveyor belt's running speed data. Several thermocouples are installed inside each temperature zone of the annealing furnace to acquire furnace temperature data for each zone. The photoelectric sensor is installed at the annealing furnace's inlet to detect whether the glass bricks have reached a preset position and outputs a trigger signal to the acquisition module, triggering the industrial camera to perform image acquisition. The acquisition module communicates with the upper-level control system. The machine communicates with the glass bricks and collects the process parameter data of the current batch. The process parameter data includes at least the product model and the thickness, density and specific heat capacity of the glass bricks corresponding to that model, as well as the annealing temperature curve setting value and cooling rate setting value. The image data, speed data, temperature data and process parameter data are aligned with the time window corresponding to a single frame image as the reference. The time window is calculated by combining the actual physical length covered by the single frame image obtained by the pre-calibrated field of view of the industrial camera on the conveyor belt with the real-time linear speed of the conveyor belt measured by the rotary encoder. After alignment, the feeding image sequence, discharging image sequence, speed sequence, temperature sequence and position tracking sequence are generated. The position tracking sequence calculates the time when the glass bricks corresponding to each frame image arrive at each temperature zone based on the pre-calibrated physical distance from the feed port to the center of each temperature zone and the linear speed of the conveyor belt.

[0020] The analysis module is used to perform analysis on several consecutively acquired frames of images from the feed image sequence. The specific process includes: Several frames of images are extracted and converted into grayscale images. In this example, the values ​​of the frames range from [3,5]. Binarized images are generated using an adaptive thresholding method. Connectivity analysis is performed on the binarized images to extract the geometric center coordinates of each connected component, resulting in a point set P={(x... i y i If the number of points n in point set P is less than the preset minimum effective number of points threshold, it is determined that the number of bricks on the current conveyor belt is insufficient for regular analysis, triggering an alarm to remind the operator to continue feeding. Once the point set P meets the quantity requirement, iterate through the points and calculate the distance between each point and its nearest neighbor, obtaining the distance set D = {d1, d2, ..., dn}. mA distance histogram is constructed based on the distance set, using a preset interval width. Each distance value in the distance set is assigned to a corresponding interval, and the frequency of distance values ​​within each interval is counted. The range R and standard deviation S of the distance set are calculated. Dd Preset range threshold R th and standard deviation threshold S Dth If R > R th or S Dd >S Dth The distance values ​​are too discrete, there are no obvious peaks in the histogram, and the judgments are disordered. When R≤R th And S Dd ≤S Dth Find the interval with the highest frequency in the histogram, take the center value of the interval as the candidate standard distance, and divide the number of distances corresponding to the interval by the total number of distance sets to obtain the concentration of the candidate standard distance. A preset concentration threshold C is set. th C th Based on actual working conditions, 70% is taken in this embodiment. If the concentration is lower than C... th This indicates the existence of high-frequency intervals, but the number of bricks corresponding to these intervals does not constitute the vast majority, indicating insufficient overall regularity in the set, thus classifying it as a disordered arrangement; if the concentration is ≥ C th If so, the candidate standard spacing is confirmed to be the unique standard spacing of the current batch of glass bricks; The standard spacing is compared with the preset minimum safe spacing threshold. The minimum safe spacing threshold is determined based on the width of the glass brick and the minimum brick spacing required by the process. If the standard spacing is less than the minimum safe spacing threshold, it means that the brick spacing is uniform, but the overall spacing is too small, which poses a risk of stacking or conveyor belt congestion and triggers an alarm. If the standard spacing is greater than the minimum safe spacing threshold, it is determined that the current glass brick is in a normal and regular arrangement. When the current arrangement is determined to be normal, the point set extracted from the initial N frames of images is aligned and averaged to obtain an average point set as a reference template. The number of points contained in this average point set is the number of reference points, n. base Record this value as a baseline for subsequent comparisons.

[0021] After entering the regular arrangement state, it switches to the fixed-time sampling detection mode. Instead of analyzing frame by frame, it extracts one frame at fixed time intervals for analysis, and extracts the point set p from the extracted frames. new Compare P new Number of points n new With the number of reference points n base If n new Greater than n base If abnormal incoming materials or foreign objects are detected, an alarm is triggered. If n new Less than nbase Starting from the current frame, several frames are continuously collected again for distance histogram analysis. If the number of points is stable and the arrangement still conforms to the rules after the complete analysis process, the baseline template and the number of baseline points are updated, and the time sampling detection mode is maintained. If the number of points is unstable, an alarm is triggered. A preset offset tolerance range and an over-limit threshold are defined. The offset tolerance range is determined based on a certain proportion of the standard spacing in the reference template; in this embodiment, 15% of the standard spacing is used as the offset tolerance range. The over-limit threshold is determined based on the percentage of locally shaking bricks allowed; in this embodiment, it is set to 20%. If n new With n base Consistent, calculate P new The offset of each point in the data is compared with the corresponding point in the baseline template. The number of points exceeding the preset offset tolerance range is counted. If the proportion of the number of points exceeding the tolerance is lower than the preset threshold, it is determined that the shaking is of individual bricks, and the current baseline template is used to continue the detection mode. If the proportion of points exceeding the tolerance exceeds the threshold, it is determined that the arrangement state has changed significantly, and the complete analysis process is re-executed to update the baseline template and the number of baseline points. If the re-analysis is triggered multiple times in a row, it is determined that the working condition continues to deteriorate and an alarm is triggered.

[0022] When determining that the brick arrangement on the conveyor belt is abnormal, a detailed analysis of the situation is performed, including the following steps: The geometric center coordinates P of each independent clique extracted through binarization and connected component analysis are used as candidate region anchor points for YOLO detection. The YOLO backbone network performs layer-by-layer convolution and downsampling on the input image to extract multi-scale feature maps. The detection head activates the detection channel only at the feature map positions corresponding to the aforementioned geometric center coordinates. Based on each geometric center, multiple prior boxes of different sizes and aspect ratios are preset around it. The size of the prior boxes is preset according to the specifications and types of glass bricks produced on the current production line. For each specification of glass brick, based on its actual physical width and length, and combined with the pixel conversion coefficients calibrated by the camera, two sets of prior box width and height values ​​are preset for horizontal and vertical placement respectively. The prior box sizes corresponding to all specifications are written into the anchor point configuration parameters of the YOLO model. It should be noted that during the firing and annealing process of glass bricks, the transparency of the glass causes the YOLO model to be unable to recognize the boundaries of the detection boxes. The detection head predicts an offset for each candidate region anchored at its geometric center, which is used to adjust the size of the prior bounding box to fit the actual brick boundary. The bounding box output expression is: , where (x i y i ) represents the coordinates of the i-th geometric center; p w and p h△x and △y represent the width and height of the pre-defined bounding box; △x and △y represent the center coordinate offset predicted by the detector head, used to correct the initial reference position; △w and △h represent the preset offset of the detector head; b x b y b represents the center coordinates of the bounding box in the final output, indicating the precise location of the detected glass brick in the image; w b h This indicates the final output boundary width and height, representing the precise dimensions of the detection box; For two horizontally adjacent bounding boxes, take the center x-coordinate and width of the left bounding box, and the center x-coordinate and width of the right bounding box. The horizontal spacing between them is the absolute value of the difference between the center x-coordinates of the two boxes minus the sum of half the widths of the two boxes. The calculation of the vertical spacing is similar. Replace the x-coordinate with the y-coordinate and the width with the height. Traverse all adjacent bounding box pairs in the current frame and calculate their horizontal and vertical spacing one by one to obtain the spacing set. Sum all the spacings to get the total spacing length of the current frame. Multiply this by the width of the conveyor belt to get the spacing area of ​​the current frame, which is the area of ​​the gap area on the conveyor belt that is not covered by the glass bricks. At the same time, sum the physical areas of all the glass bricks to get the total area of ​​the glass bricks in the current frame. The equivalent heat load is obtained by summing the heat absorbed by all glass blocks in the current frame, dividing by the time window corresponding to that frame, and then multiplying by the ratio of the total area of ​​the glass blocks to the total area of ​​the glass blocks plus the total area of ​​the spacing. The expression is as follows: , where P thermal Represents the equivalent heat load; the numerator ∑Q i Represents the sum of heat absorbed by all bricks in the current frame; Δt represents the time window; the ratio part This is the spacing correction coefficient, reflecting the coverage of glass bricks on the conveyor belt. The larger the value, the denser the glass bricks are arranged and the greater the feed rate per unit time. This results in a higher equivalent heat load, and consequently, a higher compensation power is output to the heating elements in each temperature zone in advance, overcoming the lag caused by the thermal inertia of the annealing furnace; S brick S represents the sum of the physical areas of all bricks in the current frame. gap This represents the total area of ​​the spacing between all adjacent bricks in the current frame; The heat absorption (QI) of each glass brick is calculated based on its physical area, the thickness of the glass brick corresponding to the product model, and the density and specific heat capacity corresponding to the specification category. Specifically, after YOLO outputs the specification category of each brick, it retrieves the corresponding glass brick thickness, density, and specific heat capacity from the process parameter data according to the specification category and substitutes them into the heat absorption calculation.

[0023] The control module generates adjustment commands for the heating elements and cooling fans based on the equivalent heat load. The specific process includes: A preset minimum action threshold is set. If the equivalent heat load is less than or equal to the minimum action threshold, it indicates that the current load change is small and insufficient to have a measurable impact on the furnace temperature. The current heating power and fan speed are kept constant. When the equivalent heat load is greater than the minimum action threshold, the feedforward compensation power and the target speed of the cooling fan for each temperature zone are calculated accordingly, triggering feedforward adjustment. When the equivalent heat load is negative, it is determined to be a calculation abnormality, and the feedforward output of this frame is paused, while the effective output of the previous frame remains unchanged. The preset minimum action threshold is taken as the minimum power change that the heating element of each temperature zone can stably output in a single control cycle and the minimum speed change that the cooling fan can stably execute. These are converted into the corresponding equivalent heat load changes, and the smaller of the two values ​​is used to determine the threshold.

[0024] When the equivalent heat load exceeds the minimum operating threshold, the heating element is adjusted. The specific process includes: The equivalent heat load is distributed according to the preset heat distribution coefficients for each temperature zone, resulting in the feedforward compensation power for each temperature zone. The calculation expression for the feedforward compensation power of each temperature zone is: P j =β j ×P thermal , where P j P represents the feedforward compensation power allocated to the j-th temperature zone; thermal Indicates the equivalent heat load; β j表示第 The heat distribution coefficients for j temperature zones are determined by: taking a standard-sized glass brick and calculating the actual heat absorption Q of the glass brick in the j temperature zone based on its mass and specific heat capacity as a function of temperature. j Actual heat absorption Q j The inlet temperature Tj-1 and outlet temperature Tj of the j-th temperature zone are determined based on the annealing process curve. The average specific heat capacity C of the glass brick within the inlet temperature range is estimated from the characteristic curve of the specific heat capacity of the glass brick as a function of temperature. p The actual heat absorption Q in this temperature range j The actual heat absorption is calculated by multiplying the mass m of the glass brick by the average specific heat capacity and then by the temperature difference between the temperature zones. The expression for this is: Q j =m×C p ×(Tj-Tj-1) Summing up the heat absorbed in each temperature zone yields the total heat absorbed Q across all temperature zones. total The heat distribution coefficient of a temperature zone is obtained by dividing the heat absorbed in that temperature zone by the total heat absorbed. The fitting formula is: The sum of the heat distribution coefficients of each temperature zone is equal to one.

[0025] The travel time of the current batch of glass bricks to each temperature zone is read from the position tracking sequence and used as the delay time of the feedforward signal for each temperature zone. The time when the industrial camera at the feed inlet takes the picture of the batch of glass bricks is taken as the timing start point. A delay time equal to the travel time is set for each temperature zone. The feedforward compensation power is not output during the delay time and is superimposed on the output of the temperature zone PID controller after the delay ends.

[0026] Read the temperature sequence, the PID controller compares the received actual temperature with the process set temperature of the temperature zone, subtracts the actual temperature from the set temperature to obtain the temperature deviation at the current moment, and performs proportional, integral and derivative calculations on the temperature deviation through the PID controller. The three results are superimposed to generate feedback regulation power. Before feedforward control is implemented, each temperature zone operates without feedforward compensation power. It relies on the PID controller to adjust according to the temperature deviation feedback. The output power of the PID controller in each temperature zone and the corresponding furnace temperature sequence are continuously collected. The time period with the smallest temperature fluctuation in the temperature sequence is extracted. The maximum and minimum values ​​of the PID output power in this time period are extracted. Half of the difference between the two is taken as the maximum single-step adjustment amount of the feedforward compensation power in that temperature zone.

[0027] The feedforward compensation power after step limitation is added to the feedback regulation power to form the final heating power command for this temperature zone. The feedback regulation corrects the real-time temperature deviation in a closed loop to handle small fluctuations and steady-state errors in the furnace temperature. It outputs the feedforward compensation power in advance according to changes in the feed load to handle large temperature disturbances caused by the entry of glass bricks.

[0028] As subsequent batches of glass bricks pass through the feed inlet, an industrial camera captures and analyzes images of each batch, calculating the equivalent heat load for each batch. Each batch of glass bricks corresponds to an independent set of feedforward compensation power. When a batch of glass bricks reaches each temperature zone, the corresponding power is superimposed. If the equivalent heat load of subsequent batches increases, it indicates that the batch of glass bricks needs to absorb more heat, and the feedforward compensation power for each temperature zone is increased. If the equivalent heat load of subsequent batches decreases, it indicates that the batch of glass bricks needs to absorb less heat, and the feedforward compensation power for each temperature zone is reduced accordingly. If the equivalent heat load of multiple consecutive batches is lower than the minimum action threshold, it is determined that the conveyor belt is in an unloaded or near-unloaded state, and the feedforward compensation power is gradually reduced to zero, with only PID feedback regulation remaining in each temperature zone.

[0029] Based on the preset calibration relationship between cooling fan speed and heat dissipation, the equivalent heat load of the current batch is converted into the target speed of the cooling fan. This calibration relationship is based on: measuring the heat discharged from the cooling zone per unit time by the fan at different cooling fan speeds, obtaining several sets of corresponding data points of speed and heat dissipation, and fitting the curve of the data points to obtain the functional relationship between heat dissipation and fan speed, i.e., Q. cool =f(R), where Qcool Let R represent the heat dissipation, f represent the fan speed, and f represent the fitted calibration function. Based on this, a mapping table between speed and heat dissipation is constructed. During actual control, the required heat dissipation is determined according to the equivalent heat load of the current batch. The target speed is then calculated by substituting this value into the inverse function of the calibration function. The expression is: , where R target P represents the target rotational speed of the cooling fan. thermal Indicates the equivalent heat load. This represents the inverse function of the calibration function f. The adjustment of the cooling fan speed is constrained by the cooling rate required by the process. The cooling rate setpoint obtained by the acquisition module specifies the upper and lower limits of the cooling rate of this batch of glass bricks in the cooling zone. After converting the equivalent heat load into the required heat dissipation, the control module corrects the target speed in conjunction with the cooling rate setpoint to match the heat dissipation corresponding to the fan speed with the heat dissipation required by the glass bricks, while ensuring that the cooling rate does not exceed the upper limit specified by the cooling rate setpoint.

[0030] The travel time of the current batch of glass bricks to the cooling zone inlet is read from the position tracking sequence and used as the delay time for the cooling fan feedforward signal. The timing starts from the moment the industrial camera at the feed inlet captures the image of this batch of glass bricks. The fan speed remains unchanged during this delay time. After the delay ends, the target speed command is output to the fan controller. As subsequent batches of glass bricks pass through the feed inlet in sequence, the industrial camera captures and analyzes the image of each batch, calculating the equivalent heat load for each batch. The batch-by-batch adjustment of the cooling fan speed is the same as the logic for the heating zone. Each batch of glass bricks corresponds to an independent target speed. When this batch of glass bricks arrives at the cooling zone, the corresponding speed adjustment is executed. If the load increases, the speed is increased; if the load decreases, the speed is decreased; if there is continuous no load, the speed drops back to the basic standby speed.

[0031] The verification module is used to perform targeted defect verification on abnormal glass bricks marked as risk dimensions by the analysis module in the output image sequence. The specific process includes: When determining that the brick arrangement on the conveyor belt is abnormal, YOLO outputs the bounding box and specification category of each glass brick, and simultaneously analyzes the local arrangement features of each brick to generate risk dimension markers for the glass bricks. It reads the lateral and longitudinal spacing between the glass bricks and adjacent bricks, marks bricks with spacing lower than the preset safe spacing threshold as spacing risk bricks, and marks glass bricks located at the edges of the conveyor belt as edge position risk bricks based on the position of the horizontal coordinate of the center of the glass brick bounding box in the width direction of the conveyor belt. For bricks with spacing risks, the discharge port should focus on checking whether there are micro-cracks or stress marks on the side edge area of ​​the brick. For bricks with edge risks, the discharge port should focus on checking the surface condition of the brick on the side closest to the furnace wall to determine whether there is deformation caused by unilateral overheating. The defect incidence rate of risky bricks is obtained by dividing the number of abnormal bricks marked with the same risk type in recent batches that are confirmed to have corresponding defects after directional verification at the discharge port by the total number of abnormal bricks marked with that risk type in the batch. Similarly, the defect incidence rate of normal bricks is obtained by dividing the number of normal bricks that are confirmed to have defects after defect detection at the discharge port in the same period that are not marked with any risk type by the total number of normal bricks that are not marked with any risk type in the batch. The defect detection range of normal bricks is consistent with the directional verification range of risky bricks to ensure that the defect judgment criteria of the two types of bricks are the same. The defect incidence rate is defined as the proportion of the number of bricks with defects detected within the statistical range to the total number of bricks detected within that range. In this embodiment, the number of batches is taken as 5. The defect rate of risky bricks is compared with that of normal bricks. If the defect rate of risky bricks is consistently higher than that of normal bricks, and the difference remains stable across multiple batches, the risk dimension indicates a problem with the furnace thermal field. In this embodiment, three consecutive batches are used. The temperature deviation recorded when the defective bricks pass through the corresponding temperature zone is traced back. If the temperature deviation is negative, the heat distribution coefficient of that temperature zone is increased; if the temperature deviation is positive, it is decreased. Each adjustment is a fixed percentage of the current coefficient, which is 5% in this embodiment. After adjustment, the heat distribution coefficients of all temperature zones in that zone are normalized. After the adjustment takes effect in the next batch, the defect rate comparison data of the subsequent five batches are checked. If the defect rate of risky bricks decreases after adjustment compared to before adjustment, and the difference between the decreased defect rate of risky bricks and the normal defect rate during the same period is less than the preset difference judgment threshold, the adjusted heat distribution coefficient remains unchanged. If the defect rate of risky bricks increases after adjustment compared to before adjustment, an alarm is triggered.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent temperature control system for a glass brick annealing furnace, characterized in that, include: The acquisition module is used to acquire image data, speed data, temperature data and process parameter data during the annealing process, and align the multi-source data according to the time window to generate feed image sequence and discharge image sequence; The analysis module is used to perform regularity analysis on the feed image sequence. When the arrangement is determined to be normal, a baseline template is established and the fixed-time sampling detection mode is switched. When the arrangement is determined to be abnormal, the bounding box and specification category of each glass brick are output through the detection model, the equivalent heat load is calculated and risk dimension markers are generated. The control module is used to generate feedforward compensation power for heating elements in each temperature zone and target speed commands for cooling fans based on equivalent heat load, and to combine PID feedback to adjust power and step limit output control commands. The verification module is used to verify the directional defects of glass bricks marked with risk dimensions in the output image sequence. It performs attribution analysis by comparing the defect incidence rates of risk bricks and normal bricks, and feeds the results back to the control module.

2. The intelligent temperature control system for a glass brick annealing furnace according to claim 1, characterized in that, The feed image sequence is subjected to regularity analysis, specifically including: Extract the geometric center coordinates of each glass brick from several consecutively acquired images to obtain a point set. If any of the following conditions are met, an abnormal arrangement is determined and the detection model is triggered to analyze the image frame: The number of points in the point set is less than the preset minimum effective point threshold; The range of the nearest neighbor distances of all points in the point set is greater than a preset range threshold or the standard deviation is greater than a preset standard deviation threshold. The concentration of nearest neighbor distances between points in the point set is lower than the preset concentration threshold.

3. The intelligent temperature control system for a glass brick annealing furnace according to claim 2, characterized in that, The calculation process for the nearest neighbor distance of each point in the point set specifically includes: When the number of points in the point set is greater than or equal to the preset minimum effective point threshold, the distance between each point and its nearest neighbor is calculated to obtain a distance set. A histogram is constructed based on the distance set, and the range and standard deviation of the distance set are calculated.

4. The intelligent temperature control system for a glass brick annealing furnace according to claim 3, characterized in that, The specific process for calculating the concentration degree of the nearest neighbor distances of each point in the point set includes: Find the interval with the highest frequency in the histogram, take the center value of the interval as the candidate standard interval, count the number of distances corresponding to the candidate standard interval, divide the number of distances corresponding to the interval by the total number of distances in the distance set, and obtain the concentration of the candidate standard interval.

5. The intelligent temperature control system for a glass brick annealing furnace according to claim 4, characterized in that, The specific process by which the detection model outputs the bounding box and specification category of each glass brick includes: The geometric center coordinates of the point set are used as the candidate region anchor points of the detection model. The backbone network of the detection model performs layer-by-layer convolution downsampling on the input image to extract multi-scale feature maps. The detection head only performs detection at the feature map position corresponding to the geometric center coordinates. Based on each geometric center, multiple prior boxes of different sizes and aspect ratios are preset around it. The dimensions of the prior frame are based on the physical width and length corresponding to the glass brick specifications obtained from the process parameter data, combined with the camera calibration coefficient, and two sets of width and height values ​​are preset in the horizontal and vertical directions respectively.

6. The intelligent temperature control system for a glass brick annealing furnace according to claim 5, characterized in that, The specific process of estimating the equivalent heat load and generating risk dimension markers includes: Based on the bounding box of each glass brick output by the detection model, the horizontal and vertical spacing between adjacent bounding boxes, as well as the physical area of ​​each brick, are calculated to obtain the total area of ​​the bricks and the total area of ​​the spacing in the current frame. The heat absorption of all glass bricks in the current frame is summed, divided by the time window corresponding to the frame, and then multiplied by the ratio of the total area of ​​the bricks to the total area of ​​the bricks plus the total area of ​​the spacing to obtain the equivalent heat load. The heat absorption of each brick is calculated based on its physical area, thickness, density and specific heat capacity corresponding to its specification category. Read the distance between each brick and its adjacent bricks, mark bricks with a distance lower than the preset safe distance threshold as spacing risk bricks, and mark bricks with a distance greater than the preset edge distance threshold as edge location risk bricks based on the distance between the horizontal coordinate of the center of each brick's bounding box and the center line of the conveyor belt. The preset edge distance threshold is determined based on the total width of the conveyor belt and the width of the glass brick.

7. The intelligent temperature control system for a glass brick annealing furnace according to claim 6, characterized in that, The calculation process for the feedforward compensation power of the heating elements in each temperature zone includes: The acquisition module is also used to generate a position tracking sequence. The position tracking sequence is generated by calculating the time when the glass brick arrives at each temperature zone corresponding to each frame image based on the pre-calibrated physical distance from the feed inlet to the center of each temperature zone and the conveyor belt linear speed measured in real time by the rotary encoder. The equivalent heat load is distributed according to the preset heat distribution coefficients of each temperature zone to obtain the initial feedforward compensation power of each temperature zone. The sum of the heat distribution coefficients of each temperature zone is one. The travel events of the current batch of glass bricks arriving at each temperature zone are read from the position tracking sequence and used as delay events for the feedforward signal of each temperature zone. The time when the batch of glass bricks is photographed at the feed inlet is taken as the timing start point. A delay event equal to the travel time is set for each temperature zone. The feedforward compensation power is not output during the delay event and is superimposed on the output of the PID controller of that temperature zone after the delay ends.

8. The intelligent temperature control system for a glass brick annealing furnace according to claim 7, characterized in that, The calculation process for the target speed command of the cooling fan includes: Based on the preset calibration relationship between cooling fan speed and heat dissipation, the equivalent heat load of the current batch is converted into the target speed of the cooling fan. The calibration relationship is established by measuring the heat carried away from the cooling zone by the fan per unit time at different speeds. The cooling fan speed is adjusted based on the cooling rate required by the process. The cooling rate setpoint is extracted from the process parameter data, specifying the upper and lower limits of the cooling rate of a batch of glass bricks in the cooling zone. The travel time of the current batch of glass bricks to the entrance of the cooling zone is read from the position tracking sequence and used as the delay time of the feedforward signal of the cooling fan. The timing starts from the moment when the industrial camera at the feed inlet takes the picture of the batch of glass bricks. The fan speed remains unchanged during the delay time. After the delay ends, the target speed command is output to the fan controller to ensure that the cooling air volume is adjusted synchronously when the batch of bricks enters the cooling zone.

9. The intelligent temperature control system for a glass brick annealing furnace according to claim 8, characterized in that, The specific process of combining PID feedback to adjust power and step limit output control commands includes: The acquisition module is also used to generate temperature sequences, which are generated by aligning the furnace temperatures collected in real time by thermocouples in each temperature zone according to time windows. The temperature sequence is read, and the PID controller generates feedback regulation power based on the temperature deviation between the actual temperature and the process set temperature extracted from the process parameter data. Before the feedforward control is put into operation, each temperature zone operates in pure PID control mode. The maximum and minimum values ​​of the PID output power during the period of minimum temperature fluctuation are extracted, and half of the difference between the two is taken as the maximum single-step adjustment of the feedforward compensation power of the temperature zone. The feedforward compensation power after step limitation is added to the feedback regulation power to form the heating power command of the temperature zone.

10. The intelligent temperature control system for a glass brick annealing furnace according to claim 9, characterized in that, The specific process for targeted defect verification of the abnormal glass bricks marked as risk dimensions includes: For glass bricks marked with risk dimensions, conduct targeted inspections to check for defects in specific areas based on their risk types. If the defect rate of risky bricks is higher than that of normal bricks in several consecutive batches recently, the attribution results will be fed back to the control module to correct the heat distribution coefficient of the corresponding temperature zone. If the defect rate increases after adjustment, an alarm will be triggered.

Images