Wood-plastic door production process online monitoring system

Abstract

The application discloses a kind of wood plastic door production process online monitoring systems, specifically related to wood plastic door manufacturing technical field, including: by multimodal perception module acquisition and fusion data, threshold dynamic adjustment module outputs defect identification result, composite control execution module is according to identification result to output adjustment instruction to executing mechanism, feedback correction module is according to the deviation of actual operation data and target data after execution correction determination reference value and back to threshold dynamic adjustment module, process monitoring module is according to the fusion data of determination reference value after correction corresponding generation early warning signal and feedback to multimodal perception module to adjust acquisition strategy;The application is through multimodal perception and signal-to-noise ratio self-adaptive regulation defect determination threshold, maintain defect detection accuracy in texture noise, through composite control execution and feedback correction form full-link closed loop path to curb quality drift, and through adjusting acquisition strategy, give consideration to monitoring accuracy and resource allocation efficiency.

Description

Technical Field

[0001] This invention relates to the field of wood-plastic door manufacturing technology, and more specifically, to an online monitoring system for the wood-plastic door production process. Background Technology

[0002] Wood-plastic composite doors are a type of composite door product made primarily from wood fiber materials and thermoplastic plastics. These materials are mixed, granulated, and then melt-extruded through an extruder. During the extrusion molding process, the molten material is shaped into a continuous profile by a mold. This profile is then subjected to processes such as traction, cooling, and cutting to produce a semi-finished door panel. This process involves the interaction of multiple physical quantities, including temperature field distribution, melt pressure, traction speed, and cooling rate. Fluctuations in any of these processes can directly affect the dimensional accuracy and surface quality of the final product.

[0003] With the widespread application of intelligent manufacturing technology in the wood-plastic composite door and window industry, some production lines have begun to introduce machine vision inspection devices and various sensors for production process monitoring. Machine vision systems are typically installed at the extruder exit or after the cooling and shaping section to collect image data of the door panel surface and identify common surface defects such as scratches, dents, and bubbles using image processing algorithms. Thermal imaging temperature measurement devices are used to monitor the surface temperature distribution of the profiles to determine whether the cooling and shaping is uniform. Vibration sensors are used to sense the operating status of the extruder screw and traction device. These monitoring methods, to a certain extent, achieve real-time data acquisition and visualization of the production process.

[0004] However, the raw material formula of wood-plastic composite doors contains a high proportion of wood flour fiber, and the surface of the door panel after extrusion molding has a natural wood grain structure. This texture appears as background noise in the visual image that overlaps with the defect signal in the spatial frequency domain. When the production line operates at a high speed, the imaging size of the defect in the image decreases and the residence time shortens, further increasing the difficulty of distinguishing between the defect signal and the texture noise. Under these conditions, image detection methods based on fixed judgment thresholds struggle to stably distinguish weak defect signals from background texture noise in areas with dense texture, easily leading to missed detections or false alarms, and the defect detection accuracy of existing systems will significantly decrease. Therefore, this invention proposes an online monitoring system for the wood-plastic composite door production process to solve the above problems. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: An online monitoring system for the production process of wood-plastic composite doors includes: The multimodal sensing module is used to synchronously acquire visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signals during the production process of wood-plastic doors, and perform time-series feature fusion processing to output fused feature data. The threshold dynamic adjustment module is used to comprehensively adjust the defect judgment threshold based on the ratio of signal components to noise components of various sensor signals in the fused feature data and the judgment benchmark value of the defect judgment threshold, and output the defect identification result. The composite control execution module is used to generate adjustment instructions based on the defect identification results and output the adjustment instructions to the actuators of the production equipment; The feedback correction module is used to acquire the actual operating data after the actuator executes the adjustment command, and correct the judgment benchmark value of the defect judgment threshold based on the deviation between the actual operating data and the target data, and output the corrected judgment benchmark value of the defect judgment threshold to the threshold dynamic adjustment module. The process monitoring module is used to generate an early warning signal based on the fused feature data corresponding to the judgment benchmark value of the corrected defect judgment threshold, and to feed the early warning signal back to the multimodal perception module. The multimodal sensing module is also used to adjust the acquisition strategy of visual image data, thermal imaging temperature distribution data and equipment vibration spectrum signals according to the early warning signal.

[0006] In a preferred embodiment, the specific method for performing temporal feature fusion processing is as follows: The visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signal are aligned according to the acquisition time to form three sets of sensor data at the same time. Calculate the cumulative vibration energy value of the equipment vibration spectrum signal within a preset frequency range, and use the cumulative vibration energy value as the time-series characteristic component of the equipment vibration spectrum signal; Calculate the absolute value of the rate of temperature change between adjacent acquisition times in the thermal imaging temperature distribution data, and use the absolute value of the rate of temperature change as the temporal feature component of the thermal imaging temperature distribution data. Extract the sum of squares of pixel brightness differences between consecutive frames in visual image data, and use the sum of squares of pixel brightness differences as the temporal feature component of visual image data. The temporal characteristic components of the equipment vibration spectrum signal, the temporal characteristic components of the thermal imaging temperature distribution data, and the temporal characteristic components of the visual image data are spliced ​​together according to the acquisition time to generate fused feature data.

[0007] In a preferred embodiment, the specific method for comprehensively adjusting the defect judgment threshold is as follows: Calculate the first ratio of the first signal power value to the first noise power value of the visual image data in the fused feature data, the second ratio of the second signal power value to the second noise power value of the thermal imaging temperature distribution data, and the third ratio of the third signal power value to the third noise power value of the equipment vibration spectrum signal, respectively. The difference between the first ratio and the preset first standard signal-to-noise ratio value is used as the first deviation coefficient, the difference between the second ratio and the preset second standard signal-to-noise ratio value is used as the second deviation coefficient, and the difference between the third ratio and the preset third standard signal-to-noise ratio value is used as the third deviation coefficient. The first deviation coefficient, the second deviation coefficient, and the third deviation coefficient are summed to obtain the comprehensive deviation coefficient. The adjusted defect judgment threshold is obtained by summing the judgment benchmark value of the defect judgment threshold with the comprehensive deviation coefficient.

[0008] In a preferred embodiment, the threshold dynamic adjustment module is further used to compare the adjusted defect judgment threshold with the first normalized value of the pixel gray value of the visual image data, the second normalized value of the temperature gradient value of the thermal imaging temperature distribution data, and the third normalized value of the amplitude of the equipment vibration spectrum signal, respectively. The first normalized value is the ratio of the pixel gray value to the preset first maximum reference value, the second normalized value is the ratio of the temperature gradient value to the preset second maximum reference value, and the third normalized value is the ratio of the amplitude to the preset third maximum reference value. When any one of the first normalized value, the second normalized value, and the third normalized value is greater than the adjusted defect judgment threshold, it is determined that there is a defect at the current acquisition time.

[0009] The defect identification results output by the threshold dynamic adjustment module include defect type identifier and defect location coordinates; The defect type identifier is determined by the geometric shape features of the set of pixels in the visual image data whose first normalized value is greater than the defect judgment threshold, and the defect location coordinates are defined by the bounding rectangle of the set of pixels.

[0010] In a preferred embodiment, the specific method for generating the adjustment instruction is as follows: When the defect identification result includes a defect type identifier and defect location coordinates, the preset adjustment parameter mapping table is matched according to the defect type identifier. The extrusion speed adjustment amount and traction rate adjustment amount corresponding to the defect type identifier are queried from the adjustment parameter mapping table. The direction of action of the extrusion speed adjustment amount is determined according to the relative position of the defect location coordinates in the visual image data. The extrusion speed adjustment amount and the direction of action are combined to generate an extrusion speed adjustment command, and the traction rate adjustment amount is used to generate a traction rate adjustment command.

[0011] In a preferred embodiment, the specific method for correcting the defect determination threshold is as follows: Obtain the actual operating data corresponding to a preset number of production cycle times, and form an actual operating data sequence; Obtain the target data corresponding to each actual operational data, and form a target data sequence; Calculate the difference between each data point in the actual running data sequence and the corresponding data point in the target data sequence to obtain the deviation value sequence; Calculate the arithmetic mean of the deviation numerical sequence to obtain the average deviation value; The corrected defect judgment threshold is obtained by adding the absolute value of the judgment benchmark value of the defect judgment threshold to the absolute value of the average deviation value.

[0012] In a preferred embodiment, the warning signal includes a warning level identifier and a warning type identifier, and the specific method for generating the warning signal is as follows: Obtain the fused feature data of a preset number of production cycles corresponding to the judgment benchmark value of the corrected defect judgment threshold; Calculate the first ratio sequence of signal components to noise components of visual image data, the second ratio sequence of signal components to noise components of thermal imaging temperature distribution data, and the third ratio sequence of signal components to noise components of equipment vibration spectrum signal in each fused feature data. The differences between the maximum and minimum values ​​of the first ratio sequence, the second ratio sequence, and the third ratio sequence are calculated respectively to obtain the first fluctuation amplitude, the second fluctuation amplitude, and the third fluctuation amplitude. The first fluctuation amplitude is compared with the first preset fluctuation threshold, the second fluctuation amplitude is compared with the second preset fluctuation threshold, and the third fluctuation amplitude is compared with the third preset fluctuation threshold. The sensor data types that exceed their respective preset fluctuation thresholds are recorded, and the first number of sensor data types that exceed their respective preset fluctuation thresholds is counted. When the first quantity is greater than zero, the excess amount of the first fluctuation amplitude, the second fluctuation amplitude and the third fluctuation amplitude exceeding their respective preset fluctuation thresholds are summed to obtain the total excess amount.

[0013] In a preferred embodiment, the warning level identifier is determined based on a first quantity: When the first quantity is one or two, the warning level is marked as Level 1; when the first quantity is three, the warning level is marked as Level 2; when the first quantity is greater than or equal to three and the total number of excess quantities exceeds the preset excess quantity threshold, the warning level is marked as Level 3. The warning type identifier is determined based on the sensor data type that exceeds its corresponding preset fluctuation threshold: When only visual image data exceeds the first preset fluctuation threshold, the warning type is identified as visual data anomaly. When only thermal imaging temperature distribution data exceeds the second preset fluctuation threshold, the warning type is identified as temperature data anomaly. When only equipment vibration spectrum signal exceeds the third preset fluctuation threshold, the warning type is identified as vibration data anomaly. When multiple sensor data exceed their respective preset fluctuation thresholds simultaneously, the warning type is identified as combined anomaly. Finally, the warning level identifier and warning type identifier in the warning signal are analyzed, and the sampling frequency is adjusted accordingly according to the preset sampling adjustment strategy.

[0014] The technical effects and advantages of this invention are as follows: This invention uses a multimodal sensing module to simultaneously acquire visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signals, and performs time-series feature fusion processing to output fused feature data. This allows the threshold dynamic adjustment module to comprehensively adjust the defect judgment threshold based on the ratio of signal components to noise components of various sensor signals in the fused feature data (visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signals) and the judgment benchmark value of the defect judgment threshold. This enables the defect judgment threshold to be adjusted in accordance with the actual operating state of the production line, thereby maintaining the accuracy and stability of defect judgment under dynamic conditions such as the natural wood texture on the surface of the wood-plastic door causing the defect signal to be masked by texture background noise and the high-speed operation of the production line causing the defect imaging size to shrink and the residence time to shorten.

[0015] This invention generates adjustment commands based on the defect identification results output by the threshold dynamic adjustment module through a composite control execution module, and outputs the adjustment commands to the execution mechanism of the production equipment. Simultaneously, a feedback correction module obtains the actual operating data after the execution mechanism executes the adjustment commands and corrects the judgment benchmark value of the defect judgment threshold based on the deviation between the actual operating data and the target data. The corrected judgment benchmark value of the defect judgment threshold is then output to the threshold dynamic adjustment module, enabling the defect identification results to be directly converted into adjustment actions for the extrusion speed adjustment mechanism or traction rate adjustment mechanism. The deviation between the actual output result after the adjustment action and the target requirement is fed back to the threshold dynamic adjustment module as a correction amount to correct the judgment benchmark value. Thus, when the production line is running continuously and dynamic factors such as raw material batch fluctuations, equipment wear, and changes in environmental temperature and humidity cause the actual state of the production line to drift slowly, the continuous correction effect of the closed-loop path curbs the accumulation and spread of quality fluctuations.

[0016] This invention generates an early warning signal based on the fused feature data corresponding to the corrected defect judgment threshold using a process monitoring module. This early warning signal is then fed back to a multimodal perception module. The multimodal perception module adjusts its acquisition strategies for visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signals based on the early warning signal. When the process monitoring module detects that the signal-to-noise ratio fluctuation of the visual image data in the fused feature data exceeds a preset fluctuation threshold, the early warning signal is fed back to the multimodal perception module. The multimodal perception module then increases the frame rate of the visual image data acquisition to increase the sampling density of defect morphology evolution. This allows the acquisition strategy to adaptively adjust according to the actual abnormal state of the production line, avoiding data redundancy and wasted computing resources caused by maintaining an excessively high sampling frequency when the production line is operating normally. Simultaneously, when an anomaly occurs on the production line, a higher sampling density is used to capture complete information about the defect evolution process to support subsequent adjustments and tracing operations. This ensures monitoring accuracy while achieving a reasonable allocation of system resources. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of an online monitoring system for the production process of wood-plastic door according to 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Reference Figure 1 The following examples were obtained: Example 1: An online monitoring system for the production process of wood-plastic composite doors, comprising: The multimodal sensing module is used to synchronously acquire visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signals during the production process of wood-plastic doors, and perform time-series feature fusion processing to output fused feature data. The threshold dynamic adjustment module is used to comprehensively adjust the defect judgment threshold based on the ratio of signal components to noise components of various sensor signals in the fused feature data and the judgment benchmark value of the defect judgment threshold, and output the defect identification result. The composite control execution module is used to generate adjustment instructions based on the defect identification results and output the adjustment instructions to the actuators of the production equipment; The feedback correction module is used to acquire the actual operating data after the actuator executes the adjustment command, and correct the judgment benchmark value of the defect judgment threshold based on the deviation between the actual operating data and the target data, and output the corrected judgment benchmark value of the defect judgment threshold to the threshold dynamic adjustment module. The process monitoring module is used to generate an early warning signal based on the fused feature data corresponding to the judgment benchmark value of the corrected defect judgment threshold, and to feed the early warning signal back to the multimodal perception module. The multimodal sensing module is also used to adjust the acquisition strategy of visual image data, thermal imaging temperature distribution data and equipment vibration spectrum signals according to the early warning signal.

[0020] The specific method for performing temporal feature fusion processing is as follows: The multimodal sensing module aligns visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signals according to the acquisition time, forming three sets of sensing data at the same time. The alignment operation is performed based on the sampling clock tag of each sensor. When the sampling frequency of the visual sensor is 30 frames per second, the sampling frequency of the thermal imaging sensor is 15 frames per second, and the sampling frequency of the vibration sensor is 2 kHz per second, the multimodal sensing module uses the sampling time of the visual sensor as the reference time, selects the temperature distribution data with the smallest timestamp difference from the reference time from the thermal imaging data sequence, and extracts the spectrum signal segment with the smallest timestamp difference from the reference time from the vibration data sequence, and groups the three into the same data group.

[0021] The multimodal sensing module calculates the cumulative vibration energy of the equipment's vibration spectrum signal within a preset frequency range and uses this cumulative vibration energy as the temporal characteristic component of the equipment's vibration spectrum signal. The preset frequency range is set based on the fundamental frequency and harmonics of the screw rotation of the wood-plastic door extruder. The fundamental frequency of the screw rotation is obtained by dividing the screw speed of the extruder by 60. According to experimental data, the vibration energy related to surface quality during the wood-plastic door extrusion process is mainly concentrated in the range of 1 to 5 times the fundamental frequency. The preset frequency range is taken as 1 to 5 times the fundamental frequency. For example, when the screw speed is 120 revolutions per minute, the fundamental frequency is 2 Hz, and the preset frequency range is 2 Hz to 10 Hz. The cumulative vibration energy is calculated by summing the squares of the amplitudes at each frequency point within the preset frequency range.

[0022] The multimodal sensing module calculates the absolute value of the temperature change rate between adjacent acquisition times in the thermal imaging temperature distribution data, and uses the absolute value of the temperature change rate as the temporal feature component of the thermal imaging temperature distribution data. The time interval between adjacent acquisition times is determined by the sampling frequency of the thermal imaging sensor. When the sampling frequency of the thermal imaging sensor is 15 frames per second, the time interval between adjacent acquisition times is 0.067 seconds. The temperature change rate is calculated by subtracting the temperature value of the corresponding pixel in the temperature distribution data of the previous acquisition time from the temperature value of each pixel in the temperature distribution data of the later acquisition time to obtain the temperature difference value of each pixel. The square of the sum of the squares of the temperature differences of each pixel is then taken to obtain the temperature change rate value. The absolute value of the temperature change rate is the absolute value of the temperature change rate.

[0023] The multimodal perception module extracts the sum of squares of pixel brightness differences between consecutive frames in the visual image data, and uses this sum as a temporal feature component of the visual image data. The number of consecutive frames is set based on the sampling frame rate of the visual sensor and the duration of the defect's appearance. According to experimental data, the number of frames corresponding to the duration from the emergence of a defect on the surface of a wood-plastic door to its appearance in the visual image is between 3 and 8 frames. The number of consecutive frames is set to 5 frames. The sum of squares of pixel brightness differences is calculated by taking the second frame image... The first sum of squared differences is obtained by summing the squared brightness differences of the corresponding pixels in the first frame image; the second sum of squared differences is obtained by summing the squared brightness differences of the corresponding pixels in the third frame image and the second frame image; the third sum of squared differences is obtained by summing the squared brightness differences of the corresponding pixels in the fourth frame image and the third frame image; the fourth sum of squared differences is obtained by summing the squared brightness differences of the corresponding pixels in the fifth frame image and the fourth frame image; and the total sum of squared differences is obtained by summing the first, second, third, and fourth sums of squared differences.

[0024] The multimodal sensing module stitches together the temporal feature components of the equipment vibration spectrum signal, the thermal imaging temperature distribution data, and the visual image data according to the acquisition time to generate fused feature data. The stitching method is to combine the three temporal feature components at the same acquisition time in the following order: the temporal feature component of the visual image data first, the temporal feature component of the thermal imaging temperature distribution data in the middle, and the temporal feature component of the equipment vibration spectrum signal last, into a one-dimensional vector. The dimension value of the one-dimensional vector is the sum of the dimension values ​​of the three temporal feature components. For example, when the temporal feature component of the visual image data has 1 value, the temporal feature component of the thermal imaging temperature distribution data has 1 value, and the temporal feature component of the equipment vibration spectrum signal has 1 value, the dimension value of the one-dimensional vector is 3. In this way, the temporal variation features of multi-source heterogeneous data are unified in the same data carrier for subsequent modules to call.

[0025] The specific method for comprehensively adjusting the defect judgment threshold is as follows: The threshold dynamic adjustment module calculates the first ratio of the first signal power value to the first noise power value of the visual image data, the second ratio of the second signal power value to the second noise power value of the thermal imaging temperature distribution data, and the third ratio of the third signal power value to the third noise power value of the equipment vibration spectrum signal in the fused feature data. The first signal power value is the arithmetic mean of the squares of the pixel grayscale value sequence of the visual image data; the first noise power value is the arithmetic mean of the squares of the pixel grayscale value sequence of the background region in the visual image data; the second signal power value is the arithmetic mean of the squares of the temperature value sequence of the thermal imaging temperature distribution data; the second noise power value is the arithmetic mean of the squares of the temperature value sequence of the background region in the thermal imaging temperature distribution data; the third signal power value is the arithmetic mean of the squares of the amplitude sequence of the equipment vibration spectrum signal; and the third noise power value is the arithmetic mean of the squares of the floor noise amplitude sequence of the equipment vibration spectrum signal. The calculation area for the first noise power value is the edge region of the visual image data that does not include the door panel outline, and the region is based on the image in the visual image data. The set of pixels whose grayscale gradient is less than a preset gradient threshold is defined. The preset gradient threshold is set to 30 grayscale levels per pixel based on experimental data. The calculation area for the second noise power value is the set of pixels whose temperature values ​​in the thermal imaging temperature distribution data are within a preset temperature range. The preset temperature range is set according to the external ambient temperature range of the extruder during normal operation in the equipment manual, and is set to 25 degrees Celsius to 35 degrees Celsius. The calculation area for the third noise power value is the amplitude sequence of the equipment vibration spectrum signal whose frequency values ​​are within a preset noise floor frequency range. The preset noise floor frequency range is set according to the frequency range of the mechanical background vibration of the extruder during normal operation in the equipment manual, and is set to 0 Hz to 0.8 times the fundamental frequency. The fundamental frequency is the value obtained by dividing the screw speed of the extruder by 60.

[0026] For example, when the arithmetic mean of the squares of the pixel grayscale value sequence of the visual image data is 2500 and the arithmetic mean of the squares of the pixel grayscale value sequence of the background area is 100, the first ratio is 25. When the arithmetic mean of the squares of the temperature value sequence of the thermal imaging temperature distribution data is 900 and the arithmetic mean of the squares of the temperature value sequence of the background area is 36, the second ratio is 25. When the arithmetic mean of the squares of the amplitude sequence of the device vibration spectrum signal is 1600 and the arithmetic mean of the squares of the noise floor amplitude sequence is 64, the third ratio is 25.

[0027] The threshold dynamic adjustment module uses the difference between the first ratio and the preset first standard signal-to-noise ratio value as the first deviation coefficient, the difference between the second ratio and the preset second standard signal-to-noise ratio value as the second deviation coefficient, and the difference between the third ratio and the preset third standard signal-to-noise ratio value as the third deviation coefficient. The first standard signal-to-noise ratio (SNR) value is set based on the statistical results of the baseline ratio of the signal power value to the noise power value of the visual image data when the surface of the wood-plastic door is defect-free. The value is 20 based on experimental data. The second standard SNR value is set based on the statistical results of the baseline ratio of the signal power value to the noise power value of the thermal imaging temperature distribution data when the interior of the wood-plastic door is defect-free. The value is 20 based on experimental data. The third standard SNR value is set based on the statistical results of the baseline ratio of the signal power value to the noise power value of the vibration spectrum signal of the wood-plastic door extrusion equipment during normal operation. The value is 20 based on the equipment manual and on-site test data. All three standard SNR values ​​are set to the same value of 20 because the baseline SNR levels of the three types of sensor data under their respective normal operating conditions have been experimentally determined to be stable around 20. Using the same value makes it easier to uniformly measure the contribution weight of each deviation coefficient in subsequent summation calculations. For example, when the first ratio is 25 and the first standard signal-to-noise ratio is 20, the first deviation coefficient is +5. A positive value indicates that the actual signal-to-noise ratio of the visual image data is better than the baseline level. When the second ratio is 15 and the second standard signal-to-noise ratio is 20, the second deviation coefficient is -5. A negative value indicates that the actual signal-to-noise ratio of the thermal imaging temperature distribution data is worse than the baseline level. When the third ratio is 25 and the third standard signal-to-noise ratio is 20, the third deviation coefficient is +5.

[0028] The threshold dynamic adjustment module sums the first, second, and third deviation coefficients to obtain a comprehensive deviation coefficient. This summation is achieved by algebraically adding the values ​​of the three deviation coefficients. For example, when the first deviation coefficient is +5, the second is -5, and the third is +5, the algebraic sum of the three deviation coefficients is +5, and the comprehensive deviation coefficient is also +5. A positive comprehensive deviation coefficient indicates that the overall signal-to-noise ratio of the three types of sensor data deviates positively from their respective baseline levels. Physically, this means the current production line operating state is better than the baseline state, therefore the defect judgment threshold should be increased accordingly to reduce the false alarm rate. When the comprehensive deviation coefficient is negative, it physically means the current production line operating state is worse than the baseline state, therefore the defect judgment threshold should be decreased accordingly to increase detection sensitivity.

[0029] The threshold dynamic adjustment module sums the defect judgment threshold's judgment benchmark value with the comprehensive deviation coefficient to obtain the adjusted defect judgment threshold. The judgment benchmark value for the defect judgment threshold is the initial judgment threshold pre-stored in the threshold dynamic adjustment module. This initial judgment threshold is based on the national standard for wood-plastic doors regarding the judgment criteria for qualified products in terms of appearance quality, and is set in conjunction with the statistical results of the distinguishability between qualified and defective products in the fused feature data from experimental data. Based on the experimental data, it is set to 0.5, with a dimension of one, and its corresponding comparison object is the normalized values ​​of various sensor features. For example, when the baseline value for defect determination is 0.5 and the comprehensive deviation coefficient is +0.05, the adjusted defect determination threshold is 0.55. Since the comprehensive deviation coefficient is positive, it indicates that the production line condition is better than the baseline. Therefore, the defect determination threshold is increased to 0.55, so that only more significant feature deviations are determined as defects. When the comprehensive deviation coefficient is -0.05, the adjusted defect determination threshold is 0.45. Since the comprehensive deviation coefficient is negative, it indicates that the production line condition is worse than the baseline. Therefore, the defect determination threshold is decreased to 0.45, so that even slight feature deviations are determined as defects. This achieves adaptive floating adjustment of the defect determination threshold according to the comprehensive deviation of the signal-to-noise ratio of the three types of sensor data.

[0030] The threshold dynamic adjustment module compares the adjusted defect judgment threshold with the first normalized value of the pixel gray value of the visual image data, the second normalized value of the temperature gradient value of the thermal imaging temperature distribution data, and the third normalized value of the amplitude of the equipment vibration spectrum signal. The first normalized value is the ratio of the pixel gray value to the preset first maximum reference value, the second normalized value is the ratio of the temperature gradient value to the preset second maximum reference value, and the third normalized value is the ratio of the amplitude to the preset third maximum reference value. Pixel grayscale values ​​are the brightness values ​​of each pixel in the visual image data. Their range is determined by the quantization bit depth of the visual image sensor. When the quantization bit depth is 8 bits, the pixel grayscale value ranges from 0 to 255. The preset first maximum reference value is set based on the maximum quantization value of the visual image sensor and is 255. For example, when the pixel grayscale value is 153, the first normalized value is 153 divided by 255, resulting in 0.6. Temperature gradient values ​​are the ratio of the temperature difference between adjacent pixels to the spatial distance between adjacent pixels in the thermal imaging temperature distribution data. The unit of temperature gradient value is degrees Celsius per millimeter. The preset second maximum reference value is set based on the maximum possible gradient change range of the surface temperature distribution during the wood-plastic composite door extrusion process, according to experimental data and the extruder outlet information in the equipment manual. The measured results of the surface temperature field of the profile show that the maximum possible gradient variation range is 0 to 50 degrees Celsius per millimeter. The preset second maximum reference value is 50. For example, when the temperature gradient is 30 degrees Celsius per millimeter, the second normalized value is 30 divided by 50, which gives 0.6. The amplitude is the magnitude of the amplitude corresponding to each frequency component in the vibration spectrum signal of the equipment, and the unit of amplitude is millimeters per square second. The preset third maximum reference value is set according to the maximum allowable value of the vibration amplitude of each component of the extruder during normal operation in the equipment manual. According to the equipment manual and field test data, the maximum allowable amplitude range is 0 to 5 millimeters per square second. The preset third maximum reference value is 5. For example, when the amplitude is 3 millimeters per square second, the third normalized value is 3 divided by 5, which gives 0.6.

[0031] When any one of the first, second, and third normalized values ​​is greater than the adjusted defect judgment threshold, the threshold dynamic adjustment module determines that a defect exists at the current acquisition time. The "greater than" operator is a numerical comparison operator. The adjusted defect judgment threshold is obtained after adjusting the comprehensive deviation coefficient mentioned above, and its typical value is 0.5. The comparison operation is to compare the three normalized values ​​with the adjusted defect judgment threshold one by one. For example, when the adjusted defect threshold is 0.55, the first normalized value is 0.6, the second normalized value is 0.3, and the third normalized value is 0.2, the threshold dynamic adjustment module determines that a defect exists at the current acquisition time because the first normalized value of 0.6 is greater than the defect threshold of 0.55. The defect is caused by abnormal pixel grayscale in the visual image data. When the first normalized value is 0.4, the second normalized value is 0.6, and the third normalized value is 0.2, the threshold dynamic adjustment module determines that a defect exists at the current acquisition time because the second normalized value of 0.6 is greater than the defect threshold of 0.55. The defect is caused by an abnormal temperature gradient in the thermal imaging temperature distribution data. When the first normalized value is 0.4, the second normalized value is 0.3, and the third normalized value is 0.6, the threshold dynamic adjustment module determines that there is a defect at the current acquisition time because the third normalized value of 0.6 is greater than the defect judgment threshold of 0.55. The defect is caused by an abnormal amplitude of the equipment vibration spectrum signal. When the first normalized value is 0.4, the second normalized value is 0.3, and the third normalized value is 0.2, the threshold dynamic adjustment module determines that there is no defect at the current acquisition time because all three are less than or equal to the defect judgment threshold of 0.55.

[0032] The defect identification results output by the threshold dynamic adjustment module include a defect type identifier and defect location coordinates. The defect type identifier is used to characterize the category attribute of the defect at the current acquisition time, and the defect location coordinates are used to indicate the spatial distribution of the defect on the wood-plastic door panel surface.

[0033] The defect type identifier is determined by the geometric shape features of the set of pixels in the visual image data whose first normalized value is greater than the defect judgment threshold. The pixel set is obtained as follows: the threshold dynamic adjustment module iterates through each pixel in the visual image data, compares the first normalized value of each pixel with the defect judgment threshold, and adds all pixels whose first normalized value is greater than the threshold to the pixel set. The geometric shape features of the pixel set include the extension direction feature and the compactness feature. The extension direction feature is the principal axis angle of the pixel set, which is determined by the direction of the eigenvector corresponding to the largest eigenvalue of the covariance matrix formed by the x and y coordinates of each pixel in the pixel set. The compactness feature is the ratio of the area of ​​the pixel set to the area of ​​its bounding rectangle. When the angle between the extension direction and the horizontal direction is less than 30 degrees and the compactness characteristic is less than 0.3, the defect type is identified as a scratch. This is based on experimental statistical results. According to experimental data, scratches on the surface of wood-plastic composite doors appear as elongated strip-shaped areas extending along the extrusion direction in visual images. The angle between the principal axis and the horizontal direction is usually less than 30 degrees, and the ratio of its area to the area of ​​its circumscribed rectangle is usually less than 0.3. When the angle between the extension direction and the horizontal direction is greater than or equal to 30 degrees and the compactness characteristic is less than 0.3, the defect type is identified as a pit. This is also based on experimental statistical results. According to experimental data, pits on the surface of wood-plastic composite doors appear as approximately circular or elliptical recessed areas in visual images. The angle between the principal axis and the horizontal direction is usually greater than or equal to 30 degrees, and the ratio of its area to the area of ​​its circumscribed rectangle is usually less than 0.3. When the compactness characteristic is greater than or equal to 0.3 and less than 0.8, the defect type is identified as a bubble. This is based on experimental statistical results. According to experimental data, bubble defects on and near the surface of wood-plastic doors appear as approximately circular raised areas in visual images, with the ratio of their area to the area of ​​their circumscribed rectangle typically between 0.3 and 0.8. When the compactness characteristic is greater than or equal to 0.8, the defect type is identified as other types. This is also based on experimental statistical results. According to experimental data, the ratio of the area of ​​defects other than scratches, dents, and bubbles to the area of ​​their circumscribed rectangle is typically greater than or equal to 0.8.

[0034] The defect location coordinates are defined by the bounding rectangle of the pixel set. This bounding rectangle is the smallest rectangle that completely contains all pixels in the pixel set, and its four sides are parallel to the horizontal and vertical directions of the visual image data, respectively. The defect location coordinates include the x-coordinate and y-coordinate of the top-left and bottom-right pixels of the bounding rectangle. The x-coordinate and y-coordinate of the top-left pixel are the minimum x-coordinates and y-coordinates of all pixels in the pixel set, respectively. The x-coordinate and y-coordinate of the bottom-right pixel are the maximum x-coordinates and y-coordinates of all pixels in the pixel set, respectively. For example, when the minimum x-coordinate value of all pixels in the pixel set is 100, the minimum y-coordinate value is 150, the maximum x-coordinate value is 200, and the maximum y-coordinate value is 250, then the x-coordinate value of the top-left pixel in the defect location coordinates is 100, the y-coordinate value is 150, and the x-coordinate value of the bottom-right pixel is 200, the y-coordinate value is 250. The spatial range of the circumscribed rectangle in the visual image data is defined by these four coordinate values. The threshold dynamic adjustment module combines the defect type identifier and the defect location coordinates into a defect recognition result and outputs it to the composite control execution module.

[0035] The specific method for generating adjustment instructions is as follows: When the defect identification result includes a defect type identifier and defect location coordinates, the composite control execution module matches a preset adjustment parameter mapping table based on the defect type identifier. It then queries the mapping table for the corresponding extrusion speed and traction rate adjustments. This preset mapping table is a pre-established correspondence table based on experimental data and process experience, recording the mapping values ​​between each defect type identifier and the extrusion speed and traction rate adjustments. The unit for the extrusion speed adjustment is millimeters per minute (mm / min), and the unit for the traction rate adjustment is meters per minute (m / min). According to experimental data, when the defect type identifier is a scratch, the corresponding extrusion speed adjustment is reduced by 5 mm / min, and the corresponding traction rate adjustment is reduced by 0.2 m / min. The reason for this reduction is that there is a positive correlation between the occurrence of scratch defects and excessive shear force on the profile surface caused by excessively high extrusion speeds, as observed in the experimental data. Therefore, reducing the extrusion speed and traction rate can reduce the shear force.

[0036] When the defect type is identified as pitting, the corresponding extrusion speed adjustment is increased by 8 mm / min, and the corresponding traction rate adjustment is increased by 0.3 m / min. The basis for this increase is the negative correlation between the occurrence of pitting defects and insufficient melt filling on the profile surface due to excessively slow extrusion speed, as observed in experimental data. Therefore, increasing the extrusion speed and traction rate can enhance the melt filling effect. When the defect type is identified as bubble, the corresponding extrusion speed adjustment is decreased by 3 mm / min, and the corresponding traction rate adjustment is decreased by 0.1 m / min. The basis for this decrease is the positive correlation between the occurrence of bubble defects and insufficient gas escape from the melt due to excessively fast extrusion speed, as observed in experimental data. Therefore, reducing the extrusion speed and traction rate can prolong the gas escape time. When the defect type is identified as other types, the corresponding extrusion speed adjustment is reduced by 2 mm / min, and the corresponding traction rate adjustment is reduced by 0.1 m / min. The basis for the reduction operation is the conservative adjustment strategy recommended in the equipment manual for undefined defect types, that is, to give priority to using a small reduction in speed to reduce the interference of uncertain factors. The values ​​in the adjustment parameter mapping table are all provided based on process experimental data and equipment manual. For example, the reference range of the extrusion speed adjustment is -10 mm / min to +10 mm / min, and the reference range of the traction rate adjustment is -0.5 m / min to +0.5 m / min.

[0037] The composite control execution module determines the direction of the extrusion speed adjustment based on the relative position of the defect location coordinates in the visual image data. The relative position is determined by comparing the lateral coordinate of the center point of the circumscribed rectangle corresponding to the defect location coordinates with the lateral coordinate of the centerline of the visual image data. The lateral coordinate of the centerline is the pixel width of the visual image data divided by 2. When the lateral coordinate of the center point of the circumscribed rectangle is less than the lateral coordinate of the centerline, the relative position is in the left region of the visual image data, and the direction of the extrusion speed adjustment is to the left, physically meaning that adjusting the rotation direction of the extruder screw causes the profile to shift to the left. When the lateral coordinate of the center point of the circumscribed rectangle is greater than the lateral coordinate of the centerline, the relative position is in the right region of the visual image data, and the direction of the extrusion speed adjustment is to the right, physically meaning that adjusting the rotation direction of the extruder screw causes the profile to shift to the right.

[0038] When the lateral coordinate of the center point of the circumscribed rectangle equals the lateral coordinate of the center line of the visual image data, the relative position is the central area of ​​the visual image data. In this case, the extrusion speed adjustment is directionless, meaning no lateral offset adjustment is performed. The direction is determined by the defect location coordinates, which indicate the spatial distribution of the defect on the door panel. Adjusting the direction towards the defect location ensures precise application of the adjustment to the defect area. For example, when the pixel width of the visual image data is 800 pixels, the lateral coordinate of the center line is 400 pixels, and the lateral coordinate of the center point of the circumscribed rectangle corresponding to the defect location is 250 pixels, since 250 is less than 400, the relative position is the left side of the visual image data, and the extrusion speed adjustment is directed to the left. When the lateral coordinate of the center point of the circumscribed rectangle corresponding to the defect location is 550 pixels, since 550 is greater than 400, the relative position is the right side of the visual image data, and the extrusion speed adjustment is directed to the right.

[0039] The composite control execution module combines the extrusion speed adjustment amount and the direction of action to generate an extrusion speed adjustment command, and generates a traction rate adjustment command from the traction rate adjustment amount. The combination method is to concatenate the value of the extrusion speed adjustment amount with the identification code of the direction of action to form a complete command. The identification code of the direction of action has a number 0 corresponding to left, a number 1 corresponding to right, and no number 2 corresponding to any direction. The format of the extrusion speed adjustment command is the identification code of the direction of action plus the value of the extrusion speed adjustment amount.

[0040] For example, when the extrusion speed adjustment is to decrease by 5 mm / min and the direction of action is to the left, the extrusion speed adjustment command is "0, -5", where 0 represents to the left and -5 represents a decrease of 5 mm / min. When the extrusion speed adjustment is to increase by 8 mm / min and the direction of action is to the right, the extrusion speed adjustment command is "1, +8", where 1 represents to the right and +8 represents an increase of 8 mm / min. The format of the traction rate adjustment command is the value of the traction rate adjustment itself. For example, when the traction rate adjustment is to decrease by 0.2 m / min, the traction rate adjustment command is "-0.2", and when the traction rate adjustment is to increase by 0.3 m / min, the traction rate adjustment command is "+0.3". The composite control execution module outputs the extrusion speed adjustment command and the traction rate adjustment command to the actuator of the production equipment. The actuator adjusts the speed and direction of rotation of the extruder screw according to the extrusion speed adjustment command and adjusts the speed of the traction roller according to the traction rate adjustment command, thereby achieving precise control of the process parameters in the defect area.

[0041] The specific method for determining the criterion value of the defect judgment threshold is as follows: The feedback correction module acquires the actual operating data corresponding to each of the preset number of production cycles, forming an actual operating data sequence. The production cycle is the time required for the wood-plastic door production line to complete the production of one wood-plastic door. The preset number is set based on the accuracy requirements of sample mean estimation in statistics and the actual operating cycle of the production line. According to experimental data, when the sample size is 5 to 10, its arithmetic mean has sufficient statistical stability. The preset number is set to 8.

[0042] The actual operating data consists of the actual thickness and surface flatness values ​​of the wood-plastic door panels produced by the production equipment after the equipment executes the adjustment command. The actual thickness value is measured by a laser thickness gauge located at the extruder outlet, and the actual surface flatness value is measured by a laser profilometer located at the extruder outlet. Taking a preset quantity of 8 as an example, the feedback correction module continuously acquires the actual thickness and surface flatness values ​​corresponding to each of the 8 production cycles. The 8 actual thickness and 8 actual surface flatness values ​​are arranged in the order of the production cycles to obtain the actual operating data sequence. The format of the actual operating data sequence is [actual thickness value of the 1st production cycle, actual thickness value of the 2nd production cycle, ..., actual thickness value of the 8th production cycle] and the corresponding [actual surface flatness value of the 1st production cycle, actual surface flatness value of the 2nd production cycle, ..., actual surface flatness value of the 8th production cycle].

[0043] The feedback correction module acquires the target data corresponding to each actual operating data and forms a target data sequence. The target data are the standard thickness value and standard surface flatness value of the wood-plastic door product. The standard thickness value and standard surface flatness value are preset according to the design drawings of the wood-plastic door product and the requirements for appearance quality in the national standards. The standard thickness value is 35 mm and the standard surface flatness value is 0.5 mm.

[0044] Taking a preset quantity of 8 as an example, the feedback correction module arranges the target data corresponding to each of the 8 production cycles in the same order as the actual running data sequence to obtain the target data sequence. The format of the target data sequence is [target thickness value of the 1st production cycle, target thickness value of the 2nd production cycle, ..., target thickness value of the 8th production cycle] and the corresponding [target surface flatness value of the 1st production cycle, target surface flatness value of the 2nd production cycle, ..., target surface flatness value of the 8th production cycle]. Since the target data of each production cycle are the same standard thickness value of 35 mm and standard surface flatness value of 0.5 mm, the target data sequence is: a data sequence composed of 8 35 mm and 8 0.5 mm values.

[0045] The feedback correction module calculates the difference between each data point in the actual operating data sequence and the corresponding data point in the target data sequence, obtaining a deviation value sequence. The deviation value is calculated by subtracting the i-th data point in the target data sequence from the i-th data point in the actual operating data sequence. For example, if the actual thickness values ​​in the actual operating data sequence are 35.2 mm for the first value, 34.8 mm for the second, 35.5 mm for the third, 34.9 mm for the fourth, 35.1 mm for the fifth, 34.7 mm for the sixth, 35.3 mm for the seventh, and 34.6 mm for the eighth, then the target thickness values ​​in the target data sequence are all 35 mm.

[0046] The deviation value sequence is: the first one +0.2 mm, the second one -0.2 mm, the third one +0.5 mm, the fourth one -0.1 mm, the fifth one +0.1 mm, the sixth one -0.3 mm, the seventh one +0.3 mm, and the eighth one -0.4 mm. For the actual surface smoothness value, when the actual surface smoothness value in the actual operating data sequence is the first one 0.6 mm, the second one 0.4 mm, the third one 0.7 mm, and the fourth one... When the values ​​are 0.5 mm for the first, 0.5 mm for the second, 0.3 mm for the third, 0.6 mm for the fourth, and 0.4 mm for the fifth, the surface flatness value of each target in the target data sequence is 0.5 mm. Then the deviation value sequence is: +0.1 mm for the first, -0.1 mm for the second, +0.2 mm for the third, 0 mm for the fourth, 0 mm for the fifth, -0.2 mm for the sixth, +0.1 mm for the seventh, and -0.1 mm for the eighth.

[0047] The feedback correction module calculates the arithmetic mean of the deviation value sequence to obtain the average deviation value. The calculation method is the algebraic sum of all deviation values ​​in the deviation value sequence divided by a preset quantity. Continuing with the example above, for the deviation value sequence of the actual thickness value, its algebraic sum is +0.2 + -0.2 + +0.5 + -0.1 + +0.1 + -0.3 + +0.3 + -0.4, which equals +0.1 mm. Dividing this algebraic sum of +0.1 mm by the preset quantity 8 yields the average deviation value of +0.0125 mm. For the deviation value sequence of the actual surface flatness value, its algebraic sum is +0.1 + -0.1 + +0.2 + 0 + 0 + -0.2 + ... Adding +0.1 to -0.1 results in an algebraic sum of 0 mm. Dividing this sum by the preset quantity of 8 yields an average deviation of 0 mm. The sign of the average deviation reflects the overall deviation of the actual operating data from the target data over the preset number of production cycles. A positive value indicates that the actual operating data is generally greater than the target data, a negative value indicates that the actual operating data is generally less than the target data, and a value of zero indicates that the actual operating data is generally consistent with the target data. The magnitude of the average deviation reflects the degree of deviation.

[0048] The feedback correction module adds the judgment benchmark value of the defect judgment threshold to the absolute value of the average deviation value to obtain the judgment benchmark value of the corrected defect judgment threshold. The absolute value of the average deviation value is the absolute value of the average deviation value, which is 0.0125 mm. Its physical meaning is the average deviation of the actual thickness value from the target thickness value in a preset number of production cycles. Taking a previous value of 0.5 for the defect judgment threshold as an example, the feedback correction module adds the judgment threshold value of 0.5 to the absolute value of the average deviation value of 0.0125 to obtain a corrected judgment threshold value of 0.5125. The basis for the addition operation is that when the absolute value of the average deviation value is greater than zero, it indicates that the actual operating data deviates from the target data as a whole. The deviation reflects that the current judgment threshold value fails to accurately match the actual operating state of the production line. The deviation needs to be added to the judgment threshold value as a correction amount so that the corrected judgment threshold value is closer to the actual state of the production line. When the absolute value of the average deviation value is zero, it indicates that the actual operating data is consistent with the target data as a whole. In this case, there is no need to correct the judgment threshold value. The feedback correction module outputs the corrected judgment threshold value to the threshold dynamic adjustment module as the basis for the judgment threshold value when the threshold dynamic adjustment module performs comprehensive adjustment next time.

[0049] The warning signal includes a warning level identifier and a warning type identifier. The specific method for generating a warning signal is as follows: The process monitoring module acquires fusion feature data for a preset number of production cycles corresponding to the corrected defect judgment threshold's judgment benchmark value. This preset number is the same as the preset number in the feedback correction module, both being 8. After the corrected defect judgment threshold's judgment benchmark value is output to the threshold dynamic adjustment module, the multimodal perception module collects and outputs fusion feature data corresponding to each of the subsequent 8 production cycles. Each production cycle's fusion feature data contains three values: the temporal feature component of the visual image data, the temporal feature component of the thermal imaging temperature distribution data, and the temporal feature component of the equipment vibration spectrum signal.

[0050] The process monitoring module calculates the first ratio sequence of signal components to noise components in the visual image data, the second ratio sequence of signal components to noise components in the thermal imaging temperature distribution data, and the third ratio sequence of signal components to noise components in the equipment vibration spectrum signal, respectively. The first ratio sequence is a data sequence composed of eight ratios obtained by dividing the temporal feature components of the visual image data corresponding to each of the eight production cycles by the noise component of the visual image data, arranged in the order of the production cycles. Taking the signal component values ​​of the visual image data corresponding to eight production cycles as an example (0.25, 0.28, 0.22, 0.30, 0.26, 0.24, 0.29, and 0.27 respectively), the noise component of the visual image data is taken as the square root of the arithmetic mean of the squares of the grayscale value sequence of the background region pixels, as described above. The noise component is set to 0.05. Therefore, dividing each signal component by 0.05 yields the first ratio sequence of 5.0, 5.6, 4.4, 6.0, 5.2, 4.8, 5.8, and 5.4. The second ratio sequence is a data sequence composed of eight ratios obtained by dividing each signal component by the noise component of the thermal imaging temperature distribution data corresponding to the eight production cycles, arranged in the order of the production cycles.

[0051] Taking the signal component values ​​of the thermal imaging temperature distribution data corresponding to 8 production cycles as an example, with values ​​of 0.30, 0.32, 0.28, 0.35, 0.31, 0.29, 0.33, and 0.30 respectively, the noise component of the thermal imaging temperature distribution data is taken as the square root of the arithmetic mean of the squares of the temperature value sequence in the background area, as described above. The noise component value is 0.06. Therefore, the second ratio sequence obtained by dividing each signal component by 0.06 is 5.0, 5.3, 4.7, 5.8, 5.2, 4.8, 5.5, and 5.0. The third ratio sequence is a data sequence composed of the time-series characteristic components of the equipment vibration spectrum signal corresponding to each of the 8 production cycles, which are sequentially used as signal components. The 8 ratios obtained by dividing each signal component by the noise component of the equipment vibration spectrum signal are arranged in the order of the production cycles. Taking the signal component values ​​of the equipment vibration spectrum signal corresponding to 8 production cycles as an example, the process monitoring module takes the signal component values ​​of 0.20, 0.22, 0.18, 0.25, 0.21, 0.19, 0.23 and 0.20 respectively as an example. The noise component of the equipment vibration spectrum signal is taken as the square root of the arithmetic mean of the squares of the noise floor amplitude value sequence in the previous method. The noise component value is 0.04. Then, after dividing each signal component by 0.04, the third ratio sequence is 5.0, 5.5, 4.5, 6.2, 5.2, 4.8, 5.8 and 5.0.

[0052] The process monitoring module calculates the difference between the maximum and minimum values ​​of the first, second, and third ratio sequences, respectively, to obtain the first, second, and third fluctuation amplitudes. The first fluctuation amplitude is the difference between the maximum and minimum values ​​in the first ratio sequence. Taking the first ratio sequence 5.0, 5.6, 4.4, 6.0, 5.2, 4.8, 5.8, and 5.4 as an example, the maximum value is 6.0 and the minimum value is 4.4, so the first fluctuation amplitude is 6.0 minus 4.4, which equals 1.6. Similarly, the second fluctuation amplitude is the difference between the maximum and minimum values ​​in the second ratio sequence. Taking the second ratio sequence 5.0, 5.3, 4.7, 5.8, 5.2, 4.8, 5.5, and 5.0 as an example, the maximum value is 5.8 and the minimum value is 4.7, so the second fluctuation amplitude is 5.8 minus 4.7, which equals 1.1. The third fluctuation range is the difference between the maximum value and the minimum value in the third ratio sequence. Taking the third ratio sequence of 5.0, 5.5, 4.5, 6.2, 5.2, 4.8, 5.8 and 5.0 in the example above as an example, the maximum value is 6.2 and the minimum value is 4.5. Therefore, the third fluctuation range is 6.2 minus 4.5, which is 1.7.

[0053] The process monitoring module compares the first fluctuation amplitude with the first preset fluctuation threshold, the second fluctuation amplitude with the second preset fluctuation threshold, and the third fluctuation amplitude with the third preset fluctuation threshold. It records the sensor data types that exceed their respective preset fluctuation thresholds and counts the first number of sensor data types exceeding their respective preset fluctuation thresholds. The first preset fluctuation threshold is set based on the upper limit of the fluctuation amplitude of the signal-to-noise ratio sequence of visual image data when the surface of the wood-plastic door is defect-free, and is set to 1.2 according to the experimental data. The second preset fluctuation threshold is set based on the upper limit of the fluctuation amplitude of the signal-to-noise ratio sequence of thermal imaging temperature distribution data when the interior of the wood-plastic door is defect-free, and is set to 1.0 according to the experimental data. The third preset fluctuation threshold is set based on the upper limit of the fluctuation amplitude of the vibration spectrum signal of the equipment when the extruder is operating normally, and is set to 1.4 according to the equipment manual and on-site test data.

[0054] The three preset fluctuation thresholds are set independently because the signal-to-noise ratio (SNR) fluctuation characteristics of the three types of sensor data differ. Using different thresholds for each type more accurately reflects the normal fluctuation range of each data type. Taking the first fluctuation amplitude of 1.6, the second fluctuation amplitude of 1.1, and the third fluctuation amplitude of 1.7 in the example above as examples, after comparing the first fluctuation amplitude of 1.6 with the first preset fluctuation threshold of 1.2, since 1.6 is greater than 1.2, the SNR fluctuation of the visual image data is recorded as exceeding the threshold. After comparing the second fluctuation amplitude of 1.1 with the second preset fluctuation threshold of 1.0, since 1.1 is greater than 1.0, the SNR fluctuation of the thermal imaging temperature distribution data is recorded as exceeding the threshold. After comparing the third fluctuation amplitude of 1.7 with the third preset fluctuation threshold of 1.4, since 1.7 is greater than 1.4, the SNR fluctuation of the equipment vibration spectrum signal is recorded as exceeding the threshold. All three types of sensor data are recorded as exceeding their respective preset fluctuation thresholds; therefore, the first quantity is set to 3.

[0055] When the first quantity is greater than zero, the process monitoring module sums the excess amounts of the first fluctuation amplitude, the second fluctuation amplitude, and the third fluctuation amplitude that exceed their respective preset fluctuation thresholds, obtaining the total excess amount. The excess amount is the difference between each fluctuation amplitude and its corresponding preset fluctuation threshold. When the difference is greater than zero, the difference is the excess amount; when the difference is less than or equal to zero, the excess amount is zero. For example, subtracting the first preset fluctuation threshold 1.2 from the first fluctuation amplitude of 1.6 yields a first excess amount of 0.4; subtracting the second preset fluctuation threshold 1.0 from the second fluctuation amplitude of 1.1 yields a second excess amount of 0.1; and subtracting the third preset fluctuation threshold 1.4 from the third fluctuation amplitude of 1.7 yields a third excess amount of 0.3. Summing the first excess amount of 0.4, the second excess amount of 0.1, and the third excess amount of 0.3 yields a total excess amount of 0.8.

[0056] The total excess reflects the overall deviation of the current production line's operating status from the normal fluctuation range. When the first quantity is zero, it indicates that the signal-to-noise ratio fluctuations of the three types of sensor data have not exceeded their respective normal fluctuation ranges. At this time, the process monitoring module does not generate an early warning signal. When the first quantity is greater than zero, it indicates that at least one type of sensor data has abnormal fluctuations. At this time, the process monitoring module will use the total excess and the first quantity as the basis for determining the early warning level and will feed back the early warning signal to the multimodal sensing module.

[0057] The warning level is determined based on the first quantity: The process monitoring module determines the warning level based on the first quantity. When the first quantity is one or two, the warning level is Level 1; when the first quantity is three, the warning level is Level 2; and when the first quantity is greater than or equal to three and the total amount exceeding the preset threshold, the warning level is Level 3. The preset threshold is set based on the sum of the statistical upper limits of the excess amounts of the three types of sensor data under normal operating conditions in the experimental data. According to the experimental data, under normal operating conditions, the fluctuation amplitude of each of the three types of sensor data will not exceed the corresponding preset fluctuation threshold. Therefore, the theoretical normal value of the total excess amount is zero. Considering the random fluctuation of sensor measurement noise, the preset threshold is set to 0.5 based on the field test data. When the total excess amount is greater than 0.5, it indicates that the abnormal fluctuations of multiple types of sensor data have significantly exceeded the normal range.

[0058] For example, when the first quantity is 1 and the total excess is 0.3, the warning level is marked as Level 1, indicating that only a single type of sensor data shows slight abnormal fluctuations. When the first quantity is 2 and the total excess is 0.4, the warning level is marked as Level 1, indicating that two types of sensor data show abnormal fluctuations, but each excess is relatively small. When the first quantity is 3 and the total excess is 0.4, the warning level is marked as Level 2, indicating that all three types of sensor data show abnormal fluctuations, but the overall excess is still at a low level. When the first quantity is 3 and the total excess is 1.2, because the first quantity is greater than or equal to 3 and the total excess is 1.2, which is greater than the preset excess threshold of 0.5, the warning level is marked as Level 3, indicating that all three types of sensor data show abnormal fluctuations and the overall excess is significantly large, indicating a serious abnormality in the production line operation.

[0059] The process monitoring module determines the warning type identifier based on the sensor data types that exceed their respective preset fluctuation thresholds. When only visual image data exceeds the first preset fluctuation threshold, the warning type identifier is visual data anomaly type. Visual data anomaly type indicates that the current anomaly only comes from the signal-to-noise ratio fluctuation of visual image data. The cause may be that visible defects such as scratches, pits or bubbles on the surface of the wood-plastic door cause a significant change in image grayscale, or that a sudden change in ambient light causes a change in the overall grayscale of the image.

[0060] When only thermal imaging temperature distribution data exceeds the second preset fluctuation threshold, the warning type is identified as "Temperature Data Anomaly." This indicates that the current anomaly originates solely from signal-to-noise ratio fluctuations in the thermal imaging temperature distribution data. The cause may be a temperature drift in the extruder barrel heating section or an unstable flow rate in the die cooling system, leading to a change in the temperature field distribution on the profile surface. When only equipment vibration spectrum signals exceed the third preset fluctuation threshold, the warning type is identified as "Vibration Data Anomaly." This indicates that the current anomaly originates solely from signal-to-noise ratio fluctuations in the equipment vibration spectrum signals. The cause may be wear on the extruder screw or a localized bearing failure, altering the vibration spectrum characteristics. When multiple sensor data simultaneously exceed their respective preset fluctuation thresholds, the warning type is identified as "Combined Anomaly." This indicates that the current anomaly originates from simultaneous fluctuations in the signal-to-noise ratio of at least two types of sensor data, possibly due to the combined effect of multiple anomaly factors.

[0061] For example, when the first fluctuation amplitude of 1.6 is greater than the first preset fluctuation threshold of 1.2, the second fluctuation amplitude of 1.1 is less than or equal to the second preset fluctuation threshold of 1.0, and the third fluctuation amplitude of 1.2 is less than or equal to the third preset fluctuation threshold of 1.4, only the visual image data exceeds the threshold, and the warning type is identified as the visual data anomaly type. When the first fluctuation amplitude of 1.6 is greater than the first preset fluctuation threshold of 1.2, the second fluctuation amplitude of 1.1 is greater than the second preset fluctuation threshold of 1.0, and the third fluctuation amplitude of 1.2 is less than or equal to the third preset fluctuation threshold of 1.4, both the visual image data and the thermal imaging temperature distribution data exceed the threshold, and the warning type is identified as the combined anomaly type.

[0062] The process monitoring module combines the warning level identifier and the warning type identifier into a warning signal and feeds it back to the multimodal sensing module. The multimodal sensing module analyzes the warning level identifier and the warning type identifier in the warning signal and adjusts the sampling frequency according to a preset sampling adjustment strategy. The sampling adjustment strategy is a pre-set sampling frequency adjustment rule table based on the combined state of the warning level identifier and the warning type identifier. The sampling adjustment strategy is set according to the principle that different abnormality levels and abnormality types in the wood-plastic door production process require different data acquisition density. According to experimental data and equipment manuals, when the abnormality level is low, it is only necessary to moderately increase the acquisition density of single-type sensor data to confirm whether the abnormality persists. When the abnormality level is high, it is necessary to comprehensively increase the acquisition density of all sensor data to obtain more detailed abnormality status information.

[0063] When the warning level is identified as Level 1 and the warning type is identified as visual data anomaly, the multimodal perception module increases the sampling frequency of the visual image data from the current 30 frames per second to 60 frames per second. The adjustment is based on the premise that doubling the sampling frequency corresponding to Level 1 anomalies in the experimental data is sufficient to capture key image changes during the anomaly's evolution. Meanwhile, the sampling frequency of the thermal imaging temperature distribution data remains unchanged at 15 frames per second, and the sampling frequency of the equipment vibration spectrum signal remains unchanged at 2 kHz per second. When the warning level is identified as Level 1 and the warning type is identified as temperature data anomaly, the multimodal perception module increases the sampling frequency of the thermal imaging temperature distribution data from the current 15 frames per second to 30 frames per second, while the sampling frequency of the visual image data remains unchanged at 30 frames per second, and the sampling frequency of the equipment vibration spectrum signal remains unchanged at 2 kHz per second.

[0064] When the warning level is identified as Level 1 and the warning type is identified as vibration data anomaly, the multimodal sensing module increases the sampling frequency of the equipment vibration spectrum signal from the current 2 kHz / s to 4 kHz / s, while maintaining the sampling frequency of visual image data at 30 frames per second and the sampling frequency of thermal imaging temperature distribution data at 15 frames per second. When the warning level is identified as Level 1 and the warning type is identified as a combined anomaly, the multimodal sensing module increases the sampling frequency of visual image data from 30 frames per second to 45 frames per second, the sampling frequency of thermal imaging temperature distribution data from 15 frames per second to 22 frames per second, and the sampling frequency of the equipment vibration spectrum signal from 2 kHz / s to 3 kHz / s. The adjustment range is based on the principle that a moderate increase in the sampling frequency of various sensor data under the Level 1 combined anomaly state in the experimental data is sufficient to meet the anomaly confirmation requirements; excessively high sampling frequencies will cause data redundancy and increase the processing burden.

[0065] When the warning level is identified as Level 2, regardless of the warning type, the multimodal sensing module increases the sampling frequency of visual image data from 30 frames per second to 60 frames per second, the sampling frequency of thermal imaging temperature distribution data from 15 frames per second to 30 frames per second, and the sampling frequency of equipment vibration spectrum signal from 2 kHz to 5 kHz. The adjustment range is based on the recommendation in the equipment manual that the sampling frequency of all sensor data under Level 2 abnormal conditions should be increased to more than twice the original value.

[0066] When the warning level is identified as Level 3, regardless of the warning type, the multimodal perception module increases the sampling frequency of visual image data from 30 frames per second to 90 frames per second, the sampling frequency of thermal imaging temperature distribution data from 15 frames per second to 45 frames per second, and the sampling frequency of equipment vibration spectrum signals from 2 kHz to 8 kHz. The adjustment is based on the requirement of obtaining detailed data on the entire abnormal evolution process at the highest sampling density under Level 3 abnormal conditions in experimental data to support subsequent root cause localization and process optimization decisions. After completing the sampling frequency adjustment, the multimodal perception module continues to acquire visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signals at the adjusted sampling frequency. The newly acquired data is then re-input into the threshold dynamic adjustment module for a new round of comprehensive adjustment of defect judgment thresholds and defect identification processing.

[0067] The above-mentioned models or function formulas are all dimensionless and numerical calculations. The models or function formulas are obtained by software simulation based on a large amount of collected data to obtain the most recent real situation. The preset parameters in the models or function formulas are set by those skilled in the art according to the actual situation.

[0068] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0069] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0070] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0071] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An online monitoring system for the production process of wood-plastic composite doors, characterized in that, include: The multimodal sensing module is used to synchronously acquire visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signals during the production process of wood-plastic doors, and perform time-series feature fusion processing to output fused feature data. The threshold dynamic adjustment module is used to comprehensively adjust the defect judgment threshold based on the ratio of signal components to noise components of various sensor signals in the fused feature data and the judgment benchmark value of the defect judgment threshold, and output the defect identification result. The composite control execution module is used to generate adjustment instructions based on the defect identification results and output the adjustment instructions to the actuators of the production equipment; The feedback correction module is used to acquire the actual operating data after the actuator executes the adjustment command, and correct the judgment benchmark value of the defect judgment threshold based on the deviation between the actual operating data and the target data, and output the corrected judgment benchmark value of the defect judgment threshold to the threshold dynamic adjustment module. The process monitoring module is used to generate an early warning signal based on the fused feature data corresponding to the judgment benchmark value of the corrected defect judgment threshold, and to feed the early warning signal back to the multimodal perception module. The multimodal sensing module is also used to adjust the acquisition strategy of visual image data, thermal imaging temperature distribution data and equipment vibration spectrum signals according to the early warning signal.

2. The online monitoring system for the production process of wood-plastic door according to claim 1, characterized in that, The specific method for performing temporal feature fusion processing is as follows: The visual image data, thermal imaging temperature distribution data, and equipment vibration spectrum signal are aligned according to the acquisition time to form three sets of sensor data at the same time. Calculate the cumulative vibration energy value of the equipment vibration spectrum signal within a preset frequency range, and use the cumulative vibration energy value as the time-series characteristic component of the equipment vibration spectrum signal; Calculate the absolute value of the rate of temperature change between adjacent acquisition times in the thermal imaging temperature distribution data, and use the absolute value of the rate of temperature change as the temporal feature component of the thermal imaging temperature distribution data. Extract the sum of squares of pixel brightness differences between consecutive frames in visual image data, and use the sum of squares of pixel brightness differences as the temporal feature component of visual image data. The temporal characteristic components of the equipment vibration spectrum signal, the temporal characteristic components of the thermal imaging temperature distribution data, and the temporal characteristic components of the visual image data are spliced ​​together according to the acquisition time to generate fused feature data.

3. The online monitoring system for the production process of wood-plastic door according to claim 2, characterized in that, The specific method for comprehensively adjusting the defect judgment threshold is as follows: Calculate the first ratio of the first signal power value to the first noise power value of the visual image data in the fused feature data, the second ratio of the second signal power value to the second noise power value of the thermal imaging temperature distribution data, and the third ratio of the third signal power value to the third noise power value of the equipment vibration spectrum signal, respectively. The difference between the first ratio and the preset first standard signal-to-noise ratio value is used as the first deviation coefficient, the difference between the second ratio and the preset second standard signal-to-noise ratio value is used as the second deviation coefficient, and the difference between the third ratio and the preset third standard signal-to-noise ratio value is used as the third deviation coefficient. The first deviation coefficient, the second deviation coefficient, and the third deviation coefficient are summed to obtain the comprehensive deviation coefficient. The adjusted defect judgment threshold is obtained by summing the judgment benchmark value of the defect judgment threshold with the comprehensive deviation coefficient.

4. The online monitoring system for the production process of wood-plastic door according to claim 3, characterized in that, The threshold dynamic adjustment module is also used to compare the adjusted defect judgment threshold with the first normalized value of the pixel gray value of the visual image data, the second normalized value of the temperature gradient value of the thermal imaging temperature distribution data, and the third normalized value of the amplitude of the equipment vibration spectrum signal. The first normalized value is the ratio of the pixel gray value to the preset first maximum reference value, the second normalized value is the ratio of the temperature gradient value to the preset second maximum reference value, and the third normalized value is the ratio of the amplitude to the preset third maximum reference value. If any one of the first normalized value, the second normalized value, and the third normalized value is greater than the adjusted defect judgment threshold, it is determined that there is a defect at the current acquisition time.

5. The online monitoring system for the production process of wood-plastic door according to claim 4, characterized in that, The defect identification results output by the threshold dynamic adjustment module include defect type identifier and defect location coordinates; The defect type identifier is determined by the geometric shape features of the set of pixels in the visual image data whose first normalized value is greater than the defect judgment threshold, and the defect location coordinates are defined by the bounding rectangle of the set of pixels.

6. The online monitoring system for the production process of wood-plastic door according to claim 5, characterized in that, The specific method for generating adjustment instructions is as follows: When the defect identification result includes a defect type identifier and defect location coordinates, the preset adjustment parameter mapping table is matched according to the defect type identifier. The extrusion speed adjustment amount and traction rate adjustment amount corresponding to the defect type identifier are queried from the adjustment parameter mapping table. The direction of action of the extrusion speed adjustment amount is determined according to the relative position of the defect location coordinates in the visual image data. The extrusion speed adjustment amount and the direction of action are combined to generate an extrusion speed adjustment command, and the traction rate adjustment amount is used to generate a traction rate adjustment command.

7. The online monitoring system for the production process of wood-plastic door according to claim 6, characterized in that, The specific method for determining the criterion value of the defect judgment threshold is as follows: Obtain the actual operating data corresponding to a preset number of production cycle times, and form an actual operating data sequence; Obtain the target data corresponding to each actual operational data, and form a target data sequence; Calculate the difference between each data point in the actual running data sequence and the corresponding data point in the target data sequence to obtain the deviation value sequence; Calculate the arithmetic mean of the deviation numerical sequence to obtain the average deviation value; The corrected defect judgment threshold is obtained by adding the absolute value of the judgment benchmark value of the defect judgment threshold to the absolute value of the average deviation value.

8. The online monitoring system for the production process of wood-plastic door according to claim 7, characterized in that, The warning signal includes a warning level identifier and a warning type identifier. The specific method for generating a warning signal is as follows: Obtain the fused feature data of a preset number of production cycles corresponding to the judgment benchmark value of the corrected defect judgment threshold; Calculate the first ratio sequence of signal components to noise components of visual image data, the second ratio sequence of signal components to noise components of thermal imaging temperature distribution data, and the third ratio sequence of signal components to noise components of equipment vibration spectrum signal in each fused feature data. The differences between the maximum and minimum values ​​of the first ratio sequence, the second ratio sequence, and the third ratio sequence are calculated respectively to obtain the first fluctuation amplitude, the second fluctuation amplitude, and the third fluctuation amplitude. The first fluctuation amplitude is compared with the first preset fluctuation threshold, the second fluctuation amplitude is compared with the second preset fluctuation threshold, and the third fluctuation amplitude is compared with the third preset fluctuation threshold. The sensor data types that exceed their respective preset fluctuation thresholds are recorded, and the first number of sensor data types that exceed their respective preset fluctuation thresholds is counted. When the first quantity is greater than zero, the excess amount of the first fluctuation amplitude, the second fluctuation amplitude and the third fluctuation amplitude exceeding their respective preset fluctuation thresholds are summed to obtain the total excess amount.

9. The online monitoring system for the production process of wood-plastic door according to claim 8, characterized in that, The warning level indicator is determined based on the first quantity: When the first quantity is one or two, the warning level is marked as Level 1; when the first quantity is three, the warning level is marked as Level 2; when the first quantity is greater than or equal to three and the total number of excess quantities exceeds the preset excess quantity threshold, the warning level is marked as Level 3. The warning type identifier is determined based on the sensor data type that exceeds its corresponding preset fluctuation threshold: When only visual image data exceeds the first preset fluctuation threshold, the warning type is identified as visual data anomaly. When only thermal imaging temperature distribution data exceeds the second preset fluctuation threshold, the warning type is identified as temperature data anomaly. When only equipment vibration spectrum signal exceeds the third preset fluctuation threshold, the warning type is identified as vibration data anomaly. When multiple sensor data exceed their respective preset fluctuation thresholds simultaneously, the warning type is identified as combined anomaly. Finally, the warning level identifier and warning type identifier in the warning signal are analyzed, and the sampling frequency is adjusted accordingly according to the preset sampling adjustment strategy.

Images