An online detection method for air tightness of a sealed moisture-proof tool packaging box

Abstract

The application discloses a kind of airtightness online detection methods of sealed moisture-proof tool packaging box, belong to packaging detection technical field, comprising: in the sealing area of measured packaging box, pressure sensor array, humidity sensor array and infrared thermal imaging module are arranged;Through synchronous trigger signal control, three kinds of sensors are synchronously collected, and pressure signal, relative humidity signal and temperature field image are acquired;The collected signal is preprocessed and feature is extracted, and constitutes pressure feature vector, humidity feature vector and temperature feature vector;Three kinds of feature vectors are input into bayesian fusion algorithm engine calculation comprehensive posteriori leakage probability and are converted into confidence degree;According to the comparison result of confidence degree and preset threshold, determine instruction is output to actuating mechanism, and packaging box is pushed to corresponding station.The application realizes multi-physical field signal fusion evaluation and fine grading output, and is suitable for high-speed online detection scene.

Description

Technical Field

[0001] This application belongs to the field of packaging testing technology, specifically relating to an online method for testing the airtightness of a sealed moisture-proof tool packaging box. Background Technology

[0002] Air tightness testing of packaging boxes is a crucial step in ensuring the quality of stored products. For products such as metal industrial cutting tools, which have stringent requirements for storage environments, the sealing performance of the packaging box directly determines the stability of the internal microenvironment. Insufficient air tightness will allow external humid air to seep in, accelerating product oxidation and corrosion and affecting its service life. In automated production lines, air tightness testing needs to be completed quickly and accurately online at high speeds and can be directly integrated with subsequent sorting or rework processes. This places higher demands on the efficiency and reliability of testing technology.

[0003] Existing airtightness testing solutions share certain common technical approaches. For example, some solutions use sensor modules to detect the test object and acquire electrical signals, then use processing equipment to determine the leakage area based on these signals. Others use pressure detection devices to measure the pressure value within a sealed cavity and determine airtightness by comparing the pressure difference between a standard sample and the test product under the same conditions. Still others use image acquisition devices to acquire images of the sealed cavity in real time and perform image analysis to determine airtightness. However, these solutions generally use a single type of sensor for detection, or although they involve multiple sensors, they do not perform customized sensor selection and spatial layout optimization for the specific test object, resulting in a need to improve the accuracy and reliability of the test results. Furthermore, existing solutions often use simple threshold judgments or single-index comparisons for leak determination, making it difficult to perform refined grading assessments of the sealing performance of the test object and intelligent identification of leak types.

[0004] Therefore, how to design a dedicated sensor detection architecture for the structural characteristics of specific packaging boxes, realize the synchronous acquisition and fusion evaluation of multi-physics field signals, and perform fine classification of airtightness and identification of leakage types are technical problems that urgently need to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide an online method for detecting the airtightness of sealed moisture-proof tool packaging boxes, which can effectively solve the problems of how to design a dedicated sensor detection architecture for specific packaging box structural characteristics and how to perform fine classification of airtightness and identification of leakage types.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An online method for testing the airtightness of a sealed moisture-proof tool packaging box includes: A pressure sensor array, a humidity sensor array, and an infrared thermal imaging module are arranged in the sealed area of ​​the packaging box being tested. The pressure sensor array includes multiple pressure sensor nodes distributed at the four corners of the sealing surface of the box, the humidity sensor array includes multiple humidity sensors respectively arranged on the inner and outer sides of the sealing ring, and the infrared thermal imaging module is arranged directly above the box. After the detection program is started, the pressure sensor array, the humidity sensor array and the infrared thermal imaging module are controlled by the synchronous trigger signal to collect pressure signals on the sealing surface of the box, relative humidity signals in the sealing ring area and temperature field images of the sealing area, respectively. The collected raw signals are preprocessed and feature extracted. Time-domain feature parameters are extracted from the pressure signal to form a pressure feature vector, trend feature parameters are extracted from the relative humidity signal to form a humidity feature vector, and temperature gradient distribution feature parameters are extracted from the temperature field image to form a temperature feature vector. The pressure feature vector, the humidity feature vector, and the temperature feature vector are input into the Bayesian fusion algorithm engine to calculate the comprehensive posterior leakage probability and convert the posterior leakage probability into a confidence level. Based on the comparison result between the confidence level and the preset threshold, a corresponding judgment instruction is output to the execution mechanism, so that the execution mechanism pushes the packaging box to the corresponding workstation.

[0007] Preferably, each pressure sensor node of the pressure sensor array is fixed on the four corner support seats of the detection fixture, and the support seats are provided with elastic floating elements; The pressure-sensing surface of the pressure sensor node and the sealing surface of the box are kept in elastic contact through the elastic floating member. The elastic contact is used to adaptively compensate for the flatness tolerance of the sealing surface of the packaging box. This ensures that the pressure-sensing diaphragm of each pressure sensor node forms a stable contact pressure with the sealing surface of the housing.

[0008] Preferably, the coordinates of the hotspot locations in the temperature field image are determined as follows: The temperature field data output by the infrared thermal imaging module is converted into a temperature matrix after emissivity correction and temperature calibration. For each frame of temperature matrix, calculate the temperature change of each pixel relative to the previous frame, and mark the pixels with changes exceeding a preset threshold as suspected leakage points. Within a preset detection window, the cumulative frequency of each suspected leak point is counted, and the coordinates of the pixel with the highest frequency are determined as the hotspot location coordinates, which are the location of the leak point.

[0009] Preferably, in the Bayesian fusion algorithm engine: A priori leakage probability model is established based on historical detection data statistics; Establish probability distribution models for the feature vectors of each sensor under leakage and normal conditions respectively; The conditional probability model of the pressure sensor feature vector is the product of multiple one-dimensional Gaussian distributions with independent features in each dimension. The mean and standard deviation of the one-dimensional Gaussian distribution are obtained through statistical learning of historical leakage samples. The conditional probability model of the feature vector of the humidity sensor is a three-dimensional Gaussian distribution; The conditional probability model of the feature vector of the temperature sensor is a two-dimensional Gaussian distribution; The comprehensive posterior leakage probability is calculated according to Bayes' theorem. The comprehensive posterior leakage probability is the product of the conditional leakage probability of the pressure sensor, the conditional leakage probability of the humidity sensor, the conditional leakage probability of the temperature sensor, and the prior leakage probability, divided by a normalization constant.

[0010] Preferably, the step of outputting a corresponding judgment instruction based on the comparison result between the confidence level and the preset threshold includes: When the confidence level is higher than a preset first threshold, it is judged to be of excellent quality. When the confidence level is within the preset second threshold range, it is determined to be of a qualified level; When the confidence level is within the preset third threshold range, it is determined to be a warning level; When the confidence level is lower than a preset fourth threshold, it is determined to be a leakage level; The "excellent" rating indicates that the sealing performance of the packaging box meets the factory requirements; the "qualified" rating indicates that the packaging box has acceptable defects; the "warning" rating indicates that the sealing performance of the packaging box needs to be re-inspected and confirmed; and the "leakage" rating indicates that the packaging box has leaks and should be scrapped.

[0011] Preferably, the actuator adopts a cylinder-driven push rod structure, and the extension and retraction of the cylinder is controlled by a solenoid valve; The edge computing unit sends control commands to the solenoid valve driver through a digital output port; The excellent command, qualified command, warning command, and leakage command correspond to different combinations of control voltages for the solenoid valve actuator. The solenoid valve switches different air passages according to the received control voltage combination, so that the cylinder piston rod extends to different preset stroke lengths. Different preset travel lengths correspond to the packaging box being pushed to four different target locations: the next workstation, the release channel, the re-inspection workstation, and the waste bin.

[0012] Preferably, it also includes a leak type identification step: When the mean pressure characteristic deviates from the baseline value by more than 2 standard deviations and the slope of the humidity characteristic rises by more than 0.5%RH / s, it is determined to be a leak at the sealing ring interface. When the abnormal temperature gradient region is concentrated in the edge region where the normalized distance from the hot spot to the center of the image is greater than 0.8, it is judged as a microcrack in the structure. When the abnormal temperature gradient region is concentrated in the central region where the normalized distance from the hot spot to the center of the image is less than 0.3, it is determined to be a valve leak. The leakage type identification result is bound to the packaging box serial number.

[0013] Preferably, it also includes a data upload step: The data packaging module of the edge computing unit encapsulates the detection data into JSON data packets; The detection data includes the characteristic values ​​of each sensor, fusion judgment results, airtightness level, leakage type identification results, detection timestamp, and packaging box serial number; The edge computing unit initiates an upload request to the quality data receiving endpoint of the production management system through its built-in 4G communication module; After the upload is completed, the production management system stores the current inspection data into the quality database using the packaging box serial number as the primary key.

[0014] Preferably, the data upload step further includes an abnormal retransmission mechanism: In the event of network communication failure, the edge computing unit writes the unuploaded JSON data packet into the local cache queue of the non-volatile memory and records the failure timestamp; The edge computing unit periodically detects the network communication status at preset time intervals. When the network connection is restored, it automatically reads the unuploaded JSON data packets from the local cache queue in first-in-first-out order and re-initiates the upload request. The local cache queue has a maximum capacity of 10,000 records, and when the local cache queue is full, it will automatically overwrite the oldest record.

[0015] Preferably, in the Bayesian fusion algorithm engine: The conditional probability model of the pressure sensor feature vector is the product of multiple one-dimensional Gaussian distributions with independent features in each dimension. The mean and standard deviation of each one-dimensional Gaussian distribution are obtained through statistical learning of historical leakage samples. The conditional probability model of the feature vector of the humidity sensor is a three-dimensional Gaussian distribution; The conditional probability model of the feature vector of the temperature sensor is a two-dimensional Gaussian distribution; The normalization constant is expanded using the total probability formula to be the sum of the probability of leakage state and the probability of normal state.

[0016] In summary, this application includes at least one of the following beneficial technical effects: 1. This invention arranges a pressure sensor array, a humidity sensor array, and an infrared thermal imaging module in the sealed area of ​​the packaging box under test. The pressure sensor array is distributed at the four corners of the sealing surface to obtain the pressure distribution of the sealing surface. The humidity sensor array is arranged on both the inner and outer sides of the sealing ring to monitor the humidity difference between the two sides of the sealing ring. The infrared thermal imaging module is arranged directly above the box to obtain the temperature field distribution of the sealed area. The three sensors are controlled to collect data synchronously through a synchronous trigger signal, realizing the synchronous acquisition of multi-physics field signals for the sealing structure of the packaging box. The distributed layout of the pressure sensor array at the four corners can capture local pressure anomalies at different locations on the sealing surface. The symmetrical arrangement of the humidity sensor array on the inner and outer sides can detect water vapor infiltration caused by leakage at the sealing ring interface. The infrared thermal imaging module is arranged directly above the box to cover the temperature field changes of the entire sealed area. The spatial layout of the three sensors is adapted to the sealing structure of the packaging box, overcoming the shortcomings of blind spots in single-sensor detection and the lack of customized layout of multiple sensors for specific detection objects.

[0017] 2. This invention inputs pressure feature vectors, humidity feature vectors, and temperature feature vectors into a Bayesian fusion algorithm engine to calculate the comprehensive posterior leakage probability and convert it into a confidence level. Then, based on the comparison between the confidence level and a preset threshold, it outputs corresponding judgment commands to the actuator, achieving probabilistic fusion and hierarchical output of multi-sensor features. The Bayesian fusion algorithm engine integrates the conditional probability distributions of the three sensors, avoiding the pitfalls of misjudgment caused by single-index threshold determination. The confidence level, as a continuous quantitative indicator, accurately reflects the magnitude of the leakage probability. The hierarchical comparison mechanism of the preset threshold enables refined classification of airtightness levels. Different judgment commands drive the actuator to push the packaging box to different workstations, forming a complete closed loop from fusion evaluation to physical sorting, effectively solving the problem of existing solutions' difficulty in performing refined hierarchical assessments.

[0018] 3. This invention controls three sensors to acquire data synchronously via a synchronous trigger signal. After acquisition, the raw signals are preprocessed and features are extracted. The multi-dimensional feature vectors are then uniformly input into a Bayesian fusion algorithm engine for comprehensive evaluation. The entire detection process is completed within a single sampling cycle, from signal acquisition to graded output. No step-by-step detection or multiple sampling is required, which can meet the high-speed online detection cycle requirements of automated production lines.

[0019] 4. This invention uses the confidence level output by the Bayesian fusion algorithm engine to make a four-level classification judgment. Each level corresponds to factory release, marked release, re-inspection confirmation, and scrapping. While ensuring that products with high airtightness requirements can be shipped directly, products with acceptable defects are allowed to flow through the production line without interruption. This avoids excessive scrapping caused by the traditional "all or nothing" judgment mode and achieves a reasonable balance between detection accuracy and production efficiency. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the overall process of the online airtightness testing method for the sealed moisture-proof tool packaging box of the present invention.

[0021] Figure 2 This is a flowchart of the multi-sensor synchronous acquisition and preprocessing process of the present invention.

[0022] Figure 3 This is a flowchart of the Bayesian fusion determination process of the present invention.

[0023] Figure 4 This is a flowchart of the four-level output and leakage type identification process of the present invention. Detailed Implementation

[0024] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following detailed description of specific embodiments based on the present invention is provided in conjunction with the accompanying drawings and preferred embodiments.

[0025] like Figure 1 As shown, the present invention provides an online detection method for the airtightness of a sealed moisture-proof tool packaging box. Based on a multi-sensor fusion architecture, the method simultaneously collects physical signals of the sealed area through a pressure sensor array, a humidity sensor array, and an infrared thermal imaging module, and combines a Bayesian fusion algorithm to achieve intelligent evaluation of airtightness.

[0026] In step S1, as Figure 2 As shown, the specific implementation of multi-sensor array deployment is as follows: A microelectromechanical system (MEMS) pressure sensor array, a thin-film humidity sensor array, and an infrared thermal imaging module are arranged in the sealed area of ​​the packaging box under test. The pressure sensor array includes at least 4 sensor nodes, distributed at the four corners of the sealing surface of the box. The humidity sensor array includes at least 2 sensors, respectively arranged on the inner and outer sides of the sealing ring. The infrared thermal imaging module is arranged directly above the box to acquire the temperature field distribution image of the sealed area.

[0027] Furthermore, the pressure sensor array uses a microelectromechanical system (MEMS) silicon-based pressure sensor with a preset range and preset accuracy, and is arranged at the four corners of the detection fixture.

[0028] Specifically, the pressure sensor array uses a MEMS silicon-based pressure sensor, model BMP280, with a range of 0 to 1000 Pa and an accuracy of ±1 Pa. The sensor nodes are fixed to the four corner supports of the detection fixture via threaded connections. The supports are equipped with elastic floating elements, which are cylindrical silicone rubber pads with a thickness of 2 mm and a Shore A hardness of 60 degrees. These elements are embedded in the guide sleeves of the supports, with their upper ends abutting against the bottom of the pressure sensor nodes. The compression is controlled within the range of 0.3 mm to 0.8 mm, ensuring that the sensor pressure-sensing surface maintains elastic contact with the box sealing surface. This elastic contact method can adaptively compensate for the flatness tolerance of the packaging box sealing surface, ensuring that the pressure-sensing diaphragm of each sensor node forms a stable contact pressure with the sealing surface (the contact pressure is maintained between 5 kPa and 15 kPa). The sensor nodes arranged at the four corners can obtain the pressure distribution information at the four corners of the sealing surface. When there is a local defect in the sealing ring, the pressure change at that location will be reflected in the corresponding sensor reading first.

[0029] The humidity sensor array uses a thin-film capacitive humidity sensor with a preset measurement range and preset response time, and is arranged on both the inner and outer sides of the sealing ring.

[0030] Specifically, the humidity sensor array uses a thin-film capacitive humidity sensor, model SHT30, with a measurement range of 0 to 100%RH and a response time of 1.8s. The humidity sensor inside the sealing ring is arranged inside the sealing groove, close to the internal space of the box, while the humidity sensor outside the sealing ring is arranged in the external environment of the box. The symmetrical arrangement of the two sensors can monitor the humidity difference on both sides of the sealing ring. When there is a leak at the sealing ring interface, water vapor in the outside air will penetrate into the box through the leak channel, causing the humidity inside the sealing ring to rise, while the reading of the humidity sensor outside the sealing ring remains relatively stable.

[0031] The infrared thermal imaging module uses an infrared sensor with a preset resolution and a preset field of view, and is positioned at a predetermined distance directly above the box.

[0032] Specifically, the infrared thermal imaging module uses a 64×64 pixel infrared sensor, model MLX90640, with a field of view of 110°×75°, positioned 200mm directly above the box. This module is fixed above the inspection station by a bracket, ensuring that the infrared field of view covers the entire sealed area of ​​the packaging box. The infrared thermal imaging module has a temperature detection resolution of 0.5°C, capable of detecting minute temperature changes in the sealed area. When there is a micro-leakage in the packaging box, the flow of external air can cause localized temperature anomalies in the sealed area. Infrared thermal imaging can visualize the location of the leak point. This is achieved through... The temperature field data output by the infrared sensor is converted into a temperature matrix after emissivity correction and temperature calibration. For each frame of the temperature matrix, the temperature change of each pixel relative to the previous frame is calculated. Pixels with a change exceeding a preset threshold (preferably 0.5°C, matching the temperature detection resolution of the infrared sensor) are marked as suspected leak points. Within a preset detection window (preferably 30 consecutive frames, corresponding to a detection duration of about 1 second), the cumulative frequency of each suspected leak point is counted. The coordinates of the pixel with the highest frequency are determined as the hotspot location coordinates, which are the location of the leak point, thereby realizing the spatial positioning of the leak point.

[0033] In step S2, the specific implementation of synchronization signal acquisition is as follows: After the detection program is started, the three sensor arrays synchronously acquire physical signals of the sealed area. The edge computing unit generates a synchronous trigger signal through the same clock source. This trigger signal is divided by a frequency divider to generate sampling clocks of 200Hz, 10Hz and 30fps respectively. Each sensor starts its acquisition process on the rising edge of the same trigger pulse, thereby ensuring that the acquisition start time of the three sensor arrays is consistent. The pressure sensor acquires the pressure value of the sealing surface, and the sampling frequency adopts the preset sampling frequency. The humidity sensor acquires the relative humidity value of the sealing ring area and completes the acquisition within the preset response time. The infrared thermal imaging module acquires the temperature distribution image of the sealed area and performs image acquisition using the preset frame rate.

[0034] During signal acquisition, the pressure sensor collects the sealing surface pressure signal at a preset sampling rate. After analog-to-digital conversion, the signal enters a bandpass filter with a preset bandpass frequency range to filter out low-frequency vibrations and high-frequency noise interference from the environment. The pressure sensor also collects the sealing surface pressure signal at a sampling rate of 200Hz. After being converted into a digital signal by a 12-bit analog-to-digital converter, the signal enters a bandpass filter with a bandpass frequency range of 10 to 50Hz. The lower limit frequency of the bandpass filter, 10Hz, is used to filter out low-frequency vibration interference in the factory environment, such as vibrations caused by conveyor belt operation and ground micro-vibrations caused by personnel walking. The upper limit frequency, 50Hz, is used to filter out high-frequency noise interference, such as electromagnetic noise generated by motor start-up and shutdown and power supply ripple interference. The filtered pressure signal is then sent to the edge computing unit for real-time feature extraction.

[0035] The humidity sensor collects the relative humidity signal of the sealing ring area at a sampling rate of 10Hz. Compared with the 200Hz sampling rate of the pressure sensor, the humidity sensor uses a lower sampling rate, which is determined by the physical characteristics of humidity. The diffusion speed of water vapor molecules is much lower than the propagation speed of pressure waves, and the change of humidity signal is relatively slow. An excessively high sampling rate will only introduce more thermal noise and will not be able to obtain additional effective information. The collected raw humidity data is preprocessed using a first-order hysteresis filter with a filter time constant set to 500ms to smooth the inherent electronic noise of the sensor.

[0036] The infrared thermal imaging module acquires temperature field images of the sealed area at a frame rate of 30fps. The 30fps frame rate ensures that the dynamic temperature change process of the sealed area can be captured. Each frame of the infrared image has a resolution of 64×64 pixels, with a total of 4096 temperature measurement points. The raw data output by the infrared sensor is a 16-bit temperature code value, which is converted into a Celsius temperature value after emissivity correction and temperature calibration. The infrared image sequence is stored in a circular buffer with a buffer depth of 60 frames, covering the time-series temperature data of the most recent 2 seconds.

[0037] In step S3, the specific implementation of feature extraction and preprocessing is as follows: The acquired raw signals are preprocessed. After the pressure signal is filtered by a bandpass filter, the time-domain features of the filtered pressure signal are extracted in the edge computing unit. The arithmetic mean, root mean square value and peak value within the sampling period are calculated as three time-domain feature parameters. After the humidity signal is processed by the moving average of the first-order hysteresis filter, the slope of the linear regression and the curvature of the second-order difference are calculated on the filtered humidity sequence to obtain the two trend feature parameters of humidity change slope and curvature, respectively. After the infrared image is smoothed by Gaussian and edge detected by the Sobel operator, regional statistical analysis is performed on the gradient amplitude map to calculate the average temperature difference between the sealed edge area and the center area and the coordinates of the hot spot location as two temperature gradient distribution feature parameters.

[0038] The characteristic parameters calculated after pressure signal filtering include: the arithmetic mean, root mean square (RMS) value, and peak value within the sampling period. For each sensor node in the pressure sensor array, its characteristic parameters within the sampling period are calculated independently. Taking a single sensor node as an example, the sampling period is set to 1000 ms, and the number of sampling points is 200. The formula for calculating the arithmetic mean is: ; in, Let be the pressure value at the i-th sampling point, and n be the total number of sampling points. The formula for calculating the root mean square value is: ; The peak value is the difference between the maximum and minimum pressure values ​​within the sampling period, i.e.: ; The pressure characteristic parameters of the four sensor nodes constitute a four-dimensional pressure feature vector. The subscripts 1 to 4 correspond to the four corner positions: top left, top right, bottom left, and bottom right, respectively.

[0039] The humidity signal is processed using a first-order hysteresis filter for moving average, with a preset window width. Specifically, a first-order hysteresis filter is used for moving average, with a window width of 5 sampling points. The filtered humidity sequence is denoted as... The calculation formula is as follows: ; in, These are the filter coefficients. This represents the original humidity value at time k. This is the filtered value from the previous moment. The characteristic parameters of the humidity change trend include the linear regression slope and curvature. The linear regression slope is obtained by performing a first-order linear fit on the most recent 30 filtered humidity sampling points. The slope of the fitted line is the humidity change slope. Specifically, it is calculated as follows: using the sampling point number k as the independent variable, and the filtered humidity value... As the dependent variable, the least squares method is used to fit the straight line. The fitting slope 'a' is the slope of humidity change, and its calculation formula is: ; Where m is the number of sampling points involved in the fitting. The average of the sampling point numbers. This represents the mean of the filtered humidity values. Curvature is defined as the absolute value of the second derivative of the humidity change curve, used to characterize the degree of acceleration or deceleration of humidity changes. It is calculated by performing a second-order central difference on the filtered humidity sequence, i.e.: ; in, and The first Time and the The filtered humidity value at any given time is used as the average value within the detection window as the final curvature feature value.

[0040] Feature extraction from infrared images employs the Sobel operator for edge detection and temperature gradient distribution calculation. The specific processing flow includes the following steps: First, Gaussian smoothing is applied to each frame of the infrared image, with a smoothing kernel size of 3×3 and a standard deviation of 0.8, to suppress thermal noise. Then, the Sobel operator is applied to calculate gradients in the horizontal and vertical directions. The Sobel operator's convolution kernels are as follows: ; The gradient magnitude of each pixel in the image is: ; The temperature gradient distribution characteristics include two parameters: the average temperature difference between the sealed edge region and the central region. and hotspot location coordinates The edge region is defined as the region less than or equal to 8 pixels from the image edge, the center region is defined as the region less than or equal to 12 pixels from the image center, and the hot spot is the pixel coordinate of the maximum gradient magnitude.

[0041] In step S4, the Bayesian fusion decision is implemented as follows: The extracted multidimensional feature vectors are input into the Bayesian fusion algorithm engine to calculate the posterior leakage probability. During runtime, the Bayesian fusion algorithm engine first loads the pre-trained model parameters, which include the mean vector and covariance matrix of each sensor feature under leakage and normal conditions. Then, it calculates the conditional probability values ​​of the feature vectors of the pressure sensor, humidity sensor, and temperature sensor respectively. Finally, it substitutes the three conditional probability values ​​and the prior leakage probability into the Bayesian formula to calculate the posterior leakage probability. Specifically, this includes: establishing a prior leakage probability model based on historical statistical data, calculating the conditional probability of leakage under each sensor feature condition, and using the Bayesian formula to fuse the features to obtain the comprehensive leakage probability and confidence level.

[0042] like Figure 3 As shown, the specific implementation of the Bayesian fusion algorithm includes the following steps: S401: Establish a prior probability model.

[0043] Based on historical testing data, a preset prior leakage probability is established, meaning that the packaging box has a preset leakage probability value under normal process conditions. Setting it to 0.05 means that under normal process conditions, the probability of packaging box leakage is 5%. This prior probability value comes from the statistical analysis of more than 1 million packaging boxes tested over the past 12 months.

[0044] S402: Calculate the conditional probability of each sensor.

[0045] For each sensor feature vector, establish probability distribution models for leakage conditional probability and normal conditional probability respectively. For pressure sensor feature vectors... Define the conditional probability of leakage. and normal conditional probability Since the pressure feature vector is a four-dimensional variable, assuming that the feature conditions of each dimension are independent, the conditional probability model of the pressure sensor is the product of four one-dimensional Gaussian distributions: ; in, This represents the specific value of the i-th pressure characteristic component; and These are the mean and standard deviation of the i-th pressure feature under leakage conditions, respectively. These parameters are obtained through statistical learning from historical leakage samples. As an exemplary embodiment, typical parameters trained using the maximum likelihood estimation method based on more than 1000 historical leakage samples are: the mean vector of pressure features under leakage conditions. Standard deviation vector Mean vector under normal conditions Standard deviation vector The model parameters can be incrementally updated quarterly based on actual production line data.

[0046] Similarly, for the normal state: ; Humidity feature vector It contains three feature components: the humidity value inside the sealing ring, the humidity value outside the sealing ring, and the slope of humidity change. The conditional probability model of the humidity feature vector adopts a three-dimensional Gaussian distribution. ; in, It is a three-dimensional mean vector. It is a 3×3 covariance matrix, and the parameters under normal conditions are expressed in the same form.

[0047] Temperature eigenvector It contains two characteristic components: edge-center temperature difference. and hotspot location coordinates The comprehensive index, temperature characteristics, are modeled using a two-dimensional Gaussian distribution as a conditional probability model.

[0048] S403: Bayesian fusion computation.

[0049] Calculate the overall posterior leakage probability using Bayes' theorem: ; Among them, the denominator term Let be the normalization constant, which can be expanded as follows: ; In engineering implementation, the denominator is calculated using the law of total probability, while logarithmic operations are used in the actual code to avoid underflow of the probability product.

[0050] S404: Confidence level output.

[0051] The posterior leakage probability is converted into a confidence value for subsequent grading. The formula for calculating the confidence value (Conf) is as follows: ; The confidence level ranges from 0% to 100%, with higher values ​​indicating a greater likelihood of leakage in the packaging.

[0052] In step S5, the specific implementation of the four-level output and classification is as follows: like Figure 4 As shown, based on the fusion judgment results, the airtightness of the packaging box is divided into four levels. When the overall confidence level is higher than the preset first threshold, it is judged as excellent; when the overall confidence level is within the preset second threshold range, it is judged as qualified; when the overall confidence level is within the preset third threshold range, it is judged as a warning; and when the overall confidence level is lower than the preset fourth threshold, it is judged as a leak.

[0053] The first preset threshold is set at 95%, the second preset threshold ranges from 70% to 95%, the third preset threshold ranges from 40% to 70%, and the fourth preset threshold is set at 40%. These four threshold levels are determined through statistical analysis of historical testing data: the excellent threshold of 95% corresponds to a minimum of 99% confidence level among historical normal samples, ensuring a false positive rate of less than 1% for outgoing products; the leakage threshold of 40% corresponds to a minimum of 95% confidence level among historical leakage samples, ensuring a false negative rate of less than 5%. Enterprises can adaptively adjust the thresholds within ±10% based on the specific sealing requirements of their products. This threshold classification is based on the enterprise's quality standards for different airtightness levels. An excellent level indicates that the packaging box's sealing performance meets factory requirements and can be used directly; a qualified level indicates that the packaging box has acceptable defects but still meets basic sealing requirements; a warning level indicates that the packaging box's sealing performance is questionable and requires re-inspection; and a leakage level indicates that the packaging box leaks and must be scrapped.

[0054] The specific execution method of the four-level output is as follows: When the confidence level is higher than the preset first threshold, this method outputs an excellent instruction to the actuator, and the packaging box is automatically pushed into the next station. When the confidence level is within the preset second threshold range, this method outputs a qualified instruction, and the packaging box is marked and released. When the confidence level is within the preset third threshold range, this method outputs a warning instruction, and the actuator pushes the packaging box into the re-inspection station to wait for manual review. When the confidence level is lower than the preset fourth threshold, this method outputs a leakage instruction, and the actuator pushes the packaging box into the waste bin.

[0055] The actuator adopts a cylinder-driven push rod structure, with the extension and retraction of the cylinder controlled by a solenoid valve. The edge computing unit sends control commands to the solenoid valve driver through a digital output port. The command signal is at a 24V DC level, and the actuator's response time is less than 200ms, ensuring that the packaging box can be pushed to the corresponding workstation in a timely manner. The excellent, qualified, warning, and leakage commands output by the edge computing unit correspond to different control voltage combinations of the solenoid valve driver. The solenoid valve switches different air path channels according to the received control voltage combination, causing the cylinder piston rod to extend to different preset stroke lengths. Different preset stroke lengths correspond to the packaging box being pushed to four different target positions: the next workstation, the release channel, the re-inspection station, and the scrap bin.

[0056] In step S6, the specific implementation of leak type identification is as follows: like Figure 4 As shown, when a leak or warning state is detected, the leak type is further identified based on the combination of abnormal characteristics from each sensor. A leak at the sealing ring interface corresponds to a physical failure mode where there is a gap channel between the sealing ring and the housing mating surface. In this mode, external air and moisture seep into the housing through the interface gap between the sealing ring and the housing. The pressure sensor array detects that the pressure distribution at the four corners of the sealing surface deviates from the reference value. Simultaneously, the humidity sensor inside the sealing ring detects moisture seepage leading to increased humidity, while the humidity sensor reading on the outside remains relatively stable. Therefore, the combination of abnormal pressure and abnormal humidity points to a leak at the sealing ring interface. An abnormal temperature gradient concentrated in the edge area indicates a micro-crack in the structure. This mode corresponds to a small crack in the side wall of the housing. The crack is located on the edge side wall of the housing. The external airflow caused by the leak creates a local temperature disturbance at the side wall location, thus the abnormal temperature gradient is concentrated in the image edge area. An abnormal temperature gradient concentrated in the center area indicates a valve leak. This mode corresponds to a leak in the desiccant breather valve integrated into the housing. The breather valve is located in the center area of ​​the housing. The leaking airflow creates a temperature disturbance at the center location, thus the abnormal temperature gradient is concentrated in the image center area.

[0057] The criterion for determining temperature gradient anomalies is: the temperature gradient amplitude G of a pixel exceeds the sum of the mean gradient of the entire image and twice the standard deviation, thus satisfying the condition. When this happens, the pixel is identified as a gradient anomaly; the region formed by consecutive anomalies is the temperature gradient anomaly region.

[0058] The rule for identifying the type of leakage is as follows: when the mean pressure characteristic deviates from the benchmark value by more than a preset multiple of standard deviation and the slope of the humidity characteristic increase exceeds a preset rate of change threshold, it is determined to be a leak at the sealing ring interface. Specifically, the judgment condition is that the mean pressure characteristic deviates from the benchmark value by more than 2 standard deviations. or The slope of the humidity characteristic rise exceeds 0.5%RH / s, that is... When both of the above conditions are met, it is determined that there is leakage at the sealing ring interface.

[0059] The normalized distance d from the hotspot location to the image center is calculated using the following formula: ; In the formula, (x,y) are the pixel coordinates of the hotspot location. Let be the pixel coordinates of the image center, and D be the pixel length of the image diagonal; for a 64×64 pixel infrared image, D≈90.5 pixels, and the normalized distance d ranges from 0 to 1.

[0060] When the temperature gradient is abnormal and the abnormal area is concentrated in the edge area exceeding the first preset field of view ratio, it is determined to be a structural microcrack. The first preset field of view ratio is set to 80%, that is, when the normalized distance of the hot spot position from the center of the image is greater than 0.8, it is determined to be a structural microcrack.

[0061] When the temperature gradient is abnormal and the abnormal area is concentrated in the central area below the second preset field of view ratio, it is determined to be a valve leak. The second preset field of view ratio is set to 30%, that is, when the normalized distance between the hot spot position and the center of the image is less than 0.3, it is determined to be a valve leak.

[0062] The leakage type identification results are linked to the packaging box serial number and uploaded to the MES system for analyzing the root cause of the leakage and process improvement.

[0063] The determination of the above-mentioned threshold is based on the following criteria: (1) The threshold of 2 times the standard deviation is based on the statistics of 1000 groups of sealing ring interface leakage samples. The average deviation of the pressure characteristics of the leakage samples is about 2.1 times the standard deviation. Therefore, 2 times the standard deviation is selected as the judgment boundary to take into account both the detection rate and the false alarm rate. (2) The humidity change slope threshold of 0.5%RH / s is determined based on a safety margin of 5 times the ambient humidity drift rate under normal operating conditions (usually less than 0.1%RH / s); (3) The normalized distance thresholds of 0.8 and 0.3 are determined based on the geometric proportions of the sealing structure of the packaging box. The edge of the sealing ring is about 80% of the diagonal of the image center, and the breathing valve area is about 30% of the center.

[0064] In step S7, the specific implementation of data uploading and traceability is as follows: The detection data, including the original feature values ​​of each sensor, fusion judgment results, airtightness level, leakage type identification results, along with the detection timestamp and packaging box serial number, are synchronously uploaded to the production management system to form a quality data archive, supporting full-process traceability. In this method, after the edge computing unit completes the four-level output judgment, the data packaging module of the edge computing unit encapsulates the detection data into a preset format JSON data packet and calls the built-in 4G communication module to initiate an upload request to the quality data receiving endpoint of the production management system through the HTTP RESTful API interface. After the upload is completed, the production management system stores the current detection data into the quality database with the packaging box serial number as the primary key, realizing the association storage of single packaging box detection data with historical production batch information, thereby supporting subsequent full-process quality traceability indexed by the packaging box serial number.

[0065] The data upload format is generated according to the confidence level. The "Excellent" level includes a level identifier, a confidence value, and a detection timestamp, in JSON format: {"Level": "Excellent", "Confidence": "XX.X%", "Timestamp": "YYYYMMDDThhmmss"}. The "Pass" level adds a marker to the "Excellent" level data, with the marker field being "Remarks": {"Level": "Pass", "Confidence": "XX.X%", "Timestamp": "YYYYMMDDThhmmss", "Remarks": "Marker information";} The warning level now includes a re-inspection reminder label, with the remarks field "Re-inspection station pending processing": {"Level": "Warning", "Confidence level": "XX.X%", "Timestamp": "YYYYMMDDThhmmss", "Remarks": "Re-inspection station pending processing"}. The leakage level now includes leakage determination details, including leakage type: {"Level": "Leakage", "Confidence level": "XX.X%", "Timestamp": "YYYYMMDDThhmmss", "Leakage type": "Sealing ring interface leakage / structural micro-crack / valve leakage"}.

[0066] Data uploads utilize an HTTP RESTful API interface, targeting the quality data receiving endpoint of the enterprise's MES system. Data packets are transmitted using TLS encryption to ensure the confidentiality and integrity of production data. The edge computing unit has a built-in 4G communication module, completing data upload within 300ms after detection. In case of network communication failures, the edge computing unit locally caches unuploaded data, automatically retransmitting it once the network is restored. The edge computing unit has a local cache queue with non-volatile memory. When upload fails, JSON data packets are written to the cache queue and the failure timestamp is recorded. The edge computing unit periodically checks the network communication status at preset time intervals. When network connectivity is detected, it automatically reads unuploaded data packets from the cache queue in a first-in-first-out order and re-initiates the upload request. The maximum capacity of the cache queue is set to 10,000 records. When the cache queue is full, it automatically overwrites the oldest record, ensuring that the caching mechanism does not affect subsequent detection processes due to storage space exhaustion.

[0067] The following detailed explanation of the complete workflow of the method of the present invention in an automated production line environment is illustrated through a specific application example: After the packaging box enters the inspection station from the feeding mechanism, the positioning fixture fixes the packaging box in the inspection position. The four corner sensor nodes of the pressure sensor array are tightly fitted with the four corners of the sealing surface of the box. The pressure-sensing diaphragm is aligned with the sealing groove. The inner sensor of the humidity sensor array is inserted into the inner groove of the sealing ring, while the outer sensor is exposed to the working environment. The infrared thermal imaging module is focused on the top plane of the packaging box, and the field of view covers the entire sealing area.

[0068] After the detection program is started, the pressure sensor begins to collect pressure data at a sampling rate of 200Hz, and monitors the pressure distribution at the four corners of the sealing surface in real time. The humidity sensor collects humidity data on both sides of the sealing ring at a sampling rate of 10Hz, calculates the humidity difference between the two sides, and the infrared thermal imaging module collects temperature field image sequences at a frame rate of 30fps, and calculates the temperature gradient distribution in real time.

[0069] After sampling for 1000ms, the edge computing unit extracts features from the collected data. The pressure feature extraction module calculates the arithmetic mean, root mean square value, and peak value of the four sensor nodes to generate a four-dimensional pressure feature vector. The humidity feature extraction module calculates the humidity value inside the sealing ring, the external humidity value, and the slope of humidity change to generate a three-dimensional humidity feature vector. The temperature feature extraction module calculates the temperature difference at the edge center and the location of hot spots to generate a two-dimensional temperature feature vector.

[0070] The feature vectors are input into the Bayesian fusion algorithm engine for comprehensive evaluation. The algorithm first loads the pre-trained model parameters, including the mean vector and covariance matrix of each sensor feature under leakage and normal conditions. Then, it calculates the conditional probability product of the three sensor features, combines the prior leakage probability to calculate the posterior leakage probability, and finally converts the posterior probability into a confidence value.

[0071] Suppose the detection results of a certain packaging box are as follows: the mean pressure characteristic is 5.2 Pa, which deviates from the benchmark value by 1.8 times the standard deviation; the humidity change slope is 0.3%RH / s, which does not exceed the threshold; the abnormal temperature gradient area is located in the center of the image, and the normalized distance is 0.15. The Bayesian fusion calculation yields a comprehensive posterior leakage probability of 0.28, corresponding to a confidence level of 28%. Since the confidence level of 28% is lower than the threshold of 40%, it is determined to be a leakage state. Based on the judgment condition that the abnormal temperature gradient area is located in the center and the normalized distance of 0.15 is less than 30%, the leakage type is further identified as valve leakage.

[0072] The edge computing unit sends a leak command to the actuator, and the cylinder pusher pushes the packaging box into the waste bin. At the same time, the detection data is packaged into JSON format and uploaded to the MES system. The data content includes: packaging box serial number "PKG-2024-00876", detection timestamp "20240115T143258", pressure feature vector "5.2,4.8,5.1,4.9", humidity feature vector "45.2,42.1,0.3", temperature feature vector "0.8,0.15", confidence level "28%", airtightness level "leakage", and leakage type "valve leakage".

[0073] After receiving the data, the MES system automatically updates the quality statistics report for that batch of packaging boxes and links the valve leakage record to the quality file of the desiccant breather valve supplier, providing data support for subsequent supplier evaluation and process improvement.

[0074] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0075] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment includes only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An online method for detecting the airtightness of a sealed moisture-proof tool packaging box, characterized in that, include: A pressure sensor array, a humidity sensor array, and an infrared thermal imaging module are arranged in the sealed area of ​​the packaging box being tested. The pressure sensor array includes multiple pressure sensor nodes distributed at the four corners of the sealing surface of the box, the humidity sensor array includes multiple humidity sensors respectively arranged on the inner and outer sides of the sealing ring, and the infrared thermal imaging module is arranged directly above the box. After the detection program is started, the pressure sensor array, the humidity sensor array and the infrared thermal imaging module are controlled by the synchronous trigger signal to collect pressure signals on the sealing surface of the box, relative humidity signals in the sealing ring area and temperature field images of the sealing area, respectively. The collected raw signals are preprocessed and feature extracted. Time-domain feature parameters are extracted from the pressure signal to form a pressure feature vector, trend feature parameters are extracted from the relative humidity signal to form a humidity feature vector, and temperature gradient distribution feature parameters are extracted from the temperature field image to form a temperature feature vector. The pressure feature vector, the humidity feature vector, and the temperature feature vector are input into the Bayesian fusion algorithm engine to calculate the comprehensive posterior leakage probability and convert the posterior leakage probability into a confidence level. Based on the comparison result between the confidence level and the preset threshold, a corresponding judgment instruction is output to the execution mechanism, so that the execution mechanism pushes the packaging box to the corresponding workstation.

2. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 1, characterized in that, Each pressure sensor node of the pressure sensor array is fixed on the four corner support bases of the detection fixture, and the support bases are provided with elastic floating elements. The pressure-sensing surface of the pressure sensor node and the sealing surface of the box are kept in elastic contact through the elastic floating element. The elastic contact is used to adaptively compensate for the flatness tolerance of the sealing surface of the packaging box. This ensures that the pressure-sensing diaphragm of each pressure sensor node forms a stable contact pressure with the sealing surface of the housing.

3. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 1, characterized in that, The method for determining the coordinates of hotspot locations in the temperature field image is as follows: The temperature field data output by the infrared thermal imaging module is converted into a temperature matrix after emissivity correction and temperature calibration. For each frame of temperature matrix, calculate the temperature change of each pixel relative to the previous frame, and mark the pixels with changes exceeding a preset threshold as suspected leakage points. Within a preset detection window, the cumulative frequency of each suspected leak point is counted, and the coordinates of the pixel with the highest frequency are determined as the hotspot location coordinates, which are the location of the leak point.

4. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 1, characterized in that, In the Bayesian fusion algorithm engine: A priori leakage probability model is established based on historical detection data statistics; Establish probability distribution models for the feature vectors of each sensor under leakage and normal conditions respectively; The conditional probability model of the pressure sensor feature vector is the product of multiple one-dimensional Gaussian distributions with independent features in each dimension. The mean and standard deviation of the one-dimensional Gaussian distribution are obtained through statistical learning of historical leakage samples. The conditional probability model of the feature vector of the humidity sensor is a three-dimensional Gaussian distribution; The conditional probability model of the feature vector of the temperature sensor is a two-dimensional Gaussian distribution; The comprehensive posterior leakage probability is calculated according to Bayes' theorem. The comprehensive posterior leakage probability is the product of the conditional leakage probability of the pressure sensor, the conditional leakage probability of the humidity sensor, the conditional leakage probability of the temperature sensor, and the prior leakage probability, divided by a normalization constant.

5. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 1, characterized in that, The step of outputting a corresponding judgment instruction based on the comparison result between the confidence level and the preset threshold includes: When the confidence level is higher than a preset first threshold, it is judged to be of excellent quality. When the confidence level is within the preset second threshold range, it is determined to be of a qualified level; When the confidence level is within the preset third threshold range, it is determined to be a warning level; When the confidence level is lower than a preset fourth threshold, it is determined to be a leakage level; The "excellent" rating indicates that the sealing performance of the packaging box meets the factory requirements; the "qualified" rating indicates that the packaging box has acceptable defects; the "warning" rating indicates that the sealing performance of the packaging box needs to be re-inspected and confirmed; and the "leakage" rating indicates that the packaging box has leaks and should be scrapped.

6. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 5, characterized in that, The actuator adopts a cylinder-driven push rod structure, and the extension and retraction of the cylinder is controlled by a solenoid valve; The edge computing unit sends control commands to the solenoid valve driver through a digital output port; The excellent command, qualified command, warning command, and leakage command correspond to different combinations of control voltages for the solenoid valve actuator. The solenoid valve switches different air passages according to the received control voltage combination, so that the cylinder piston rod extends to different preset stroke lengths. Different preset travel lengths correspond to the packaging box being pushed to four different target locations: the next workstation, the release channel, the re-inspection workstation, and the waste bin.

7. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 1, characterized in that, It also includes a leak type identification step: When the mean pressure characteristic deviates from the baseline value by more than 2 standard deviations and the slope of the humidity characteristic rises by more than 0.5%RH / s, it is determined to be a leak at the sealing ring interface. When the abnormal temperature gradient region is concentrated in the edge region where the normalized distance from the hot spot to the center of the image is greater than 0.8, it is judged as a microcrack in the structure. When the abnormal temperature gradient region is concentrated in the central region where the normalized distance from the hot spot to the center of the image is less than 0.3, it is determined to be a valve leak. The leakage type identification result is bound to the packaging box serial number.

8. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 1, characterized in that, It also includes the data upload step: The data packaging module of the edge computing unit encapsulates the detection data into JSON data packets; The detection data includes the characteristic values ​​of each sensor, the fusion judgment result, the airtightness level, the leakage type identification result, the detection timestamp, and the packaging box serial number; The edge computing unit initiates an upload request to the quality data receiving endpoint of the production management system through its built-in 4G communication module; After the upload is completed, the production management system stores the current inspection data into the quality database using the packaging box serial number as the primary key.

9. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 8, characterized in that, The data upload step also includes an abnormal retransmission mechanism: In the event of network communication failure, the edge computing unit writes the unuploaded JSON data packet into the local cache queue of the non-volatile memory and records the failure timestamp; The edge computing unit periodically detects the network communication status at preset time intervals. When the network connection is restored, it automatically reads the unuploaded JSON data packets from the local cache queue in first-in-first-out order and re-initiates the upload request. The local cache queue has a maximum capacity of 10,000 records, and when the local cache queue is full, it will automatically overwrite the oldest record.

10. The online airtightness detection method for a sealed moisture-proof tool packaging box according to claim 4, characterized in that, In the Bayesian fusion algorithm engine: The conditional probability model of the pressure sensor feature vector is the product of multiple one-dimensional Gaussian distributions with independent features in each dimension. The mean and standard deviation of each one-dimensional Gaussian distribution are obtained through statistical learning of historical leakage samples. The conditional probability model of the feature vector of the humidity sensor is a three-dimensional Gaussian distribution; The conditional probability model of the feature vector of the temperature sensor is a two-dimensional Gaussian distribution; The normalization constant is expanded using the total probability formula to be the sum of the probability of leakage state and the probability of normal state.

Images