Particle counting monitoring device and method combining dual pump system with displacement sensing technology

Abstract

The present invention relates to the field of particle counting monitoring technology, and discloses a particle counting monitoring device and method that combines a dual-pump system with displacement sensing technology. The device includes a dual-pump module, a displacement sensing module, a particle counting module, and an adaptive control module. The suction pump and the blowing pump of the dual-pump module work independently, and are coordinated to start and stop by a synchronous controller to stably collect and discharge particles. The displacement sensing module collects the displacement data of the sampling head in real time. The particle counting module calculates the particle distribution density based on the displacement and sampling data using an area density conversion algorithm. The adaptive control module reads the identification information of the sampling head chip and uses an AI algorithm model to determine the optimal sampling mode. The invention has accurate detection and strong adaptability, can effectively monitor particles on the surface of objects, and can be widely used in multiple industries such as electronics and pharmaceuticals to ensure product quality.

Description

Technical Field

[0001] The present invention relates to the technical field of particle counting monitoring, and in particular to a particle counting monitoring device and method combining a dual pump system with displacement sensing technology. Background Art

[0002] Monitoring surface particles is crucial in many industrial and scientific fields. For example, in the electronics manufacturing industry, the presence of tiny particles on chip surfaces can cause circuit shorts and signal transmission anomalies, seriously impacting chip performance and product quality. In the pharmaceutical industry, particle contamination on drug packaging or production equipment can lead to drug contamination, threatening patient health.

[0003] Traditional particle counting and monitoring methods have numerous limitations. Early optical microscopy methods, while capable of visually observing particles, were extremely inefficient and difficult to automate and detect over large areas. Filter-based weighing methods only provide information on the total particle count, but fail to capture the distribution of particles across the surface. For surfaces with uneven particle distribution, these results may not accurately reflect the actual situation.

[0004] Some existing particle counting monitoring devices use a single-pump system, which can easily lead to unstable airflow during the particle intake and exhaust processes. This single pump handles both intake and exhaust, and during this switching process, the airflow pressure and flow rate fluctuate, resulting in incomplete particle collection or residual particles within the device, affecting the accuracy of test results. Furthermore, single-pump systems struggle to adapt to the diverse demands for particle intake and exhaust speeds in different testing scenarios.

[0005] The application of displacement sensing technology in particle counting monitoring also has its limitations. The displacement sensors used in some devices have low precision and are unable to accurately measure the minute displacement of the sampling head. This creates significant difficulties in precisely determining the particle sampling location and calculating the particle density per unit area. Furthermore, existing displacement sensing modules often do not work well with particle counting modules, resulting in poor data fusion and an inability to provide accurate and effective displacement data support for particle counting.

[0006] Most existing monitoring devices lack adaptive sampling mode selection. Different surface morphologies, such as flat, curved, and irregular, require different sampling parameters to ensure detection accuracy. However, traditional devices typically use fixed sampling modes and are unable to automatically adjust sampling parameters based on the type of sampling head and the actual surface being detected, resulting in inaccurate detection results and low efficiency.

[0007] As various industries continue to increase their requirements for product quality, there is an urgent need for a particle counting monitoring device and method that can overcome the above-mentioned defects, achieve more efficient, accurate and intelligent particle detection, and meet the needs of modern industrial production and scientific research. Summary of the Invention

[0008] The object of the present invention is to provide a particle counting monitoring device and method combining a dual pump system with displacement sensing technology to solve the problems raised in the above background technology.

[0009] To achieve the above-mentioned object, the present invention provides the following technical solutions: a particle counting monitoring device combining a dual-pump system with displacement sensing technology, comprising a dual-pump module, a displacement sensing module, a particle counting module and an adaptive control module;

[0010] The dual-pump module includes an air suction pump and an air blowing pump, which respectively perform particle suction and discharge operations through independent channels, and the start and stop timing of the two are coordinated by a synchronous controller;

[0011] The displacement sensing module is used to collect displacement data of the sampling head in real time, including moving distance, speed and direction;

[0012] The particle counting module calculates the particle distribution density per unit area using an area density conversion algorithm based on the displacement data and the particle sampling data;

[0013] The adaptive control module obtains identification information through a chip embedded in the sampling head, and determines the optimal sampling mode of the sampling head based on the identification information through a preset AI algorithm model.

[0014] Preferably, the displacement sensing module is specifically implemented as follows:

[0015] Use laser displacement sensor or capacitive displacement sensor to detect the relative position between the sampling head and the object surface in real time;

[0016] During the sampling process, the displacement data is segmented according to the time series, the cumulative value of each segment is calculated and fused with the particle sampling number to generate a particle distribution density map;

[0017] If the fluctuation amplitude of a certain segment of displacement data exceeds the preset fluctuation threshold, it is marked as an abnormal displacement segment and triggers the adaptive control module to recalibrate the sampling parameters.

[0018] Preferably, the synchronous control process of the dual pump module includes:

[0019] When the suction pump is started, the blowing pump is started after a delay of T1 to ensure that the suction airflow is stable before performing particle discharge;

[0020] During the operation of the air pump, its exhaust pressure value is monitored in real time. If the deviation between the exhaust pressure and the preset pressure value exceeds the threshold, a pump pressure abnormality signal is generated and fed back to the adaptive control module to adjust the air pump power.

[0021] Preferably, the particle counting implementation method of the particle counting module includes:

[0022] Step A1: receiving real-time displacement data and particle sampling data from the displacement sensing module, and dividing the displacement data into continuous time series segments based on a preset time interval;

[0023] Step A2: Based on the moving distance, speed, and direction of each displacement segment, the time series segments are converted into corresponding surface detection regions through a region mapping algorithm, and the area of ​​each detection region is calculated. The region density conversion algorithm is implemented as follows:

[0024] Divide the displacement data into several sampling units, each unit corresponds to a surface area;

[0025] Count the number of particle samples in each unit and calculate the local particle density based on the unit area;

[0026] The local particle density of all units is weighted averaged to generate the overall particle distribution density value, and abnormal units with discrete values ​​exceeding the preset range are eliminated;

[0027] Step A3: Count the number of particle samples in each detection area and calculate the local particle density value based on the area of ​​the corresponding area;

[0028] Step A4: Perform spatial interpolation processing on the local particle density values ​​of adjacent detection areas to generate a particle distribution density map covering the entire detection range;

[0029] Step A5: Identify abnormal areas in the particle distribution density map where the density difference between adjacent areas exceeds a preset threshold, and perform smoothing correction on the particle density values ​​in the abnormal areas based on the spatial continuity characteristics of historical sampling data;

[0030] Step A6: Output the corrected particle distribution density map and particle statistics per unit area.

[0031] Preferably, the specific operation mode of the adaptive control module is:

[0032] When replacing the sampling head, the identification code stored in the chip is read through radio frequency identification technology, including the sampling head type, applicable surface morphology and sampling parameter range;

[0033] The identification code is input into the AI ​​algorithm model to match the corresponding suction pump power curve, displacement sensor sensitivity and particle count correction coefficient, and automatically switch to the optimal sampling mode.

[0034] Preferably, the sampling head includes three types: flat type, curved type and flexible type;

[0035] The planar sampling head is made of rigid material and is suitable for flat surfaces. Its sampling parameters include a fixed aspiration rate and a linear displacement calibration coefficient.

[0036] The curved surface sampling head has a built-in adjustable bracket and is suitable for regular curved surfaces. Its sampling parameters include dynamic suction rate and curvature compensation coefficient;

[0037] The flexible sampling head is made of silicone and is suitable for irregular surfaces. Its sampling parameters include adaptive suction power and displacement fluctuation tolerance range.

[0038] Preferably, the AI ​​algorithm model adopts a neural network algorithm based on dynamic weight allocation, which specifically includes the following steps:

[0039] Step B1: Encode the sampling head type in the identification code into a three-dimensional orthogonal vector ,in Boolean activation values ​​corresponding to planar, curved, and flexible types respectively;

[0040] Step B2: Constructing surface morphology feature matrix ,in is the surface roughness grade scalar, is the curvature eigenvector, 、 is a type-dependent adaptive coefficient that satisfies ;

[0041] Step B3: Parameter fusion is performed through a dual-channel convolution kernel, where:

[0042] Power curve parameters ;

[0043] Sensitivity coefficient ;

[0044] Correction factor ;

[0045] in represents a three-dimensional tensor convolution, represents the Hadamard product, represents the activation function, represents the linear rectification function, represents the hyperbolic tangent function, is the temperature coefficient, Respectively represent the weight matrices related to the calculation of power curve parameters, sensitivity coefficients, and correction coefficients; Represents a vector With the matrix Vertical splicing, Represents a vector With the matrix The Kronecker product, Represents a vector With the matrix Horizontal splicing; represents the bias term;

[0046] Step B4: Establish a dynamic loss function ,in The deviation value between the output parameter and the preset parameter library; optimize the weight matrix through back propagation , so that the output parameters deviate from the preset parameter library Minimize the weighted sum of squares of is the weight coefficient.

[0047] Preferably, an abnormality feedback module is also included, and its operation mode is as follows:

[0048] During the particle counting process, if N consecutive abnormal displacement segments or particle density values ​​exceeding the preset reasonable range are detected, a system abnormality signal is generated, where N is a preset positive integer constant;

[0049] The system abnormal signal triggers the adaptive control module to suspend the current sampling process and start the self-test program to test the dual pump module, displacement sensor module and particle counting module item by item.

[0050] Preferably, the specific steps of the self-test program include:

[0051] Perform pressure pulse tests on the suction pump and the blow pump. If the pulse response time exceeds the preset threshold, it is determined to be a pump failure.

[0052] Perform zero drift calibration on the displacement sensor module. If the drift value after calibration still exceeds the tolerance range, the sensor is considered to be faulty.

[0053] Perform a standard particle sample test on the particle counting module. If the counting error exceeds the allowable value, reload the particle density conversion algorithm.

[0054] Preferably, a particle counting monitoring method based on the particle counting monitoring device combining the above dual pump system with displacement sensing technology comprises the following steps:

[0055] The suction pump and the blowing pump in the dual-pump module are used to perform particle suction and discharge operations respectively through independent channels, and the start and stop timing of the two are coordinated by a synchronous controller;

[0056] The displacement sensing module collects the displacement data of the sampling head in real time, including the moving distance, speed and direction, and transmits the displacement data to the particle counting module;

[0057] The particle counting module calculates the particle distribution density per unit area using the regional density conversion algorithm based on the received displacement data and particle sampling data;

[0058] An adaptive control module is used to obtain identification information through a chip embedded in the sampling head, and a preset AI algorithm model is used to determine the optimal sampling mode of the sampling head based on the identification information.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The particle counting monitoring device and method proposed in the present invention, which combines a dual-pump system with displacement sensing technology, have many significant beneficial effects compared to traditional technologies. In terms of detection accuracy, the design of the dual-pump module greatly improves the stability of particle collection and discharge. The suction pump and the blow pump operate through independent channels, and the start and stop timings are coordinated by a synchronous controller, avoiding the problem of unstable airflow in a single-pump system. After the suction pump is started, the blow pump is delayed for T1 time to ensure that the particles are discharged after the suction airflow stabilizes, preventing particles from remaining in the device or being sucked in again, and ensuring that each collected particle can be accurately counted. When the blow pump is running, the exhaust pressure is monitored in real time. If the pressure deviation exceeds the threshold, the adaptive control module promptly adjusts the blow pump power to further ensure the smoothness and stability of particle discharge, making the particle counting results more reliable.

[0061] The displacement sensing module uses a laser or capacitive displacement sensor to collect real-time, high-precision displacement data of the sampling head, including distance, speed, and direction. During the sampling process, the displacement data is segmented and processed according to time series and fused with the particle count. The resulting particle distribution density map intuitively and accurately reflects the distribution of particles on the surface. If the fluctuation amplitude of a segment of displacement data exceeds a preset threshold, it is promptly marked as an abnormal displacement segment, triggering the adaptive control module to recalibrate the sampling parameters. This effectively avoids particle counting errors caused by abnormal displacement and improves detection accuracy.

[0062] The particle counting module's regional density conversion algorithm has been carefully designed to comprehensively consider both displacement data and particle sampling data. By dividing the displacement data into sampling units, calculating the local particle density of each unit and performing a weighted average, while simultaneously eliminating outliers, the calculated overall particle distribution density value is more accurate. When generating the particle distribution density map, the local particle density values ​​of adjacent detection areas are spatially interpolated, and the particle density values ​​in outlier areas are smoothed based on the spatial continuity characteristics of historical sampling data. This further improves the accuracy and reliability of the map and provides users with more precise particle distribution information.

[0063] The adaptive control module uses the identification information of the chip embedded in the sampling head to determine the optimal sampling mode through a preset AI algorithm model, demonstrating powerful intelligent adaptability. When replacing different types of sampling heads (flat, curved, and flexible), it can automatically match the corresponding suction pump power curve, displacement sensor sensitivity, and particle count correction factor. The flat sampling head is suitable for flat surfaces. Its fixed suction rate and linear displacement calibration coefficient ensure efficient and accurate sampling on flat surfaces. The curved sampling head has a built-in adjustable bracket, and its dynamic suction rate and curvature compensation coefficient enable optimized sampling based on the characteristics of regular curved surfaces. The flexible sampling head is made of silicone, and its adaptive suction power and displacement fluctuation tolerance range make it well adapted to irregular surfaces. This adaptive capability greatly improves the device's adaptability to different detection scenarios, ensuring high-precision and high-efficiency detection regardless of the complexity of the detection surface morphology.

[0064] Furthermore, the abnormality feedback module and self-test routine provide strong assurance for the device's stable operation. During the particle counting process, if N consecutive abnormal displacement segments or particle density values ​​outside the preset acceptable range are detected, a system abnormality signal is immediately generated, triggering the adaptive control module to suspend sampling and initiate a self-test routine. Each component of the dual pump module, displacement sensor module, and particle counting module is individually tested to promptly identify and resolve potential issues such as pump failure, sensor failure, and algorithm errors, ensuring the device is always in optimal working condition and reducing detection errors and production delays caused by equipment failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 This is a working principle diagram of the particle counting monitoring device of the present invention;

[0066] Figure 2 Flowchart for particle counting module counting and graph generation;

[0067] Figure 3 Schematic diagram of the working characteristics and detection of different types of sampling heads;

[0068] Figure 4Schematic diagram for calculating and optimizing sampling parameters for AI algorithm models. DETAILED DESCRIPTION

[0069] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0070] See also Figures 1-4 The present invention provides a particle counting monitoring device and method that combines a dual-pump system with displacement sensing technology. The particle counting monitoring device mainly consists of a dual-pump module, a displacement sensing module, a particle counting module, and an adaptive control module.

[0071] The dual-pump module is responsible for both particle intake and exhaust within the device. It comprises an air suction pump and an air blow pump, each operating through independent channels, with their start and stop timing precisely coordinated by a synchronous controller. During operation, the air suction pump activates first, drawing particle-laden gas into the device. The air blow pump then activates, expelling the absorbed particles, ensuring continuous and stable particle sampling.

[0072] The displacement sensing module is responsible for collecting real-time displacement data on the sampling head. The data collected covers movement distance, speed, and direction. During the sampling process, the module uses specific sensors to monitor the relative position of the sampling head and the surface of the object in real time. For example, when inspecting particles on the surface of an industrial product, the displacement sensing module can continuously capture the displacement information of the sampling head as it moves across the surface. This data is then transmitted to the particle counting module, providing critical foundational data for subsequent particle counting and distribution density calculations.

[0073] The particle counting module receives displacement data and particle sampling data from the displacement sensing module and calculates the particle distribution density per unit area using an area density conversion algorithm. It first processes the received data, determines the corresponding detection area based on the displacement data, and then, based on the number of particles sampled within each area, calculates the particle distribution density per unit area through a series of calculations, providing users with intuitive particle distribution information.

[0074] The adaptive control module obtains identification information from a chip embedded in the sampling head. When the sampling head is installed on the device, the module reads the information in the chip and, using a pre-set AI algorithm model, determines the optimal sampling mode for the sampling head based on this identification information. This allows the device to adapt to different types of sampling heads and detection scenarios, improving detection accuracy and efficiency.

[0075] The present invention will be further described below in conjunction with Examples 1 to 6:

[0076] Example 1:

[0077] In this embodiment, the specific implementation method of the synchronous control of the displacement sensing module and the dual pump module is further described.

[0078] The displacement sensing module uses a laser displacement sensor to detect the relative position of the sampling head and the object surface in real time. The laser displacement sensor emits a laser beam and measures the time it takes for the laser beam to reflect back to the sensor. By leveraging the constant speed of light, the sensor accurately calculates the distance between the sampling head and the object surface. When detecting particles on a flat surface, the laser displacement sensor can capture the trajectory of the sampling head in real time, obtaining information on its movement distance, speed, and direction.

[0079] During the sampling process, the displacement data is segmented and processed according to the time series. Assuming the sampling time is 10 minutes, with each minute being a segment, the displacement data of these 10 minutes is divided into 10 segments. The cumulative value of each segment of displacement is then calculated. For example, in the first period of time, the sampling head moved 5 cm in the X direction and 3 cm in the Y direction. The cumulative displacement value calculated by the Pythagorean theorem is approximately 5.83 cm. At the same time, the number of particle samples in each period of time is recorded, and the cumulative displacement value is fused with the number of particle samples. Using specialized data processing software, a particle distribution density map is generated with the cumulative displacement value as the horizontal axis and the number of particle samples as the vertical axis, which intuitively displays the particle distribution at different locations.

[0080] If the fluctuation amplitude of a segment of displacement data exceeds a preset fluctuation threshold—for example, if the preset fluctuation threshold is 1 cm and the sampling head displacement fluctuates by 1.5 cm within a certain period—this segment is marked as an abnormal displacement segment. Once marked as an abnormal displacement segment, the adaptive control module is triggered to recalibrate the sampling parameters. Based on a pre-set algorithm, the adaptive control module adjusts parameters such as the sampling head's movement speed and sampling frequency to ensure the accuracy of subsequent sampling data.

[0081] For the synchronous control of the dual pump modules, when the suction pump starts, the blowing pump starts after a delay of T1 time. The T1 time is set based on actual experiments and experience. For example, after many tests, it was found that when T1 is set to 5 seconds, it can ensure that the suction airflow is stable before performing the particle discharge operation. During the operation of the blowing pump, its exhaust pressure value is monitored in real time by a pressure sensor. If the deviation between the exhaust pressure and the preset pressure value exceeds the threshold, for example, the preset pressure value is 50kPa and the threshold is ±5kPa, when the exhaust pressure is detected to be 56kPa, a pump pressure abnormality signal is generated and fed back to the adaptive control module. After receiving the signal, the adaptive control module adjusts the power of the blowing pump according to the preset algorithm to ensure stable operation of the blowing pump and ensure that the particles can be smoothly discharged from the device.

[0082] In practical applications, such as detecting particles on chip surfaces in electronic chip manufacturing plants, the synchronized control of the displacement sensing module and dual-pump module ensures stable and accurate detection. The delicate surface of the chip requires high precision in displacement detection, a requirement that laser displacement sensors can meet. Furthermore, the stable operation of the dual-pump module ensures the effective collection and removal of particles from the chip surface, preventing residual particles from affecting test results.

[0083] Example 2:

[0084] This embodiment describes in detail a particle counting implementation method of a particle counting module.

[0085] When the particle counting module performs particle counting, the steps include:

[0086] Step A1: Receive real-time displacement data and particle sampling data from the displacement sensing module and segment the displacement data into continuous time series segments based on a preset time interval. For example, assume the preset time interval is 10 seconds. During this 10-second period, the displacement sensing module continuously collects displacement data from the sampling head. For example, during this 10-second period, the sampling head moves 3 cm in the X direction and 2 cm in the Y direction, and the particle sampling device collects 10 particles. The particle counting module processes this 10-second period of data as a segment.

[0087] Step A2: Based on the distance, speed, and direction of each displacement segment, a regional mapping algorithm is used to convert the time series segments into corresponding surface detection regions. The area of ​​each detection region is then calculated. The regional density conversion algorithm is implemented as follows: The displacement data is divided into several sampling units, assuming that each sampling unit corresponds to a square surface region with a side length of 1 cm. The number of particles sampled within each unit is counted. For example, if 3 particles are collected within a sampling unit with an area of ​​1 cm², the local particle density of that unit is 3 particles / cm². The local particle densities of all units are weighted averaged to generate an overall particle distribution density value. During the weighted averaging process, different units are weighted according to their importance, for example, units near the detection center are given higher weights. Furthermore, outliers are eliminated if the discrete value exceeds a preset range. Assuming the preset range is ±50% of the local particle density average, if the local particle density of a unit is 10 particles / cm² and the overall average is 3 particles / cm², the unit density is out of range and is eliminated.

[0088] Step A3: Count the number of particles sampled within each detection zone and calculate the local particle density based on the area of ​​the corresponding zone. For example, if a detection zone is 5 square centimeters and 15 particles are collected, the local particle density for that zone is 3 particles / square centimeter.

[0089] Step A4: Spatially interpolate the local particle density values ​​of adjacent detection areas to generate a particle distribution density map covering the entire detection range. This spatial interpolation algorithm supplements the particle density information in the intermediate locations between adjacent areas, making the map smoother and more accurate in reflecting the particle distribution across the entire detection area.

[0090] Step A5: Identify abnormal regions in the particle distribution density map where the density difference between adjacent regions exceeds a preset threshold. Smooth the particle density values ​​in these abnormal regions based on the spatial continuity of the historical sampling data. Assuming the preset threshold is 2 particles / cm², if the particle density difference between two adjacent regions reaches 3 / cm², the region is identified as abnormal. Using the particle density variation patterns surrounding this region in the historical sampling data, adjust the particle density values ​​in these abnormal regions to better reflect actual conditions.

[0091] Execute step A6: Output the corrected particle distribution density map and particle count per unit area. This data can be displayed visually to the user on a display screen or transmitted to other devices for further analysis. In practical applications, such as detecting foreign particles on the surface of food packaging in a food processing workshop, the particle counting module's workflow can accurately determine the particle distribution on the packaging surface, helping companies promptly identify product quality issues.

[0092] Example 3:

[0093] In the entire particle counting monitoring device, the adaptive control module and the sampling head work together to enable the device to automatically adjust to the optimal sampling mode according to different detection objects and scenarios, thereby ensuring the accuracy and efficiency of the detection results.

[0094] When the sampling head needs to be replaced, the adaptive control module plays a key role. It uses radio frequency identification (RFID) technology to read the identification code stored on a chip embedded in the sampling head. RFID technology uses radio frequency signals to read the identification code contactlessly, offering convenient operation, fast reading speed, and high accuracy. The identification code contains a wealth of information, including the sampling head type, applicable surface morphology, and sampling parameter range. For example, when a new sampling head is installed, the adaptive control module quickly reads the identification code and determines that the sampling head is curved. This information provides crucial information for subsequent adjustments to the sampling mode.

[0095] In actual application scenarios, different types of sampling heads are suitable for different detection surfaces. The planar sampling head is made of rigid material and has a strong and stable structure, which enables it to maintain good stability and sealing when attached to a flat surface. When performing particle detection on the surface of a printed circuit board, the planar sampling head can fit tightly to the surface of the circuit board, ensuring effective particle collection and reducing external interference. Its sampling parameters include a fixed aspiration rate and a linear displacement calibration coefficient. The fixed aspiration rate is set based on research on the particle characteristics of the printed circuit board surface and actual detection experience, for example, it is set to 4 liters / minute. Such an aspiration rate can ensure that the particles on the circuit board surface are fully aspirated into the detection device without causing damage to the circuit board or affecting the detection accuracy due to excessive suction. The linear displacement calibration coefficient is designed to compensate for the displacement error that may occur when the sampling head moves on the plane. It is obtained through analysis and calibration of a large amount of experimental data to ensure the accuracy of the displacement data, thereby improving the accuracy of the particle distribution density calculation.

[0096] The curved sampling head features a built-in adjustable bracket, allowing it to flexibly adapt to regular curved surfaces. For example, when inspecting the curved surface of an automobile engine cylinder block, the adjustable bracket can be adjusted to the cylinder block's curved shape. Before inspection, the operator can manually or automatically adjust the bracket's angle and position to match the cylinder block surface, ensuring the sampling head's optimal fit and comprehensiveness. Sampling parameters include a dynamic aspiration rate and a curvature compensation coefficient. The dynamic aspiration rate adjusts in real time based on the surface's location and curvature. In areas with greater curvature, where particle distribution may be more complex, the aspiration rate is appropriately increased to ensure sufficient particle collection. In areas with less curvature, the aspiration rate is reduced to avoid oversampling or surface damage. The curvature compensation coefficient compensates for inspection errors caused by the curved surface. By accurately measuring and analyzing the cylinder block's surface geometry and integrating extensive experimental data, an appropriate curvature compensation coefficient is determined, enabling the calculation of particle distribution density to more accurately reflect the actual particle situation on the curved surface.

[0097] The flexible sampling head is made of silicone, which offers excellent flexibility and elasticity, allowing it to adapt to various irregular surfaces. This advantage is particularly evident when inspecting particles on complex surfaces. It can closely conform to the various concave and convex shapes of the surface, capturing every area where particles may be present. Its sampling parameters include adaptive suction power and displacement fluctuation tolerance. Adaptive suction power automatically adjusts based on surface conditions. In areas with prominent features, suction power is increased to ensure particle collection; in concave areas, suction power is reduced to prevent surface damage or excessive impurities from interfering with detection results. The displacement fluctuation tolerance allows for a certain range of fluctuations in the sampling head's displacement. Due to the irregularities of the surface, it is difficult for the sampling head to remain absolutely stable during movement. Setting a reasonable displacement fluctuation tolerance prevents minor fluctuations from being misidentified as anomalies, ensuring the stability and reliability of the detection process.

[0098] After obtaining the sampling head's identification code, the adaptive control module inputs it into the AI ​​algorithm model. Based on the information in the identification code, the AI ​​algorithm model quickly matches the corresponding aspirator pump power curve, displacement sensor sensitivity, and particle count correction factor, automatically switching to the optimal sampling mode. For example, for a curved sampling head, the AI ​​algorithm model adjusts the aspirator pump power curve based on the relevant information in the identification code. In areas with greater curvature, the aspirator pump power is increased to improve the aspiration rate; in areas with less curvature, the aspirator pump power is reduced to ensure sampling stability. Simultaneously, the displacement sensor sensitivity is improved, enabling the displacement sensor module to more accurately detect subtle changes in the sampling head's displacement on the curved surface, providing more reliable displacement data for accurate calculation of particle distribution density. Furthermore, the particle count correction factor is adjusted to compensate for the impact of curved surface shape on particle counts, ensuring that the final detection results accurately reflect the particle distribution on the curved surface. In this way, the adaptive control module works in conjunction with different types of sampling heads, enabling the particle counting monitoring device to efficiently and accurately complete particle detection tasks on various surfaces, meeting the detection needs of different industries.

[0099] Example 4:

[0100] This embodiment is used to describe a neural network algorithm based on dynamic weight allocation in an AI algorithm model. This algorithm plays a core role in determining the optimal sampling mode in the entire particle counting monitoring device.

[0101] In step B1, the sampling head type in the identification code is encoded as a three-dimensional orthogonal vector .in, 、 、 The Boolean activation values ​​corresponding to the flat, curved, and flexible sampling heads. For example, if the flat sampling head is currently used, then , , ; If it is a curved sampling head, then , , ; For flexible sampling head, , , This encoding method can present the characteristics of different types of sampling heads in a concise and easy-to-calculate vector form, providing basic data support for subsequent steps.

[0102] Step B2: Constructing the surface morphology feature matrix .here, It represents the surface roughness grade scalar, which is obtained through a professional surface roughness measuring instrument. The larger the value, the rougher the surface of the object. For example, after measurement, the surface roughness grade scalar of an object is is 5. It is the curvature eigenvector, which describes the curvature of the object surface and can be calculated by analyzing the geometric shape of the object surface. 、 As type-dependent adaptation coefficients, ,in is a vector The modulus of Calculated. Assume ,but , , The value of The results and actual experience are used to determine it, so as to ensure that the surface morphology feature matrix can be accurately constructed so that it can accurately reflect the comprehensive characteristics of the object surface.

[0103] Go to step B3 and perform parameter fusion through dual-channel convolution kernel. , It is a three-dimensional tensor convolution operation, which can perform convolution operations on data in three-dimensional space and extract the features of the data. It is a weight matrix related to the calculation of power curve parameters. Its value is optimized through a large number of experiments and data training, and determines the degree of influence of input data on power curve parameters. Represents a vector With the matrix The longitudinal splicing integrates the sampling head type information and surface morphological feature information. It is a bias term used to adjust the calculation results to avoid model deviation. As an activation function, a common one is the Sigmoid function, which can map the calculation results to interval, so that the model has nonlinear expression capabilities and the calculation results of power curve parameters are more in line with actual needs.

[0104] Sensitivity coefficient , is the Hadamard product, i.e., the multiplication of corresponding elements; is the weight matrix associated with the calculation of sensitivity coefficients; Represents a vector With the matrix Kronecker product; It is a linear rectification function. When the calculation result is greater than 0, the original value is output. When it is less than 0, 0 is output. This can filter out effective information and enhance the model's ability to capture key information, thereby accurately calculating the sensitivity coefficient that meets the actual detection scenario.

[0105] Correction factor , It is the hyperbolic tangent function, which can normalize the calculation results to interval, making the correction coefficient more reasonable. It is the temperature coefficient, which can adjust the convergence speed and stability during model training. Its optimal value is usually determined through multiple experiments during the model training process. is the weight matrix associated with the calculation of the correction coefficient; Represents a vector With the matrix For horizontal splicing, the correction coefficient is calculated by comprehensively considering the sampling head type and surface morphology characteristics.

[0106] In step B4, a dynamic loss function is established .here, 、 、 These are the deviations between the output values ​​of the power curve parameters, sensitivity coefficients, and correction coefficients and the corresponding standard values ​​in the preset parameter library. The preset parameter library is accumulated through extensive experiments and practical applications, and contains ideal parameter values ​​for different sampling head types and surface morphologies. 、 、 is the weight coefficient, which is set according to the importance of different parameters in the detection process. For example, in a high-precision detection scenario, the weight of the power curve parameter deviation may be increased. . Through the back propagation algorithm, the weight matrix is ​​continuously adjusted 、 、 , so that the loss function Minimization, or minimizing the weighted sum of squared deviations between the output parameters and the preset parameter library, optimizes the model and improves its accuracy and adaptability. In the field of electronic device manufacturing, when performing particle detection on chip surfaces, this AI algorithm model can precisely adjust sampling parameters based on the complex conditions of the chip surface, improving the accuracy and efficiency of particle detection.

[0107] Example 5:

[0108] During the particle counting process, the abnormal feedback module continuously monitors the data of the displacement sensor module and the particle counting module. If N consecutive abnormal displacement segments or particle density values ​​outside the preset reasonable range are detected, a system abnormality signal will be generated. Here, N is a positive integer constant pre-set based on actual application scenarios and experience. For example, when performing particle detection on the surface of a precision optical lens, due to the extremely high requirements for the surface quality of the lens, the particle distribution density should be strictly controlled within a certain range. Assume that the setting is . If the displacement sensing module detects that the displacement fluctuation amplitude of the sampling head exceeds the preset fluctuation threshold for three consecutive times, for example, the preset fluctuation threshold is 0.5 mm, and the actual displacement fluctuation amplitude detected for three consecutive times is 0.6 mm, 0.7 mm, and 0.65 mm respectively, or the particle density value calculated by the particle counting module exceeds the preset reasonable range for three consecutive times, assuming that the preset reasonable range of particle density is 0-5 particles / square centimeter, and the actual calculated particle density values ​​are 6 particles / square centimeter, 7 particles / square centimeter, and 6.5 particles / square centimeter for three consecutive times, then the abnormal feedback module will quickly generate a system abnormality signal.

[0109] This system anomaly signal immediately triggers the adaptive control module to pause the current sampling process. Upon receiving the signal, the module immediately stops the suction and blowdown pumps, and simultaneously halts data collection from the particle counter and displacement sensor modules to prevent further accumulation of erroneous data. It then initiates a self-test, performing a check on each of the dual pump module, displacement sensor module, and particle counter module.

[0110] When testing the dual-pump module, a pressure pulse test is performed by sending pressure pulse signals of a specific frequency and intensity to the suction pump and the blow pump. For example, a pressure pulse signal with a frequency of 20 Hz and an intensity of 60 kPa is sent. If the pulse response time of the suction pump or the blow pump exceeds the preset threshold (assuming the preset threshold is 0.4 seconds, and the actual pulse response time detected for a pump is 0.6 seconds), then the pump is determined to be faulty. Pump failure may be caused by impeller wear, pipe blockage, or poor sealing. Once a pump failure is discovered, timely repair or replacement of the relevant components is required to ensure the normal operation of the dual-pump module and the stable suction and discharge of particles.

[0111] Perform zero drift calibration on the displacement sensor module. Ideally, when the sampling head is stationary, the output value of the displacement sensor module should be 0. However, in actual use, factors such as sensor aging and environmental interference may cause the output value to deviate, i.e., zero drift. Use professional calibration equipment to calibrate the displacement sensor module. If the drift value after calibration still exceeds the tolerance range, for example, if the tolerance range is set to ±0.05 mm and the drift value after calibration is 0.1 mm, the sensor is considered to have failed. Sensor failure will seriously affect the accuracy of the displacement data, and thus affect the calculation results of particle counts and distribution density, so a new sensor needs to be replaced in a timely manner.

[0112] Perform a standard particle sample test on the particle counting module. Prepare a standard sample with a known number and distribution of particles, place it in the detection device, and let the particle counting module perform the detection. If the counting error exceeds the allowable value, for example, the allowable counting error is ±3%, and the actual counting error reaches 5%, reload the particle density conversion algorithm. Reloading the algorithm can attempt to repair counting error problems caused by algorithm operation errors, data anomalies, or unreasonable parameter settings, to ensure that the particle counting module can accurately calculate the particle distribution density. In the pharmaceutical industry, when performing particle detection on the surface of drug packaging, the effective operation of the abnormal feedback module can promptly discover problems in the detection process, ensure the quality and safety of drugs, and prevent unqualified products from entering the market.

[0113] Example 6:

[0114] During the pressure pulse test of the dual-pump module, a dedicated pressure pulse generator sends pressure pulse signals to the suction and blow pumps. The frequency and intensity of the pressure pulse signals are carefully set based on the pump model, specifications, and design requirements. For example, for a certain suction and blow pump model, after technical analysis and experimental verification, the pressure pulse frequency is set to 15 Hz and the intensity to 55 kPa. After the pressure pulse signal is sent, the pump's pulse response time is accurately recorded using high-precision pressure sensors and time measurement equipment. Wear on mechanical components within the pump, such as the impeller and bearings, or blockages or leaks in the pipelines, can reduce the pump's responsiveness and increase the pulse response time. If the pulse response time exceeds the preset threshold—for example, if the preset threshold is 0.5 seconds and the actual pulse response time for a pump is 0.7 seconds—the pump is considered faulty. Once a pump fault is confirmed, maintenance personnel will perform repairs based on the specific situation. If the impeller is worn, replace it with a new one; if the pipe is blocked, clean the pipe; if the seal is damaged and causes leakage, replace the seal to ensure that the pump body returns to normal working condition and ensure the efficiency and stability of particle transportation.

[0115] For the zero drift calibration of the displacement sensing module, a professional calibration platform is used. The displacement sensing module is installed on the calibration platform, and when the calibration platform is in a stationary state, the output value of the displacement sensing module is measured. Under normal circumstances, the output value at this time should be 0, but due to the influence of various factors, deviations may occur. If the drift value after calibration still exceeds the tolerance range, for example, the tolerance range is set to ±0.08 mm, and the drift value measured after calibration is 0.12 mm, the sensor is judged to have failed. Sensor failure will cause errors in the collected displacement data, resulting in deviations in the subsequent calculated particle distribution density, affecting the accuracy of the test results. Therefore, once the sensor is determined to have failed, it is necessary to replace it with a new sensor of the same model and specification in a timely manner to ensure the reliability of the displacement data.

[0116] When performing a standard particle sample test on a particle counting module, a standard particle sample is selected. The particle count and distribution of the standard particle sample have been rigorously measured and verified, ensuring high accuracy and reliability. The standard particle sample is placed in the detection device, and the particle counting module is activated for testing. The counting error is calculated by comparing the particle counting module's test results with the actual count of the standard particle sample. If the counting error exceeds the allowable value (assuming the allowable counting error is ±4% and the actual counting error reaches 6%), the particle density conversion algorithm is reloaded. This algorithm reloading process includes reinitializing the algorithm parameters and reading the latest calibration data. After reloading the algorithm, the standard particle sample is tested again to verify whether the counting error is within the allowable range. If it still does not meet the requirements, further inspection of the algorithm's code logic or data transmission process may be necessary, or the particle counting module hardware may need to be inspected and repaired to ensure that the particle counting module can accurately calculate particle distribution density. In the aerospace component manufacturing industry, when performing particle testing on component surfaces, rigorous self-test procedures effectively safeguard the accuracy and reliability of the detection device, ensuring that component quality meets high standards and safeguards the safe operation of aerospace equipment.

[0117] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0118] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A particle counting monitoring device combining a dual pump system with displacement sensing technology, characterized in that: Includes dual pump module, displacement sensing module, particle counting module and adaptive control module; The dual-pump module includes an air suction pump and an air blowing pump, which respectively perform particle suction and discharge operations through independent channels, and the start and stop timing of the two are coordinated by a synchronous controller; The displacement sensing module is used to collect displacement data of the sampling head in real time, including moving distance, speed and direction; The particle counting module calculates the particle distribution density per unit area using an area density conversion algorithm based on the displacement data and the particle sampling data; The adaptive control module obtains identification information through a chip embedded in the sampling head, and determines the optimal sampling mode of the sampling head based on the identification information through a preset AI algorithm model; The specific operation mode of the adaptive control module is: When the sampling head is replaced, the identification code stored in the chip is read through radio frequency identification technology, including the sampling head type, applicable surface morphology and sampling parameter range; The identification code is input into the AI ​​algorithm model to match the corresponding suction pump power curve, displacement sensor sensitivity and particle count correction coefficient, and automatically switch to the optimal sampling mode.

2. The particle counting monitoring device according to claim 1, characterized in that The specific implementation of the displacement sensing module is as follows: Use laser displacement sensor or capacitive displacement sensor to detect the relative position between the sampling head and the object surface in real time; During the sampling process, the displacement data is segmented according to the time series, the cumulative value of each segment is calculated and fused with the particle sampling number to generate a particle distribution density map; If the fluctuation amplitude of a certain segment of displacement data exceeds the preset fluctuation threshold, it is marked as an abnormal displacement segment and triggers the adaptive control module to recalibrate the sampling parameters.

3. The particle counting monitoring device according to claim 1, characterized in that The synchronous control process of the dual pump module includes: When the suction pump is started, the blowing pump is started after a delay of T1 to ensure that the suction airflow is stable before performing particle discharge; During the operation of the air pump, its exhaust pressure value is monitored in real time. If the deviation between the exhaust pressure and the preset pressure value exceeds the threshold, a pump pressure abnormality signal is generated and fed back to the adaptive control module to adjust the air pump power.

4. The particle counting monitoring device according to claim 1, characterized in that The particle counting implementation method of the particle counting module includes: Step A1: receiving real-time displacement data and particle sampling data from the displacement sensing module, and dividing the displacement data into continuous time series segments based on a preset time interval; Step A2: Based on the moving distance, speed, and direction of each displacement segment, the time series segments are converted into corresponding surface detection regions through a region mapping algorithm, and the area of ​​each detection region is calculated. The region density conversion algorithm is implemented as follows: Divide the displacement data into several sampling units, each unit corresponds to a surface area; Count the number of particle samples in each unit and calculate the local particle density based on the unit area; The local particle density of all units is weighted averaged to generate the overall particle distribution density value, and abnormal units with discrete values ​​exceeding the preset range are eliminated; Step A3: Count the number of particle samples in each detection area and calculate the local particle density value based on the area of ​​the corresponding area; Step A4: Perform spatial interpolation processing on the local particle density values ​​of adjacent detection areas to generate a particle distribution density map covering the entire detection range; Step A5: Identify abnormal areas in the particle distribution density map where the density difference between adjacent areas exceeds a preset threshold, and perform smoothing correction on the particle density values ​​in the abnormal areas based on the spatial continuity characteristics of historical sampling data; Step A6: Output the corrected particle distribution density map and particle statistics per unit area.

5. The particle counting monitoring device according to claim 1, characterized in that: The sampling head includes three types: flat type, curved type and flexible type; The planar sampling head is made of rigid material and is suitable for flat surfaces. Its sampling parameters include a fixed aspiration rate and a linear displacement calibration coefficient. The curved surface sampling head has a built-in adjustable bracket and is suitable for regular curved surfaces. Its sampling parameters include dynamic suction rate and curvature compensation coefficient; The flexible sampling head is made of silicone and is suitable for irregular surfaces. Its sampling parameters include adaptive suction power and displacement fluctuation tolerance range.

6. The particle counting monitoring device according to claim 5, characterized in that: The AI ​​algorithm model adopts a neural network algorithm based on dynamic weight allocation, which specifically includes the following steps: Step B1: Encode the sampling head type in the identification code into a three-dimensional orthogonal vector ,in Boolean activation values ​​corresponding to planar, curved, and flexible types respectively; Step B2: Constructing surface morphology feature matrix ,in is the surface roughness grade scalar, is the curvature eigenvector, 、 is a type-dependent adaptive coefficient that satisfies ; Step B3: Parameter fusion is performed through a dual-channel convolution kernel, where: Power curve parameters ; Sensitivity coefficient ; Correction factor ; in represents a three-dimensional tensor convolution, represents the Hadamard product, represents the activation function, represents the linear rectification function, represents the hyperbolic tangent function, is the temperature coefficient, Respectively represent the weight matrices related to the calculation of power curve parameters, sensitivity coefficients, and correction coefficients; Represents a vector With the matrix Vertical splicing, Represents a vector With the matrix The Kronecker product, Represents a vector With the matrix Horizontal splicing; represents the bias term; Step B4: Establish a dynamic loss function ,in The deviation value between the output parameter and the preset parameter library; optimize the weight matrix through back propagation , so that the output parameters deviate from the preset parameter library Minimize the weighted sum of squares of is the weight coefficient.

7. The particle counting monitoring device according to claim 1, characterized in that: It also includes an exception feedback module, which operates as follows: During the particle counting process, if N consecutive abnormal displacement segments or particle density values ​​exceeding the preset reasonable range are detected, a system abnormality signal is generated, where N is a preset positive integer constant; The system abnormal signal triggers the adaptive control module to suspend the current sampling process and start the self-test program to test the dual pump module, displacement sensor module and particle counting module item by item.

8. The particle counting monitoring device according to claim 7, characterized in that: The specific steps of the self-test procedure include: Perform pressure pulse tests on the suction pump and the blow pump. If the pulse response time exceeds the preset threshold, it is determined to be a pump failure. Perform zero drift calibration on the displacement sensor module. If the drift value after calibration still exceeds the tolerance range, the sensor is considered to be faulty. Perform a standard particle sample test on the particle counting module. If the counting error exceeds the allowable value, reload the particle density conversion algorithm.

9. A particle counting monitoring method based on the device according to any one of claims 1 to 8, characterized in that: The following steps are involved: The suction pump and the blowing pump in the dual-pump module are used to perform particle suction and discharge operations respectively through independent channels, and the start and stop timing of the two are coordinated by a synchronous controller; The displacement sensing module collects the displacement data of the sampling head in real time, including the moving distance, speed and direction, and transmits the displacement data to the particle counting module; The particle counting module calculates the particle distribution density per unit area using the regional density conversion algorithm based on the received displacement data and particle sampling data; Adopting an adaptive control module to obtain identification information through the chip embedded in the sampling head, and using a preset AI algorithm model to determine the optimal sampling mode of the sampling head based on the identification information; The specific operation mode of the adaptive control module is: When the sampling head is replaced, the identification code stored in the chip is read through radio frequency identification technology, including the sampling head type, applicable surface morphology and sampling parameter range; The identification code is input into the AI ​​algorithm model to match the corresponding suction pump power curve, displacement sensor sensitivity and particle count correction coefficient, and automatically switch to the optimal sampling mode.

Images