Component mounting quality detection method for chip mounter

Abstract

The invention relates to the technical field of image data processing, and discloses an element mounting quality detection method for a chip mounter, which comprises the following steps of: synchronously acquiring images before and after element mounting and a vacuum pressure signal of a mounting suction nozzle, compensating vibration by utilizing an inherent substrate background area in the images, obtaining an optical position deviation through image difference and mass center calculation, and calculating the surface mounting quality of the element according to the optical position deviation. And finally, logically associating the deviation with the change characteristic of the pressure signal, and classifying the physical process of the mounting event. According to the invention, through the collaborative diagnosis of optical detection and physical signals, the physical reason of the mounting event can be clearly distinguished; the inherent limitation caused by vibration interference and unclear fault attribution during high-speed operation of a traditional detection mode is avoided, and the method has good adaptability to incoming material elements with different appearances.

Description

Technical Field

[0001] This invention relates to a method for inspecting the placement quality of components in a pick-and-place machine, belonging to the field of image data processing technology. Background Technology

[0002] In surface mount technology, using machine vision to inspect the position of components after they have been mounted is a common technique to ensure the reliability of circuit board functions. It usually works by identifying and measuring features of the components or their target pads, and then assessing the quality by calculating the difference between the geometric position obtained and the theoretical value.

[0003] However, given the current demand for large-scale and low-cost production in the electronics manufacturing industry, especially when dealing with a massive number of general-purpose passive components from multiple sources with batch-to-batch variations in appearance, the aforementioned detection methods rely on the identification of the component's inherent characteristics. This necessitates dealing with interference from non-critical information such as component color, silkscreen printing, and reflections, which in turn increases the complexity and cost of the detection algorithms and supporting lighting systems. This cost increase is particularly pronounced when pursuing higher detection accuracy, creating a persistent contradiction with the low-cost application scenarios of general-purpose components.

[0004] Specifically, existing technologies face a dilemma in practical applications: ensuring reliable detection requires investing in hardware and software costs disproportionate to the economic value of the detected object; conversely, it necessitates tolerating higher false positive rates due to variations in incoming materials or mechanical vibrations of the equipment during high-speed operation. Furthermore, simple optical image analysis often yields similar results for two physically distinct faults: significant component misalignment and complete component loss due to nozzle failure. This fails to provide attribution judgment for rapid troubleshooting in the production process. Therefore, the technical problem this invention aims to solve is to establish a low-cost detection method that eliminates the influence of component appearance variations and equipment vibrations, and whose results can clearly diagnose the physical process of placement events. Summary of the Invention

[0005] This invention provides a component placement quality inspection method for a pick-and-place machine. Its main purpose is to solve the problem of how to establish a low-cost inspection method that can eliminate the influence of component appearance changes and equipment operation vibrations during the inspection process, and whose inspection results can clearly diagnose the physical process of placement events.

[0006] To achieve the above objectives, this invention provides a component placement quality inspection method for a pick-and-place machine. This method establishes diagnostic logic that arbitrates optical inspection results using physical process signals to classify component placement events under high-speed operating conditions. The method includes: before the component is placed onto the target pad on the printed circuit board, simultaneously capturing a first image containing the target pad and the background area of ​​the printed circuit board substrate adjacent to the target pad using an imaging device, as a dynamic real-time reference image; after component placement is completed, capturing a second image containing the component, the target pad, and the substrate at the same position as the captured first image using the imaging device. The second image of the background area is obtained, and the vacuum pressure signal of the mounting nozzle is monitored synchronously during the component mounting process to obtain the pressure time-domain curve. Based on the substrate background area in the first and second images, the subpixel displacement between images caused by equipment vibration is calculated and compensated to generate an aligned second image. The first image and the aligned second image are differentially processed to generate a residual image. The centroid position of the bright area in the residual image is calculated to determine the planar mounting position deviation of the component. Based on the value of the planar mounting position deviation and the characteristics of the pressure time-domain curve, the physical process of the component mounting event is classified.

[0007] Preferably, the step of capturing the first image is performed when the placement head carrying the component descends to a predetermined imaging height, which allows the imaging device to be in the same focal plane as the final placement position without contacting the printed circuit board.

[0008] Preferably, before differential processing, the method further includes: before performing the mounting task, generating an ideal pad mask that defines the geometry of the target pad based on pre-stored computer-aided design data corresponding to the target pad; performing a logical AND operation between the first image and the ideal pad mask to avoid any image information in the first image located outside the geometry of the target pad; generating a cleaned reference image; the subsequent differential processing is performed using the cleaned reference image and the aligned second image; and performing a logical AND operation between the first image and the inversion of the ideal pad mask to detect whether there are foreign objects outside the target pad area, and outputting a warning signal when foreign objects are detected.

[0009] Preferably, the step of classifying the physical process of component placement events specifically includes: if the planar placement position deviation is greater than a predetermined offset threshold, and the pressure change rate of the pressure time-domain curve exceeds a predetermined transition slope threshold at the placement time, then the event is classified as high-offset placement; if the residual image shows that the pixel values ​​of the entire target pad area are higher than the brightness threshold, and the pressure value of the pressure time-domain curve remains within the adsorption pressure range after the placement time, then the event is classified as component loss or empty placement.

[0010] Preferably, the method further includes a step of diagnosing element coplanarity defects, performed after generating the residual image, and includes: calculating the intensity standard deviation of all pixels in the residual image whose pixel values ​​are higher than a background noise threshold. When the strength standard deviation If the predetermined coplanarity defect threshold is exceeded, the component is determined to have a coplanar defect.

[0011] Preferably, the method can also process optically transparent elements and includes: calculating the geometric residual energy and texture residual energy of the residual image; when the geometric residual energy is lower than a predetermined geometric threshold and the texture residual energy is higher than a predetermined texture threshold, determining that the current element is an optically transparent element, and switching to perform the following steps to evaluate the mounting quality: comparing the texture residual energy of the residual image with a reference texture energy set for the optically transparent element of this model, and evaluating the mounting quality based on the difference between the two.

[0012] Preferably, the method further includes a step of diagnosing the upstream solder paste printing process: at the initial stage of a production batch, the average pixel intensity of residual images generated by multiple components with planar mounting position deviations within the allowable tolerance range is recorded, and a residual intensity mean and residual intensity standard deviation are statistically calculated to jointly constitute the residual intensity historical fingerprint of the batch. For each residual image generated in subsequent production, its average pixel intensity is compared with the residual intensity historical fingerprint. When the average pixel intensity deviates from the residual intensity mean by more than a predetermined multiple of the residual intensity standard deviation, it is determined that there is an abnormality in the solder paste printing process.

[0013] Preferably, the step of calculating and compensating for subpixel-level displacement between images caused by device vibration specifically includes: extracting background image blocks corresponding to the substrate background area from the first image and the second image respectively, calculating the subpixel-level displacement vector between the two background image blocks using a phase correlation algorithm, and digitally translating the second image according to the displacement vector so as to align the substrate background areas of the two images before performing differential processing.

[0014] Preferably, the step of determining that there is an abnormality in the solder paste printing process further includes determining the type of abnormality: if the average pixel intensity is lower than the mean residual intensity and its deviation exceeds a predetermined multiple of the standard deviation of the residual intensity, it is determined that there is too much solder paste; if the average pixel intensity is higher than the mean residual intensity and its deviation exceeds a predetermined multiple of the standard deviation of the residual intensity, it is determined that there is too little solder paste or missing printing.

[0015] Preferably, the method further includes the step of performing systematic process control: continuously recording and storing a series of planar placement position deviation data determined by component placement, analyzing the statistical trend of the deviation data to determine whether there is a unidirectional offset trend that characterizes mechanical drift of the placement head or a periodic offset pattern that characterizes abnormal feeder status, and outputting instructions for predictive maintenance or equipment calibration based on the determined offset trend or offset pattern.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: 1. This invention establishes a detection method that logically correlates optical imaging detection with physical process signals. By acquiring images before and after placement while simultaneously monitoring the vacuum pressure signal of the placement nozzle, the detection system can not only determine the positional deviation of components on a two-dimensional plane, but also classify the physical process of the placement event itself based on the changing characteristics of the pressure signal. This allows for a clear distinction between similar optical results caused by completely different physical reasons, namely, high-offset placement and component loss or blank placement. This collaborative diagnostic mechanism of optical and physical information avoids the blind spots in fault attribution of traditional purely optical detection, making the location of equipment problems more accurate and providing a direct basis for subsequent process adjustments or equipment maintenance.

[0017] 2. This invention captures the target pad and its adjacent substrate background area, which is part of the printed circuit board body, in two imaging processes before and after mounting. This background area is used as a dynamic reference system, and its relative displacement between the two imaging processes is used to calculate and compensate for the mechanical vibration caused by the high-speed operation of the equipment. This mechanism allows subsequent image differential processing to be free from the interference of vibration artifacts, thereby obtaining residual information that truly reflects the changes in component position. This online dynamic calibration method based on the image's own information solves the common industry problem of decreased detection reliability in high-speed mounting scenarios without adding any additional hardware, enabling high-precision online process monitoring to be achieved without sacrificing equipment operating efficiency.

[0018] 3. The core of the detection logic employed in this invention lies in extracting residual information by comparing images of the same position before and after mounting. This verifies the correctness of the occlusion relationship between the component and the target pad, rather than identifying the optical or geometric features of the component itself. This relationship verification method completely decouples the detection results from appearance differences such as the component's color, characters, reflective properties, and dimensional tolerances between different batches. Therefore, this method demonstrates adaptability and stability when handling low-cost, mass-produced general-purpose materials from different suppliers with low appearance consistency. It avoids the tedious work of frequently adjusting algorithm parameters or rebuilding templates due to material differences in existing technologies, thus improving the overall operating efficiency and flexibility of the production line. The invention also possesses the capability to extend diagnostics and control to both upstream and downstream processes. On one hand, by introducing computer-aided design data to generate ideal pad masks, it can not only purify reference images to address pre-existing pad defects but also detect foreign objects on the substrate before mounting, thus expanding the functionality from post-construction quality inspection to pre-construction substrate inspection. On the other hand, by continuously recording and statistically analyzing a series of mounting deviation data, it can determine whether there is systematic mechanical drift or periodic material supply instability in the equipment, thereby outputting instructions for predictive maintenance. This elevates quality control from post-construction single-point error correction to pre-construction process prevention and system-level health status monitoring. Attached Figure Description

[0019] Figure 1 This is a flowchart of the component mounting quality collaborative diagnosis method of the present invention.

[0020] Figure 2 This is a state machine diagram for the diagnostic classification logic of a single mounting event in this invention.

[0021] Figure 3 This is a model diagram of the core functions and extended use cases of the detection system of this invention.

[0022] Figure 4 This is a diagram showing the distributed deployment and communication architecture of the mounting quality inspection system of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0024] This invention discloses a component placement quality inspection method for a pick-and-place machine. This method is based on the logical correlation between optical imaging detection and physical process signals. It includes: simultaneously acquiring image information and the vacuum pressure signal of the placement nozzle before and after component placement; calculating and compensating for mechanical vibrations introduced by equipment operation using the substrate background area in the image; obtaining optical position deviation through image difference and centroid calculation; and finally logically correlating the magnitude of this deviation with the variation characteristics of the pressure signal to classify and diagnose the physical causes of placement events. In surface mount applications, detection methods relying on static templates are difficult to handle variables such as uneven lighting, circuit board bending, and differences in the appearance of incoming components under actual working conditions. To address this challenge, this solution is configured to capture a first image containing the target pad and the adjacent substrate background area using an imaging device before the component is placed onto the target pad on the printed circuit board. The timing of this step is deterministically determined by setting the placement head carrying the component to descend to a predetermined imaging height, which allows the optical position deviation to be determined. The objective lens focal plane of the imaging device coincides with the plane where the component is finally mounted on the pad, while maintaining a safe distance from the circuit board. The resulting first image serves as a reference image encompassing all real-time environmental and workpiece disturbance factors, providing a reference frame for subsequent differential processing to effectively suppress common-mode noise. Furthermore, during high-speed operation of the pick-and-place machine, the machine's own mechanical micro-vibrations can cause relative displacement between images within a small time interval between two imaging operations, resulting in pseudo-residual signals. Therefore, this method employs an online dynamic alignment mechanism based on the image background. After component mounting, the imaging device captures a second image containing the component, target pad, and substrate background area at the same position as the first image. A deterministic subpixel-level displacement compensation step is then performed. This procedure involves extracting predefined background image blocks corresponding to the substrate background area from both the first and second images, and using a phase correlation algorithm to calculate the subpixel-level translation vector between the two background image blocks. The second image is digitally translated based on the displacement vector to generate an aligned second image. This process compensates for the displacement introduced by vibration, so that subsequent differential processing mainly reflects the image changes introduced by component mounting.

[0025] To reduce reliance on component body feature recognition, this method uses image difference and centroid localization to determine deviations. The process involves pixel-level difference processing of the first image and the aligned second image to generate a residual image. As an implementation, to avoid interference from potential pre-existing defects on the target pads, a reference image cleanup and pre-inspection step can be performed. Before the placement task, this step generates an ideal pad mask that defines the geometry of the target pad based on pre-stored computer-aided design data. The first image is then logically ANDed with this ideal pad mask to generate a cleaned reference image for subsequent difference processing. Simultaneously, a logical operation can be performed between the inverted first image and the mask. This system is used to detect foreign objects outside the solder pad area and output an early warning. After generating a residual image, the system calculates the centroid position of the highlighted area in the residual image and determines the component's planar placement position deviation based on the deviation of this centroid position from the preset ideal center position. This verification method reduces the sensitivity of the detection results to appearance differences such as component body color, characters, and reflective properties. Simple optical analysis may present similar detection results for faults with different physical causes. To achieve fault attribution, this method establishes a diagnostic logic that arbitrates the optical detection results using physical process signals. This logic simultaneously monitors the vacuum pressure signal of the placement nozzle to obtain the pressure time-domain curve while performing optical imaging. The system classifies the physical process of placement events based on the determined numerical value of the planar placement position deviation and the characteristics of the pressure time-domain curve. The deterministic procedure is as follows: If the planar placement position deviation is greater than a predetermined offset threshold and the pressure change rate of the pressure time-domain curve at the placement time exceeds a predetermined transition slope threshold, the event category is determined to be high offset placement. In contrast, if the residual image shows that the pixel value of the entire target pad area is higher than the brightness threshold and the pressure value of the pressure time-domain curve remains within the adsorption pressure range after the placement time, the event category is determined to be component loss or empty placement.

[0026] The procedure for calculating and compensating for subpixel-level displacement and obtaining the time-domain pressure curve characteristics was further determined. Specifically, before executing the phase correlation algorithm, a substrate background area image block not less than four times the projected area of ​​the component to be mounted and excluding pads or high-contrast lines is first extracted from the first and second images. A 5x5 Gaussian low-pass filter is applied to this image block to suppress image noise. Then, a Hanning window function is applied to reduce spectral leakage. Subsequently, in the calculated phase correlation matrix, a two-dimensional Gaussian surface fitting is performed on the maximum point and its 3x3 neighborhood to obtain the subpixel-accurate translation vector. At the same time, the raw pressure signal acquired by the vacuum pressure sensor at a frequency of not less than 1kHz is passed through a Savitzky-Gore filter with a window length of 9 and a polynomial order of 3. This is a digital filter that directly performs least-squares fitting in the time domain to smooth the data and obtain a smooth pressure curve. The placement moment is defined as the moment when the placement head Z-axis servo motor controller sends a signal indicating that the preset placement height has been reached. Finally, the pressure change rate at the moment of mounting is deterministically calculated by the following formula, which is obtained by calculating the pressure change rate across the mounting point. The average slope within an asymmetric time window is used to obtain a stable physical quantity that is insensitive to noise.

[0027] Furthermore, the method can also be applied to extended process diagnostics. On one hand, it can be used to diagnose coplanar defects in components. The steps involve calculating the intensity standard deviation of all pixels in the residual image whose values ​​are higher than the background noise threshold after generating the residual image. When the strength standard deviation When the predetermined coplanarity defect threshold is exceeded, the component is determined to have a coplanar defect. The basis for this judgment is that the edges of tilted components will produce a grayscale gradient in the residual image, thereby increasing the intensity standard deviation. On the other hand, this method can be used for online monitoring of the upstream solder paste printing process. The procedure is as follows: in the initial stage of a production batch, the average pixel intensity of the residual images generated by multiple components with positional deviations within the allowable range is recorded, and the mean residual intensity is statistically calculated. with standard deviation These elements together constitute the residual intensity historical fingerprint of this batch. For residual images in subsequent production, their average pixel intensity is compared with this historical fingerprint. When the intensity deviates from the mean... Exceeding the standard deviation When the solder paste printing process is found to be abnormal at a predetermined multiple, it is determined that there is an abnormality in the solder paste printing process. Finally, through continuous recording and statistical trend analysis of a series of mounting deviation data, it is possible to determine whether there is unidirectional mechanical drift or periodic feeding instability in the equipment. The results of this statistical trend analysis can be used to output predictive maintenance instructions to achieve early intervention for systemic deviations. In addition, the procedures for handling optical transmission elements and performing systemic process control have been further clarified. Among them, the texture residual energy is obtained by calculating its angular second moment based on the gray-level co-occurrence matrix of the residual image, which is a statistical matrix depicting the frequency of gray-level pairs under specific spatial relationships in the image. This angular second moment serves as a quantitative index of image texture uniformity, and its calculation formula is as follows: ,in, The values ​​are normalized elements in the gray-level co-occurrence matrix; the statistical trend analysis of the deviation data is achieved by maintaining an exponentially weighted moving average control chart for the X and Y axis deviation sequences of each mounting head. This is a trend monitoring method that assigns higher weight to recent data, and its sequence values... From the recursive formula Calculate that, among which For the current deviation, The previous period sequence value, weighting factor Set to 0.2, when the sequence value Five consecutive times exceeding the standard deviation of historically stable processes The determined control limit of three standard deviations, i.e. If so, it is determined that there is a systematic deviation trend that characterizes abnormal equipment status.

[0028] Example 1: In a continuously operating surface mount production line for consumer electronics, the pick-and-place machine operates at its rated speed limit, mounting tens of thousands of common resistive and capacitive components from different suppliers. These components exhibit batch-to-batch variations in color, surface gloss, and dimensional tolerances. Simultaneously, the high-speed operation of the equipment induces continuous, high-frequency mechanical micro-vibrations on the machine. This condition presents a dual challenge to a vision inspection system based on static template feature matching: the system must not only cope with the recognition difficulties caused by the inconsistency in component appearance but also handle the image acquisition position drift caused by vibration, leading to an increased false positive rate and forcing production processes to adjust accordingly. The process involves a trade-off between operational efficiency and quality control. In this scenario, for each component placement, at the instant the placement head carrying the component descends and is about to contact the printed circuit board, the imaging device captures a first image. This image records the target pad under the current lighting and circuit board conditions, and simultaneously captures the adjacent substrate background area, which serves as a dynamic reference for subsequent placement. After the component placement is completed, the nozzle lifts, and the imaging device captures a second image at the same position. Subsequently, an online dynamic alignment mechanism based on the image background is triggered. The system calculates the subpixel-level translation vector introduced by mechanical vibration by comparing the substrate background area in the first and second images. The vector is then used to perform digital translation compensation on the second image, generating an aligned second image. The capture of the first image provides a dynamic temporal and spatial reference for the vibration compensation step, while the vibration compensation step provides a calibrated and aligned data input for subsequent differential processing. This combination of mechanisms enables the reference for image comparison to remain stable under high-speed motion conditions.

[0029] Subsequently, the system performs pixel-level differential processing. Generate a residual image In this residual image, the difference information about the appearance of the component body is canceled out as a common background in the two imaging processes, leaving only the bright residuals representing unmasked pad areas or component overflow areas caused by the placement offset. This transforms a problem of complex feature recognition on variable objects into a problem of locating the geometric centroid of a signal containing only differential information. Based on this, the system further compares the calculated planar placement position deviation with the synchronously acquired pressure time-domain curve. The system uses logical associations based on the characteristics of each component. When a batch of feeders experiences a material shortage at the end, causing the nozzle to suck up empty material, the optical part detects a large residual covering the entire pad, while the pressure signal part detects that there is no pressure transition before and after placement. Based on this, the system classifies the physical process of the event as component loss or empty placement and outputs corresponding replenishment instructions. Ultimately, the production line can continue to operate at its rated speed, and the number of detection-related downtime events caused by component appearance differences and equipment vibration decreases. Furthermore, the system can distinguish between defects caused by positional misalignment and defects caused by material supply interruption, providing operators with actionable maintenance guidelines. The basis of quality control shifts from the identification of the variable object itself to the verification of the correctness of the physical relationship between the placement behavior and the target position, as well as the synchronous monitoring of the physical state of the actuator.

[0030] Example 2: To quantitatively evaluate the detection stability of the method of the present invention in dealing with component appearance differences and equipment mechanical vibration interference, this comparative experiment was established. The purpose was to compare the performance of the method of the present invention with a conventional detection method based on component body contour feature recognition under controlled conditions, thereby verifying the effectiveness of the present invention through data. The experiment was conducted on a standard pick-and-place machine. The PCB carrier platform of this machine integrated a programmable vibration table to apply controlled mechanical micro-vibrations. Two visual inspection methods were used for data acquisition and analysis. Method A, the method of the present invention, is based on dynamic real-time reference images and differential residual centroid calculation. This method includes an online dynamic alignment mechanism based on the image background. Method B served as the control group, using a conventional method based on static CAD. The test method for matching the edge and corner features of components using templates used three batches of passive components: batch A, a dark matte resistor of specification 0402; batch B, a gray high-gloss resistor of specification 0402 with characters printed on its surface; and batch C, a light brown ceramic capacitor of specification 0603. A key parameter in the test was the vibration frequency, which was set to simulate the typical vibration state of the pick-and-place machine during high-speed operation. The main factor affecting the value of this parameter was the motion harmonics of the spindle and crossbeam of the equipment. The technical trade-off was that the set frequency needed to be sufficient to interfere with conventional detection method B, but not to exceed the realistic range of industrial scenarios. Based on this decision logic, the vibration frequency was set to 100Hz, and the amplitude was calibrated to produce a peak displacement of 25 micrometers within the imaging field of view.

[0031] The test procedure is as follows: For each component in batches A, B, and C, 100 placement operations were performed under both vibration-free and 100Hz vibration conditions. During each placement, a random planar offset within the range of 0 to 60 micrometers was applied to the placement position, and a high-precision laser displacement sensor recorded the actual offset. Methods A and B simultaneously analyzed the images after each placement, calculating their respective planar placement position deviation measurements. If the difference between the measured value and the actual offset exceeded 15 micrometers, it was considered a misjudgment. If the average measurement error of each group is plotted, it can be observed that under all test conditions, the error value corresponding to Method A remained at a low level below 5 micrometers, while the error value of Method B... The error increases dramatically under specific conditions. Specifically, when testing component B with a glossy surface under vibration-free conditions, the average measurement error of method B is 18.7 micrometers, with a corresponding false positive rate of 21%, while the error of method A is only 3.3 micrometers, with a false positive rate of 0%. When 100Hz vibration is introduced, the average measurement error of method B when dealing with component B further increases to 35.2 micrometers, with a false positive rate as high as 58%, while the average measurement error of method A under the same conditions remains at 4.8 micrometers, with a false positive rate of only 2%. For components A and C with regular appearance, under 100Hz vibration, the average measurement errors of method B also reach 24.1 micrometers and 26.5 micrometers respectively, while method A remains below 4.4 micrometers.

[0032] The mechanism behind the aforementioned data discrepancies lies in the fact that the detection logic of Method B relies on the stable extraction of component edge features. Therefore, when faced with surface highlights and character interference in batch B, or dynamic blur introduced by vibration, its feature extraction process will produce a large deviation. In contrast, Method A performs pixel-level differential processing on images before and after mounting, and its detection results are largely independent of the optical characteristics of the component itself. At the same time, its built-in online dynamic alignment mechanism uses the substrate background in the image as a reference and compensates for vibration displacement before performing differential processing, thereby suppressing the influence of vibration on the measurement results.

[0033] Example 3: This example combines Figures 1 to 4 This describes a method for inspecting the placement quality of components in a pick-and-place machine, such as... Figure 1As shown, the method begins with capturing first and second images before and after component placement online, while simultaneously monitoring the vacuum pressure signal of the placement nozzle. The core processing path of the method includes: using the substrate background area in the before and after images to perform compensation for vibration-induced displacement to avoid mechanical vibration interference; then performing image differential processing to generate a residual image reflecting position changes; and calculating the centroid position to determine the planar placement deviation of the component. This optical position deviation information and the pressure signal change characteristics acquired simultaneously are input into the collaborative diagnosis and classification stage. This stage, based on logical association rules, ultimately classifies the physical process of the placement event into component loss or empty placement, placement qualification or high offset placement. In addition, the method also includes two extended diagnosis and control loops: one is to diagnose the upstream solder paste printing process and output early warning signals when abnormalities occur; the other is to perform systematic process control by performing statistical trend analysis on historical deviation data and output predictive maintenance or equipment calibration instructions accordingly.

[0034] like Figure 2 As shown, the process is triggered by the event that the placement head descends to the predetermined imaging height, causing the system to enter a waiting placement state. The subsequent capture of the first image event causes the system to switch to a dynamic reference captured state. When the placement and synchronous monitoring events occur, the system enters the core analysis state and sequentially performs a series of actions in this state, including capturing the second image, compensating for vibration displacement, generating a residual image, calculating positional deviation, and acquiring pressure curve characteristics. Finally, the system determines the final state and exits based on a set of mutually exclusive logical conditions. If the residual image is bright and the pressure value is maintained within the adsorption range, the system enters the final state of component loss or blank placement. If the deviation is greater than the offset threshold and the pressure change rate is greater than the slope threshold, the system enters the final state of high-offset placement. If the deviation is less than the threshold and the pressure is normal, the system enters the final state of qualified placement.

[0035] like Figure 3 As shown, the pick-and-place machine, as the main executor, together with the operator / maintenance system as external participants, performs the core function of quality inspection. This is presented as a collection of interconnected sub-functional use cases. These use cases include synchronous acquisition of image and pressure signals and compensation for vibration-induced displacement as basic inputs and compensation mechanisms; logical association diagnosis and classification as core diagnostic logic; and diagnostic capabilities for upstream solder paste printing processes, systematic process control, diagnosis of component coplanar defects, and handling of optically transmissive components.

[0036] like Figure 4As shown, the core of the entire system is the pick-and-place machine, which acts as a device node. Its data processing revolves around an industrial control computer, which acts as an execution node. This computer is responsible for receiving and processing data from two types of physical devices: one is imaging devices that transmit image information via an image data bus, and the other is vacuum pressure sensors that transmit pressure signals via a sensor bus. Inside the computer, three core software components are deployed to perform detection tasks: the core analysis engine is responsible for basic data processing, the collaborative diagnosis and classification module performs core logical judgments, and the extended diagnosis module handles advanced functions such as process diagnosis. In order to achieve systematic data management and remote monitoring, the collaborative diagnosis and classification module communicates with a central server through the standard TCP / IP protocol. The database in this server is specifically used to persistently store historical fingerprint and deviation data for long-term analysis. Finally, the engineer's workstation serves as the human-machine interaction terminal. The monitoring dashboard component running on it also obtains data from the server via the TCP / IP protocol, providing the operator with an interface to monitor and manage the entire detection system.

[0037] Example 4: Before starting a new printed circuit board assembly task, a systematic offline calibration procedure is performed for the new type of circuit board and components used in the task to determine a series of key thresholds in the detection method of this invention. This calibration procedure is performed on a pick-and-place machine with the detection system installed. The initial state is to prepare five printed circuit board samples from the same production batch to be tested, and a sufficient number of matching components to be placed. The first step of the procedure is to calibrate the planar placement position deviation threshold and the pressure transition slope threshold. The operator controls the pick-and-place machine to perform twenty normal placements and twenty placements with a preset large offset on the sample board in low-speed mode. The large offset is set to be greater than 50% of the component size. The system synchronously records the planar placement position deviation value and pressure time domain curve generated for each placement event. The pressure change rate of the pressure curve at the placement moment was calculated, thus obtaining two datasets: a normal placement group and a high offset placement group. The distribution of the planar placement position deviation values ​​in the two datasets was analyzed, and a value between the upper limit of the deviation distribution in the normal placement group and the lower limit of the deviation distribution in the high offset placement group was selected as the planar placement position deviation threshold. Similarly, the distribution of the pressure change rate in the two datasets was analyzed, and a slope value that can distinguish whether the component has successfully detached from the nozzle was selected as the pressure transition slope threshold.

[0038] The second step of the procedure is to calibrate the brightness threshold for missing or unplaced components. The operator controls the pick-and-place machine to perform ten unplacement operations without picking up components. The system analyzes the residual image generated by each unplacement operation, in which the entire target pad area is brightened, records its average pixel intensity, and multiplies the minimum average pixel intensity value obtained from these ten measurements by a safety factor of 0.9. The result is set as the brightness threshold. The third step of the procedure is to calibrate the coplanarity defect threshold. The operator places a thin sheet of known thickness on one side of the target pad on a sample board to introduce a defined tilt angle during component placement. This placement is repeated ten times, and the standard deviation of the pixel intensity of the residual image generated for each placement is calculated. The minimum standard deviation obtained from these ten measurements is set as the coplanarity defect threshold. In subsequent production, any strength standard deviation... Placement events exceeding this threshold will be identified as having coplanarity defects. After calibrating all the aforementioned thresholds, a trial production run of one hundred times will be initiated. During this trial production run, the system will use the calibrated thresholds to monitor the quality of each placement, and only for placement events judged to be of acceptable quality will the average pixel intensity of their residual images be recorded. After the trial production run is completed, the system will perform statistical analysis on the average pixel intensity data of all the recorded acceptable samples, and calculate its arithmetic mean, which will be used as the mean residual intensity of the batch. At the same time, its standard deviation is calculated. These two values and Together, they constitute the residual strength history fingerprint used to diagnose the upstream solder paste printing process.

[0039] Example 5: When processing the production of an optical transmissive element, specifically a glass-encapsulated diode, it was observed that the geometric residual energy was too low due to the low shielding of this type of element. This made it impossible to obtain an effective positional deviation using the calculation method based on the geometric residual centroid. Therefore, before putting it into formal production, a reference texture energy calibration procedure was performed on this type of element. In the calibration procedure, ten samples of this type of optical transmissive element were taken, and a high-precision coordinate measuring device was used to place each sample at the theoretical center position of the target pad. For each of these ten placements, the detection system captured the first and second images before and after mounting, and generated a residual image. Subsequently, the system calculated and recorded the texture residual energy of the residual images corresponding to these ten placements, and defined the arithmetic mean of these ten measurement results as the reference texture energy of this type of optical transmissive element. This reference texture energy value was stored in the system's material database and associated with the part number of the element.

[0040] In the subsequent online automated production process, when the system identifies the currently mounted component as an optical transmissive element of this model, it calls its calibrated reference texture energy value. After generating the residual image of this mounting, the system does not calculate its centroid position, but instead calculates its texture residual energy and compares the measured energy value with the called reference texture energy value. If the difference between the two exceeds a tolerance range set based on the statistical distribution of the aforementioned calibration experimental data, the system determines that there is a quality defect in this mounting. This calibration and detection process enables the method of the present invention to be applied to the quality detection of this type of optical transmissive element.

[0041] Example 6: In the initial stage of continuous production operation of the system, in order to establish a baseline reference model that can characterize the normal mechanical state of the equipment for subsequent long-term health monitoring, the system executes the following procedure: For each placement head, it dynamically maintains a sliding window containing the most recent one hundred placement deviation data, and continuously calculates the arithmetic mean of the deviation data in the X and Y axes within the window. The core parameter of this model, namely the maintenance warning threshold, is set to 50% of the allowable placement tolerance. When the absolute value of the sliding average deviation data of any placement head exceeds the maintenance warning threshold for ten consecutive calculations, the baseline reference model determines that the equipment state has deviated from the normal baseline and triggers a predictive maintenance command to record the drift state and data of the placement head for subsequent calibration.

[0042] During continuous system operation, its health status monitoring logic also includes an immediate intervention procedure for defects introduced by upstream processes. When the system performs a pre-inspection step of substrate quality before component placement, and obtains a non-zero result by performing a logical AND operation between the dynamic real-time reference image and the inverted image of the ideal pad mask, this indicates that there is a pre-stored foreign object outside the target pad area. The system immediately stops the placement operation on the target pad and stores the unique identification code of the printed circuit board and the image coordinates of the foreign object in the quality traceability log. Subsequently, the pick-and-place machine skips the placement task at that point and continues to perform placement operations at other locations. The combination of this baseline model and the anomaly handling procedure enables the system not only to evaluate the placement quality at a single point, but also to assess the long-term trend of its own operating status based on historical data, and to intervene immediately when defects are introduced by upstream processes.

[0043] 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.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for inspecting the placement quality of components in a pick-and-place machine, characterized in that, This method establishes a diagnostic logic that arbitrates optical detection results using physical process signals to classify component placement events under high-speed operating conditions. The method includes: before the component is placed onto the target pad on the printed circuit board, simultaneously capturing a first image containing the target pad and the background area of ​​the printed circuit board substrate adjacent to the target pad using an imaging device, serving as a dynamic real-time reference image; after component placement, capturing a second image containing the component, the target pad, and the substrate background area at the same position as the first image capture, and simultaneously monitoring the vacuum pressure signal of the placement nozzle during component placement to obtain a pressure time-domain curve; based on the substrate background area in the first and second images, calculating and compensating for sub-pixel-level displacement between images introduced by device vibration to generate an aligned second image; performing differential processing on the first and aligned second images to generate a residual image; calculating the centroid position of the highlighted area in the residual image to determine the planar placement position deviation of the component; and classifying the physical process of the component placement event based on the value of the planar placement position deviation and the characteristics of the pressure time-domain curve.

2. The component mounting quality inspection method for a pick-and-place machine according to claim 1, characterized in that, The step of capturing the first image is performed when the placement head carrying the component descends to a predetermined imaging height, which allows the imaging device to be in the same focal plane as the final placement position without contacting the printed circuit board.

3. The component mounting quality inspection method for a pick-and-place machine according to claim 1, characterized in that, Before differential processing, the method further includes: before performing the mounting task, generating an ideal pad mask that defines the geometry of the target pad based on pre-stored computer-aided design data corresponding to the target pad; performing a logical AND operation between the first image and the ideal pad mask to generate a cleaned reference image; the subsequent differential processing is performed using the cleaned reference image and the aligned second image; and performing a logical AND operation between the first image and the inversion of the ideal pad mask to detect whether there are foreign objects outside the target pad area, and outputting a warning signal when foreign objects are detected.

4. The component mounting quality inspection method for a pick-and-place machine according to claim 1, characterized in that, The steps for classifying the physical process of component placement events specifically include: if the planar placement position deviation is greater than a predetermined offset threshold, and the pressure change rate of the pressure time-domain curve exceeds a predetermined transition slope threshold at the placement time, then the event is classified as high offset placement; if the residual image shows that the pixel values ​​of the entire target pad area are higher than the brightness threshold, and the pressure value of the pressure time-domain curve remains within the adsorption pressure range after the placement time, then the event is classified as component loss or empty placement.

5. A method for inspecting the placement quality of components in a pick-and-place machine according to claim 1, characterized in that, The method also includes a step of diagnosing component coplanarity defects, performed after the residual image is generated, and includes: calculating the intensity standard deviation of all pixels in the residual image whose pixel values ​​are higher than a background noise threshold. When the strength standard deviation If the predetermined coplanarity defect threshold is exceeded, the component is determined to have a coplanar defect.

6. A method for inspecting the placement quality of components in a pick-and-place machine according to claim 1, characterized in that, The method can also handle optically transparent elements and includes: calculating the geometric residual energy and texture residual energy of the residual image; when the geometric residual energy is lower than a predetermined geometric threshold and the texture residual energy is higher than a predetermined texture threshold, determining that the current element is an optically transparent element, and switching to perform the following steps to evaluate the mounting quality: comparing the texture residual energy of the residual image with a reference texture energy set for the optically transparent element of this model, and evaluating the mounting quality based on the difference between the two.

7. A method for inspecting the placement quality of components in a pick-and-place machine according to claim 1, characterized in that, The method also includes a step of diagnosing the upstream solder paste printing process: at the initial stage of a production batch, the average pixel intensity of residual images generated by multiple components with planar mounting position deviations within the allowable tolerance range is recorded, and a residual intensity mean and residual intensity standard deviation are statistically calculated to form the residual intensity historical fingerprint of the batch. For each residual image generated in subsequent production, its average pixel intensity is compared with the residual intensity historical fingerprint. When the average pixel intensity deviates from the residual intensity mean by more than a predetermined multiple of the residual intensity standard deviation, it is determined that there is an abnormality in the solder paste printing process.

8. A method for inspecting the placement quality of components in a pick-and-place machine according to claim 1, characterized in that, The steps for calculating and compensating for subpixel-level displacement between images caused by equipment vibration specifically include: extracting background image blocks corresponding to the substrate background area from the first image and the second image respectively; calculating the subpixel-level displacement vector between the two background image blocks using a phase correlation algorithm; and digitally translating the second image based on the displacement vector to align the substrate background areas of the two images before performing differential processing.

9. A method for inspecting the placement quality of components in a pick-and-place machine according to claim 7, characterized in that, The step of determining that there is an abnormality in the solder paste printing process also includes determining the type of abnormality: if the average pixel intensity is lower than the mean residual intensity and its deviation exceeds a predetermined multiple of the standard deviation of the residual intensity, it is determined that there is too much solder paste; if the average pixel intensity is higher than the mean residual intensity and its deviation exceeds a predetermined multiple of the standard deviation of the residual intensity, it is determined that there is too little solder paste or missing printing.

10. A method for inspecting the placement quality of components in a pick-and-place machine according to claim 1, characterized in that, The method also includes a step of performing systematic process control: continuously recording and storing a series of planar placement position deviation data determined by component placement, analyzing the statistical trend of the deviation data to determine whether there is a unidirectional offset trend that characterizes mechanical drift of the placement head or a periodic offset pattern that characterizes abnormal feeder status, and outputting instructions for predictive maintenance or equipment calibration based on the determined offset trend or offset pattern.

Images