Memory bank visual sensor testing method and device

Abstract

The invention relates to a memory bank visual sensor testing method and device, and the device comprises a supporting seat which is provided with two guide rails arranged at an interval; the display control mechanism is located above the supporting seat, a visual inspection mechanism is arranged below the display control mechanism, and the display control mechanism is electrically connected with the visual inspection mechanism; the turn-over clamping mechanism comprises a tool clamp and sliding blocks in sliding fit with the guide rails, the two ends of the tool clamp are in rotating fit with the sliding blocks respectively, the sliding blocks can slide in the direction close to or away from the visual inspection mechanism, the tool clamp can rotate relative to the sliding blocks, and the tool clamp can rotate relative to the sliding blocks. And the method is further used for detecting top surface defects and bottom surface defects of the memory bank to be detected. The automatic detection of the two sides of the memory bank can be completed in a single detection device without manual turning, so that the detection efficiency is improved, and the stability of image acquisition and the accuracy of a detection result are ensured.

Description

Technical Field

[0001] This invention relates to the technical field of memory module testing, and in particular to a method and apparatus for testing memory modules using a visual sensor. Background Technology

[0002] Currently, with the rapid development of the electronic information industry, memory modules, as core components of computers and related electronic devices, are experiencing continuous expansion in production scale, placing higher demands on the efficiency and accuracy of product testing. Existing memory module testing technologies mostly rely on fixed-angle visual inspection devices, which can only perform appearance and soldering quality inspection on one side. They often cannot simultaneously address defects on the other side, thus requiring manual flipping of the memory module for re-inspection. This method not only complicates the testing process and reduces efficiency, but also easily causes physical damage such as bending and scratches to the memory module during manual flipping, affecting subsequent testing accuracy and product yield. Furthermore, the instability and randomness of manual operation not only increase labor costs but also easily introduce testing errors, making it difficult to meet the demands of modern memory module production lines for large-scale, high-consistency, and high-precision automated testing. Summary of the Invention

[0003] To address the issues of memory module testing being limited to single-sided testing, requiring manual flipping which is inefficient and prone to damage, this application provides a memory module visual sensor testing method and apparatus.

[0004] A memory module vision sensor testing device, using a memory module vision sensor testing method, the testing device comprising: The support base is provided with two spaced-apart guide rails; A display control mechanism is located above the support base, and a vision detection mechanism is provided below the display control mechanism. The display control mechanism is electrically connected to the vision detection mechanism. A flipping clamping mechanism is used to clamp and flip the memory module to be tested, and is located below the visual inspection mechanism. The flipping clamping mechanism includes a tooling fixture and a sliding block that slides with the guide rail. The two ends of the tooling fixture are respectively rotatably engaged with the sliding block. The sliding block can slide towards or away from the visual inspection mechanism, thereby causing the memory module to be tested on the tooling fixture to slide to the memory module placement position or inspection position. The tooling fixture can rotate relative to the sliding block, thereby being used to detect top and bottom defects of the memory module to be tested.

[0005] By adopting the above technical solution, and by setting a combination structure of guide rails, sliding blocks and tooling fixtures on the support base, the memory module can be moved smoothly from the placement position to the detection position, and the automatic flipping and detection process can be completed in conjunction with the vision inspection mechanism, thereby effectively supporting the automated detection requirements of the method implementation.

[0006] Preferably, the sliding block includes an integrated transmission part and a connecting part. The transmission part slides with the guide rail, and the connecting part rotates with the tooling fixture. One of the connecting parts is connected to a rotary motor electrically connected to the display control mechanism. The drive end of the rotary motor passes through the connecting part and connects to the tooling fixture, thereby controlling the memory module under test on the tooling fixture to flip over.

[0007] By adopting the above technical solution, a transmission part and a connecting part are set inside the sliding block, and a rotary motor is used to realize the flipping drive of the tooling fixture. This enables the memory module under test to achieve stable automatic flipping, ensuring the continuity and consistency of the top and bottom surface detection, thereby further improving the overall automation level and ease of operation of the device.

[0008] A method for testing a memory module visual sensor, applied in a memory module visual sensor testing device, the method comprising: When it is detected that the memory module to be tested has been placed in the flipping clamping mechanism, a unique workpiece identifier is assigned to the memory module to be tested; When the memory bar is detected to be in the detection position, the light source and vision system of the vision inspection mechanism are controlled to be triggered synchronously to obtain the corresponding top surface inspection image, and the top surface inspection image is associated with and stored with the unique identifier of the workpiece. The flipping clamping mechanism is controlled to perform a flipping action based on a preset time compensation parameter, and after the flipping is completed, the attitude compensation algorithm is called to dynamically adjust the imaging area and / or focus parameters of the visual detection mechanism. When it is detected that the vision inspection mechanism has completed dynamic adjustment, the vision inspection mechanism is controlled to acquire the corresponding bottom surface inspection image, and the bottom surface inspection image is associated with and stored with the unique identifier of the same workpiece; The top and bottom surface inspection images, which are associated with the same unique identifier of the workpiece, are input into the defect fusion and discrimination model. The global feature vectors extracted from the top and bottom surface inspection images are then merged and calculated to output the corresponding global inspection results.

[0009] By adopting the above technical solution, and by setting up a flipping clamping mechanism and a vision inspection mechanism in the detection device for coordinated control, an automatic detection process for the top and bottom surfaces of memory modules is realized, avoiding the damage and inefficiency caused by manual flipping. This ensures detection accuracy while improving overall detection efficiency and stability.

[0010] Preferably, the step of assigning a unique workpiece identifier to the memory module under test when it is detected that the memory module under test has been placed in the flipping clamping mechanism includes: The system acquires the timestamp information generated when it detects that the memory module under test has been placed in the flipping clamping mechanism, as well as the preset transfer sequence number of the memory module under test, and generates a corresponding initial identifier based on the timestamp information and the transfer sequence number. The detection position information of the previous detection cycle is indexed, and the initial identifier is encoded according to the detection position information of the previous detection cycle to generate and assign a corresponding unique identifier for the memory bar to be tested.

[0011] By adopting the above technical solution, a unique identifier is generated when the memory module under test is placed in the flipping clamping mechanism. This identifier is then combined with a timestamp and a transfer serial number for recording. This ensures that each memory module can be uniquely tracked throughout the entire testing process, avoiding confusion and omissions caused by manual management, and making the testing process more reliable and traceable.

[0012] Preferably, the step of encoding the initial identifier based on the detection position information of the previous detection cycle to generate and assign a corresponding unique identifier for the memory module under test includes: Based on the detection position information of the previous detection cycle, determine the top surface detection time, the actual completion time of flipping, and the expected completion time of the previous detection cycle; The first posture parameter at the time of the top surface detection is obtained, and the second posture parameter at the time of the actual completion of the flipping is obtained. Based on the first posture parameter and the second posture parameter, the corresponding position deviation code is determined. The position deviation code is used to participate in the workflow of dynamically adjusting the imaging area and / or focus parameters of the vision detection mechanism. Calculate the time deviation between the actual completion time of the flipping and the expected completion time, determine the corresponding time deviation code based on the time deviation value, and use the time deviation code to determine the preset time compensation parameter; The initial identifier is encoded according to the position deviation code and the time deviation code to generate and assign a unique workpiece identifier to the memory module under test.

[0013] By adopting the above technical solution, the detection position information of the previous detection cycle is introduced into the process of generating the unique identifier of the workpiece, and the deviation between the attitude parameters and the time parameters is used for encoding. This enables the inheritance of historical information of the detection object, allowing the detection system to make adaptive corrections based on historical deviations, thereby ensuring the continuity of detection and the accuracy of the compensation mechanism.

[0014] Preferably, before the step of controlling the flipping clamping mechanism to perform the flipping action based on the preset time compensation parameter, the following steps are included: Whether the value mapped by the time deviation code is within a preset deviation threshold range; If not, the preset time compensation parameters of the previous detection cycle will be used. If so, the preset time compensation parameters of the previous detection cycle are corrected according to the time deviation code to generate new preset time compensation parameters.

[0015] By adopting the above technical solution, the time deviation code is thresholded before the flipping operation, and it is decided whether to continue or modify the compensation parameter. This can effectively avoid meaningless frequent corrections, so that the compensation mechanism can ensure both accuracy and stability, thereby improving the robustness and detection efficiency of the flipping control.

[0016] Preferably, the step of correcting the preset time compensation parameter of the previous detection cycle according to the time deviation code to generate a new preset time compensation parameter includes: Determine the first and second weighting coefficients for the sum; Multiply the preset time compensation parameter of the previous detection cycle by the first weighting coefficient to generate the corresponding first correction value; The value mapped by the time deviation code is multiplied by the second weighting coefficient to generate the corresponding second correction value; The first correction value and the second correction value are added together to generate a new preset time compensation parameter.

[0017] By adopting the above technical solution, and by setting weighting coefficients to weight and fuse the time compensation parameters and deviation values ​​of the previous detection cycle to form new compensation parameters, the compensation mechanism can maintain a smooth transition during the correction process. This not only avoids large fluctuations in parameters but also gradually approaches the ideal compensation effect, thereby achieving higher synchronization accuracy of the flipping action.

[0018] Preferably, the step of dynamically adjusting the imaging area and / or focus parameters of the visual detection mechanism by invoking the attitude compensation algorithm includes: Based on the position deviation code, the corresponding translation offset and tilt angle are determined, wherein both the translation offset and the tilt angle can take the value of zero; The imaging area of ​​the visual inspection mechanism is dynamically adjusted according to the translation offset. The focusing parameters of the visual inspection mechanism are dynamically adjusted according to the tilt angle.

[0019] By adopting the above technical solution, the translation offset and tilt angle are calculated based on the position deviation encoding during the attitude compensation process, and the imaging area and focus parameters of the vision inspection mechanism are dynamically adjusted. This ensures that the image acquisition of the memory stick after flipping remains centered and clear, thereby significantly reducing the detection error caused by attitude deviation.

[0020] Preferably, in the step of merging and calculating the global feature vectors extracted from the top surface detection image and the bottom surface detection image to output the corresponding global detection result, the defect fusion discrimination model includes at least a multi-level feature fusion model and a defect discrimination model, including: Based on the top surface detection image, the corresponding appearance geometric features are extracted, and based on the bottom surface detection image, the corresponding surface state features are extracted. Based on the appearance geometric features and the surface condition features, registration and matching are performed according to the spatial position corresponding to the unique identifier of the workpiece to generate a corresponding registration feature group; The registered feature group is input into the multi-level feature fusion model to perform weighted fusion of geometric features and texture features to generate a global feature vector. The global feature vector is input into the defect discrimination model, and the corresponding global detection result is output.

[0021] By adopting the above technical solution, the appearance geometric features and surface state features are extracted separately in the double-sided image processing stage, and the registration and matching are completed under the correspondence of the workpiece's unique identifier. Then, the weighted fusion is performed through a multi-level feature fusion model, which can ensure the spatial and semantic unity of the feature information of the top and bottom surfaces, and improve the accuracy and comprehensiveness of the final judgment.

[0022] Preferably, in the step of inputting the global feature vector into the defect discrimination model and outputting the corresponding global detection result, the global detection result includes at least a preliminary screening model and a secondary discrimination model, including: The global feature vector is input into a preliminary screening model based on rule thresholds to identify whether there are abnormal features that exceed the corresponding abnormal threshold. The global feature vector is input into a secondary discriminant model based on a machine learning model to classify the abnormal features as defect types. Based on the spatial location information contained in the global feature vector, the specific location coordinates corresponding to the abnormal feature are determined; Based on the defect type and specific location coordinates of each of the aforementioned abnormal features, the global detection result of the memory module under test is output accordingly.

[0023] By adopting the above technical solution, a mechanism combining preliminary screening and secondary discrimination is used when the fused feature vector is input into the discrimination model. This can quickly screen out obvious anomalies first, then perform fine classification on complex defects, and output the defect location and category in combination with spatial information, thereby ensuring that the detection results are both comprehensive and have high-precision positioning capabilities.

[0024] In summary, this application includes at least one of the following beneficial technical effects: 1. This application utilizes a combination structure of guide rails, sliding blocks, and tooling fixtures on a support base. This allows the memory module under test to move smoothly between its placement and inspection positions, and to acquire a top-side image under the action of a vision inspection mechanism. After the top-side inspection is completed, the tooling fixture is flipped by the rotation of the sliding blocks, so that the bottom side of the memory module faces the vision inspection mechanism, thereby acquiring a bottom-side image. Throughout the process, the display control mechanism and the vision inspection mechanism maintain an electrical connection and synchronous control, ensuring that the flipping action and image acquisition are coordinated in time and position. This design allows the inspection system to automatically inspect both sides of the memory module in a single inspection device without manual flipping. This not only improves inspection efficiency but also avoids the risks of bending and scratching the memory module caused by manual flipping, while ensuring the stability of image acquisition and the accuracy of inspection results. This meets the demands of modern production lines for high-efficiency, high-consistency, and high-precision inspection.

[0025] 2. This application achieves continuous image acquisition and automated defect detection of the top and bottom surfaces of memory modules by setting up a flipping clamping mechanism and a vision inspection mechanism in the testing device, introducing a unique workpiece identification management mechanism, and combining light source triggering, time compensation, and posture compensation algorithms. Specifically, at the initial stage of detection, each memory module to be tested is assigned a unique identifier, enabling it to be accurately tracked and matched throughout the entire detection process, avoiding the risk of misalignment or missed detection caused by manual operation. After the top surface image is acquired, the system uses preset time compensation parameters to drive the flipping clamping mechanism to complete the flipping, and calls the posture compensation algorithm to dynamically correct the imaging area and focus of the vision inspection mechanism, thereby ensuring that the bottom surface image after flipping remains stable and clear. Finally, the top and bottom surface images under the same workpiece unique identifier are input into the defect fusion discrimination model, global features are extracted and fused for judgment, and a complete detection result is output. This application effectively overcomes the problems of existing technologies that can only detect one side, require manual flipping, and have the risk of damage, realizing automated, non-destructive, and high-precision detection of both sides of memory modules, greatly improving detection efficiency and consistency, and reducing the uncertainty caused by manual intervention. Attached Figure Description

[0026] Figure 1 This is a flowchart of a memory module vision sensor testing method according to one embodiment of this application.

[0027] Figure 2 This is a schematic diagram of the structure of a memory module vision sensor testing device according to one embodiment of this application. Figure 1 .

[0028] Explanation of reference numerals in the attached drawings: 1. Support base; 2. Display control mechanism; 3. Vision inspection mechanism; 4. Guide rail; 5. Tooling fixture; 6. Sliding block; 61. Transmission part; 62. Connecting part; 7. Rotary motor. Detailed Implementation

[0029] The present application will be further described in detail below with reference to the accompanying drawings.

[0030] In one embodiment, such as Figure 1 As shown, this application discloses a memory module vision sensor testing device and a memory module vision sensor testing method. The testing device includes: Support base 1, wherein the support base 1 is provided with two spaced guide rails 4; The display control mechanism 2 is located above the support base 1, and a vision inspection mechanism is located below the display control mechanism 2. The display control mechanism 2 is electrically connected to the vision inspection mechanism. A flipping clamping mechanism is used to clamp and flip the memory module to be tested, and is located below the visual inspection mechanism. The flipping clamping mechanism includes a tooling fixture 5 and a sliding block 6 that slides with the guide rail 4. The two ends of the tooling fixture 5 are respectively rotatably engaged with the sliding block 6. The sliding block 6 can slide towards or away from the visual inspection mechanism 3, thereby causing the memory module to be tested on the tooling fixture 5 to slide to the memory module placement position or inspection position. The tooling fixture 5 can rotate relative to the sliding block 6, thereby being used to detect the top and bottom defects of the memory module to be tested.

[0031] In this embodiment, the support base 1 is used to support the structural and electrical components of the entire device. A display control mechanism 2 is mounted above the support base 1. This mechanism integrates a display screen and a control circuit board, capable of displaying images and detection results acquired through visual inspection in real time. It also serves as a human-machine interface for inputting detection commands and parameter configurations. A visual inspection mechanism 3 is mounted below the display control mechanism 2. The visual inspection mechanism 3 is electrically connected to the display control mechanism 2 and includes a light source, lens assembly, image sensor, and signal acquisition circuitry. It can acquire images from the memory module upon triggering a control command and transmit the data to the control module for analysis and processing.

[0032] In terms of mechanical structure, the upper surface of the support base 1 is provided with two spaced guide rails 4 to define the movement path of the flipping clamping mechanism. The flipping clamping mechanism includes a tooling fixture 5 and a sliding block 6. The sliding block 6 is slidably engaged with the guide rails 4 and can reciprocate in the direction of approaching or moving away from the vision inspection mechanism 3. The two ends of the tooling fixture 5 are respectively rotatably engaged with the two sliding blocks 6, which not only ensures the sliding stability of the fixture in the horizontal direction, but also allows the tooling fixture to flip around the sliding blocks in the vertical plane. With this design, the memory module to be tested can first be sent to the placement position or inspection position by the movement of the sliding block 6, and then the flipping action of the tooling fixture 5 completes the switching between the top and bottom surfaces, enabling the vision inspection mechanism 3 to automatically inspect both sides of the memory module.

[0033] Furthermore, such as Figure 1 As shown, the sliding block 6 includes an integrated transmission part 61 and a connecting part 62. The transmission part 61 is slidably engaged with the guide rail 4, and the connecting part 62 is rotatably engaged with the tooling fixture 5. One of the connecting parts 62 is connected to a rotary motor 7 that is electrically connected to the display control mechanism. The drive end of the rotary motor passes through the connecting part and is connected to the tooling fixture 5, thereby controlling the memory stick under test on the tooling fixture 5 to flip over.

[0034] In this embodiment, the sliding block 6 adopts an integrated design, internally comprising two functional areas: a transmission part 61 and a connecting part 62. The transmission part 61 is located at the lower end of the sliding block 6 and forms a sliding engagement with the guide rail 4 on the support base 1, ensuring that the sliding block 6 can move smoothly along the length direction of the guide rail 4, thereby driving the tooling fixture 5 to reciprocate between the detection position and the placement position. The connecting part 62 is located at the upper end of the sliding block 6 and is used to rotate with the tooling fixture 5, ensuring that the tooling fixture 5 can rotate around the axis of the connecting part 62 while maintaining stable clamping of the memory module to be tested.

[0035] A rotary motor 7 is installed at one end of the connecting part 62. The rotary motor 7 is electrically connected to the display control mechanism 2 via a cable and is controlled by the output of the control program. When a flipping action is required, the display control mechanism 2 sends a drive signal to the rotary motor 7. The drive end of the rotary motor 7 passes through the connecting part 62 and is fixedly connected to the rotating shaft of the tooling fixture 5. When the motor rotates, it directly drives the tooling fixture 5 to rotate, realizing the flipping switching of the memory module between the top and bottom surfaces. Through this design, the sliding block 6 can both provide translational movement along the guide rail to complete the transport of the memory module and complete the flipping of the tooling fixture 5 through the drive of the rotary motor, thereby ensuring the automation and continuity of the entire inspection process.

[0036] like Figure 2As shown, a memory module visual sensor testing method is applied in a memory module visual sensor testing device. The device includes a flipping clamping mechanism for holding and flipping the memory module under test, and a visual inspection mechanism located above the flipping clamping mechanism. The flipping clamping mechanism is an execution structure capable of fixing and flipping the memory module under test during the testing process. It must ensure clamping stability to avoid shaking or displacement during testing or flipping, and also have rotation or flipping capabilities to allow the memory module under test to smoothly switch between the top and bottom surfaces, thereby achieving double-sided image acquisition. The visual inspection mechanism is an imaging system installed above the flipping clamping mechanism for acquiring images of the memory module. It typically consists of a light source, lens, image sensor, and data processing unit. It can acquire clear image information through illumination equalization and focus adjustment, and transmit it to an analysis module for defect identification. The testing method includes: S10. When it is detected that the memory module under test has been placed in the flipping clamping mechanism, a unique workpiece identifier is assigned to the memory module under test. The unique workpiece identifier is a unique data identifier assigned to each memory module under test in the testing process. It is usually generated by timestamp, transfer sequence number or testing cycle parameter to ensure that the images of the top and bottom surfaces and the subsequent discrimination results can correspond one-to-one with the same memory module, avoiding confusion and omissions in the testing process.

[0037] S20. When the memory module is detected to be in the detection position, the light source and vision system of the vision inspection mechanism are triggered synchronously to obtain the corresponding top surface inspection image, and the top surface inspection image is associated with the workpiece's unique identifier and stored. S30. When the flipping clamping mechanism is controlled to perform a flipping action based on preset time compensation parameters, and after the flipping is completed, the attitude compensation algorithm is called to dynamically adjust the imaging area and / or focus parameters of the visual inspection mechanism. The preset time compensation parameter refers to the time correction amount set to compensate for the actuator action delay or synchronization error when controlling the flipping clamping mechanism to perform the flipping action. This parameter can keep the flipping action and the image acquisition of the visual inspection mechanism in sync, thereby improving the stability of the inspection. The attitude compensation algorithm is an algorithm that corrects the image offset and blur caused by clamping eccentricity, attitude tilt or jitter after the flipping action is completed. By adjusting the imaging area or automatically correcting the focus, the bottom image is kept clear and centered, thereby ensuring the accuracy of subsequent defect identification.

[0038] S40. When it is detected that the vision inspection mechanism has completed dynamic adjustment, control the vision inspection mechanism to acquire the corresponding bottom surface inspection image, and associate and store the bottom surface inspection image with the unique identifier of the same workpiece. S50. When the top surface inspection image and the bottom surface inspection image, both associated with the same unique identifier of the workpiece, are input into the defect fusion discrimination model, the global feature vectors extracted from the top and bottom surface inspection images are merged and calculated to output the corresponding global inspection result. The defect fusion discrimination model is an intelligent analysis model that can integrate information from both top and bottom surface images. It typically includes feature extraction, feature fusion, and discrimination modules, and can perform global analysis based on two-sided data to output the detection conclusion and defect category of the memory module.

[0039] In this embodiment, the detection process is uniformly scheduled by the display control mechanism. First, when the memory module is placed into the flipping clamping mechanism, the trigger signal output by the position sensor is transmitted to the display control mechanism. The display control mechanism immediately generates a unique identifier for the workpiece and writes it into the database as an index for subsequent images and results. Next, the position detection signal confirms that the memory module has entered the detection position. The display control mechanism sends a synchronous trigger command to the light source driving circuit and the vision acquisition unit. The light source is lit at a set intensity, and the image sensor completes the top surface image acquisition and stores the data stream along with the unique identifier. After the top surface detection is completed, the display control mechanism sends an action signal to the flipping motor driving unit according to the currently preset time compensation parameters. The flipping clamping mechanism begins to flip. During this process, the time compensation parameters are used to correct motor delay and system latency. The system eliminates differences in timing; when the flipping action ends and the flipping position signal is fed back to the display control mechanism, the system immediately calls the attitude compensation algorithm to correct the ROI of the imaging area of ​​the vision inspection mechanism and the focal length parameters of the autofocus module in real time, so that the bottom image of the memory module is kept within the optimal acquisition range; after the attitude compensation is completed and the signal is output, the display control mechanism triggers the light source and image sensor to work synchronously again, acquire the bottom image and store it with the same unique identifier; finally, the image processing module sends the top and bottom images corresponding to the unique identifier into the defect fusion discrimination model. The model performs double-sided feature extraction, multi-layer feature fusion and classification discrimination, and returns the fusion calculation results to the display control mechanism and outputs the global detection results in the display control mechanism, realizing the complete workflow of automated double-sided detection of memory modules.

[0040] Furthermore, when it is detected that the memory module under test has been placed in the flipping clamping mechanism, the step of assigning a unique workpiece identifier to the memory module under test includes: S101. Obtain the timestamp information generated when the memory module under test is detected to have been placed in the flipping clamping mechanism, as well as the preset transfer sequence number of the memory module under test. Generate the corresponding initial identifier based on the timestamp information and the transfer sequence number. The timestamp information refers to the time record automatically generated by the system when the detection system recognizes that the memory module has been correctly placed in the flipping clamping mechanism. This time record can reflect the specific placement time with millisecond-level precision, thus providing accurate time characteristics for the subsequent generation of unique identifiers. The transfer sequence number refers to the sequential number preset for each memory module in the production or inspection process. This number can reflect the sequential relationship of the memory module in the production line or inspection batch, thus ensuring that memory modules in different batches and different positions can be accurately distinguished at the data level. The initial identifier refers to the preliminary identifier data composed of the timestamp information and the transfer sequence number at the beginning of the inspection stage. It does not yet contain information from the historical inspection cycle, but it can serve as the basis for generating the unique identifier of the workpiece.

[0041] S102. Index the detection position information of the previous detection cycle, and encode the initial identifier based on the detection position information of the previous detection cycle to generate and assign a corresponding unique identifier for the memory module under test. The detection position information of the previous detection cycle refers to the parameters related to the position of the memory module recorded by the system in the previous round of detection. This information usually includes the position coordinates or attitude data formed by the memory module during the top surface detection, flipping action, and bottom surface detection, which is used to reflect the operating status and displacement of the equipment in the previous cycle. The unique identifier of the workpiece is a complete identifier data generated by further integrating the detection position information of the previous detection cycle on the basis of the initial identifier. It can transmit historical position information and individual characteristics between different detection cycles, thereby realizing accurate traceability and correspondence of detection data.

[0042] In this embodiment, when the detection system detects that a memory module to be tested has been placed in position via the position sensor on the flipping clamping mechanism, a position trigger signal is transmitted to the display control mechanism. The display control mechanism immediately calls the system time to generate high-precision timestamp information and simultaneously reads the corresponding circulation sequence number of the memory module from the production line database or the detection scheduling system. The display control mechanism combines the timestamp information and the circulation sequence number to form an initial identifier and temporarily writes this initial identifier into the data buffer as the workpiece identifier basis for the current detection cycle. After completing the generation of the initial identifier, the display control mechanism simultaneously accesses the historical detection database and indexes the detection position information related to the flipping clamping mechanism in the previous detection cycle. This information is output by a displacement encoder, angle sensor, or image positioning algorithm and typically includes the trigger coordinates of the top surface detection, the completion coordinates of the flipping execution, and attitude angle parameters. The display control mechanism encodes and combines the retrieved location information with the initial identifier to form a complete unique identifier for the workpiece, which includes a timestamp, transfer sequence number, and historical location information. This unique identifier is then written into the database and allocated to the current memory module. In subsequent steps such as top surface image acquisition, flipping control, and bottom surface inspection, all data streams and inspection results are indexed by this unique identifier, achieving a one-to-one correspondence between inspection data and physical memory modules and historical traceability.

[0043] For example, during the detection cycle, when the sensor detects the placement of a memory module, the display control mechanism generates a timestamp "20250908143015" and retrieves the transfer serial number "A103" from the production line database, combining them to generate the initial identifier "A103_20250908143015". Subsequently, the system indexes the position information detected in the previous cycle as X=128.6mm, Y=52.1mm, and tilt angle=1.2°, and appends it to the initial identifier to obtain a complete unique identifier for the workpiece. "A103_20250908143015_P128.6_52.1_T1.2"; This identifier is invoked and matched during both the top and bottom detection phases, thus ensuring that the detection results are strictly bound to the memory module itself.

[0044] Furthermore, the step of encoding the initial identifier based on the detection position information of the previous detection cycle to generate and assign a corresponding unique identifier for the memory module under test includes: S1021. Based on the detection position information of the previous detection cycle, determine the top surface detection time, the actual completion time of flipping, and the expected completion time of the previous detection cycle. S1022. Obtain the first posture parameters at the time of top surface detection, obtain the second posture parameters at the time of actual completion of flipping, and determine the corresponding position deviation code based on the first posture parameters and the second posture parameters. The position deviation code is used to participate in the workflow of dynamically adjusting the imaging area and / or focus parameters of the vision inspection mechanism. S1023. Calculate the time deviation between the actual completion time and the expected completion time of flipping, and determine the corresponding time deviation code based on the time deviation value. The time deviation code is used to determine the preset time compensation parameter. S1024. Encode the initial identifier according to the position deviation code and the time deviation code to generate and assign a unique workpiece identifier for the memory module under test.

[0045] In this embodiment, the display control mechanism first accesses the data log stored in the previous detection cycle. Based on the detection trigger record and motor feedback signal, it parses the top surface detection trigger time, the actual completion time of the flipping execution, and the system-preset expected completion time of the flipping, thereby obtaining key time reference information. Subsequently, the display control mechanism extracts the first posture parameter corresponding to the top surface detection time and the second posture parameter corresponding to the actual completion time of the flipping through the image posture recognition unit and the angle sensor of the flipping execution mechanism, respectively. It then generates a position deviation code through differential calculation. This code reflects the translational offset and tilt angle of the memory module before and after the flipping, and is used to dynamically correct the imaging area and / or focus parameters in subsequent visual detection stages to ensure the stability and clarity of the acquired images. Next, the system calculates the time deviation value using the time difference between the completion time of the flipping execution and the expected completion time, and maps it to a time deviation code. This code represents the deviation between the mechanism's action rhythm and the system's control timing, and can be used as a basis for updating or correcting the flipping action time compensation parameters. Finally, the display control mechanism appends the position deviation code and time deviation code to the previously generated initial identifier to form a complete unique identifier for the workpiece containing a timestamp, transfer sequence number, position deviation, and time deviation. This unique identifier is written to the database and allocated to the current memory module to be tested, so that the subsequent top surface inspection, flipping action, and bottom surface inspection processes can all call this identifier, realizing a one-to-one correspondence between the detection data and the workpiece body and a closed-loop control of the adaptive compensation mechanism.

[0046] Furthermore, before the step of controlling the flipping clamping mechanism to perform the flipping action based on preset time compensation parameters, the following steps are included: S21. Check whether the value mapped by the time deviation code is within the preset deviation threshold range; call the detection data parsing unit to read the time deviation code generated by the previous detection cycle and map it into a time deviation amount in numerical form; then display the control mechanism to compare the time deviation amount with the preset deviation threshold range to determine whether it is within the allowable range. S22. If not, the preset time compensation parameters of the previous detection cycle will be used. When the comparison result shows that the time deviation does not exceed the threshold range, the display control mechanism will directly use the preset time compensation parameters of the previous detection cycle and send them to the flip drive unit as the execution reference for the current cycle, thereby avoiding unnecessary parameter jitter. S23. If so, the preset time compensation parameter of the previous detection cycle is corrected according to the time deviation code to generate a new preset time compensation parameter. If the comparison result shows that the time deviation exceeds the threshold range, the system will start the parameter correction algorithm, and the control mechanism will correct the time compensation parameter of the previous detection cycle according to the value corresponding to the time deviation code, generate a new preset time compensation parameter, and write the parameter into the control register and database. At the same time, it will be transmitted to the flip drive unit as the control basis for the current cycle, so as to ensure that the timing of the flip action can dynamically follow the changes in the detection environment and the mechanism response.

[0047] In this embodiment, for example, in a certain detection, the time deviation obtained by the system decoding is +45ms, while the preset threshold range is ±20ms. Since this value exceeds the threshold range, the display control mechanism corrects the time compensation parameter of 120ms in the previous cycle to 120ms + 45ms = 165ms, and writes this corrected value as the new preset time compensation parameter into the system, so that the flipping action and the triggering of image acquisition are strictly synchronized. If the time deviation is only +10ms, the system determines that it is within the threshold range and directly uses 120ms from the previous cycle as the current compensation parameter, thereby avoiding frequent system adjustments that cause control instability.

[0048] Furthermore, the step of correcting the preset time compensation parameters of the previous detection cycle based on the time deviation code to generate new preset time compensation parameters includes: S231. Determine the first weighting coefficient and the second weighting coefficient, which sum to 1; according to the system's preset smoothing correction rules, determine a set of first weighting coefficients and second weighting coefficients, the sum of which is fixed to 1, for weighted allocation between historical parameters and real-time deviation values; S232. Multiply the preset time compensation parameter of the previous detection cycle by the first weighting coefficient to generate the corresponding first correction value; the display control mechanism multiplies the preset time compensation parameter stored in the previous detection cycle by the first weighting coefficient to obtain the first correction value, which represents the historical control timing component retained in the new compensation parameter; S233. Multiply the value mapped by the time deviation code by the second weighting coefficient to generate the corresponding second correction value; The system multiplies the value mapped by the time deviation code by the second weighting coefficient to generate the second correction value, which reflects the adaptive correction amount of the new compensation parameter to the current detection environment and the mechanism response deviation. S234. Add the first correction value and the second correction value to generate a new preset time compensation parameter. Add the first correction value and the second correction value to obtain the new preset time compensation parameter, and write the parameter into the database and the drive register as the control reference when the flipping clamping mechanism performs the flipping action in this cycle, thereby realizing the organic combination of smooth transition and dynamic compensation.

[0049] Furthermore, the step of dynamically adjusting the imaging area and / or focus parameters of the visual inspection mechanism by invoking the pose compensation algorithm includes: S301. Based on the position deviation code, determine the corresponding translation offset and tilt angle, wherein both the translation offset and tilt angle can be zero. S302. Dynamically adjust the imaging area of ​​the visual inspection mechanism according to the translation offset; S303. The focusing parameters of the visual inspection mechanism are dynamically adjusted according to the tilt angle.

[0050] In this embodiment, after the flipping clamping mechanism completes the flipping action and the position sensor feeds back the positioning signal, the display control mechanism immediately enters the attitude compensation process. First, the display control mechanism calls the position deviation code in the memory. This code is generated by the difference in attitude parameters between the previous detection cycle and the current flipping process, and parses out the numerical translational offset and tilt angle. The translational offset, in the form of ΔX and ΔY, represents the displacement of the memory bar in the horizontal and vertical directions, and the tilt angle, Δθ, represents the rotation angle of the memory bar around the reference axis. When ΔX, ΔY, and Δθ are all zero, it means that the current workpiece is completely consistent with the ideal detection position and attitude, and no compensation is required. Subsequently, the system transmits ΔX and ΔY as input parameters to the ROI control unit of the image acquisition module. The ROI control unit converts these parameters into pixel coordinate correction values, and moves the image window by corresponding pixels in the horizontal and vertical directions by driving the imaging control circuit or the offset command of the image buffer, thereby realizing the dynamic adjustment of the imaging area. This ensures that the memory bar under test is always located in the center area of ​​the image, avoiding feature loss caused by edge truncation or off-field displacement.

[0051] Next, the display control mechanism transmits Δθ as an input parameter to the autofocus control unit. The autofocus unit calculates the corresponding focal plane offset based on the value of Δθ. When a positive tilt angle is detected, it drives a stepper motor to move the lens forward along the optical axis to shorten the focal length; when a negative tilt angle is detected, it drives the stepper motor to move the lens backward to lengthen the focal length, thereby compensating for focus deviation caused by workpiece tilt. For some systems employing digital focusing, Δθ is mapped to a correction parameter of the image sharpness gradient function. After acquiring the initial image, the autofocus algorithm quickly scans the focal plane and selects the focal length corresponding to the maximum value of the sharpness function as the final focus result, thus ensuring that image details remain sharp at the pixel level.

[0052] Furthermore, in the step of merging and calculating the global feature vectors extracted from the top and bottom surface detection images to output the corresponding global detection results, the defect fusion and discrimination model includes at least a multi-level feature fusion model and a defect discrimination model, including: S401. Extract the corresponding appearance geometric features based on the top surface detection image, and extract the corresponding surface state features based on the bottom surface detection image; S402. Based on the appearance geometric features and surface condition features, perform registration and matching according to the spatial position corresponding to the workpiece's unique identifier to generate the corresponding registration feature group. S403. Input the registered feature group into the multi-level feature fusion model, perform weighted fusion of geometric features and texture features, and generate a global feature vector. S404. Input the global feature vector into the defect discrimination model and output the corresponding global detection result.

[0053] In this embodiment, firstly, the image processing module performs edge detection and geometric modeling algorithms on the top surface detection image to extract appearance geometric features related to the memory module's gold finger contour, component arrangement, and overall dimensions. Simultaneously, it performs texture analysis and brightness distribution calculations on the bottom surface detection image to extract surface state features such as solder joint morphology, welding uniformity, and contamination spots. These features are all represented in vector form and labeled as primary feature data under the current workpiece's unique identifier. Subsequently, the display control mechanism performs spatial registration and matching of the top surface appearance geometric features and the bottom surface state features based on the location information corresponding to the workpiece's unique identifier. This ensures a one-to-one correspondence between the two types of features within the same component area of ​​the same workpiece, thereby generating a complete registration feature group. This feature group is logically bound to the workpiece's unique identifier.

[0054] Next, the system inputs the registered feature set into a multi-level feature fusion model. This model first normalizes the geometric features at a low level, then performs principal component analysis or convolutional feature extraction on the texture features at a high level. It then uses weighted coefficients to fuse features from different levels, obtaining a global feature vector that comprehensively represents the overall state of the workpiece. These weighted coefficients can be pre-set fixed values ​​or adaptively generated by the attention mechanism during model training. Finally, the display control mechanism inputs the global feature vector into a defect discrimination model. This model includes a rapid screening submodule based on threshold rules and a fine-grained classification submodule based on machine learning. In the initial screening stage, the system quickly removes obviously abnormal samples. Then, in the classification stage, it determines the specific type and distribution location of the defects and finally outputs the global detection result. This result includes the workpiece's qualification status, defect category, and defect location coordinates.

[0055] Furthermore, in the step of inputting the global feature vector into the defect discrimination model and outputting the corresponding global detection result, the global detection result includes at least a preliminary screening model and a secondary discrimination model, including: S4041. Input the global feature vector into the preliminary screening model based on rule thresholds to identify whether there are abnormal features that exceed the corresponding abnormal threshold. S4042. Input the global feature vector into a secondary discriminant model based on a machine learning model to classify the defect type of the abnormal features; S4043. Based on the spatial location information contained in the global feature vector, determine the specific location coordinates corresponding to the abnormal features; S4044. Based on the defect type and specific location coordinates of each abnormal feature, output the global detection results of the memory module under test.

[0056] In this embodiment, the global feature vector is fed into a preliminary screening model based on rule-based thresholds. This model includes a threshold comparator and a logic judgment unit to detect whether there are feature components that exceed a preset abnormal threshold, such as excessive deviation in the grayscale value of solder joints or the width of the gold finger edge exceeding the allowable range. When a feature is determined to exceed the threshold, it is marked as an abnormal feature and a flag bit is generated. Subsequently, the system inputs the same global feature vector into a secondary discrimination model based on a machine learning model. This model can be a support vector machine, a convolutional neural network, or an ensemble learning classifier, used to perform fine-grained classification of the defect types of the marked abnormal features, thereby determining whether they belong to specific defect types such as notches, cold solder joints, contamination spots, or device misalignment.

[0057] After completing defect classification, the display control mechanism further invokes the spatial mapping unit to map the defect features to the workpiece surface coordinate system using the spatial location information contained in the global feature vector. This calculates the specific location coordinates of the abnormal features on the memory module, typically outputting millimeter-level physical coordinates or pixel-level image coordinates for subsequent result tracing and localization. Finally, the system integrates the defect type and corresponding location coordinates of each abnormal feature to generate a complete global inspection result. This result includes not only the pass / fail determination information of the memory module under test but also a detailed record of the defect category and spatial location. The result is then output to the display control mechanism or uploaded to the production line quality management system.

[0058] For example, in one inspection, the analysis of the global feature vector shows that the width deviation of the gold finger in a certain area exceeds the threshold, and it is marked as an abnormal feature by the initial screening model; the secondary discrimination model further identifies its type as "gold finger defect"; the spatial mapping unit calculates its coordinates as X=126.4mm, Y=48.9mm based on the feature position parameters; the system's final output result is "unqualified", and the inspection interface displays the defect category as gold finger defect, and the defect location as (126.4, 48.9). Through the above process, the inspection system can quickly screen, accurately classify, and clearly locate defects, achieving efficient, intelligent, and traceable global inspection results.

[0059] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A memory stick vision sensor testing apparatus, comprising: The test device comprises: A support seat (1) provided with two spaced guide rails (4); A display control mechanism (2) located above the support seat (1), and a visual detection mechanism (3) located below the display control mechanism (2), wherein the display control mechanism (2) is electrically connected with the visual detection mechanism (3); A turnover clamping mechanism for clamping and turning over the memory stick to be tested, and located below the visual detection mechanism, the turnover clamping mechanism comprises a tool clamp (5) and a sliding block (6) in sliding fit with the guide rail (4), the two ends of the tool clamp (5) are in rotary fit with the sliding block (6), the sliding block (6) can slide towards or away from the visual detection mechanism (3), so as to slide the memory stick to be tested on the tool clamp (5) to a memory stick placement position or a detection position, and the tool clamp (5) can rotate relative to the sliding block (6), so as to detect the top and bottom defects of the memory stick to be tested.

2. The memory visual sensor testing device of claim 1, wherein, The sliding block (6) comprises a transmission part (61) and a connecting part (62) arranged integrally, the transmission part (61) is in sliding fit with the guide rail (4), and the connecting part (62) is in rotary fit with the tool clamp (5), wherein a connecting part (62) is connected with a rotary motor (7) electrically connected with the display control mechanism, the driving end of the rotary motor passes through the connecting part and is connected with the tool clamp (5), so as to control the memory stick to be tested on the tool clamp (5) to turn over.

3. A memory stick vision sensor testing method, characterized by, The test method applied to the memory stick visual sensor test device of any one of claims 1-2, the test method comprising: When it is detected that the memory stick to be tested has been placed on the turnover clamping mechanism, a workpiece unique identification is assigned to the memory stick to be tested; When it is detected that the memory stick has been placed in the detection position, the light source and the visual system of the visual detection mechanism are controlled to be triggered synchronously to obtain a corresponding top surface detection image, and the top surface detection image is stored in association with the workpiece unique identification; Based on a preset time compensation parameter, the turnover clamping mechanism is controlled to perform a turnover action, and after the turnover is completed, a posture compensation algorithm is called to dynamically adjust the imaging area and / or focusing parameter of the visual detection mechanism; When it is detected that the visual detection mechanism has been dynamically adjusted, the visual detection mechanism is controlled to obtain a corresponding bottom surface detection image, and the bottom surface detection image is stored in association with the same workpiece unique identification; The top surface detection image and the bottom surface detection image associated with the same workpiece unique identification are jointly input into a defect fusion discrimination model to perform merging calculation based on the global feature vectors extracted from the top surface detection image and the bottom surface detection image, and then a corresponding global detection result is output.

4. The memory visual sensor test method of claim 3, wherein, In the step of assigning a workpiece unique identification to the memory stick to be tested when it is detected that the memory stick to be tested has been placed on the turnover clamping mechanism, the step comprises: acquire timestamp information generated when it is detected that the to-be-tested memory bank has been placed in the turnover clamping mechanism, and a preset flow sequence number of the to-be-tested memory bank, generate a corresponding initial identification according to the timestamp information and the flow sequence number; index detection position information of a previous detection cycle, encode the initial identification according to the detection position information of the previous detection cycle, and generate and assign a corresponding workpiece unique identification for the to-be-tested memory bank.

5. The memory stick visual sensor test method of claim 4, wherein, In the step of encoding the initial identification according to the detection position information of the previous detection cycle to generate and assign a corresponding workpiece unique identification for the to-be-tested memory bank, the step includes: determining a top surface detection time, a turnover actual completion time and an expected completion time of the previous detection cycle according to the detection position information of the previous detection cycle; acquiring a first attitude parameter of the top surface detection time, acquiring a second attitude parameter of the turnover actual completion time, and determining a corresponding position deviation code according to the first attitude parameter and the second attitude parameter, the position deviation code being used to participate in a work flow of dynamically adjusting an imaging area and / or a focusing parameter of the visual detection mechanism; calculating a time deviation value between the turnover actual completion time and the expected completion time, and determining a corresponding time deviation code according to the time deviation value, the time deviation code being used to determine a preset time compensation parameter; encoding the initial identification according to the position deviation code and the time deviation code to generate and assign a corresponding workpiece unique identification for the to-be-tested memory bank.

6. The memory stick visual sensor test method of claim 5, wherein, Before the step of controlling the turnover clamping mechanism to perform a turnover action based on the preset time compensation parameter, the step includes: determining whether a value mapped by the time deviation code is located within a preset deviation threshold interval; if not, the preset time compensation parameter of the previous detection cycle is used; if yes, the preset time compensation parameter of the previous detection cycle is modified according to the time deviation code to generate a new preset time compensation parameter.

7. The memory stick visual sensor test method of claim 6, wherein, In the step of modifying the preset time compensation parameter of the previous detection cycle according to the time deviation code to generate a new preset time compensation parameter, the step includes: determining a first weight coefficient and a second weight coefficient that are both 1; multiplying the preset time compensation parameter of the previous detection cycle by the first weight coefficient to generate a corresponding first modified value; multiplying a value mapped by the time deviation code by the second weight coefficient to generate a corresponding second modified value; adding the first modified value and the second modified value to generate a new preset time compensation parameter.

8. The memory stick visual sensor test method of claim 5, wherein, In the step of calling a posture compensation algorithm to dynamically adjust the imaging area and / or the focusing parameter of the visual detection mechanism, the step includes: determining a translation offset and an inclination angle according to the position deviation code, wherein the translation offset and the inclination angle can both take a zero value; dynamically adjusting the imaging area of the visual detection mechanism according to the translation offset; dynamically adjusting the focusing parameter of the visual detection mechanism according to the inclination angle.

9. The memory stick visual sensor test method of claim 3, wherein, In the step of merging and calculating the global feature vectors extracted based on the top surface detection image and the bottom surface detection image, and then outputting the corresponding global detection result, the defect fusion discrimination model at least includes a multi-level feature fusion model and a defect discrimination model, which includes: Based on the top surface detection image, corresponding appearance geometric features are extracted, and based on the bottom surface detection image, corresponding surface state features are extracted; According to the appearance geometric features and the surface state features, registration matching is performed according to the spatial positions corresponding to the unique identification of the workpiece, to generate corresponding registration feature groups; The registration feature groups are input into the multi-level feature fusion model, the geometric features and the texture features are weighted and fused to generate global feature vectors; The global feature vectors are input into the defect discrimination model to output the corresponding global detection result.

10. The memory stick visual sensor testing method of claim 9, wherein, In the step of inputting the global feature vectors into the defect discrimination model to output the corresponding global detection result, the global detection result at least includes a preliminary screening model and a secondary discrimination model, which includes: The global feature vectors are input into the preliminary screening model based on the rule threshold to identify whether there is an abnormal feature exceeding the corresponding abnormal threshold; The global feature vectors are input into the secondary discrimination model based on the machine learning model to classify the defect types of the abnormal features; Based on the spatial position information contained in the global feature vectors, the specific position coordinates corresponding to the abnormal features are determined; According to the defect types and the specific position coordinates of each abnormal feature, the global detection result of the to-be-tested memory bar is output correspondingly.

Images