A Method and System for Controlling Methylene Blue Spot Sample Tests Based on a Multimodal Large Model

Abstract

This invention relates to the field of methylene blue testing and discloses a control method and system for methylene blue spot testing based on a multimodal large model. The system includes a housing with a liquid addition module for adding methylene blue solution, a stirring bottle for preparing the suspension, a stirring module for stirring the suspension, a spotting paper, and a rotation module for rotating the spotting paper. An electrically operated telescopic rod is rotatably connected inside the housing. The piston end of the electric telescopic rod has a glass rod for dipping into the suspension. A pushing module for rotating the electric telescopic rod is also located inside the housing. The system also includes a detection camera located above the spotting paper, which houses an intelligent test control calculation module for running the system program and achieving closed-loop control. The intelligent test control calculation module includes: a multi-source data access module; an event trigger alignment and preprocessing module; a multimodal large model inference module; and a guardrail closed-loop control module. This invention can obtain stable, reproducible control and objective judgment results.

Description

Technical Field

[0001] This invention relates to the field of methylene blue testing, and specifically to a method and system for controlling methylene blue spot testing based on a multimodal large model. Background Technology

[0002] Methylene blue spot testing is a common colorimetric / colorimetric detection process. It typically involves spotting the reaction solution onto test paper to form a spot, and the result is determined based on the color of the spot and its outer light blue halo. As detection tasks develop towards large-scale, standardized, and equipment-based processes, the steps of spotting, adding materials, stirring, imaging acquisition, and result interpretation are gradually evolving from manual operation to automation and image analysis methods, in order to obtain more stable detection processes and more consistent interpretation results.

[0003] Existing solutions typically employ an actuator to complete the feeding and mixing, and a dipping-type sampling mechanism to spot the reaction solution onto preset points on the test strip. Subsequently, a camera and light source are used to acquire images of the spot (or short-time images). Then, based on methods such as fixed thresholds, color space discrimination, edge / region feature extraction, and template matching, the features of the spot area and its outer edge halo are analyzed to provide a positive / negative or corresponding grade judgment result. Some systems will preset controls on the sampling, photography sequence, and feeding steps, and handle abnormalities by manual review or manual re-processing.

[0004] In practical applications, the dipping-type sampling process is easily affected by factors such as immersion depth, immersion time, lifting speed, and sampling height / contact time, resulting in fluctuations in the sampling amount, droplet morphology, and landing point, and may lead to phenomena such as dripping, continuous dripping, splashing, deviation, and contamination. At the same time, color halo characteristics are often sensitive to differences in paper batches, paper surface texture, and reflective interference, which limits the stability of image interpretation based on fixed parameters or a single rule under different working conditions. Therefore, in scenarios with minimal human intervention, it is often difficult to obtain stable, reproducible experimental control and objective, consistent judgment results under conditions where sampling process fluctuations and multi-source interference coexist. Summary of the Invention

[0005] Based on the above description, the present invention provides a method and system for controlling methylene blue spot sampling tests based on a multimodal large model, in order to solve the problem in the prior art that it is difficult to stably achieve reproducible control and objective consistency judgment of methylene blue spot sampling tests under the combined effects of spotting process fluctuations and multi-source imaging interference.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: a methylene blue spot test control system based on a multimodal large model, including a shell, a liquid addition module for adding methylene blue solution, a stirring bottle for mixing suspension, a stirring module for stirring suspension, a spotting paper and a rotation module for driving the spotting paper to rotate.

[0007] An electric telescopic rod two is rotatably connected inside the outer shell. The piston end of the electric telescopic rod two is provided with a glass rod for dipping the suspension. The outer shell is also provided with a push module for pushing the electric telescopic rod two to rotate.

[0008] It also includes a detection camera located above the sample paper, which is equipped with an intelligent test control calculation module for running system programs and realizing closed-loop control;

[0009] The intelligent test control calculation module includes:

[0010] The multi-source data access module is used to acquire visual data corresponding to the droplet spots and process data corresponding to the spotting process, and write timestamps for the visual data and process data respectively.

[0011] The event-triggered alignment and preprocessing module is used to extract frame sequences or short video clips containing preset durations before and after the sampling event from the visual data, using the "sampling event time" as the alignment benchmark, and then align the frame sequences or short video clips with the process data according to the timestamp and encapsulate them into unified input data.

[0012] The multimodal large model inference module is used to input unified input data into the multimodal large model according to the preset structured output specification and output structured inference results.

[0013] The guardrail closed-loop control module is used to perform confidence gating and consistency verification based on structured reasoning results, and to generate control commands based on the anomaly-action mapping relationship to drive the execution of one of the following actions: adjusting sampling parameters, adjusting imaging parameters, and resampling;

[0014] The recording and reporting module is used to associate and record unified input data, structured reasoning results, control instructions and their triggering reasons, and output a traceable report containing key evidence frames. The structured reasoning results include at least: color halo judgment results and confidence levels, spot sampling validity judgment results, anomaly types and confidence levels, control suggestion fields and evidence fields.

[0015] The above technical solution enables automated methylene blue testing without excessive human intervention. It allows for the simultaneous acquisition and timestamping of visual data of the droplet and data from the spotting process. The system extracts and aligns frame sequences before and after the spotting event, forming a unified input. A multimodal large model then outputs a structured result containing confidence levels for color halo assessment, spotting validity, anomaly types, and control recommendations. Furthermore, based on confidence gating and consistency verification, control commands are generated to drive actions such as adjusting spotting parameters, adjusting imaging parameters, or resampling, forming a closed-loop error correction. Simultaneously, key evidence frames, control commands, and triggering causes are linked and stored, thereby improving the stability of interpretation and the reproducibility of control under conditions of both spotting fluctuations and imaging interference.

[0016] Based on the above technical solution, the present invention can be further improved as follows.

[0017] Furthermore, the intelligent test control calculation module is electrically or communicatively connected to the liquid addition module, the stirring module, the rotation module, the electric telescopic rod II, the pushing module, and the detection camera, respectively.

[0018] The liquid addition module includes a storage tank on the outer shell, a liquid addition pipe connected to the inside of the storage tank, the bottom end of the liquid addition pipe being located inside the lower part of the storage tank, and a miniature metering pump connected to the inside of the liquid addition pipe.

[0019] The outer shell is provided with a protective shell, and a support frame is provided on the outside of the protective shell. The liquid addition tube is snapped onto the support frame, and the end of the liquid addition tube away from the storage tank is located at the top of the stirring bottle. The outer shell is provided with a limiting sleeve, and the stirring bottle is inserted into the inside of the limiting sleeve.

[0020] The rotating module includes a servo motor located inside the housing. The output shaft of the servo motor passes through the top of the housing and is fixed to a rotating disk. The sample paper is placed on the top of the rotating disk. The rotating disk has an ergonomic groove for suspending the sample paper. The diameter of the ergonomic groove is smaller than the diameter of the sample paper. A pressure ring for pressing the outer ring of the sample paper is placed on the rotating disk.

[0021] Furthermore, the stirring module includes an electric telescopic rod 1 disposed inside the outer shell, and the protective shell is located between the outer sides of the electric telescopic rod 1 and the electric telescopic rod 2. The top of the protective shell has two movable holes that allow the piston ends of the electric telescopic rod 1 and the electric telescopic rod 2 to pass through and move.

[0022] The piston end of the electric telescopic rod is provided with an extension arm, and the top of the extension arm away from the electric telescopic rod is provided with a servo motor 2. The output shaft of the servo motor 2 passes through the extension arm and is provided with a stirring rod, which is located inside the stirring bottle.

[0023] The pushing module includes an electric telescopic push rod and a bearing seat located inside the outer shell. A rotating rod is rotatably connected to the bearing seat via a bearing. The rotating rod is fixed to the bottom of the electric telescopic rod. A connecting sleeve is fixed to the outside of the rotating rod. A push plate is provided on the outside of the connecting sleeve. A long strip-shaped sliding hole is opened on the push plate. A limit sliding rod is slidably connected in the sliding hole.

[0024] The limiting slide rod consists of a cylindrical rod, a limiting plate located at one end of the cylindrical rod, and a nut detachably connected to the other end of the cylindrical rod. The cylindrical rod is slidably connected in the sliding hole. The opposite sides of the limiting plate and the nut are both in contact with the outer side of the push plate, and the contact surfaces are both smooth.

[0025] The piston end of the electric telescopic push rod is hinged to the limiting plate by a pin. The push plate is provided with a sensing plate on the side away from the connecting sleeve. The housing is also provided with a proximity sensor that senses the sensing plate.

[0026] The piston end of the electric telescopic rod 2 is fixed with a connecting plate. The glass rod is located on the side of the connecting plate away from the electric telescopic rod 2. The glass rod moves between the sample paper and the stirring bottle. The detection camera is located on the outside of the protective shell.

[0027] Furthermore, the event trigger alignment and preprocessing module is used to sample event timestamps. Using this as a baseline, time windows are extracted from visual data. The frame sequence within is used as the visual part of the unified input data, in which and The preset positive value;

[0028] Furthermore, the process data section of the unified input data should include at least three of the following sampling parameters and their timestamps: immersion depth, immersion duration, lifting speed, sampling height, contact time, and landing point coordinates. Optional parameters include feeding amount and feeding sequence, stirring status, point number, and positioning calibration information.

[0029] The preprocessing includes at least one of the following: white balance / color calibration, reflection suppression, exposure normalization, spot ROI positioning and cropping, in order to reduce the impact of paper texture, ambient light changes or reflections on color fringing interpretation.

[0030] Furthermore, the structured reasoning result satisfies a preset field set and includes at least:

[0031] Color halo determination fields: color halo present / absent / uncertain and corresponding confidence level;

[0032] Sampling validity field: valid / invalid;

[0033] Anomaly type field: Output at least one anomaly type and its corresponding confidence level from the following: dripping, continuous dripping, splashing, offset, contamination, reflection interference, paper defect, and dripping crosstalk;

[0034] Control suggestion field: includes at least one of the following: the parameter to be adjusted, the adjustment direction (increase / decrease / hold), and the resampling wait time or the number of resampling times; the control suggestion field is used by the guardrail closed-loop control module to directly parse and generate control commands;

[0035] Evidence fields: include at least one of the keyframe index and the ROI coordinates of the droplet or halo candidate region.

[0036] Furthermore, the guardrail closed-loop control module includes:

[0037] The confidence level gating submodule is used when the confidence level of the hue determination field is less than a threshold. Or the confidence level of the exception type field is less than the threshold. When, trigger delay Post-resampling at least Second-rate;

[0038] The consistency verification submodule is used to verify the consistency of the data. The color halo determination result of the second resampling is subject to majority voting or consistency determination. If the consistency does not meet the preset conditions, an uncertain conclusion is output and at least one of changing the point or changing the paper is triggered.

[0039] The fallback judgment submodule is used to conservatively verify the existence of color halo by calling the rule judgment based on color difference / boundary region contrast when the confidence gating is triggered and an uncertain conclusion is still output, and write the fallback judgment result and the triggering reason into the traceable report.

[0040] Furthermore, the guardrail closed-loop control module includes anomaly-action mapping relationships, and contains at least the following three types of anomaly mapping rules:

[0041] When the anomaly type is dripping, a control command is generated to reduce the lifting speed or increase the spotting residence time, and the spotting is redone.

[0042] When the anomaly type is splash, a control command is generated to change the sampling location or reduce the sampling height, and the sampling is redone.

[0043] When the anomaly type is reflective interference, control instructions are generated to adjust imaging parameters and mask the reflective area, and resampling is triggered.

[0044] Furthermore, the recording and reporting module associates and stores evidence of the exception type, the hit mapping rule, the keyframe index, and the corresponding control command.

[0045] This invention also provides a control method for methylene blue spot sampling tests based on a multimodal large model. The method, applying the aforementioned control system for methylene blue spot sampling tests based on a multimodal large model, includes the following steps:

[0046] S1. Prepare a 10.0 g / L ± 0.1 g / L methylene blue solution. Weigh fine aggregate and clean water into a stirring bottle and stir to form a suspension. Add the initial methylene blue solution and continue stirring. Then, drive the glass rod to dip into the mixed suspension and drop it onto the spotting paper in the air by the cooperation of the push module and the electric telescopic rod II. This triggers the spotting event and acquires the process data corresponding to the spotting event. At the same time, visual data corresponding to the drop spot is collected.

[0047] S2. Based on the timing of the sampling event, the visual data is captured by event triggering and aligned with the process data according to the timestamp to generate unified input data;

[0048] S3. Input the unified input data into the multimodal large model and output the structured reasoning results including color halo determination and its confidence level, spot sampling validity, anomaly type and its confidence level, control suggestion field and evidence field;

[0049] S4. Based on the structured reasoning results, perform confidence gating and consistency verification, and generate control commands to drive the equipment to perform at least one of parameter adjustment, resampling, resampling, changing sampling point, or changing paper.

[0050] S5. After executing the control command, repeat steps S1 to S4 until the termination condition is met.

[0051] S6. Generate and output a traceable report containing key evidence frames.

[0052] Furthermore, the spotting events in S1 include the glass rod dipping into the suspension, the glass rod moving, the glass rod spotting on the spotting paper, and the collection of spotting droplets.

[0053] The S2 includes point sample event timestamps. Use the time window as a reference The frame sequence; and perform at least one preprocessing on the frame sequence: white balance / color calibration, exposure normalization, reflection suppression or reflection mask marking, and spot ROI positioning and cropping based on point number or positioning calibration information;

[0054] The repeated process in S5 starts from the point sample event triggered in S1 and ends in step S4.

[0055] Furthermore, S4 includes when the confidence level for color halo determination is less than a threshold. Or the confidence level of the anomaly type is less than the threshold. Time, delay At least after execution Secondary resampling; for The resampling results are subject to majority voting or consistency determination; if the consistency does not meet the preset conditions, at least one of the following is triggered: change the sampling point or change the paper, and the sampling is redone.

[0056] The S4 also includes error correction control based on the anomaly type: when the anomaly type is dripping, the sample is redone after adjusting the parameters by decreasing the lifting speed or increasing the sample dwell time; when the anomaly type is splashing, the sample is redone after changing the spot position or decreasing the spot height; when the anomaly type is reflective interference, the exposure is reduced and reflective mask marking is performed before resampling; and a maximum number of redoes or maximum number of resamplings is set for each error correction action.

[0057] The traceable report output by S6 includes at least: time window identifiers corresponding to unified input data, key frame indexes and ROI coordinates, sampling parameters and their timestamps, structured inference results, confidence gating trigger reasons, consistency judgment results, anomaly-action mapping hit items, and control instruction logs, and associates key frame indexes with corresponding trigger reasons and control instructions.

[0058] Compared with the prior art, the technical solution of this application has the following beneficial technical effects:

[0059] 1. This invention achieves automatic identification and error correction of abnormal and uncertain situations through "multi-source data fusion with point sample event alignment + multi-modal large model structured reasoning + guardrail closed-loop control", thereby obtaining more stable and reproducible control and more consistent objective judgment results under complex working conditions;

[0060] 2. This invention introduces a confidence gating and consistency verification mechanism, which can automatically trigger resampling or parameter adjustment in low confidence or interference scenarios. It also forms a traceable evidence chain by recording the association between key evidence frames and triggering reasons, reducing misjudgments and human subjective differences. Furthermore, it relies on a semi-automated testing system to achieve quantitative addition, automatic mixing, fixed-point sampling, and visual recognition, thereby improving reliability and auditability under conditions with minimal human intervention. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the overall structure of a methylene blue spot sampling test control system based on a multimodal large model, provided in an embodiment of the present invention.

[0062] Figure 2 This is a cross-sectional structural diagram of an embodiment of the present invention;

[0063] Figure 3 This is a schematic diagram of the connection structure of the electric telescopic rod II in an embodiment of the present invention;

[0064] Figure 4 This is an embodiment of the present invention. Figure 3 A magnified view of part A in the middle;

[0065] Figure 5 This is a block diagram of the module connection of the methylene blue spot sample test control system based on a multimodal large model in an embodiment of the present invention;

[0066] Figure 6 This is an internal functional block diagram of the intelligent test control calculation module in an embodiment of the present invention;

[0067] Figure 7 This is a schematic diagram of the closed-loop control and data flow logic of the methylene blue spot test control system in an embodiment of the present invention;

[0068] Figure 8 This is a flowchart illustrating the methylene blue spot sample test control method based on a multimodal large model in Embodiment 2 of the present invention.

[0069] Reference numerals: 1. Outer casing; 2. Liquid filling module; 21. Storage tank; 22. Liquid filling pipe; 23. Micro metering pump; 24. Support frame;

[0070] 31. Limiting sleeve; 32. Stirring bottle; 33. Servo motor 1; 34. Rotating disk; 35. Sample paper; 36. Pressure ring;

[0071] 4. Protective shell; 51. Electric telescopic rod one; 52. Servo motor two; 53. Extension arm; 54. Stirring rod;

[0072] 6. Push module; 61. Electric telescopic push rod; 62. Bearing seat; 63. Rotating rod; 64. Connecting sleeve; 65. Push plate; 66. Sliding hole; 67. Sensing plate; 68. Proximity sensor; 69. Limiting slide rod;

[0073] 71. Two electric telescopic poles; 72. Glass rod; 73. Connecting plate; 8. Detection camera. Detailed Implementation

[0074] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0075] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0076] Example 1:

[0077] refer to Figure 1 and Figure 5The methylene blue spotting test control system based on a multimodal large model includes a housing 1, which is equipped with a liquid addition module 2 for adding methylene blue solution, a stirring bottle 32 for mixing the suspension, a stirring module for stirring the suspension, a spotting paper 35, and a rotation module for rotating the spotting paper 35. A quantitative amount of methylene blue solution is added to the stirring bottle 32 via the liquid addition module 2, and then mixed by the stirring module. An electric telescopic rod 71 is rotatably connected inside the housing 1. The piston end of the electric telescopic rod 71 is equipped with a glass rod 72 for dipping into the suspension. A push module 6 is also provided inside the housing 1 to drive the electric telescopic rod 71 to rotate, allowing the glass rod 72, after dipping into the suspension, to move above the spotting paper 35. A detection camera 8 is also included above the spotting paper 35 to monitor the spotting paper 35 and its surrounding environment, ensuring real-time acquisition of the test results.

[0078] The detection camera 8 is equipped with an intelligent test control computing module for running system programs and implementing closed-loop control. This module is electrically or communicatively connected to the liquid addition module 2, the stirring module rotation module, the electric telescopic rod 71, the pushing module 6, and the detection camera 8. It outputs control commands and receives operational status information. In this embodiment, the intelligent test control computing module can be implemented using a single-board computer, an embedded industrial control computer, or an edge computing board with AI computing power. It includes a processor, memory, and a communication interface. The communication interface can be a wired interface (e.g., RS485, CAN, UART, USB, Ethernet) or a wireless interface (e.g., Wi-Fi, Bluetooth) to adapt to the controllers of different execution modules. To achieve closed-loop control, the intelligent test control computing module sends control commands to each execution module. The control commands include at least: liquid volume / liquid timing commands for the liquid addition module 2, rotation speed / duration commands for the stirring module, target angle / target position commands for the rotation module, extension / speed / position holding time commands for the electric telescopic rod 71, and rotation angle / speed / return-to-zero commands for the push module 6. Simultaneously, each execution module returns operating status information to the intelligent test control calculation module. This operating status information includes at least: current execution status (idle / executing / completed / fault), measured values ​​of key parameters (e.g., current stroke, current angle, current speed), and abnormal information (e.g., blockage, overload, loss of synchronization, communication anomaly). Status information can be reported periodically (e.g., every 10ms to 100ms) or by events (e.g., upon reaching the desired position) to meet the real-time requirements of closed-loop control.

[0079] In addition, the liquid addition module 2, stirring module, rotation module, electric telescopic rod 71, and pushing module 6 each have at least one controllable parameter and are set with a corresponding status feedback interface. The status feedback interface outputs process data containing timestamps to the intelligent test control calculation module. The process data includes at least one of the following: liquid addition volume and sequence, stirring speed or stirring duration, rotation angle or point number, and displacement stroke, extension speed, or arrival time of electric telescopic rod 71. In addition, to ensure that multi-source data can be aligned, the intelligent test control calculation module adds a timestamp to the feedback data from each module. The system clock of the intelligent test control calculation module can be used as a reference to record the arrival time of each status feedback record. Among them, multi-source data refers to the visual data corresponding to the droplet and the process data corresponding to the spotting process. The alignment accuracy is improved by setting a fixed reporting cycle and recording communication delay statistics. Furthermore, the process data can be organized in the form of time-series logs in the storage structure. Each log contains at least: module identifier, parameter name, parameter value, timestamp, and running status code.

[0080] In one optional embodiment, a controllable light source module is included, which works in conjunction with the detection camera 8. This module has an interface for adjusting the light source level or brightness and is connected to an intelligent test control calculation module. This allows the intelligent test control calculation module to perform at least one of the following operations when acquiring droplet images: reducing exposure or adjusting the light source level, to suppress reflective interference. In this embodiment, the controllable light source module can use a ring LED, a strip LED, or a surface light source structure, installed around the lens of the detection camera 8 or above the sample paper 35. Its brightness or level can be controlled via PWM dimming, constant current level switching, or a digital interface. The intelligent test control calculation module adjusts the brightness based on the high-brightness saturation in the acquired image. The system uses indicators such as area occupancy, local contrast, or visibility of color halo boundaries to determine the presence of reflective interference. When reflective interference is detected, the intelligent test control calculation module performs at least one of the following measures: reducing the exposure time or gain through the control interface of the detection camera 8; reducing the brightness of the light source or switching to a low level through the light source control interface; and further generating mask marks for the reflective areas, reducing the weight of the masked areas or removing them in subsequent image analysis or multimodal inference to reduce the impact of reflection on color halo determination. After completing the exposure / light source adjustment, the intelligent test control calculation module triggers resampling to acquire the droplet image to ensure that the visual data entering inference and evidence storage meets the readability requirements.

[0081] refer to Figure 2The liquid addition module 2 includes a storage tank 21 mounted on the outer casing 1. The storage tank 21 has a liquid addition pipe 22 connected to its interior. The bottom end of the liquid addition pipe 22 is located below the interior of the storage tank 21. A micro metering pump 23 connected to the interior of the liquid addition pipe 22 is mounted on the liquid addition pipe 22. The micro metering pump can obtain the amount of liquid flowing in real time to achieve quantitative addition. The outer casing 1 has a protective shell 4, and a support frame 24 is located on the outside of the protective shell 4. The liquid addition pipe 22 is snapped onto the support frame 24. The liquid addition pipe 22 is located at... One end of the storage tank 21 is located at the top of the stirring flask 32, allowing the liquid addition tube 22 to deliver liquid into the stirring flask 32; a limiting sleeve 31 is provided on the outer shell 1, and the stirring flask 32 is inserted into the inner side of the limiting sleeve 31 to limit the placement of the stirring flask 32; specifically, the micro metering pump 23 can be a MNX-100-S laboratory small metering pump, and through the pumping of the micro metering pump 23, the methylene blue solution in the storage tank 21 can be quantitatively discharged into the stirring flask 32 through the liquid addition tube 22 for addition.

[0082] refer to Figure 2 The rotating module includes a servo motor 33 housed within the outer casing 1. The output shaft of the servo motor 33 passes through the top of the outer casing 1 and is fixed to a rotating disk 34, allowing the servo motor 33 to drive the rotating disk 34 to rotate. A sample paper 35 is placed on top of the rotating disk 34. The rotating disk 34 has an suspending groove for suspending the sample paper 35. The diameter of the suspending groove is smaller than the diameter of the sample paper 35. A pressure ring 36 is placed on the rotating disk 34 to press the outer ring of the sample paper 35, allowing the sample paper 35 to be placed on the rotating disk 34 via the pressure ring 36. The middle portion of the sample paper is suspended due to the suspending groove, allowing the suspension to drip onto the sample paper 35. Normal diffusion; it should be noted that, under normal circumstances, the total diameter of the droplets will not exceed 15mm. The rotation angle of the servo motor 33 driving the rotating disk 34 is controlled based on 20mm. The rotation angle is determined according to the actual size of the rotating disk 34 and the sample paper 35. This embodiment provides a size example. In this embodiment, the diameter of the sample paper 35 and the rotating disk 34 is 120mm. A sample area is set at 22.56°. A total of 12 areas can be set, and each area does not interfere with the others, ensuring the feasibility of the sample test. The set sample areas are based on the actual test conditions, as long as the droplets in each area do not interfere with each other.

[0083] refer to Figure 1 and Figure 2The stirring module includes an electric telescopic rod 51 housed within the outer casing 1. A protective casing 4 is located between the outer sides of the electric telescopic rod 51 and the electric telescopic rod 71. The top of the protective casing 4 has two movable holes that allow the piston ends of the electric telescopic rods 51 and 71 to move through, respectively. This allows the protective casing 4 to shield and protect the electric telescopic rods 51 and 71, and also allows the piston ends of the electric telescopic rods 51 and 71 to move normally without being restricted by the protective casing 4. The piston end of the electric telescopic rod 51 is equipped with an extension arm 5. 3. This allows the extension arm 53 to be driven to move by the electric telescopic rod 51. A servo motor 52 is provided on the side of the top of the extension arm 53 away from the electric telescopic rod 51. The output shaft of the servo motor 52 passes through the extension arm 53 and is equipped with a stirring rod 54. The stirring rod 54 is located inside the stirring bottle 32. The stirring rod 54 includes a rod body and a stirring blade at the bottom of the rod body. The stirring blade can increase the stirring efficiency, so that the suspension can be mixed more quickly. In use, the servo motor 52 drives the stirring rod 54 to rotate, and the stirring rod 54 stirs the liquid in the stirring bottle 32, so that it can be mixed evenly.

[0084] refer to Figure 3 and Figure 4 The pushing module 6 includes an electric telescopic push rod 61 and a bearing seat 62 located inside the outer casing 1. A rotating rod 63 is rotatably connected to the bearing seat 62 via a bearing. The rotating rod 63 is fixed to the bottom of the electric telescopic rod 71, allowing the electric telescopic rod 71 to rotate via the rotating rod 63. A connecting sleeve 64 is fixed to the outside of the rotating rod 63, and a push plate 65 is provided on the outside of the connecting sleeve 64. When the push plate 65 swings, it can drive the rotating rod 63 to rotate via the connecting sleeve 64. An elongated sliding hole 66 is provided on the push plate 65, and a limiting slide rod 69 is slidably connected in the sliding hole 66. The limiting slide rod 69 is made of a circular... The cylindrical rod consists of a column rod, a limiting plate at one end of the cylindrical rod, and a nut detachably connected to the other end of the cylindrical rod. The cylindrical rod is slidably connected in the sliding hole 66. The opposite sides of the limiting plate and the nut are in contact with the outer side of the push plate 65, and the contact surfaces are smooth. Through the limiting cooperation between the limiting plate and the nut, the cylindrical rod can slide in the sliding hole 66, and then the push plate 65 can be driven to swing through the linkage of the limiting slide rod 69. The smooth surface of the opposite sides of the nut and the limiting plate can greatly reduce the friction when driving the push plate 65 to swing, so that it can swing normally.

[0085] refer to Figure 4The piston end of the electric telescopic push rod 61 is hinged to the limiting plate via a pin. The limiting plate can swing horizontally through the hinge of the pin. Specifically, this allows the electric telescopic push rod 61 to push the limiting plate to slide along the surface of the push plate 65 through the hinge of the pin. A sensing plate 67 is provided on the side of the push plate 65 away from the connecting sleeve 64. A proximity sensor 68 is also provided inside the housing 1 to sense the sensing plate 67, so that the proximity sensor 68 can sense the sensing plate 67 and trigger a stop signal. Specifically, when the sensing plate 67 approaches the proximity sensor 68 and triggers the sensing signal of the proximity sensor 68, the glass rod 72 is located inside the stirring flask 32. When the electric telescopic push rod 61 drives the push plate 65, it moves the glass rod 72. When the electric telescopic push rod 61 reaches the set stroke positioning value, the glass rod 72 is positioned above the sampling paper 35. This setting allows the glass rod 72 to sample the suspension in the stirring bottle 32 normally through the accurate sensing of the proximity sensor 68. Due to the high precision of the induction trigger, there will be no contact with the stirring rod 54. The process of moving the glass rod 72 above the sampling paper 35 does not require high precision; it is only necessary to ensure that the glass rod 72 is above the sampling paper 35. Furthermore, the specific model of the proximity sensor 68 can be an SJ30-A2 inductive slotted proximity switch, and the sensing piece 67 is a metal piece to realize the function of induction control. In addition, any sensor that can meet the effect of real-time and accurate proximity sensing can be used.

[0086] refer to Figure 2 and Figure 3 A connecting plate 73 is fixed to the piston end of the electric telescopic rod 71. The glass rod 72 is located on the side of the connecting plate 73 away from the electric telescopic rod 71, so that the electric telescopic rod 71 can drive the glass rod 72 to move together through the connecting plate 73, and the glass rod 72 can follow the electric telescopic rod 71 to perform circular motion under the rotation of the rotating rod 63 through the connecting plate 73. The detection camera 8 is located on the outside of the protective shell 4, so that the detection camera 8 can be installed above the spotting paper 35, so that it can acquire visual information about the spotting paper 35 and its surroundings.

[0087] In this embodiment, the controllable parameters of each module are set and executed by the driver or controller. The controllable parameters of the liquid addition module 2 include single liquid addition volume, liquid addition rate, number of liquid additions, and adjacent liquid addition intervals. Its status feedback interface outputs the liquid addition start time, liquid addition end time, cumulative liquid addition volume, instantaneous flow rate, or pump steps, etc. Specifically, the quantitative addition effect is achieved by setting the parameters of the micro metering pump 23. The controllable parameters of the stirring module include stirring speed, stirring duration, and stirring start / stop time. Its status feedback interface outputs speed feedback, running time, overload status, etc. By setting the speed of the servo motor 2 52, the stirring rate of the stirring rod 54 can be controlled. The controllable parameters of the rotation module include target angle, angular velocity, and homing / positioning strategy. Its status feedback interface outputs the current angle, target angle, angular velocity, angular velocity, and homing / positioning strategy, etc. The angle arrival mark, point number, or encoder count are determined by the parameter settings of servo motor 33. The controllable parameters of electric telescopic rod 71 include telescopic stroke, telescopic speed, and position holding time (i.e., the time to maintain contact or hover after reaching the position). Its status feedback interface outputs the current stroke, speed estimate, position mark, and position time. The controllable parameters of push module 6 include the rotation angle and speed of electric telescopic rod 71. Its status feedback interface outputs the current rotation angle, position mark, and whether the return to zero was successful. The fixed-point positioning function is achieved by proximity sensor 68 in conjunction with sensing plate 67. Electric telescopic push rod 61 drives push plate 65 to move glass rod 72. By setting the stroke value of electric telescopic push rod 61, glass rod 72 can move normally above the sample paper 35.

[0088] refer to Figure 6 The intelligent experimental control computing module includes: a multi-source data access module, an event trigger alignment and preprocessing module, a multimodal large model inference module, a guardrail closed-loop control module, and a recording and reporting module. The multi-source data access module acquires visual data corresponding to the droplets and process data corresponding to the sampling process, and writes timestamps to both the visual and process data. By adding a unified time stamp to data from different sources, a prerequisite can be provided for establishing data correspondences around the same sampling event.

[0089] In this embodiment, the visual data can be a continuous image stream, discrete frame sequence, or short video clip acquired by the detection camera 8; the process data can be a parameter data stream generated during the spotting action, including one or more of the following process parameters: the depth of the glass rod 72 immersed in the stirring bottle 32, the immersion time, the speed at which the electric telescopic rod 71 drives the glass rod 72 to lift upward, the spotting height of the glass rod 72 when it moves onto the spotting paper 35, the contact time between the glass rod 72 and the spotting paper 35, and the coordinates of the spotting landing point.

[0090] Furthermore, the event trigger alignment and preprocessing module uses the "spotting event moment" as the alignment benchmark to extract frame sequences or short video clips containing preset durations before and after the spotting event from the visual data, and then aligns the frame sequences or short video clips with the process data according to the timestamp and encapsulates them into unified input data. Through the above settings, the spotting action itself and the spot formation process can be included in the same analysis unit; preferably, the spotting event moment is recorded as... ;Should The timing of the droplet contact with the bearing surface can be determined by the spotting action trigger signal, the pose positioning signal, the contact detection signal, or visual detection. To improve system stability, the action trigger signal or the positioning signal output by the actuator is preferred. The basis for determination; furthermore, the event trigger alignment and preprocessing module is configured to: use point sample event times. Using this as a baseline, time windows are extracted from visual data. The frame sequence within is used as the visual part of the unified input data, in which and The preset positive value; where, Used in the preparation stage before spotting. Used for the diffusion and visualization stage after spotting; in a preferred implementation The settings can be adjusted based on the device's response speed, camera frame rate, and sampling motion rhythm. The parameters can be set according to the diffusion time of droplets on paper-based materials, the duration of color halo appearance, and interpretation requirements. The above parameters can be obtained through pre-experiment calibration or dynamically adjusted according to specific test objects. The process data section of the unified input data should include at least three of the following: immersion depth, immersion time, lifting speed, sample height, contact time, and landing point coordinates, as well as their corresponding timestamps. Optional parameters include feeding amount and feeding sequence, stirring status, point number, and positioning calibration information. These parameters can characterize the sample application process status and are used in conjunction with visual features for subsequent reasoning.

[0091] To reduce the impact of external environment and carrier differences on image interpretation, preprocessing includes at least one of the following: white balance / color calibration, reflection suppression, exposure normalization, and spot ROI localization and cropping, to reduce the influence of paper texture, ambient light changes, or reflections on color halo interpretation. Preferably, the preprocessed image area is concentrated on the spot and the local area where color halo may appear, to reduce interference from irrelevant backgrounds. For example, the candidate center position of the spot can be determined first based on the point number or positioning calibration information. Reconstruct the local region The data is cropped so that the subsequent model focuses mainly on the areas related to droplets and color halos. After the above alignment and preprocessing, the unified input data contains both visual and process information, which can reflect the process status and results of the same sampling event. This unified input data is preferred as the standardized input for the subsequent multimodal large model inference module.

[0092] The multimodal large model inference module is used to input unified input data into the multimodal large model according to a preset structured output specification and output structured inference results. Unlike the method of only outputting natural language descriptions, this embodiment emphasizes that the output results should be structured field results that can be directly parsed by subsequent control logic. Furthermore, the structured inference results satisfy the preset field set and include at least the following: color halo determination field: color halo exists / does not exist / is uncertain and the corresponding confidence level; spot validity field: valid / invalid; anomaly type field: output at least one anomaly type and the corresponding confidence level from dripping, continuous dripping, splashing, offset, contamination, reflection interference, paper defect, and drip spot crosstalk; control suggestion field: at least one of the parameters to be adjusted, adjustment direction (increase / decrease / maintain), and resampling waiting time or resampling number; evidence field: at least one of the keyframe index and ROI coordinates of drip spot or color halo candidate regions.

[0093] Among them, the control suggestion field is used by the guardrail closed-loop control module to directly parse and generate control commands; specifically, the multimodal large model is not only used to assist in judgment, but also serves as a source of structured decision output, providing direct input for subsequent automatic control; for example, in a specific implementation, the structured reasoning result can be represented as: ;in, This indicates the result of the color halo assessment. Indicates the confidence level of the color halo determination. This indicates the result of the sampling validity determination. Indicates the exception type. Indicates the confidence level of the anomaly type. This indicates a control suggestion field. This indicates the evidence field.

[0094] refer to Figure 6 and Figure 7The guardrail closed-loop control module is used to perform confidence gating and consistency verification based on structured reasoning results, and to generate control commands based on the anomaly-action mapping relationship to drive the equipment to perform at least one of the following actions: adjust sampling parameters, adjust imaging parameters, resample, redo sampling, change sampling location, and change paper, thereby forming a closed loop of "acquisition-reasoning-execution-reacquisition". Through this closed-loop structure, the process or acquisition conditions can be corrected in time when anomalies occur, instead of passively accepting unstable results after a single sampling. Among them, the paper replacement operation reminds the test personnel to manually replace the paper. The reminder method can be set by setting a signal light or directly by text reminder on the main control computer, which can be achieved through preset programs and simple instructions.

[0095] Furthermore, the guardrail closed-loop control module includes: a confidence gating submodule, a consistency verification submodule, and a fallback decision submodule. By constraining the inference results within the framework of gating, consistency, and fallback verification, the reliability of automatic decision-making under complex working conditions can be improved. Specifically, the confidence gating submodule is used when the confidence level of the color halo determination field is less than a threshold. Or the confidence level of the exception type field is less than the threshold. When, trigger delay Post-resampling at least Next; among them, and For a pre-set confidence threshold, For delay parameters, The parameter is the number of resampling operations. Therefore, in this embodiment, when the reliability of the model output is insufficient, the system will not directly use the result as the final conclusion, but will re-acquire the data related to the current sampling event by resampling after a delay, so as to avoid misjudgment caused by single acquisition error.

[0096] The consistency verification submodule is used to verify the consistency of the data. The color halo determination result of the second resampling is subject to majority voting or consistency determination. If the consistency does not meet the preset conditions, an uncertain conclusion is output and at least one of changing the sampling point or changing the paper is triggered. The consistency determination can adopt majority voting, proportional threshold determination, or other equivalent consistency evaluation methods; for example, if The proportion of identical color halo determination results appearing in the resampling results is not less than the preset consistency threshold. If the consistency requirement is met, then the consistency requirement is considered met; otherwise, the stability of the result under the current point position, current paper surface, or current imaging conditions is considered insufficient, and an uncertain conclusion should be output, and further operations such as changing the point position or changing the paper should be performed.

[0097] The fallback decision submodule is used to conservatively verify the existence of color halo by calling rule discrimination based on color difference / boundary region contrast when the confidence gating is triggered and an uncertain conclusion is still output. The fallback decision result and the triggering reason are written into a traceable report. In other words, when the model is uncertain and a stable judgment cannot be formed after resampling, the system can also call the rule discrimination method for conservative verification.

[0098] In one alternative implementation, the catch-all determination can be achieved by comparing the color difference between the central region and the outer annular region of the droplet. Let the average color feature of the central region be... The average color characteristic of the outer region is The color difference can then be characterized as: Simultaneously, it can also be combined with the boundary area contrast index. Joint discrimination is performed to improve the stability of conservative verification.

[0099] Furthermore, the guardrail closed-loop control module includes anomaly-action mapping relationships, and contains at least the following three types of anomaly mapping rules: when the anomaly type is dripping, a control command is generated to reduce the lifting speed or increase the sampling dwell time, and a resampling is triggered; when the anomaly type is splashing, a control command is generated to change the sampling position or reduce the sampling height, and a resampling is triggered; when the anomaly type is reflective interference, a control command is generated to adjust the imaging parameters and mask the reflective area, and a resampling is triggered.

[0100] For droplet-related anomalies, the system can improve the droplet transfer process by reducing the lifting speed and extending the sample dwell time; for splashing anomalies, the system can reduce uncontrolled diffusion caused by impact by reducing the sample height and reselecting the sampling point; for reflective interference anomalies, the system can reduce the risk of visual misjudgment by reducing exposure, adjusting the light source level, or applying mask markings to the bright areas; the above control actions can be executed individually or in combination; in a preferred implementation, the anomaly-action mapping relationship can be pre-stored in the memory in the form of a mapping table, and the guardrail closed-loop control module can look up the table according to the anomaly type, anomaly confidence level, and current process parameter status to generate specific control instructions.

[0101] refer to Figure 6 and Figure 7The recording and reporting module is used to associate and record unified input data, structured inference results, control commands and their triggering reasons, and output a traceable report containing key evidence frames. Through this setting, key evidence in the collection, inference and execution stages can be linked together. Preferably, the traceable report includes at least the time window identifier corresponding to the unified input data, key frame index and ROI coordinates, sampling parameters and their timestamps, structured inference results, confidence gating triggering reasons, consistency judgment results, anomaly-action mapping hit items and control command logs. By associating keyframe indices with corresponding triggering causes and control commands, a complete chain of evidence can be formed. For example, in a sampling event, when the multimodal large model outputs an anomaly type of "drip" and the confidence level is higher than the threshold, the guardrail closed-loop control module can directly call the anomaly-action mapping rule to generate control commands to reduce the lifting speed and increase the sampling dwell time, while triggering a resampling. The recording and reporting module then associates and stores the anomaly type, the hit mapping rule, the corresponding keyframe index, and the actual control command issued. Another example is when the multimodal large model outputs an uncertain color halo determination result, and the confidence level of the color halo determination field is lower than the threshold. At that time, the confidence gating submodule can first trigger a delay. After at least The system performs a second sampling; if the subsequent consistency verification results still do not meet the preset conditions, the system outputs an uncertain conclusion and further triggers a change of sampling point or paper change operation; if necessary, the fallback judgment submodule can be called to perform conservative verification, and the relevant triggering reasons can be written into the traceability report.

[0102] refer to Figure 7 In summary, this embodiment organically combines multi-source data access, event trigger alignment and preprocessing, multimodal large model inference, guardrail closed-loop control, and recording and reporting to realize a closed-loop control process of "data acquisition - structured inference - control correction - re-acquisition verification" around the same sampling event. This improves the control stability and judgment objectivity of methylene blue sampling tests under the combined effects of sampling process fluctuations and multi-source imaging interference.

[0103] Example 2:

[0104] refer to Figure 1 and Figure 8 This embodiment also provides a control method for methylene blue spot sampling tests based on a multimodal large model, which applies the methylene blue spot sampling test control system based on a multimodal large model in Embodiment 1, and includes the following steps:

[0105] S1. Prepare a 10.0 g / L ± 0.1 g / L methylene blue solution. Weigh fine aggregate and clean water into a stirring flask 32 and stir to form a suspension. Add the initial methylene blue solution and continue stirring. Specifically, the servo motor 52 drives the stirring rod 54 to stir the suspension in the stirring flask 32 to mix it. The speed of the servo motor 52 is adjustable from 600 r / min ± 60 r / min to 400 r / min ± 40 r / min. Then, by cooperating with the push module 6 and the electric telescopic rod 71, drive the glass rod 72 to dip into the mixed suspension and drop it onto the spotting paper 35 at the suspended position. This triggers the spotting event and obtains the corresponding data. The process data is collected simultaneously, along with visual data corresponding to the droplets. Specifically, the electric telescopic rod 71 drives the glass rod 72 into the stirring bottle 32 for dipping, then drives the glass rod 72 to rise above the stirring bottle 32, and then the electric telescopic push rod 61 drives the push plate 65 to swing. Through the limiting cooperation of the limiting slide rod 69 and the sliding hole 66, the push plate 65 can drive the connecting sleeve 64 to rotate together, thereby allowing the electric telescopic rod 71 to drive the glass rod 72 to rotate together, so that the glass rod 72 moves to the top of the spotting paper 35. Finally, the electric telescopic rod 71 drives the glass rod 72 to move down to spot the sample, leaving a droplet on the spotting paper 35 for the detection camera 8 to collect.

[0106] It should be noted that in this embodiment, the method does not perform isolated interpretation of a single spot image, but rather constructs a closed-loop process around the same spotting action: "spotting event triggering - multi-source data acquisition - unified input construction - structured reasoning - guardrail control - resampling verification - report output"; preferably, the first The secondary closed-loop process can be characterized as follows: ;in, Indicates the first The unified input data corresponding to each sampling event. Indicates the first Substructured reasoning results Indicates the first Secondary control commands; an example is given below to illustrate this, where uniform input data is generated after the first sampling. Structured reasoning results are obtained through multimodal large model reasoning. If an abnormal dripping condition is detected, a control command to reduce the lifting speed is generated. After executing the control command, sampling is performed again to form new unified input data. This allows the system to enter the next closed loop.

[0107] In step S1, a methylene blue solution of 10.0 g / L ± 0.1 g / L is first prepared. This concentration range is used to control the deviation of the basic concentration of the mother liquor, so that the subsequent spot color and halo appearance have good comparability. Then, 200 g of fine aggregate and 500 ml ± 5 ml of clean water are weighed and added to the stirring flask 32 for stirring to form a suspension. On this basis, the initial methylene blue solution is added and stirring is continued to ensure that the suspension and methylene blue solution are fully mixed. The methylene blue solution in the storage tank 21 is continuously and quantitatively added through the micro metering pump 23. In this embodiment... Each time, add 5 ml of methylene blue solution, stir for 1 minute, and then spot the sample until a color halo appears. Stop adding the solution but do not stop stirring. Preferably, continuous stirring is used to reduce the impact of particle sedimentation and local concentration fluctuations on the amount of sample taken and the spotting results. After mixing is completed, the glass rod 72 is driven by the cooperation of the push module 6 and the electric telescopic rod 71 to pick up the mixed suspension and drop it onto the spotting paper 35 in the suspended area. The spotting paper 35 in the suspended area is a spotting area that is supported in a predetermined position and kept at a distance from the background structure below, so as to reduce the impact of background contact on the morphology of the drop spot and the imaging quality.

[0108] Furthermore, the spotting event in step S1 can be a process event formed by completing a spotting action around the glass rod 72. Specifically, the spotting event in S1 includes the glass rod 72 dipping into the suspension, the glass rod 72 moving, the glass rod 72 spotting on the spotting paper 35, and the collection of spotting droplets. Thus, the spotting event includes both the droplet transfer process and the collection process after the spotting is formed. In step S1, process data corresponding to the spotting event is also acquired simultaneously, and visual data corresponding to the spotting are collected. Preferably, the process data includes at least three of the following: immersion depth, immersion duration, lifting speed, spotting height, spotting dwell time, contact time, and landing point coordinates, along with their timestamps. It may also optionally include at least one of the following: stirring status, spot number, positioning and calibration information, and liquid addition sequence.

[0109] In some implementations, process data can be represented as: ;in, Indicates the first Item process parameters or parameter vector, Represents the corresponding timestamp; visual data can be represented as: ;in, Indicates the first Frame image, This represents the corresponding acquisition timestamp; by uniformly writing timestamps to process data and visual data, a foundation can be provided for subsequent event triggering and time alignment; for example, in a non-restricted example, This can include "immersion depth = 5.0 mm, timestamp = 10.120 s", "lifting speed = 3.2 mm / s, timestamp = 10.480 s", and "sampling height = 1.5 mm, timestamp = 10.930 s". The data may include multiple frames of images acquired continuously at 10.800s, 10.840s, and 10.880s; the timing relationship between each process parameter and the corresponding image frame can be determined by the timestamps.

[0110] S2. Using the sampling event time as a reference, the visual data is subjected to event-triggered clipping and aligned with the process data by timestamp to generate unified input data; in step S2, the visual data is subjected to event-triggered clipping and aligned with the process data by timestamp to generate unified input data; preferably, the sampling event time is denoted as The timing can be determined by the spotting trigger signal, the glass rod 72 positioning signal, the contact detection signal, or the moment when the droplet first contacts the spotting paper 35 as identified by the visual side; to improve the stability of time synchronization, the actuator control signal or positioning signal is preferably used as the time synchronization signal. The basis for determination.

[0111] Furthermore, S2 includes point-sampled event moments. Use the time window as a reference Frame sequence; where, and As a preset positive value, Used in the preparation stage before spotting. Used to cover the diffusion and color halo development stages after spot application; in some embodiments, The time interval can be 0.05s to 1.5s. The time should be between 0.2s and 5s; for cases where the color halo appears slowly, the time can be increased appropriately. Based on the time window, visual fragments corresponding to the point sample event can be obtained: For example, in a non-restrictive example, if the point sample event time... And set , The captured visual time window is At this point, only image frames whose timestamps fall within this interval are retained as the visual input for this spotting event.

[0112] Step S2 also includes aligning the visual segments with the process data according to timestamps; when the sampling frequency of the process data is inconsistent with the 8-frame rate of the detection camera, nearest neighbor time alignment, linear interpolation alignment, or statistical aggregation based on time windows can be used for processing. Taking the lifting speed as an example, the average lifting speed within the time window can be calculated: ;in, For example, in a non-limiting example, if three discrete lifting speed values ​​are collected within the time window, namely 3.0 mm / s, 3.4 mm / s, and 3.2 mm / s, then it can be approximately considered that... The average lifting speed is used as one of the process features in the unified input data. To reduce the impact of paper batch differences, ambient light changes, surface reflections, and background textures on subsequent interpretation, step S2 further includes preprocessing the frame sequence. Specifically, at least one preprocessing is performed on the frame sequence: white balance / color calibration, exposure normalization, reflection suppression or reflection mask marking, and spot ROI localization and cropping based on point number or positioning calibration information. Among these, white balance or color calibration is used to unify the image color reference; exposure normalization is used to reduce the overall brightness drift of the image; reflection suppression or reflection mask marking is used to reduce the interference of bright reflective areas on the interpretation of spot boundaries and color halo areas; spot ROI localization and cropping is used to focus the model's attention on the spot and its surrounding color halo candidate areas.

[0113] In a preferred embodiment, candidate centers of droplets can be determined based on spot numbers or location calibration information. And construct a local clipping region: ;in, and These are the width and height of the ROI, respectively; for example, in a non-restricted example, if the candidate center coordinates are... And set , Then we can get Subsequent image processing and model inference are performed only on this region, thereby reducing interference from the background region.

[0114] For image frames with reflective interference, a brightness threshold can also be used. Constructing a reflective mask: Reflective masks can be used to shield bright areas during the preprocessing stage, or as additional information in subsequent multimodal large model inputs; for example, in a non-limiting example, if a reflective threshold is set... (Corresponding to an 8-bit grayscale image), when the brightness value of a certain pixel... At that time, This indicates that the location is marked as a reflective area; when At that time, This indicates that the location is not a reflective area.

[0115] After event triggering, timing alignment, and preprocessing, uniform input data is formed: ;in, This represents the preprocessed visual segment. This represents the aligned process data. Indicates the moment of a point event. Metadata refers to information that includes at least one of the following: location number, paper number, test batch number, or time window identifier; for example, in a non-limiting example, It can include "ROI frame sequence with point number A03 and time window of [11.800s, 13.000s]", "average lifting speed 3.2mm / s", "spot height 1.5mm", "event time 12.000s", etc., and use it as the current input of the multimodal large model.

[0116] S3. Input the unified input data into the multimodal large model and output the structured inference results, including color halo determination and its confidence level, spot sample validity, anomaly type and its confidence level, control suggestion field and evidence field. In step S3, the unified input data is input into the multimodal large model to output the structured inference results. The structured inference results are not free text results, but structured results that meet the preset field specifications so that they can be directly called by the subsequent guardrail control logic. Specifically, S3 inputs the unified input data into the multimodal large model and outputs the structured inference results, including color halo determination and its confidence level, spot sample validity, anomaly type and its confidence level, control suggestion field and evidence field. Among them, the color halo determination field may include existence, non-existence or uncertainty; the spot sample validity field may include validity or invalidity; the anomaly type field includes at least one of dripping, continuous dripping, splashing, offset, contamination, reflective interference, paper surface defect or drip spot crosstalk; the control suggestion field includes at least one of the parameters to be adjusted, adjustment direction, resampling waiting time, and resampling number; the evidence field includes at least one of the keyframe index, ROI coordinates, and anomaly triggering basis.

[0117] In a preferred embodiment, the structured reasoning result can be expressed as: ;in, This indicates the result of the color halo assessment. Indicates the confidence level of the color halo determination. This indicates the result of the sampling validity determination. Indicates the exception type. Indicates the confidence level of the anomaly type. This indicates a control suggestion field. This represents the evidence field; for example, in a non-restricted example, the model could output R={existence, 0.87, valid, drip, 0.82, "reduce lifting speed and increase sampling dwell time", "keyframe 12, ROI=[270,370]\times[200,280]"}; in this case, the system can directly generate corresponding control instructions based on the anomaly type "drip" and the control suggestion field; furthermore, the system can perform format verification on the structured inference results output by the multimodal large model; when there are missing fields, abnormal field values, or logical conflicts in fields, the result can be marked as non-compliant output and enter the resampling process, re-inference process, or conservative control process.

[0118] S4. Based on the structured inference results, perform confidence gating and consistency verification, and generate control commands to drive the equipment to perform at least one of parameter adjustment, resampling, resampling, changing the sampling point, or changing the paper; In step S4, based on the structured inference results, perform confidence gating, consistency verification, and error correction control, and generate control commands to drive the equipment to perform at least one of parameter adjustment, resampling, resampling, changing the sampling point, or changing the paper; Further, S4 includes when the confidence level of the color halo determination is less than a threshold. Or the confidence level of the anomaly type is less than the threshold. Time, delay At least after execution Secondary resampling; for The resampling results are subject to majority voting or consistency determination; if the consistency does not meet the preset conditions, at least one of changing the sampling point or changing the paper is triggered and the sampling is redone.

[0119] Specifically, when the following conditions are met At this time, the system does not directly adopt the current reasoning result, but delays it. At least after execution Secondary resampling; delay Used to provide stabilization time for spot diffusion, halo development, or imaging parameter adjustment; in some embodiments, A value of 50ms to 2s is acceptable. A value of 2 to 5 is acceptable; for example, in a non-restrictive example, if the confidence level of the halo determination obtained in this inference is... Set threshold Therefore, The system does not directly use the result, but instead delays it. Execute after Secondary resampling; for The results of the second resampling can be processed using majority voting or consensus determination methods. Let... The result of the secondary halo determination is: The dominant result is denoted as Then the consistency score can be defined as: ;in, For characteristic functions; when When, the consistency is determined to meet the preset conditions; when If the consistency does not meet the preset conditions, at least one of changing the sampling point or changing the paper is triggered, and the sampling is redone. For example, in a non-restrictive example, if the color halo determination results of the three resamplings are "present, present, absent", then the dominant result is "present", and the consistency score is 1. If a consistency threshold is preset. If, then consistency is considered to be satisfied; if If the consistency condition is not met, the process of changing the location or changing the paper needs to continue.

[0120] Furthermore, S4 also includes error correction control based on the type of anomaly: when the anomaly type is dripping, the parameter adjustment of reducing the lifting speed or increasing the spotting dwell time is executed and the spotting is redone; when the anomaly type is splashing, the spotting is redone after changing the spotting position or reducing the spotting height; when the anomaly type is reflective interference, the exposure is reduced or the light source level is adjusted and reflective mask marking is performed before resampling; and a maximum number of redoes or maximum number of resamplings is set for each error correction action.

[0121] Specifically, when the exception type is "drip", the following adjustments can be made: ;in, Adjustment amount for lifting speed. This refers to the adjustment amount of the spotting residence time; by decreasing the lifting speed and / or increasing the spotting residence time, it is beneficial to improve the stability of the droplet transfer from the glass rod to the spotting paper; for example, in a non-limiting example, if the current lifting speed is 3.0 mm / s and the spotting residence time is 0.20 s, then after determining that the droplet is stuck, the lifting speed can be adjusted to 2.5 mm / s and the spotting residence time can be adjusted to 0.30 s, and then the spotting can be repeated.

[0122] When the exception type is splash, the following adjustments can be made: ;in, This is the amount of adjustment for the sampling height. This indicates that a spot change is triggered. Reducing the spot height can decrease the droplet impact energy, and changing the spot location can reduce the impact of the contaminated area on subsequent interpretation. For example, in a non-limiting example, if the current spot height is 2.0 mm, the spot height can be lowered to 1.2 mm after splashing is detected. If contamination has occurred in the area adjacent to the original spot location, the spot location can be changed from A03 to A04 before redoing the spotting.

[0123] When the anomaly type is reflective interference, the following adjustments can be made: ;in, Adjust the amount of exposure. This refers to the amount of light source adjustment; for example, in a non-limiting example, if the current exposure parameter is 100, the system can reduce the exposure parameter to 80 after recognizing reflective interference.

[0124] To prevent the process from entering an infinite loop, a maximum number of retakes or a maximum number of resampling operations is set for each error correction action; preferably, the maximum number of resampling operations can be set separately. Maximum number of resampling attempts Maximum number of times to switch positions and maximum number of paper replacements When any upper limit is reached and a stable determination result is still not obtained, the system outputs a termination status and writes the triggering reason into a traceable report.

[0125] It should also be noted that, in order to confirm the stability of the halo, the formal judgment should only begin when none of the above-mentioned abnormal types are present, that is, when the halo can appear normally and there are no other influencing factors. Specifically, after the halo appears, the halo should be spotted once every 1 minute. When the halo appears 5 times consecutively, that is, when the halo lasts for 5 minutes without disappearing, it can be determined as the adsorption endpoint. Furthermore, if the halo disappears within the first 4 minutes, 5 ml of quantitative methylene blue solution should be added and the test should continue. If the halo disappears at the 5th minute, only 2 ml of methylene blue solution needs to be added and the test should continue. When it is determined to formally enter the stability test of the halo, the spotting paper 35 can be replaced to allow the spotting spot sufficient space.

[0126] S5. After executing the control command, repeat steps S1 to S4 until the termination condition is met. In step S5, after executing the control command, repeat steps S1 to S4 until the termination condition is met. Here, repeated execution means that after the test device has completed parameter adjustment, resampling preparation, redoing sampling preparation, spot change preparation, or paper change preparation according to the previous round of control command, a new round of sampling event triggering, acquisition, reasoning, and control process is restarted. Further, the repeated execution process in S5 starts from the triggering of the sampling event in S1 and ends in step S4. That is, each cycle begins with the triggering of a new sampling event and ends with the structured reasoning result of that round being used to generate control commands. When the termination condition is not met, proceed to the next cycle. When the termination condition is met, proceed to step S6.

[0127] In some implementations, the termination condition includes at least one of the following: the color halo is clearly determined and the corresponding confidence level is not lower than a threshold. The anomaly type is clearly defined and the consistency after processing meets preset conditions; the maximum number of resampling times, maximum number of rework times, maximum number of point changes, or maximum number of paper changes is reached; or the preset batch termination condition is met; accordingly, the stop criterion can be expressed as: ;in, This represents the current number of resampling attempts. This represents the current number of resamples; for example, in a non-restricted example, if a round of decision results in... And consistency score At the same time, set , Then it satisfies The system can determine that the termination condition has been met and terminate the loop; otherwise, if 5 resampling operations have been performed and the settings are met, the loop will terminate. Even if the current result is still unstable, the process can be terminated and the termination reason can be output when the upper limit is reached.

[0128] S6. Generate and output a traceable report containing key evidence frames. In S6, a traceable report containing key evidence frames is generated and output. The traceable report is used to associate and record sampling events, unified input data, structured inference results, control actions and their triggering reasons, thereby forming an auditable full-process evidence chain. Furthermore, the traceable report output by S6 includes at least: time window identifiers corresponding to unified input data, key frame indexes and ROI coordinates, sampling parameters and their timestamps, structured inference results, confidence gating triggering reasons, consistency judgment results, anomaly-action mapping hit items, and control instruction logs. The key frame indexes are associated with the corresponding triggering reasons and control instructions to form an auditable full-process evidence chain.

[0129] In a preferred embodiment, the record item of the traceable report can be represented as: ;in, This indicates that the input data should be uniform. Represents the results of structured reasoning. This indicates the actual control command executed, `reason` indicates the triggering reason, `window_id` indicates the time window identifier, and `model_ver` indicates the model version information; for example, in a non-restricted example, This can be recorded as: "window_id=W20251119-01; unified input data corresponds to point A03; structured inference result is 'droplet hanging, confidence level 0.82'; control command is 'lifting speed reduced from 3.0mm / s to 2.5mm / s, sampling dwell time increased from 0.20s to 0.30s'; triggering reason is 'abnormal droplet hanging and confidence level higher than threshold'; model version is V1.0". This allows for complete backtracking of the entire process of a single experiment. Preferably, the traceability report should establish at least the following relationships: the correspondence between keyframe index and anomaly type, the correspondence between anomaly type and control command, the correspondence between control command and subsequent resampling or redo results, and the correspondence between termination status and triggering reason. Through the above-mentioned relational presentation method, backtracking, auditing, and verification of the entire sampling experiment process can be achieved.

[0130] Therefore, by combining suspension preparation, glass rod spotting, event-triggered acquisition, time window interception, preprocessing, multimodal large model structured reasoning, confidence gating, consistency verification, anomaly correction control, cyclic execution, and evidence chain report output, this embodiment can achieve reproducible control and objective consistency judgment of methylene blue spotting experiments under the conditions of fluctuations in the spotting process and interference from multiple sources of imaging.

[0131] It should be understood that the above embodiments are only one of the preferred embodiments of the present invention and are not intended to limit the present invention. Equivalent substitutions or conventional modifications made by those skilled in the art to the parameter values, threshold settings, field naming, exception type expansion, control action amplitude, and report structure of each step without departing from the spirit and substance of the present invention shall all fall within the protection scope of the present invention.

Claims

1. A control system for methylene blue spot sampling tests based on a multimodal large model, characterized in that, Includes a shell (1), on which is provided a liquid addition module (2) for adding methylene blue solution, a stirring bottle (32) for mixing suspension, a stirring module for stirring suspension, a spotting paper (35) and a rotation module for rotating the spotting paper (35); The outer shell (1) is rotatably connected to an electric telescopic rod two (71). The piston end of the electric telescopic rod two (71) is provided with a glass rod (72) for dipping the suspension. The outer shell (1) is also provided with a push module (6) for pushing the electric telescopic rod two (71) to rotate. It also includes a detection camera (8) located above the sample paper (35), and the detection camera (8) is equipped with an intelligent test control calculation module for running system programs and realizing closed-loop control; The intelligent test control calculation module includes: The multi-source data access module is used to acquire visual data corresponding to the droplet spots and process data corresponding to the spotting process, and write timestamps for the visual data and process data respectively. The event trigger alignment and preprocessing module is used to extract frame sequences or short video clips containing preset durations before and after the point sampling event from the visual data, using the "point sampling event time" as the alignment benchmark, and then align the frame sequences or short video clips with the process data according to the timestamp and encapsulate them into unified input data. The multimodal large model inference module is used to input unified input data into the multimodal large model according to the preset structured output specification and output structured inference results. The guardrail closed-loop control module is used to perform confidence gating and consistency verification based on structured reasoning results, and to generate control commands based on the anomaly-action mapping relationship to drive the execution of one of the following actions: adjusting sampling parameters, adjusting imaging parameters, and resampling; The recording and reporting module is used to associate and record unified input data, structured reasoning results, control instructions and their triggering reasons, and output a traceable report containing key evidence frames. The structured reasoning results include at least: color halo judgment results and confidence levels, spot sampling validity judgment results, anomaly types and confidence levels, control suggestion fields and evidence fields.

2. The methylene blue spot sampling test control system based on a multimodal large model according to claim 1, characterized in that, The intelligent test control calculation module is electrically or communicatively connected to the liquid addition module (2), the stirring module rotation module, the electric telescopic rod II (71), the pushing module (6), and the detection camera (8); The liquid addition module (2) includes a storage tank (21) on the outer shell (1), a liquid addition pipe (22) connected to the inside of the storage tank (21), the bottom end of the liquid addition pipe (22) is located below the inside of the storage tank (21), and a micro metering pump (23) connected to the inside of the liquid addition pipe (22). The outer shell (1) is provided with a protective shell (4), and a support frame (24) is provided on the outside of the protective shell (4). The liquid adding tube (22) is snapped onto the support frame (24), and the end of the liquid adding tube (22) away from the storage tank (21) is located at the top of the stirring bottle (32). The outer shell (1) is provided with a limiting sleeve (31), and the stirring bottle (32) is inserted into the inside of the limiting sleeve (31). The rotating module includes a servo motor (33) located inside the housing (1). The output shaft of the servo motor (33) passes through the top of the housing (1) and is fixed to a rotating disk (34). The dotted paper (35) is placed on the top of the rotating disk (34). The rotating disk (34) has an overhead groove for suspending the dotted paper (35). The diameter of the overhead groove is smaller than the diameter of the dotted paper (35). A pressure ring (36) for pressing the outer ring of the dotted paper (35) is placed on the rotating disk (34).

3. The methylene blue spot sampling test control system based on a multimodal large model according to claim 2, characterized in that, The stirring module includes an electric telescopic rod 1 (51) disposed inside the outer shell (1), and the protective shell (4) is located between the outer sides of the electric telescopic rod 1 (51) and the electric telescopic rod 2 (71). The top of the protective shell (4) is provided with two movable holes through which the piston ends of the electric telescopic rod 1 (51) and the electric telescopic rod 2 (71) respectively pass. The piston end of the electric telescopic rod (51) is provided with an extension arm (53). The top of the extension arm (53) away from the electric telescopic rod (51) is provided with a servo motor (52). The output shaft of the servo motor (52) passes through the extension arm (53) and is provided with a stirring rod (54). The stirring rod (54) is located inside the stirring bottle (32). The push module (6) includes an electric telescopic push rod (61) and a bearing seat (62) located inside the outer shell (1). A rotating rod (63) is rotatably connected to the bearing seat (62) via a bearing. The rotating rod (63) is fixed to the bottom of the electric telescopic rod (71). A connecting sleeve (64) is fixed to the outside of the rotating rod (63). A push plate (65) is provided on the outside of the connecting sleeve (64). A long strip-shaped sliding hole (66) is opened on the push plate (65). A limit sliding rod (69) is slidably connected in the sliding hole (66). The limiting slide bar (69) consists of a cylindrical rod, a limiting plate located at one end of the cylindrical rod, and a nut detachably connected to the other end of the cylindrical rod. The cylindrical rod is slidably connected in the sliding hole (66). The opposite sides of the limiting plate and the nut are both in contact with the outer side of the push plate (65), and the contact surfaces are both smooth. The piston end of the electric telescopic push rod (61) is hinged to the limiting plate by a pin. The push plate (65) is provided with a sensing plate (67) on the side away from the connecting sleeve (64). The housing (1) is also provided with a proximity sensor (68) that senses the sensing plate (67). The piston end of the electric telescopic rod 2 (71) is fixed with a connecting plate (73), and the glass rod (72) is located on the side of the connecting plate (73) away from the electric telescopic rod 2 (71). The glass rod (72) moves between the spotting paper (35) and the stirring bottle (32). The detection camera (8) is located outside the protective shell (4).

4. The methylene blue spot sample test control system based on a multimodal large model according to claim 1, characterized in that, The event trigger alignment and preprocessing module is used to sample event moments. Using this as a baseline, time windows are extracted from visual data. The frame sequence within is used as the visual part of the unified input data, in which and The preset positive value; Furthermore, the process data section of the unified input data should include at least three of the following sampling parameters and their timestamps: immersion depth, immersion duration, lifting speed, sampling height, contact time, and landing point coordinates. Optional parameters include feeding amount and feeding sequence, stirring status, point number, and positioning calibration information. The preprocessing includes at least one of the following: white balance / color calibration, reflection suppression, exposure normalization, spot ROI positioning and cropping, in order to reduce the impact of paper texture, ambient light changes or reflections on color fringing interpretation.

5. The methylene blue spot sampling test control system based on a multimodal large model according to claim 4, characterized in that, The structured reasoning result satisfies a preset set of fields and includes at least: Color halo determination fields: color halo present / absent / uncertain and corresponding confidence level; Sampling validity field: valid / invalid; Anomaly type field: Output at least one anomaly type and its corresponding confidence level from the following: dripping, continuous dripping, splashing, offset, contamination, reflection interference, paper defect, and dripping crosstalk; Control suggestion field: includes at least one of the following: the parameter to be adjusted, the adjustment direction (increase / decrease / hold), and the resampling wait time or the number of resampling times; the control suggestion field is used by the guardrail closed-loop control module to directly parse and generate control commands; Evidence fields: include at least one of the keyframe index and the ROI coordinates of the droplet or halo candidate region.

6. The methylene blue spot sampling test control system based on a multimodal large model according to claim 5, characterized in that, The guardrail closed-loop control module includes: The confidence level gating submodule is used when the confidence level of the hue determination field is less than a threshold. Or the confidence level of the exception type field is less than the threshold. When, trigger delay Post-resampling at least Second-rate; The consistency verification submodule is used to verify the consistency of the data. The color halo determination result of the second resampling is subject to majority voting or consistency determination. If the consistency does not meet the preset conditions, an uncertain conclusion is output and at least one of changing the point or changing the paper is triggered. The fallback judgment submodule is used to conservatively verify the existence of color halo by calling the rule judgment based on color difference / boundary region contrast when the confidence gating is triggered and an uncertain conclusion is still output, and write the fallback judgment result and the triggering reason into the traceable report.

7. The methylene blue spot sampling test control system based on a multimodal large model according to claim 6, characterized in that, The guardrail closed-loop control module includes anomaly-action mapping relationships, and contains mapping rules for at least the following three types of anomalies: When the anomaly type is dripping, a control command is generated to reduce the lifting speed or increase the spotting residence time, and the spotting is redone. When the anomaly type is splash, a control command is generated to change the sampling location or reduce the sampling height, and the sampling is redone. When the anomaly type is reflective interference, control instructions are generated to adjust imaging parameters and mask the reflective area, and resampling is triggered. Furthermore, the recording and reporting module associates and stores evidence of the exception type, the hit mapping rule, the keyframe index, and the corresponding control command.

8. A control method for methylene blue spot sampling tests based on a multimodal large model, employing the methylene blue spot sampling test control system based on a multimodal large model as described in any one of claims 1-7, characterized in that, Includes the following steps: S1. Prepare a 10.0 g / L ± 0.1 g / L methylene blue solution. Weigh fine aggregate and clean water into a stirring bottle (32) and stir to form a suspension. Add the initial methylene blue solution and continue stirring. Then, drive the glass rod (72) to dip into the mixed suspension and drop it onto the spotting paper (35) in the air by the cooperation of the push module (6) and the electric telescopic rod (71). That is, trigger the spotting event and obtain the process data corresponding to the spotting event, and at the same time collect the visual data corresponding to the spotting. S2. Based on the timing of the sampling event, the visual data is captured by event triggering and aligned with the process data according to the timestamp to generate unified input data; S3. Input the unified input data into the multimodal large model and output the structured reasoning results including color halo determination and its confidence level, spot sampling validity, anomaly type and its confidence level, control suggestion field and evidence field; S4. Based on the structured reasoning results, perform confidence gating and consistency verification, and generate control commands to drive the equipment to perform at least one of parameter adjustment, resampling, resampling, changing sampling point, or changing paper. S5. After executing the control command, repeat steps S1 to S4 until the termination condition is met. S6. Generate and output a traceable report containing key evidence frames.

9. The method for controlling methylene blue spot sampling experiments based on a multimodal large model according to claim 8, characterized in that, The spotting events in S1 include the glass rod (72) dipping into the suspension, the glass rod (72) moving, the glass rod (72) spotting on the spotting paper (35), and the collection of spotting droplets; The S2 includes point sample event timestamps. Use the time window as a reference The frame sequence; and perform at least one preprocessing on the frame sequence: white balance / color calibration, exposure normalization, reflection suppression or reflection mask marking, and spot ROI positioning and cropping based on point number or positioning calibration information; The repeated process in S5 starts from the point sample event triggered in S1 and ends in step S4.

10. The method for controlling methylene blue spot sampling experiments based on a multimodal large model according to claim 9, characterized in that, S4 includes when the confidence level for halo determination is less than a threshold. Or the confidence level of the anomaly type is less than the threshold. Time, delay At least after execution Secondary sampling; right The results of the second resampling are then subject to majority voting or consistency determination. When the consistency does not meet the preset conditions, trigger at least one of the following: change the sampling point or change the paper, and redo the sampling. The S4 also includes error correction control based on the anomaly type: when the anomaly type is dripping, the parameter adjustment of reducing the lifting speed or increasing the spotting dwell time is executed and the spotting is redone; when the anomaly type is splashing, the spotting position is changed or the spotting height is reduced and the spotting is redone; when the anomaly type is reflective interference, the exposure is reduced and reflective mask marking is performed and then resampling is performed. And set a maximum number of replays or maximum number of resampling for each error correction action; The traceable report output by S6 includes at least: time window identifiers corresponding to unified input data, key frame indexes and ROI coordinates, sampling parameters and their timestamps, structured inference results, confidence gating trigger reasons, consistency judgment results, anomaly-action mapping hit items, and control instruction logs, and associates key frame indexes with corresponding trigger reasons and control instructions.

Images