Calibration method and system for sampling positioning accuracy based on sample library

Abstract

The present application discloses a method and system for calibrating sampling positioning accuracy based on a sample library. The method includes: controlling a sampling tool loaded on a sampling arm to start working based on a preset initial sampling path, obtaining multi-source state features of the sampling environment at preset intervals; normalizing the multi-source state features to form a multi-source state feature vector, inputting the vector into a pre-trained state traction deviation prediction model, and outputting a position deviation pair, including the position deviation of the sampling tool's current path point and the center position deviation of the target sampling container; based on the position deviation pair, dynamically correcting and updating the initial sampling path, periodically monitoring the sampling environment, and repeatedly predicting the position deviation pair until the sampling arm is positioned at the center point of the target sampling container. Thus, the efficiency and accuracy of sampling positioning calibration are improved.

Description

Technical Field

[0001] The present invention relates to the field of sampling control technology, and in particular to a method and system for calibrating sampling positioning accuracy based on a sample library. Background Art

[0002] With the continuous advancement of science and technology, sample repositories, as a commonly used precision sample storage and sampling facility, are widely used due to their ability to provide long-term, stable cryogenic storage of biological samples and enable rapid and accurate quantitative sampling when needed. In cryogenic sample repositories, the sampling arm (also known as a box and tube picker) is a key automated device responsible for accurately removing the required sample containers from the sample repositories to meet subsequent experimental or analytical needs.

[0003] However, in actual applications, the environment of the sample library is complex and changeable, and the sampling arm is easily interfered with by various factors during the sampling process, such as environmental vibration, accumulation of mechanical transmission errors, and sample container loading offset. These factors may cause the sampling arm to deviate when positioning the sample container, thereby affecting the accuracy and efficiency of sampling. Especially in long-term continuous operation or high-frequency sampling operations, the error will be significantly amplified, which will have an adverse effect on the experimental results. Traditional calibration methods often rely on manual intervention, and calibration is performed by visually observing whether the sampling arm is accurately moved to the position of each sample container. This method is not only time-consuming and labor-intensive, but the accuracy of the calibration is also affected by human factors, and it is difficult to meet the high requirements of high-throughput experiments for sampling accuracy and efficiency. Summary of the Invention

[0004] The present application provides a calibration method for sampling positioning accuracy based on a sample library, thereby achieving high-precision adaptive calibration and improving the efficiency of sampling positioning calibration.

[0005] This application provides a method for calibrating sampling positioning accuracy based on a sample library, including:

[0006] S101, controlling the sampling tool loaded on the sampling arm to start working based on a preset initial sampling path, and obtaining multi-source state characteristics of the sampling environment at preset time intervals;

[0007] S102, normalizing the multi-source state features to form a multi-source state feature vector, inputting it into a pre-trained state traction deviation prediction model, and outputting a position deviation pair, including the position deviation of the current path point of the sampling tool and the center position deviation of the target sampling container;

[0008] S103, based on the position deviation pair, dynamically correct and update the initial sampling path, periodically monitor the sampling environment, and repeat steps S102 to S103 until the sampling arm is positioned at the center point of the target sampling container.

[0009] Preferably, the preset initial sampling path is determined in the following manner:

[0010] The initial calibration position of the sampling arm on the workbench and the theoretical center position of the target sampling container are obtained from the preset sample library to generate an initial sampling path.

[0011] Preferably, the multi-source state characteristics of the sampling environment include state characteristics of the sampling arm and state characteristics of the target sampling container;

[0012] The state characteristics of the sampling arm include command speed and acceleration, and joint torque fluctuations;

[0013] The state characteristics of the target sampling container include vibration characteristics.

[0014] Preferably, the pre-trained state traction deviation prediction model is used to predict the real-time position deviation pair of the sampling arm and the target sampling container based on the input multi-source state feature vector, and is used to dynamically calibrate the initial sampling path. The state traction deviation prediction model is obtained in the following manner:

[0015] A1. Collect a large number of multi-source state features and their corresponding actual deviation pairs from different sampling environments and different sampling paths. Use the actual deviation pairs to label the multi-source state features as a training sample set.

[0016] A2. Use the training sample set to train the pre-selected neural network structure, continuously optimize the model parameters, and generate the final state traction deviation prediction model.

[0017] Preferably, the actual deviation pair includes the actual position deviation of the sampling arm and the actual center position deviation of the target sampling container; the actual position deviation is set to: the difference between the actual position of the sampling arm at the current moment and the position at the corresponding time point in the preset initial path; the actual center position deviation is set to: the difference between the actual center position of the target sampling container and the theoretical center position.

[0018] Preferably, in S103, dynamically correcting and updating the initial sampling path based on the position deviation pair specifically includes:

[0019] Sum the position of the sampling arm in the initial sampling path at the current moment and the actual position deviation, and update the position information of the initial sampling path at the current moment to correct the deviation;

[0020] The deviation between the theoretical center position and the actual center position of the target sampling container is summed, the end point position of the initial sampling path is updated and corrected, and the theoretical center position of the target sampling container is updated together.

[0021] Preferably, the target vibration feature is generated according to a preset visual key feature extraction mechanism, and the visual key feature extraction mechanism specifically includes:

[0022] S201, using a preset image acquisition device to acquire a sequence of frame images within a current monitoring period, using a preset contour extraction and screening algorithm to obtain a target mouth region of a target sampling container in each frame, locating a set of edge pixel coordinates of the target mouth region, fitting a circular contour using a least squares method, and calculating the actual center coordinates of the target mouth region;

[0023] S202, based on the image frame sequence of the current monitoring period, calculating the displacement vector of the actual center coordinates of the target mouth area in the consecutive frames, and converting the pixel displacement into the actual physical displacement according to the camera calibration parameters of the image acquisition device to generate a displacement sequence;

[0024] S203, performing principal component analysis on the displacement sequence, extracting the main vibration direction as the vibration direction value, and setting the standard deviation of the displacement sequence in the main vibration direction as the vibration amplitude;

[0025] S204, performing a fast Fourier transform on the projection sequence of the displacement sequence onto the main direction, calculating the power spectrum density, and extracting the main frequency as the vibration regularity value;

[0026] S205 , combining the vibration direction value, the vibration amplitude, and the vibration regularity value to form the vibration characteristics of the target sampling container.

[0027] Preferably, before S103, the method further includes:

[0028] S301, based on the effective path of the sampling arm in the initial sampling path at the current moment, obtaining an effective ratio value, where the effective path is set to the remaining path in the initial sampling path that the sampling arm has not yet reached, and the effective ratio value is set to the ratio of the effective path length to the initial sampling path length;

[0029] S302: Using the effective proportion value, obtain a correction dependency factor, multiply the correction dependency factor by the position deviation pair, and obtain a new position deviation pair to replace the original position deviation pair.

[0030] Preferably, the correction dependency factor is calculated according to the following formula:

[0031]

[0032] in, is the correction dependency factor, is the effective proportion value, and k is the preset attenuation slope coefficient, which is used to control the correction amplitude.

[0033] The present application also provides a calibration system for sampling positioning accuracy based on a sample library, the system comprising: an acquisition module, a prediction module, and a correction module;

[0034] The acquisition module is used to control the sampling tool loaded on the sampling arm to start working based on a preset initial sampling path, and obtain multi-source state characteristics of the sampling environment at preset time intervals;

[0035] The prediction module is used to normalize the multi-source state features to form a multi-source state feature vector, which is input into a pre-trained state traction deviation prediction model and outputs a position deviation pair, including the position deviation of the sampling tool's current path point and the center position deviation of the target sampling container;

[0036] The correction module is used to dynamically correct and update the initial sampling path based on the position deviation, periodically monitor the sampling environment, and repeatedly jump to the prediction module until the sampling arm is positioned at the center point of the target sampling container.

[0037] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0038] By building a pre-stored reference coordinate system in the sample library and combining the multi-source state characteristics of the sampling arm during dynamic operation (including mechanical errors, environmental vibrations, etc.), the trained state traction deviation prediction model is used to calculate position deviation pairs in real time, dynamically updating the sampling path to achieve high-precision adaptive calibration and real-time compensation for dynamic deviations. In addition, through comprehensive multi-state characteristics, the state traction deviation prediction model is used to predict deviations, eliminating the need for real-time error monitoring during the sampling process, reducing computational redundancy and resources, and improving the efficiency of sampling positioning calibration.

[0039] On a laboratory workbench, the target sampling container may experience micron-level displacement due to environmental vibration (such as equipment motor operation, tabletop resonance) or human interference. Vibration feature extraction based on high-frequency image sequences can capture the container's instantaneous position changes in real time, providing data support for dynamic calibration of the sampling path. By extracting vibration direction, amplitude, and dominant frequency from multiple dimensions, the intensity and pattern of vibration interference on the container, as well as the dynamic relationship, can be accurately quantified. Combined with matching analysis of the sampling arm's acceleration and the container's dominant vibration frequency, the implicit vibration relationship between the sampling arm and the container can be derived, improving the accuracy of deviation prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of a flow chart of a method for calibrating sampling positioning accuracy based on a sample library according to an embodiment of the present invention;

[0041] Figure 2 This is a structural block diagram of a calibration system for sampling positioning accuracy based on a sample library according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] To facilitate understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0043] It should be noted that the terms “vertical”, “horizontal”, “up”, “down”, “left”, “right” and similar expressions used in this document are for illustrative purposes only and do not represent the only implementation method.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains; the terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0045] In dynamic experimental scenarios, the relative position of the sampling tool and the target sampling container is easily disturbed by multiple factors such as environmental vibration, accumulated mechanical transmission errors, and container loading offsets. Existing static calibration schemes are unable to compensate for dynamic deviations in real time, resulting in offsets in the sampling process. In particular, the errors are significantly amplified during long-term continuous operation or high-frequency sampling operations. In addition, traditional methods rely on manual intervention for calibration, making it difficult to combine historical reference coordinates stored in the sample library with real-time status characteristics to achieve intelligent sampling positioning, affecting the accuracy and efficiency of high-throughput experiments.

[0046] Example 1: Figure 1 It is a flow chart of a method for calibrating sampling positioning accuracy based on a sample library according to an embodiment of the present invention.

[0047] like Figure 1 As shown, a calibration method for sampling positioning accuracy based on a sample library includes the following steps:

[0048] S101 , controlling the sampling tool loaded on the sampling arm to start working based on a preset initial sampling path, and obtaining multi-source state characteristics of the sampling environment at preset time intervals.

[0049] Specifically, the preset initial sampling path is determined as follows:

[0050] Based on the current sampling environment scenario, the initial calibration position of the sampling arm on the workbench (mechanical origin reference coordinates) and the theoretical center position of the target sampling container (the theoretical position corresponding to the corresponding position of the target sampling container loaded on the carrier, that is, the theoretical center position of the target sampling container, is converted from the carrier coordinate system to achieve association with the sample library) are obtained from the preset sample library to generate the initial sampling path.

[0051] For example, when the sampling arm is reset to the mechanical origin, the position information of the sampling center of the sampling tool and the position information of the center of the target sampling container are used as the starting and ending points of the initial sampling path, respectively. Based on the starting and ending points, a sampling path is planned (using a path interpolation method, or referring to path generation methods in related prior art, which will not be described in detail in this invention), generating an initial sampling path consisting of a sequence of path points. The sampling tool can be a needle, a straw, etc.

[0052] It should be noted that the position information is set as three-dimensional coordinates, including the position information of the reference coordinate system on the X, Y, and Z axes. The construction of the reference coordinate system is set according to the actual experimental environment. However, the sampling arm and the target sampling container need to be in the same reference coordinate system to determine the initial static position information. This method can be implemented based on the coordinate system conversion in the relevant existing technology, and the present invention will not elaborate on this.

[0053] In some embodiments, the preset time interval is determined according to the acquisition frequency. For example, the preset time interval is set to 2 seconds. The multi-source state characteristics of the sampling environment include state characteristics of the sampling arm and state characteristics of the target sampling container.

[0054] Specifically, the state characteristics of the sampling arm include command speed and acceleration (real-time speed and acceleration commands output by the motion controller, used to analyze the path deviation trend caused by the inertia of the robotic arm), joint torque fluctuation (real-time torque values ​​of each joint drive motor are collected by the torque sensor and normalized into a percentage form, which is used to detect abnormal vibrations caused by mechanical jamming or external resistance. The variance value or standard deviation can be taken to reflect the degree of torque fluctuation within the monitoring period); the state characteristics of the target sampling container include: vibration characteristics (based on image sequence analysis, sampled by a high-frequency camera, determined according to the pixel sequence of the edge features of the target sampling container, and the vibration direction, vibration amplitude and vibration regularity value of the edge features can be extracted based on the pixel sequence as vibration characteristics).

[0055] S102, normalize the multi-source state features to form a multi-source state feature vector, input it into the pre-trained state traction deviation prediction model, and output the position deviation pair, which is expressed as: ( ), is the position deviation of the current path point of the sampling tool, is the center position deviation of the target sampling container.

[0056] In some embodiments, a pre-trained state traction deviation prediction model is used to predict the real-time position deviation of the sampling arm and the target sampling container based on the input multi-source state feature vectors, and is used to dynamically calibrate the initial sampling path. The model is obtained by:

[0057] A1. Collect a large number of multi-source state features and their corresponding actual deviation pairs from different sampling environments on different sampling paths. Use the actual deviation pairs to label the multi-source state features as a training sample set.

[0058] The actual deviation pair includes the actual position deviation of the sampling arm and the actual center position deviation of the target sampling container. The actual position deviation is set as: the difference between the actual position of the sampling arm at the current moment and the position at the corresponding time point in the preset initial path, reflecting the impact of mechanical transmission errors (such as gear clearance and belt slack) and external interference (such as load changes and vibration) on positioning accuracy. The actual position of the sampling arm at the current moment is obtained based on the installed encoder, which is not described in detail in the present invention. The actual center position deviation is set as: the difference between the actual center position of the target sampling container and the theoretical center position. The actual center position can be determined using a visual positioning system. The visual positioning system can be configured as an image acquisition device or a laser tracker to calibrate the actual position of the target sampling container.

[0059] It should be noted that the collection of training sample sets can utilize a large amount of historical laboratory sampling positioning calibration data, and the historical data can be labeled. Alternatively, an experimental scenario can be set up. In the experimental environment, the state characteristics of the sampling arm and the target sampling container can be adjusted, and the corresponding actual deviation pairs can be recorded.

[0060] A2. Use the training sample set to train the pre-selected neural network structure, continuously optimize the model parameters, and generate the final state traction deviation prediction model.

[0061] Therefore, through the combination of multi-source state features, the training model learns the implicit relationship between the mechanical vibration transmission between the sampling arm and the target container on the work surface, which can more comprehensively predict the calibration information based on the sampling environment.

[0062] S103, based on the position deviation pair, dynamically correct the initial sampling path, generate a calibration sampling path, update the initial sampling path, periodically monitor the sampling environment, and repeat steps S102 to S103 until the sampling arm is positioned at the center point of the target sampling container.

[0063] Specifically, the position of the sampling arm in the initial sampling path at the current moment and the actual position deviation are summed, the position information of the initial sampling path at the current moment is updated and corrected, the theoretical center position of the target sampling container and the actual center position deviation are summed, the end point position of the initial sampling path is updated and corrected, and the theoretical center position of the target sampling container is updated together.

[0064] It can be understood that by utilizing the multi-source state characteristics of the sampling environment, the most likely position deviation of the current sampling arm and the target sampling container based on the original theoretical position can be inferred, thereby calibrating the original path and generating a real-time path that conforms to the current environment. The encoder data set by the sampling arm and the sample library reference coordinates constitute the basis for static calibration, and the visual positioning and vibration sensor data provide the basis for dynamic compensation. The two realize deviation coupling calculation through the prediction model.

[0065] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0066] By constructing a reference coordinate system pre-stored in the sample library and combining the multi-source state characteristics of the sampling arm in dynamic operation (including mechanical errors, environmental vibrations, etc.), the trained state traction deviation prediction model is used to calculate the position deviation in real time, dynamically update the sampling path, and achieve high-precision adaptive calibration, thereby realizing real-time compensation for dynamic deviations. In addition, through comprehensive multi-state characteristics, the state traction deviation prediction model is used to predict deviations, eliminating the need for real-time error monitoring during the sampling process, reducing computational redundancy and resources, and improving the efficiency of sampling positioning calibration.

[0067] Example 2: The method for extracting the state characteristics of the target sampling container in Example 1 is further limited to avoid being limited to the deviation analysis of the sampling arm. In actual experimental scenarios, the vibration of the target sampling container is not only affected by environmental factors, but may also produce a synergistic effect with the movement of the sampling arm. Simply obtaining the vibration characteristics of the target sampling container based on the sampling arm deviation analysis may not be comprehensive and accurate enough, and cannot fully reflect the dynamic relationship between the two, thereby affecting the accuracy of sampling positioning, and ignoring the collaborative deviation characteristics between the sampling arm and the target sampling container. Therefore, the state characteristics of the target sampling container are introduced into the multi-source state characteristics, and together with the state characteristics of the sampling arm, constitute the multi-source state characteristics.

[0068] Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0069] In some embodiments, the vibration feature in the state feature of the target sampling container is generated according to a preset visual key feature extraction mechanism, which specifically includes:

[0070] S201, using a preset image acquisition device (installed above the carrier, used to acquire the mouth images of all sampling containers in the carrier, the installation position is set according to the actual scene and needs, and it is sufficient to acquire the mouth images of all sampling containers in the carrier), collect the frame image sequence in the current monitoring period, use the preset contour extraction and screening algorithm to obtain the target mouth area of ​​the target sampling container in each frame image, locate the edge pixel coordinate set of the target mouth area, fit the circular contour through the least squares method, and calculate the actual center coordinates of the target mouth area.

[0071] Specifically, the preset contour extraction and screening algorithm is:

[0072] B1. Convert each frame image into a grayscale image, use the mouth image recognition model to extract all mouth areas in the grayscale image, and obtain the unique position code of each mouth area in the carrier.

[0073] Specifically, the mouth image recognition model can be trained by using a large number of carrier images collected by image acquisition devices, and the mouth area of ​​the collected carrier images can be labeled. The labeled samples can be used for continuous training and learning to learn the image feature information of the mouth area of ​​the sampling container, thereby realizing the extraction of the mouth area. This will not be elaborated in the present invention.

[0074] Among them, all mouth areas are obtained in the grayscale image, and the unique position code of each mouth area in the grayscale image is determined, which is expressed as (a, b), where a is the row number of the mouth area in the grayscale image, and b is the column number of the mouth area in the grayscale image.

[0075] B2. Based on the loading position code of the target sampling container in the carrier (the pre-determined loading position code, i.e., the row and column number of the target sampling container in the carrier, expressed as (m, n), where m is the number of rows of the target sampling container and n is the number of columns of the target sampling container), compare it with the unique position codes of all mouth areas to determine the target mouth area of ​​the target sampling container.

[0076] It can be understood that both the unique position code and the loading position code are used to indicate the distribution position characteristics of the sampling containers in the carrier.

[0077] S202: Based on the image frame sequence of the current monitoring period, calculate the displacement vector of the actual center coordinates of the target mouth area in consecutive frames. According to the camera calibration parameters (pixel-to-physical scale conversion), convert the pixel displacement into actual physical displacement (the method of converting image pixel coordinates into actual physical coordinates can refer to relevant existing technologies and is not described in detail in this invention), and generate a displacement sequence.

[0078] S203 , performing principal component analysis on the displacement sequence, extracting the main vibration direction (consistent with the direction of maximum variance) as the vibration direction value, and setting the standard deviation of the displacement sequence in the main vibration direction as the vibration amplitude.

[0079] Specifically, step S203 includes:

[0080] C1. Convert the displacement sequence into M 2's displacement matrix X, M is the number of frames, calculate the covariance matrix C;

[0081] Among them, the displacement matrix X= , covariance matrix C = , and are the displacement components of the kth frame in the X and Y directions, M is the total number of frames in the detection cycle, is the transpose of the displacement matrix, and M 2 matrix converted to 2 M matrix, the dimension of the covariance matrix C is 2 2. Describe the variance and covariance in the X and Y directions: , represents the variance, represents the covariance.

[0082] C2. Solve the eigenvalues ​​of the covariance matrix C 、 ( ≥ ) and the corresponding eigenvector 、 , calculate the main direction of vibration as Vibration direction value: , is the vibration direction value;

[0083] Among them, the maximum eigenvalue Indicates that the data is in the main direction The variance on the eigenvalue reflects the concentration of vibration energy. Indicates that data is in the secondary direction The variance on , eigenvector Pointing in the main direction of vibration, They represent the main eigenvectors The components in the X and Y directions, is the vibration direction value, which indicates the angle between the main vibration direction and the X-axis.

[0084] C3. Project the displacement to the main direction , calculate the standard deviation to get the vibration amplitude: ;

[0085] in, , represents the displacement vector of the kth frame, is the main direction feature vector, M is the total number of frames, is the vibration amplitude, and exemplarily, the displacement vector is projected onto the main direction , and get the projection sequence , the standard deviation of the projection sequence reflects the degree of dispersion of the vibration in the main direction, that is, the vibration amplitude.

[0086] Through principal component analysis and data dimensionality reduction, the main vibration direction is identified, the main vibration mode is extracted from the two-dimensional displacement data, the secondary interference is suppressed, and redundant calculations are reduced. The direction and amplitude directly reflect the vibration characteristics of the target sampling container.

[0087] S204 , performing a fast Fourier transform (FFT) on the projection sequence of the displacement sequence onto the main direction, calculating the power spectrum density, and extracting the main frequency (the frequency corresponding to the peak of the power spectrum density) as the vibration regularity value.

[0088] For example, the frequency domain signal is obtained by calculating FFT , the power spectral density is: , the sampling rate is 200 Hz, the power spectrum density length N is 256 (number of frames), and the peak of the power spectrum density appears at 25 Hz (corresponding to the motor operating frequency), which is used as the vibration regularity value to reflect the vibration characteristics of the target sampling container at this time.

[0089] S205 , combining the vibration direction value, the vibration amplitude, and the vibration regularity value to form the vibration characteristics of the target sampling container.

[0090] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0091] On a laboratory workbench, the target sampling container may experience micron-level displacement due to environmental vibration (such as equipment motor operation, tabletop resonance) or human interference. Vibration feature extraction based on high-frequency image sequences can capture the container's instantaneous position changes in real time, providing data support for dynamic calibration of the sampling path. By extracting the vibration direction, amplitude, and dominant frequency from multiple dimensions, the intensity and pattern of vibration interference on the container, as well as the dynamic relationship, can be accurately quantified. For example, low-frequency vibration (such as the impact of equipment startup and shutdown) may cause the container to drift slowly, while high-frequency vibration (such as motor resonance) may induce periodic micro-tremors. This data provides a classification basis for the compensation algorithm. Combined with the matching analysis of the sampling arm's acceleration and the container's dominant vibration frequency, the implicit vibration relationship between the sampling arm and the container can be derived, improving the accuracy of deviation prediction.

[0092] Traditional static calibration relies only on initial position information and cannot cope with real-time disturbances during the experiment. Through vibration feature compensation, the sampling tool can dynamically adjust the path points; in scenarios such as cell culture and PCR sampling, micron-level deviations at the container mouth may cause the sampling tool to hit the tube wall or liquid splashing. Vibration compensation can avoid such accidents and improve the sampling success rate; in the continuous sampling scenario of well plates, the vibration compensation algorithm can automatically adapt to the batch position deviations of the well plate (such as carrier installation errors), reducing the frequency of manual recalibration.

[0093] In experimental environments, complex vibration sources (e.g., centrifuges, shakers, and sampling tools coexisting) can make it difficult to distinguish interference types manually. By extracting features based on vibration direction and frequency, we can precisely identify the source of interference and accurately represent the real-time vibration characteristics of the target sampling container.

[0094] Example 3: When approaching a target, the traditional solution may frequently adjust the deviation correction due to environmental noise or sensor jitter, resulting in path oscillation. Frequent path corrections increase the computational load of the motion controller. Instantaneous errors in visual positioning or encoders may lead to sudden and large deviation corrections, causing mechanical shock.

[0095] Therefore, the embodiments of the present application are optimized based on the above embodiments.

[0096] In some embodiments, before step S103, the method further includes:

[0097] S301, based on the effective path of the sampling arm in the initial sampling path at the current moment, obtain the effective proportion value, the effective path is set to the remaining path in the initial sampling path that the sampling arm has not yet reached, and the effective proportion value is set to the ratio of the effective path length to the initial sampling path length.

[0098] S302: Using the effective proportion value, obtain a correction dependency factor, multiply the correction dependency factor by the position deviation pair, and obtain a new position deviation pair to replace the original position deviation pair.

[0099] Specifically, the correction dependency factor is calculated according to the following formula:

[0100]

[0101] in, is the correction dependency factor, is the effective ratio value, k is the attenuation slope coefficient (the larger k is, the faster the attenuation is), which is used to control the correction amplitude.

[0102] Therefore, the effective ratio The closer it is to 1, the longer the remaining path is, and the correction needs to respond quickly. The closer it is to 0, the shorter the remaining path is, and over-adjustment needs to be suppressed to achieve a higher degree of dependence when the distance from the target sampling container is far (allowing a large range of correction and quickly eliminating the initial sampling path deviation). When approaching the target sampling container, the dependence gradually decreases and oscillations are suppressed (suppressing high-frequency small oscillations and avoiding mechanical shocks, such as inertial shocks caused by sudden stops or reverse adjustments, to ensure a smooth approach to the end point).

[0103] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages:

[0104] By calculating the effective proportion value and the correction dependency factor, dynamic adjustment of the degree of dependence on correction at different stages is achieved. By dynamically adjusting the correction dependency, the path oscillation phenomenon when approaching the target sampling container is effectively suppressed, so that the sampling arm can approach the end point more smoothly, improving the accuracy of sampling and reducing frequent path corrections, thereby reducing the computing load of the motion controller and improving the operating efficiency and stability of the system; it also suppresses sudden and large-scale corrections, reduces the occurrence of mechanical shock, protects the joints and transmission components of the sampling arm, reduces the failure rate of the equipment, and extends the service life of the equipment.

[0105] In scenarios such as cell culture and PCR sampling that require high sampling accuracy, reducing path oscillation and mechanical impact can avoid accidents such as the sampling tool hitting the tube wall or liquid splashing, thereby improving the success rate of sampling.

[0106] Embodiment 4: In some embodiments, the state traction deviation prediction model further includes several deviation category prediction models, each deviation category prediction model corresponds to a deviation category feature label, and the acquisition method includes:

[0107] S401 , based on each training sample in the training sample set, extract the values ​​corresponding to the joint torque fluctuation and the vibration regularity value in its multi-source state feature vector to form a key feature vector.

[0108] S402: Perform cluster analysis (such as K-means or DBSCAN) based on all key feature vectors to classify them into several sample classes. Each sample class includes at least one key feature vector. The average of the key feature vectors is used as the deviation class feature label of the sample class.

[0109] S403: Based on the training samples corresponding to all key feature vectors in each sample class, a pre-selected neural network structure is trained to obtain a deviation category prediction model, and a corresponding deviation category feature label is assigned.

[0110] Specifically, in step S02, the multi-source state feature vectors are input into a pre-trained state traction deviation prediction model, which also includes:

[0111] Extract the multi-source state feature vector to obtain the key feature vector, calculate the Euclidean distance value between the key feature vector and the deviation category feature label corresponding to each deviation category prediction model, input the multi-source state feature vector into the deviation category prediction model with the smallest Euclidean distance value, and output the position deviation pair.

[0112] In a dynamic environment, the deviation between the sampling arm and the sampling container may be caused by the coupling of multiple factors (such as mechanical errors and vibrations). The general model has insufficient prediction accuracy for specific deviation patterns, and is particularly prone to failure in scenarios where multiple disturbances coexist. During model training, based on high-dimensional feature inputs, the inference delay of a single large model is high. Therefore, through cluster analysis, the key features with the strongest correlation with deviation in the multi-source state features (such as joint torque fluctuations and vibration regularity values) are divided into different categories. Each category corresponds to a specific deviation pattern (such as mechanical error dominance and high-frequency vibration dominance). Only features with a significant impact on the deviation are retained to improve clustering efficiency and classification accuracy. Each cluster center represents a feature combination of a typical deviation scenario. As an example, Category A: high torque fluctuation + low-frequency vibration → mechanical transmission error dominates; Category B: low torque fluctuation + high-frequency vibration → environmental disturbances dominate.

[0113] The vibration regularity value of the target sampling container and the torque fluctuation value of the sampling arm are extracted as key feature vectors to perform preliminary classification of the messy sample data, establish targeted differentiated predictions of specific deviation patterns for subsequent model predictions, improve prediction accuracy, avoid performance fluctuations of a single model in complex scenarios, and enhance stability when multiple disturbances coexist.

[0114] Furthermore, an embodiment of the present invention also provides a calibration system for sampling positioning accuracy based on a sample library.

[0115] Figure 2 This is a structural block diagram of a calibration system for sampling positioning accuracy based on a sample library according to an embodiment of the present invention.

[0116] like Figure 2 As shown, the calibration system for sampling positioning accuracy based on the sample library includes: an acquisition module, a prediction module, and a correction module.

[0117] Specifically, the acquisition module is used to control the sampling tool loaded on the sampling arm to start working based on a preset initial sampling path, and obtain the multi-source state characteristics of the sampling environment at preset time intervals; the prediction module is used to normalize the multi-source state characteristics to form a multi-source state feature vector, which is input into a pre-trained state traction deviation prediction model, and output a position deviation pair, including the position deviation of the current path point of the sampling tool and the center position deviation of the target sampling container; the correction module is used to dynamically correct and update the initial sampling path based on the position deviation pair, periodically monitor the sampling environment, and repeatedly jump to the prediction module until the sampling arm is positioned at the center point of the target sampling container.

[0118] It should be noted that other specific implementation contents of the embodiment of the present invention may refer to the above-mentioned calibration method for sampling positioning accuracy based on a sample library.

[0119] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations are readily apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A calibration method for sampling positioning accuracy based on a sample library, characterized in that: include: S101, controlling the sampling tool loaded on the sampling arm to start working based on a preset initial sampling path, and obtaining multi-source state characteristics of the sampling environment at preset intervals, including state characteristics of the sampling arm and state characteristics of the target sampling container, wherein the state characteristics of the sampling arm include command speed and acceleration, and joint torque fluctuations, and the state characteristics of the target sampling container include vibration characteristics, wherein the vibration characteristics are obtained by: using an image acquisition device to acquire a frame image sequence within a current monitoring period, using a contour extraction and screening algorithm to obtain a target mouth area of ​​the target sampling container in each frame image, and determining the actual center coordinates of the target mouth area; calculating a displacement vector of the actual center coordinates of the target mouth area in consecutive frames, and converting pixel displacement into actual physical displacement according to camera calibration parameters of the image acquisition device to generate a displacement sequence; performing principal component analysis on the displacement sequence, extracting the main vibration direction as the vibration direction value, and setting the standard deviation of the displacement sequence in the main vibration direction as the vibration amplitude; performing fast Fourier transform on a projection sequence of the displacement sequence onto the main direction, calculating the power spectral density, and extracting the main frequency as the vibration regularity value; and combining the vibration direction value, vibration amplitude, and vibration regularity value to form the vibration characteristics of the target sampling container; S102: Normalize the multi-source state features to form a multi-source state feature vector, input it into a pre-trained state traction deviation prediction model, and output a position deviation pair, including the position deviation of the current path point of the sampling tool and the center position deviation of the target sampling container; based on the effective path of the sampling arm in the initial sampling path at the current moment, obtain an effective proportion value, where the effective path is set to the remaining path in the initial sampling path that the sampling arm has not yet reached, and the effective proportion value is set to the ratio of the effective path length to the initial sampling path length; use the effective proportion value to obtain a correction dependency factor, multiply the correction dependency factor by the position deviation pair, and obtain a new position deviation pair to replace the original position deviation pair; S103, based on the position deviation pair, dynamically correct and update the initial sampling path, periodically monitor the sampling environment, and repeat steps S102 to S103 until the sampling arm is positioned at the center point of the target sampling container.

2. The method for calibrating sampling positioning accuracy based on a sample library according to claim 1, wherein: The preset initial sampling path is determined as follows: The initial calibration position of the sampling arm on the workbench and the theoretical center position of the target sampling container are obtained from the preset sample library to generate an initial sampling path.

3. The method for calibrating sampling positioning accuracy based on a sample library according to claim 1, wherein: The pre-trained state traction deviation prediction model is used to predict the real-time position deviation of the sampling arm and the target sampling container based on the input multi-source state feature vector, and is used to dynamically calibrate the initial sampling path. The state traction deviation prediction model is obtained in the following manner: A1. Collect a large number of multi-source state features and their corresponding actual deviation pairs from different sampling environments and different sampling paths. Use the actual deviation pairs to label the multi-source state features as a training sample set. A2. Use the training sample set to train the pre-selected neural network structure, continuously optimize the model parameters, and generate the final state traction deviation prediction model.

4. The method for calibrating sampling positioning accuracy based on a sample library according to claim 3, wherein: The actual deviation pair includes the actual position deviation of the sampling arm and the actual center position deviation of the target sampling container; the actual position deviation is set to: the difference between the actual position of the sampling arm at the current moment and the position at the corresponding time point in the preset initial path; the actual center position deviation is set to: the difference between the actual center position of the target sampling container and the theoretical center position.

5. The method for calibrating sampling positioning accuracy based on a sample library according to claim 3, wherein: In S103, the dynamic correction and update of the initial sampling path based on the position deviation pair specifically includes: Sum the position of the sampling arm in the initial sampling path at the current moment and the actual position deviation, and update the position information of the initial sampling path at the current moment to correct the deviation; The deviation between the theoretical center position and the actual center position of the target sampling container is summed, the end point position of the initial sampling path is updated and corrected, and the theoretical center position of the target sampling container is updated together.

6. The method for calibrating sampling positioning accuracy based on a sample library according to claim 1, wherein: The correction dependency factor is calculated according to the following formula: ； in, is the correction dependency factor, is the effective proportion value, and k is the preset attenuation slope coefficient, which is used to control the correction amplitude.

7. A calibration system for sampling positioning accuracy based on a sample library, which executes the calibration method for sampling positioning accuracy based on a sample library according to any one of claims 1 to 6, characterized in that: The system includes: an acquisition module, a prediction module, and a correction module; The acquisition module is used to control the sampling tool loaded on the sampling arm to start working based on a preset initial sampling path, and obtain multi-source state characteristics of the sampling environment at preset time intervals; The prediction module is used to normalize the multi-source state features to form a multi-source state feature vector, which is input into a pre-trained state traction deviation prediction model and outputs a position deviation pair, including the position deviation of the sampling tool's current path point and the center position deviation of the target sampling container; The correction module is used to dynamically correct and update the initial sampling path based on the position deviation, periodically monitor the sampling environment, and repeatedly jump to the prediction module until the sampling arm is positioned at the center point of the target sampling container.

Images