An adaptive coordinate calibration method for samples inside a liquid nitrogen tank

Abstract

This invention discloses an adaptive coordinate calibration method for samples inside a liquid nitrogen tank, belonging to the field of control systems and automation technology. The method includes acquiring multi-source environmental data inside the liquid nitrogen tank and processing the multi-source environmental data to generate multi-source sensor data; generating dynamic weight allocation parameters based on environmental state characteristics in the multi-source environmental data; performing fusion compensation calculations on the multi-source sensor data according to the dynamic weight allocation parameters to generate calibration coordinates; acquiring the actual grasping position of the sample during storage and retrieval, calculating the deviation value between the actual grasping position and the calibration coordinates; and updating the weight calculation rules of the dynamic weight allocation parameters according to the deviation value. This invention employs a closed-loop feedback control mechanism combined with adaptive compensation technology for environmental disturbances, enabling high-precision dynamic calibration of sample coordinates in complex environments inside a liquid nitrogen tank, improving the positioning accuracy and reliability during sample storage and retrieval.

Description

Technical Field

[0001] This invention relates to the field of control systems and automation technology, and in particular to an adaptive coordinate calibration method for samples inside a liquid nitrogen tank. Background Technology

[0002] Liquid nitrogen tanks are ultra-low temperature storage devices widely used in biomedicine, life sciences, and other fields for the long-term preservation of biological samples. To automate sample access and management, robotic arms and other automated devices are typically employed. Precise coordinate calibration is a prerequisite for the automated device to accurately grasp and place samples, directly impacting the reliability of the automated system and the safety of the samples. Coordinate calibration aims to determine the precise position of the sample in the three-dimensional space within the tank, providing a target point for the motion planning of the control system.

[0003] Currently, in automated systems, coordinate positioning is typically achieved using methods based on vision, laser, or a fusion of multiple sensors. For example, a camera deployed inside a liquid nitrogen tank captures image features of a sample rack, which is then combined with data from a ranging sensor to calculate the three-dimensional coordinates. Under ideal conditions, these methods can provide relatively accurate positioning information, enabling the robotic arm to perform storage and retrieval operations.

[0004] However, existing calibration methods have significant limitations when dealing with the unique dynamic working environment inside liquid nitrogen tanks. When the tank lid is opened, the contact between outside air and the cryogenic nitrogen gas inside instantly forms a dense fog, severely interfering with the imaging quality of the vision sensor. Simultaneously, the movement of the robotic arm and the disturbance of the liquid nitrogen cause minute vibrations in the tank and sample holder. Existing technical solutions often use fixed weights and compensation parameters to process sensor data, making it difficult to adapt to this complex environment with real-time changes in fog concentration and vibration intensity, resulting in significant errors in calibration coordinates. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an adaptive coordinate calibration method for samples inside a liquid nitrogen tank. By employing a closed-loop feedback control mechanism and combining it with adaptive compensation technology for environmental disturbances, this method can achieve high-precision dynamic calibration of sample coordinates in complex environments within a liquid nitrogen tank, thereby improving the positioning accuracy and reliability during sample storage and retrieval.

[0006] The above objectives can be achieved through the following approach:

[0007] An adaptive coordinate calibration method for samples inside a liquid nitrogen tank includes: acquiring multi-source environmental data inside the liquid nitrogen tank and processing the multi-source environmental data to generate multi-source sensor data; generating dynamic weight allocation parameters based on environmental state characteristics in the multi-source environmental data; performing fusion compensation calculation on the multi-source sensor data according to the dynamic weight allocation parameters to generate calibration coordinates; acquiring the actual grasping position of the sample and calculating the deviation value between the actual grasping position and the calibration coordinates; and updating the weight calculation rules of the dynamic weight allocation parameters according to the deviation value.

[0008] Optionally, the step of acquiring multi-source environmental data inside the liquid nitrogen tank and processing the multi-source environmental data to generate multi-source sensor data includes: acquiring infrared ranging data and image feature data, and combining them to form original position data; acquiring tank mechanical vibration data and real-time temperature inside the tank; and performing temperature shielding processing on the original position data and the tank mechanical vibration data based on the real-time temperature inside the tank to generate multi-source sensor data.

[0009] Optionally, generating dynamic weight allocation parameters based on environmental state features in the multi-source environmental data includes: extracting fog density parameters and vibration frequency parameters from the multi-source environmental data; and combining the fog density parameters and vibration frequency parameters as environmental state features to generate the dynamic weight allocation parameters.

[0010] Optionally, the step of performing fusion compensation calculation on the multi-source sensor data according to the dynamic weight allocation parameters to generate calibration coordinates includes: weighting and fusing the infrared ranging data and image feature data according to the sensor weight allocation ratio in the dynamic weight allocation parameters to generate preliminary fused coordinates; performing vibration displacement compensation on the preliminary fused coordinates using vibration compensation data in the multi-source sensor data according to the vibration compensation intensity in the dynamic weight allocation parameters to generate dynamically compensated coordinates; and iteratively correcting the dynamically compensated coordinates to generate calibration coordinates.

[0011] Optionally, the step of iteratively correcting the dynamically compensated coordinates to generate calibration coordinates includes: obtaining a digital grid model defining the ideal coordinates of the sample storage location; using the dynamically compensated coordinates as initial coordinate estimates and comparing them with the ideal coordinates in the digital grid model to calculate a correction vector; applying the initial coordinate estimates to the correction vector and iteratively correcting until convergence to generate calibration coordinates.

[0012] Optionally, the step of updating the weight calculation rule of the dynamic weight allocation parameter according to the deviation value includes: obtaining an error threshold and a concentration threshold; increasing the weight of the vibration frequency parameter of the dynamic weight allocation parameter when the deviation value exceeds the error threshold; and decreasing the weight of the image feature data of the dynamic weight allocation parameter when the fog density parameter in the environmental state features exceeds the concentration threshold.

[0013] Optionally, the step of calculating the deviation value by comparing the actual grasping position with the calibration coordinates includes: performing three-dimensional spatial vector analysis on the actual grasping position and the calibration coordinates to obtain a position error vector; and decoupling the components of the position error vector to obtain the deviation value.

[0014] Optionally, the step of performing three-dimensional spatial vector analysis on the actual grasping position and the calibration coordinates to obtain the position error vector includes: calculating the difference between the coordinate components of the actual grasping position and the calibration coordinates to obtain the deviation value of each coordinate axis; and combining the deviation values ​​of each coordinate axis into a vector to obtain the position error vector.

[0015] Optionally, applying the initial coordinate estimate to the correction vector and performing iterative correction until convergence to generate calibration coordinates includes: performing vector addition on the initial coordinate estimate and the correction vector to obtain an updated coordinate estimate; and repeatedly performing vector correction and convergence judgment on the updated coordinate estimate to generate calibration coordinates.

[0016] Based on the same inventive concept, this invention also provides an adaptive coordinate calibration system for samples inside a liquid nitrogen tank. The system includes: a multi-source environmental sensing module for acquiring multi-source environmental data inside the liquid nitrogen tank and processing the multi-source environmental data to generate multi-source sensor data; a dynamic weight generation module for generating dynamic weight allocation parameters based on environmental state characteristics in the multi-source environmental data; a coordinate calibration module for performing fusion compensation calculations on the multi-source sensor data according to the dynamic weight allocation parameters to generate calibration coordinates; a deviation feedback analysis module for acquiring the actual grasping position of the sample and calculating the deviation value by comparing the actual grasping position with the calibration coordinates; and a weight adaptive update module for updating the weight calculation rules of the dynamic weight allocation parameters according to the deviation value.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] This invention achieves intelligent fusion and compensation of sensor data by real-time monitoring of multi-source environmental data within a liquid nitrogen tank and dynamically adjusting weight parameters based on environmental characteristics. This effectively suppresses interference from environmental factors such as dense fog and vibration on coordinate measurements, ensuring that coordinate calibration results maintain high stability and reliability even under complex and changing working conditions.

[0019] This invention establishes a closed-loop feedback system from calibration calculation to actual operation. By comparing the actual grasping position with the calibration coordinates, the deviation value is quantified, and this deviation value is used to update the weight calculation rules in reverse. This enables the calibration system to have the ability to learn and evolve, continuously adapt to changes in equipment status and environmental drift, and maintain long-term calibration accuracy.

[0020] The method of this invention can dynamically adjust the weights of different sensors according to environmental characteristics. For example, in a dense fog environment, it can reduce the reliance on visual images and instead rely on other sensors. This adaptability greatly enhances the calibration system's tolerance to harsh environments, avoids the problem of interrupting the entire automated process due to the failure or performance degradation of a single sensor, and improves the robustness of the system.

[0021] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating an adaptive coordinate calibration method for samples inside a liquid nitrogen tank, according to an embodiment of the present invention.

[0024] Figure 2 This is a diagram illustrating the effect of temperature shielding correction in an embodiment of the present invention.

[0025] Figure 3 This is a digital grid model diagram according to an embodiment of the present invention.

[0026] Figure 4 This is a diagram illustrating the coordinate iteration correction convergence process in an embodiment of the present invention.

[0027] Figure 5This is a diagram illustrating the adaptive update effect of the weight calculation rules in an embodiment of the present invention.

[0028] Figure 6 This is a schematic diagram of the structure of an adaptive coordinate calibration system for samples inside a liquid nitrogen tank, according to an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Reference Figure 1 One embodiment of the present invention proposes an adaptive coordinate calibration method for samples inside a liquid nitrogen tank. It adopts a closed-loop feedback control mechanism and combines environmental disturbance adaptive compensation technology to achieve high-precision dynamic calibration of sample coordinates in complex environments inside a liquid nitrogen tank, thereby improving the positioning accuracy and reliability during sample storage and retrieval.

[0031] The method described in this embodiment specifically includes:

[0032] Acquire multi-source environmental data inside the liquid nitrogen tank, and process the multi-source environmental data to generate multi-source sensor data;

[0033] Based on the environmental state characteristics in the multi-source environmental data, dynamic weight allocation parameters are generated.

[0034] The multi-source sensor data is fused and compensated according to the dynamic weight allocation parameters to generate calibration coordinates.

[0035] Obtain the actual grab position of the sample storage, and calculate the deviation value by comparing the actual grab position with the calibration coordinates;

[0036] Based on the deviation value, update the weight calculation rules of the dynamic weight allocation parameters.

[0037] Specifically, the system first integrates sensor information from multiple sources to initially perceive the spatial position of the sample rack inside the liquid nitrogen tank. Its core innovation lies in the fact that it does not process all sensor information equally. Instead, it dynamically generates a set of weighting parameters by analyzing key environmental characteristics such as mist and vibration within the tank in real time. These parameters determine the reliability of different sensor data and the required compensation intensity under the current environment. The system uses these dynamic parameters to intelligently weight, fuse, and compensate the multi-source sensor data, generating an optimal calibration coordinate. More importantly, this method introduces a feedback mechanism based on actual operational results. It compares the actual position grasped by the automated equipment with the system-calibrated coordinates, quantifying the deviation value. This deviation value is not only used for a single evaluation but also serves as an input signal to drive the system's self-learning, updating and optimizing the calculation rules for the weighting parameters themselves.

[0038] Optionally, the step of acquiring multi-source environmental data within the liquid nitrogen tank and processing the multi-source environmental data to generate multi-source sensor data includes:

[0039] Acquire infrared ranging data and image feature data, and combine them to form raw position data;

[0040] Acquire mechanical vibration data of the tank and real-time temperature inside the tank;

[0041] Based on the real-time temperature inside the tank, the original position data and the mechanical vibration data of the tank are subjected to temperature shielding processing to generate multi-source sensor data.

[0042] Specifically, firstly, infrared ranging data and image feature data are acquired and combined to form the raw position data. Operationally, an infrared ranging sensor measures the straight-line distance between the probe and a preset reference point on the sample holder; this measurement is the infrared ranging data. Simultaneously, an industrial camera captures an image containing the sample holder, and an image recognition algorithm locates one or more feature points of the sample holder with well-defined geometric shapes, extracting their pixel coordinates in the image to form image feature data. Using the infrared ranging data as depth information, combined with the camera's intrinsic parameter model, the two-dimensional pixel coordinates in the image can be converted into three-dimensional spatial coordinates, thus initially determining the position of the sample holder; these three-dimensional spatial coordinates are the raw position data. Secondly, mechanical vibration data of the tank and real-time temperature inside the tank are acquired. An accelerometer installed on the tank structure collects vibration signals caused by external equipment or internal liquid nitrogen disturbances in real time; this signal is the tank's mechanical vibration data. Simultaneously, thermocouple temperature sensors deployed near the measurement area continuously measure the near-minus 196 degrees Celsius low-temperature environment inside the tank to obtain the real-time temperature inside the tank. Finally, temperature shielding processing is applied to the original position data and the mechanical vibration data of the tank based on the real-time temperature inside the tank, generating multi-source sensor data. Temperature shielding is a data compensation algorithm designed to eliminate the negative impact of extremely low temperatures on sensor measurement accuracy. Since the physical characteristics of the sensors drift with drastic temperature changes, correction based on real-time temperature is necessary. For the original position data, the errors originate from the infrared ranging sensor and the industrial camera image sensor. The correction process can be expressed by the following formula:

[0043] P c =P r +ΔP(T),

[0044] Among them, P c These are the corrected position coordinates, i.e., the result of the original position data after temperature shielding processing. P rThis is the raw position data obtained directly from calculation. T is the real-time temperature inside the tank. ΔP(T) is a temperature-related compensation vector. The value of this vector is determined by establishing a temperature error model or lookup table based on calibration experimental data at different low temperatures before the sensor leaves the factory. The temperature error model establishment process is as follows: Experimental data acquisition: Simulate the tank environment using a low-temperature constant temperature chamber (-200℃~25℃). At five temperature points—-196℃, -180℃, -160℃, -140℃, and 25℃—obtain the true value using a high-precision calibration stage (≤0.01mm). Simultaneously acquire the raw sensor data, collecting 50-100 sets of data for each point to calculate the average error. Data fitting: Fit the "temperature-average error" to a 1st-3rd order polynomial using the least squares method (selecting the lowest RMSE order), or generate a lookup table at 10℃ intervals. Validation and deployment: Validate using temperature points not involved in the fitting (e.g., -192℃). An error ≤0.1mm is considered acceptable. Write the model parameters into the hardware and calibrate and update every 6 months. The mechanical vibration data of the tank are corrected as follows:

[0045] A c =A r +ΔA(T),

[0046] Among them, A c These are the corrected vibration data, A r The data consists of raw vibration data collected by an accelerometer, and ΔA(T) is the compensation value obtained by querying a pre-stored temperature-vibration error relationship based on the real-time temperature T inside the tank. The corrected position coordinates after temperature shielding and the corrected vibration data together constitute the multi-source sensor data used for subsequent fusion calculations. The temperature shielding correction effect is as follows: Figure 2 As shown.

[0047] Optionally, generating dynamic weight allocation parameters based on environmental state features in the multi-source environmental data includes:

[0048] Fog density parameters and vibration frequency parameters are extracted from the multi-source environmental data;

[0049] The fog density parameter and the vibration frequency parameter are combined as environmental state features to generate the dynamic weight allocation parameter.

[0050] Specifically, two key environmental state parameters—fog density and vibration frequency—were extracted from previously acquired and temperature-shielded multi-source environmental data. The fog density parameter was extracted by analyzing image feature data. When the liquid nitrogen tank lid is opened, the contact between the warm, humid outside air and the low-temperature nitrogen inside the tank instantly generates dense fog, severely affecting image quality. By processing real-time image sequences, such as calculating image contrast, sharpness, or gradient variance, the impact of fog on visual information acquisition can be quantified. The more blurred the image, the lower the calculated sharpness value, and vice versa; thus, a fog density parameter inversely proportional to visual visibility can be generated. The vibration frequency parameter was extracted by analyzing the tank's mechanical vibration data. The time-domain vibration signal acquired by the accelerometer was converted to the frequency domain using signal processing techniques such as Fast Fourier Transform, thereby analyzing its spectral characteristics. The main frequencies of concentrated vibration energy or the frequencies with the largest amplitudes were identified and used as the vibration frequency parameter characterizing the current mechanical disturbance state. Subsequently, the extracted fog density and vibration frequency parameters are combined to form a multi-dimensional environmental state feature. This environmental state feature comprehensively describes the main interference sources affecting the coordinate measurement accuracy at the current moment, namely the severity of visual occlusion and the instability of mechanical structure displacement. Finally, based on this environmental state feature, dynamic weight allocation parameters are generated through a pre-defined weight allocation model.

[0051] Optionally, the step of performing fusion compensation calculation on the multi-source sensor data according to the dynamic weight allocation parameters to generate calibration coordinates includes:

[0052] Based on the sensor weight allocation ratio in the dynamic weight allocation parameters, the infrared ranging data and image feature data are weighted and fused to generate preliminary fused coordinates;

[0053] Based on the vibration compensation intensity in the dynamic weight allocation parameters, vibration compensation data from the multi-source sensor data is used to perform vibration displacement compensation on the preliminary fused coordinates to generate dynamically compensated coordinates.

[0054] The dynamically compensated coordinates are iteratively corrected to generate calibration coordinates.

[0055] Specifically, firstly, based on the sensor weight allocation ratio in the dynamic weight allocation parameters, the infrared ranging data and image feature data from the multi-source sensor data are weighted and fused to generate preliminary fused coordinates. In this step, the system calculates two independent estimated values ​​of the sample frame position coordinates using the infrared ranging data and image feature data, respectively. Then, based on the sensor weight allocation ratio previously determined by the fog density parameter, these two coordinate estimates are weighted and averaged. This process can be expressed by the following formula:

[0056] Pf =w img ·P img +w ir ·P ir ,

[0057] Among them, P f These are the generated preliminary fusion coordinates. P img These are the three-dimensional spatial coordinates calculated from image feature data. P ir These are the three-dimensional spatial coordinates calculated from the distance data of the sample holder's reference points. img and w ir These are the weight coefficients obtained from the dynamic weight allocation parameters, corresponding to the image data and the infrared ranging data respectively, and w img The value decreases as the fog density increases. Next, based on the vibration compensation intensity in the dynamic weight allocation parameters, vibration compensation data from multi-source sensors is used to perform vibration displacement compensation on the preliminary fused coordinates, generating dynamically compensated coordinates. The vibration compensation data is the mechanical vibration data of the tank after temperature shielding treatment, which can be converted into a real-time displacement compensation vector through integration or model analysis. The system combines this displacement vector with the vibration compensation intensity obtained from the dynamic weight allocation parameters to correct the preliminary fused coordinates, thereby offsetting the positional deviation caused by mechanical vibration. The calculation is as follows:

[0058] P c =P f +S vib ·ΔP vib ,

[0059] Among them, P c These are the dynamically compensated coordinates. P f These are the preliminary fusion coordinates obtained in the previous step. S vib The vibration compensation intensity is obtained from the dynamic weight allocation parameters. This value is positively correlated with the vibration frequency parameter and represents the magnitude of the compensation. ΔP vib The real-time vibration displacement vector is calculated from the vibration compensation data. Finally, the dynamically compensated coordinates are iteratively corrected to generate the final calibration coordinates.

[0060] Optionally, the iterative correction of the dynamically compensated coordinates to generate calibration coordinates includes:

[0061] Obtain a digital grid model that defines the ideal coordinates of the sample storage location;

[0062] The dynamically compensated coordinates are used as the initial coordinate estimates and compared with the ideal coordinates in the digital grid model to calculate the correction vector.

[0063] The initial coordinate estimate is applied to the correction vector, and the correction is iteratively performed until convergence, generating the calibration coordinates.

[0064] Specifically, the first step is to obtain a digital grid model that defines the ideal coordinates of the sample storage location. The digital grid model is as follows: Figure 3 As shown. This model is a pre-established, high-precision 3D coordinate database that precisely describes the theoretical center coordinates of each storage hole on the sample rack inside the liquid nitrogen tank in digital form. This model can be generated based on the sample rack design drawings or established through a one-time precise calibration offline using high-precision measuring equipment. It represents the "absolute true" reference system for the sample storage location. The correction process begins by using the dynamically compensated coordinates generated in the previous step as the initial coordinate estimate. The system compares this initial coordinate estimate with all ideal coordinates in the digital grid model, typically using a nearest neighbor algorithm to determine the ideal coordinate that is spatially closest to the initial coordinate estimate. Once the corresponding ideal coordinate is found, the system calculates a correction vector pointing from the current initial coordinate estimate to its corresponding ideal coordinate. Subsequently, the system applies this correction vector to the initial coordinate estimate, iteratively correcting until convergence, ultimately generating the calibration coordinates. This iterative process is as follows: The initial coordinate estimate is added to the calculated correction vector to obtain an updated coordinate estimate. This operation can be expressed as:

[0065] P est (k+1)=P est (k)+α·(P ideal -P est (k)),

[0066] Among them, P est (k+1) is the updated coordinate estimate obtained in the (k+1)th iteration. P est (k) is the coordinate estimate for the k-th iteration. In the first iteration, P est (0) represents the input coordinates after dynamic compensation. P ideal It is found from the digital grid model at the k-th iteration that is related to P. est (k) The closest ideal coordinates. α is a step size factor or learning rate, ranging from 0 to 1, used to control the magnitude of each correction to ensure stable convergence of the iterative process. Expression P ideal -P est(k) represents the correction vector calculated in step k. The system will repeat the vector correction and convergence judgment process for this updated coordinate estimate. Convergence is determined when the Euclidean distance between the coordinate estimates obtained from two consecutive iterations is less than a preset minimum threshold, or when the magnitude of the correction vector is less than that threshold. The coordinate iteration correction convergence process is as follows: Figure 4 As shown. The iteration process terminates, and the estimated coordinates at this point are output as the final calibration coordinates.

[0067] Optionally, the weight calculation rule for updating the dynamic weight allocation parameters based on the deviation value includes:

[0068] Obtain the error threshold and concentration threshold;

[0069] When the deviation value exceeds the error threshold, the weight of the vibration frequency parameter of the dynamic weight allocation parameter is increased;

[0070] When the fog density parameter in the environmental state features exceeds the concentration threshold, the image feature data weight of the dynamic weight allocation parameter is reduced.

[0071] First, the system needs to pre-set an error threshold and a concentration threshold. The error threshold is calibrated experimentally as follows: Initial setting: Combining the robotic arm's repeatability (e.g., ±0.1mm) and inherent sensor error (e.g., infrared ±0.05mm), take 1.5-2 times the maximum hardware error (e.g., 0.2mm), and then add sample safety redundancy (e.g., sample tube hole position tolerance) for adjustment. Experimental calibration: Use a high-precision calibration stage (≤0.001mm) to simulate different vibration / fog conditions, statistically analyze the maximum deviation within the 95% confidence interval, and set the threshold to 1.2 times this value (to avoid false triggering / insufficient triggering). Optional optimization: During long-term operation, fine-tune based on historical deviation data (e.g., if 90% of the deviation is stable at 0.1-0.15mm, it can be fine-tuned from 0.2mm to 0.18mm). The concentration threshold is calibrated experimentally as follows: Prerequisite: First, quantify "fog density" into image clarity (normalized 0-1, 0 for no fog, 1 for complete blur). Experimental Calibration: A fog generator is used to simulate different fog densities to test the success rate of image feature extraction (e.g., baseline detection rate). When the success rate drops sharply from 95% to below 80%, or the image coordinate transformation error exceeds 0.5mm, the fog density at this point is the concentration threshold (e.g., 0.6). Equipment Adaptation: Recalibration is required when changing cameras / lenses (e.g., the threshold can be increased to 0.7 for high-pixel cameras). The error threshold defines the maximum acceptable deviation between the calibrated coordinates and the actual grasping position, serving as the standard for judging the success of a single calibration. The concentration threshold defines the critical level of fog density that affects the performance of the visual sensor, used to assist in judging the source of error. After a complete sample access operation, the system compares the acquired deviation value with these thresholds and triggers the corresponding weight calculation rule update. The adaptive update effect of the weight calculation rule is as follows: Figure 5 As shown. The update logic is divided into two cases: The first case is when the calculated deviation value exceeds the preset error threshold, indicating that the current calibration coordinate accuracy is insufficient. In this case, the system will analyze and consider mechanical vibration to be an important potential factor causing the deviation, requiring enhanced suppression of the vibration effect. To this end, the system will increase the weight of the vibration frequency parameter in the dynamic weight allocation parameters. This is not a direct modification of the single vibration compensation intensity, but rather an adjustment of the underlying generation rules. For example, the calculation rule for the vibration compensation intensity can be expressed as:

[0072] S vib =γ·g(F vib ),

[0073] Among them, S vib It is the vibration compensation intensity, F vibThe vibration frequency parameter is γ. g() is a function that calculates the basic compensation intensity based on the vibration frequency parameter, while γ is an adjustable sensitivity coefficient that represents the weight of the vibration frequency parameter. When the deviation exceeds the threshold, the system updates this rule, for example, by increasing the value of γ, so that in subsequent calculations, the same vibration frequency parameter will generate a larger vibration compensation intensity, thus more effectively offsetting the vibration displacement. The second scenario is when the system detects that the fog density parameter in the environmental state features also exceeds the preset concentration threshold at the same time as the deviation exceeds the threshold. This strongly indicates that severe distortion of visual information is the main cause of calibration failure. Therefore, the system updates the rule to reduce the weight of image feature data in the dynamic weight allocation parameter. This is achieved by adjusting the calculation rule responsible for generating the image data weights. For example, this rule can be expressed as:

[0074] w = f(D) fog ,β),

[0075] Where w is the weight assigned to the image feature data, and D fog Here, β is the fog density parameter, f() is the function that calculates the weights based on the fog density, and β is an adjustable parameter for this rule. When the above conditions are met, the system will adjust the value of β so that the function f() is applied to the same fog density parameter D. fog This will output a lower image feature data weight w. This ensures that in similar dense fog conditions in the future, the system will rely more on the less affected infrared ranging data. The vibration frequency parameter weight of the dynamic weight allocation parameter is improved through the following quantization algorithm: Triggering condition: Let the actual grasping deviation value be D, and the error threshold be D. th D th The optimal value is 0.2 mm, determined by experimental calibration, when D > D. th When this occurs, the update of the vibration frequency parameter weights is initiated. The update formula is as follows: For the vibration compensation sensitivity coefficient γ, the initial value γ = 0.5, with a range of 0.1 to 2.0, is adjusted according to the following rules:

[0076] γ new =γ old +k γ ·(DD th )

[0077] Where, k γ This is a proportionality coefficient, with 0.2 being the preferred value for engineering projects. It can be fine-tuned according to the equipment model, and γ... new Not exceeding the upper limit value γ max =2.0, to avoid overcompensation; at the same time, when the mist density D inside the can is... fog Exceeding the concentration threshold P th When the value is 0.6, the image feature data weight parameter β is simultaneously reduced. The initial value of β is 1.0, with a range of 0.1 to 1.0. The update formula is as follows:

[0078] β new =β old -k β ·(D fog -P th ),

[0079] Where, k β β is the proportionality constant. new Not lower than the lower limit β min =0.1, preserving the reference value of the basic image.

[0080] Optionally, the step of calculating the deviation value by comparing the actual grasping position with the calibrated coordinates includes:

[0081] The actual grasping position and the calibration coordinates are subjected to three-dimensional spatial vector analysis to obtain the position error vector;

[0082] The position error vector is decoupled by components to obtain the deviation value.

[0083] Specifically, the first step is to perform a three-dimensional spatial vector analysis between the actual grasping position and the calibration coordinates to obtain the position error vector. The system needs to acquire two key coordinate data points. The first is the calibration coordinates, which, after a series of fusion, compensation, and correction processes, ultimately provide the three-dimensional coordinates of the target grasping point to the automated equipment (such as a robotic arm). The second is the actual grasping position, which represents the final spatial position of the robotic arm's end effector when it successfully completes the sample grasping task inside the liquid nitrogen tank. This data is typically fed back by the encoder of the robotic arm's servo system after the grasping action is completed and confirmed as successful. These two coordinates must be represented in the same unified coordinate system. By performing a vector subtraction operation on these two three-dimensional spatial points, a vector pointing from the calibration coordinates to the actual grasping position can be obtained. This vector is the position error vector, which intuitively represents the spatial difference between the prediction and the actual position. This calculation can be expressed as:

[0084] V err =P act -P cal ,

[0085] Among them, V err It is the calculated position error vector. P act It is the three-dimensional coordinate vector of the actual grasping position fed back by the robotic arm. P cal This is the three-dimensional coordinate vector of the calibration coordinates output by the system. The second step is to decouple the position error vector into its components to obtain the deviation value. Component decoupling refers to projecting and decomposing the position error vector onto each coordinate axis of its three-dimensional coordinate system to obtain the independent components of the vector along each axis. If the position error vector V... errThe components are (Δx, Δy, Δz), which represent the independent error magnitudes in the X, Y, and Z directions, respectively. The system then calculates the final deviation value based on these decoupled components. The deviation value is a scalar used to assess the severity of the overall error, typically calculated as the Euclidean norm of the position error vector, i.e., its geometric length in three-dimensional space. This calculation can be expressed as:

[0086] D = ||V err ||,

[0087] Where D is the final deviation value. ||V err || represents the position error vector V err Calculate its norm or modulus. This deviation value D will serve as the direct basis for subsequent comparison with the error threshold.

[0088] Optionally, the step of performing three-dimensional spatial vector analysis on the actual grasping position and the calibration coordinates to obtain the position error vector includes:

[0089] The difference between the actual grasping position and the calibrated coordinates is calculated to obtain the deviation value of each coordinate axis.

[0090] The position error vector is obtained by combining the deviation values ​​of each coordinate axis into a vector.

[0091] Specifically, the first step is to calculate the difference in coordinate components between the actual grasping position and the calibrated coordinates to obtain the deviation values ​​for each coordinate axis. Operationally, the system acquires the three-dimensional coordinate vector P of the actual grasping position. act Its components are represented as (x act ,y act ,z act ), and the three-dimensional coordinate vector P for calibration coordinates. cal Its components are represented as (x cal ,y cal ,z cal Subsequently, the corresponding components of these two vectors are subtracted one by one to obtain the magnitude of the deviation in the three orthogonal directions of X, Y, and Z. These calculated scalar values ​​are the deviation values ​​for each coordinate axis. Then, the deviation values ​​for each coordinate axis are combined to obtain the final position error vector. This step takes the three independent coordinate axis deviation values ​​calculated in the previous step as components of a new vector in a three-dimensional Cartesian coordinate system, thus constructing a complete spatial vector. This vector points from the calibration coordinate point to the actual grasping position point, and its direction and magnitude accurately describe the spatial difference between the two. The entire process can be represented by a unified vector subtraction formula:

[0092] V err =P act -Pcal =(x act -x cal ,y act -y cal ,z act -z cal ),

[0093] Among them, V err This is the final position error vector. P act It is the actual gripping position coordinate fed back from devices such as the robotic arm encoder. P cal These are the target calibration coordinates calculated by this method. (x) act -x cal ),(y act -y cal ), and (z act -z cal These are the calculated deviation values ​​along the X, Y, and Z axes, respectively.

[0094] Optionally, applying the initial coordinate estimate to the correction vector and iteratively correcting it until convergence to generate calibration coordinates includes:

[0095] The initial coordinate estimate is added to the correction vector to obtain the updated coordinate estimate.

[0096] The updated coordinate estimates are repeatedly subjected to vector correction and convergence judgment to generate calibration coordinates.

[0097] Specifically, the system first performs vector addition on the initial coordinate estimate and the correction vector to obtain the updated coordinate estimate. In any step of the iteration, the system treats the current coordinate estimate as a spatial vector, and simultaneously treats the correction vector pointing to the corresponding ideal coordinates in the digital grid model as a spatial vector. By adding these two vectors, a new vector that is spatially closer to the ideal coordinate point is obtained, and its endpoint is the updated coordinate estimate. Mathematically, this operation is a simple vector addition, ensuring that the direction and magnitude of the coordinate correction have clear physical meaning. Then, the vector correction and convergence check are repeatedly performed on the updated coordinate estimate to generate calibration coordinates. This means that the updated coordinate estimate generated in the previous step will be used as the initial coordinate estimate for the next iteration, and the system will recalculate the correction vector and perform vector addition again. This loop repeats continuously, forming an iterative sequence, so that the coordinate estimate approaches its ideal position in the digital grid model in each iteration. This loop is not infinite but is accompanied by a convergence check. After each iteration, the system checks whether the distance between two consecutive coordinate estimates is less than a preset minimum threshold. When this condition is met, it indicates that the coordinate estimate is sufficiently stable and very close to the ideal position, and the iterative process converges and terminates. At this point, the coordinate estimate is formally determined as the final calibration coordinate.

[0098] Based on the same inventive concept, such as Figure 6 As shown, the present invention also provides an adaptive coordinate calibration system for samples inside a liquid nitrogen tank, the system comprising:

[0099] The multi-source environmental sensing module is used to acquire multi-source environmental data inside the liquid nitrogen tank and process the multi-source environmental data to generate multi-source sensor data.

[0100] The dynamic weight generation module is used to generate dynamic weight allocation parameters based on the environmental state characteristics in the multi-source environmental data.

[0101] The coordinate calibration module is used to perform fusion compensation calculations on the multi-source sensor data according to the dynamic weight allocation parameters, and generate calibration coordinates.

[0102] The deviation feedback analysis module is used to obtain the actual grasping position of the sample storage, and calculate the deviation value by comparing the actual grasping position with the calibration coordinates;

[0103] The weight adaptive update module is used to update the weight calculation rules of the dynamic weight allocation parameters according to the deviation value.

[0104] To verify the feasibility of this invention in practice, it was applied to a large-scale automated biobank. This biobank utilizes multiple large liquid nitrogen tanks and associated automated robotic arm systems to ensure the long-term safe storage and efficient automated access of millions of precious biological samples. However, in actual operation, the dense fog generated by liquid nitrogen evaporation and the mechanical vibration from surrounding equipment severely affect the robotic arm's precise positioning of the samples, resulting in a high failure rate and the risk of sample damage. This invention is applied to the coordinate calibration stage of its automated access system to improve its positioning accuracy and stability under complex working conditions. To verify the effectiveness of this invention, a three-month continuous test was conducted on a liquid nitrogen tank automated access system in the third quarter of 2024. During the test, the system fully recorded multi-source sensor data, coordinate calibration process data, and final actual grasping deviation data under different fog concentrations and vibration intensities. In this embodiment, the invention first acquires multi-source environmental data using infrared ranging sensors, industrial cameras, accelerometers, and thermocouple temperature sensors deployed within the liquid nitrogen tank. For example, in a sample retrieval task, the system acquired distance data from the sample rack reference point and image data of feature points, combining them to form the original position data. Simultaneously, an accelerometer collected data on the mechanical vibration of the tank caused by the liquid nitrogen replenishment pump, and a thermocouple measured the real-time temperature inside the tank at -195.8℃. Based on this real-time temperature, the system immediately performed temperature shielding processing on the original position and vibration data, compensated for sensor drift using a preset temperature error model, and generated high-precision multi-source sensor data, providing a reliable foundation for subsequent calculations. In the dynamic weight allocation stage, the system analyzes environmental characteristics in real time. After an opening operation on the morning of August 10, 2024, dense fog quickly formed inside the tank. The system calculated a high fog density parameter by analyzing image clarity. At the same time, the vibration generated by the robotic arm movement was captured by the accelerometer, and the system extracted the main vibration frequency parameters after Fourier transform. Based on these two parameters, the system dynamically generates weight allocation parameters: due to the influence of fog, the weight assigned to image feature data is reduced to 0.3, while the weight assigned to infrared ranging data is increased to 0.7; simultaneously, a moderate intensity of vibration compensation is set according to the vibration frequency. In the coordinate fusion compensation calculation, the system first performs weighted fusion of the temperature-shielded infrared ranging data and image feature data to obtain preliminary fused coordinates. Subsequently, using the vibration compensation data, displacement compensation is performed on the preliminary fused coordinates according to the set compensation intensity to generate dynamically compensated coordinates. Finally, the system uses this coordinate as an initial estimate and compares it with the ideal coordinates in the pre-stored digital grid model.For example, the dynamically compensated coordinates are (X: 150.4mm, Y: 320.7mm, Z: -550.9mm), and their closest ideal coordinates in the digital grid model are (X: 150.0mm, Y: 321.0mm, Z: -551.0mm). The system uses an iterative correction algorithm to gradually converge the coordinate estimates, ultimately generating highly accurate calibration coordinates (X: 150.1mm, Y: 320.9mm, Z: -551.0mm). Closed-loop feedback and self-learning mechanisms are the core advantages of this invention. In a grasping operation on September 5, 2024, the system's calibration coordinates deviated from the actual grasping position fed back by the robotic arm by more than a preset error threshold. System analysis revealed that both the fog density parameter and vibration frequency parameter were at high levels when this deviation occurred. Based on preset rules, the system determined that this error was mainly caused by high-frequency vibration, and therefore automatically updated the weight calculation rules, increasing the weight coefficient γ of the vibration frequency parameter when generating vibration compensation intensity. In subsequent tests under similar operating conditions, the system's calibration accuracy was significantly improved, with the deviation reduced by approximately 40%. Data comparison shows that the method of this invention has significant advantages in calibration accuracy and environmental adaptability. In dense fog, traditional methods can result in calibration deviations exceeding 5mm, while this invention, by dynamically reducing image weights, controls the deviation to within 2mm. Under strong vibration conditions, the vibration compensation mechanism of this method can reduce the positioning error caused by vibration from approximately 3mm to below 1mm. After one month of adaptive learning optimization, the system's average grasping success rate increased from 95% to over 99.5%.

[0105] Table 1. Multi-source sensor data and environmental adaptive fusion data.

[0106]

[0107] Table 2. Coordinate calibration and iterative correction data table

[0108]

[0109] Table 3 Comparison of Adaptive Update Effects of Weight Rules

[0110]

[0111]

[0112] As can be seen from the data in Tables 1 to 3 above, this invention demonstrates excellent adaptability and high precision in practical applications. Table 1 clearly shows how the system dynamically adjusts the fusion weights of image and infrared ranging data based on real-time changes in fog density and vibration frequency. For example, at 10:31:05, when the fog density reaches 0.78, the image data weight is significantly reduced to 0.3, ensuring that the calibration results are not excessively affected by low-quality images. The data in Table 2 reveals the step-by-step optimization of the coordinate calibration process of this invention. Through vibration compensation and iterative correction, the final calibration coordinates are very close to the ideal physical coordinates, with the deviation from the actual grasping position generally controlled within 0.3mm, showing extremely high absolute accuracy. In the high-vibration scenario at 10:31:40, although the initial deviation is large, the system can still generate high-precision coordinates through correction. Table 3 strongly demonstrates the effectiveness of the closed-loop feedback and self-learning mechanism. After a large deviation occurred on September 5th, the system updated the weight calculation rules. In tests conducted under similar conditions on September 12, the deviation value significantly decreased from 0.71mm to 0.42mm, representing an accuracy improvement of over 40%. This demonstrates that the present invention can not only cope with single environmental changes but also learn from historical data to continuously optimize its model, thereby maintaining stable and reliable performance in long-term operation and greatly improving the operational efficiency and security of automated biobanks.

[0113] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0114] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for adaptive coordinate calibration of samples inside a liquid nitrogen tank, characterized in that, The method includes: Acquire multi-source environmental data inside the liquid nitrogen tank and process the multi-source environmental data to generate multi-source sensor data, including: acquiring infrared ranging data and image feature data, which are combined to form original position data; acquiring tank mechanical vibration data and real-time temperature inside the tank; and performing temperature shielding processing on the original position data and tank mechanical vibration data based on the real-time temperature inside the tank to generate multi-source sensor data. Based on the environmental state characteristics in the multi-source environmental data, dynamic weight allocation parameters are generated, including: extracting fog density parameters and vibration frequency parameters from the multi-source environmental data; and combining the fog density parameters and vibration frequency parameters as environmental state characteristics to generate the dynamic weight allocation parameters. The multi-source sensor data is fused and compensated according to the dynamic weight allocation parameters to generate calibration coordinates. This includes: weighting and fusing the infrared ranging data and image feature data according to the sensor weight allocation ratio in the dynamic weight allocation parameters to generate preliminary fused coordinates; compensating the preliminary fused coordinates for vibration displacement using vibration compensation data from the multi-source sensor data according to the vibration compensation intensity in the dynamic weight allocation parameters to generate dynamically compensated coordinates; and iteratively correcting the dynamically compensated coordinates to generate calibration coordinates. Obtain the actual grab position of the sample storage, and calculate the deviation value by comparing the actual grab position with the calibration coordinates; Based on the deviation value, update the weight calculation rules of the dynamic weight allocation parameters.

2. The adaptive coordinate calibration method for samples inside a liquid nitrogen tank according to claim 1, characterized in that, The step of iteratively correcting the dynamically compensated coordinates to generate calibration coordinates includes: Obtain a digital grid model that defines the ideal coordinates of the sample storage location; The dynamically compensated coordinates are used as the initial coordinate estimates and compared with the ideal coordinates in the digital grid model to calculate the correction vector. The initial coordinate estimate is applied to the correction vector, and the correction is iteratively performed until convergence, generating the calibration coordinates.

3. The adaptive coordinate calibration method for samples inside a liquid nitrogen tank according to claim 1, characterized in that, The weight calculation rule for updating the dynamic weight allocation parameters based on the deviation value includes: Obtain the error threshold and concentration threshold; When the deviation value exceeds the error threshold, the weight of the vibration frequency parameter of the dynamic weight allocation parameter is increased; When the fog density parameter in the environmental state features exceeds the concentration threshold, the image feature data weight of the dynamic weight allocation parameter is reduced.

4. The adaptive coordinate calibration method for samples inside a liquid nitrogen tank according to claim 1, characterized in that, The step of calculating the deviation value by comparing the actual grasping position with the calibrated coordinates includes: The actual grasping position and the calibration coordinates are subjected to three-dimensional spatial vector analysis to obtain the position error vector; The position error vector is decoupled by components to obtain the deviation value.

5. The adaptive coordinate calibration method for samples inside a liquid nitrogen tank according to claim 4, characterized in that, The step of performing three-dimensional spatial vector analysis on the actual grasping position and the calibration coordinates to obtain the position error vector includes: The difference between the actual grasping position and the calibrated coordinates is calculated to obtain the deviation value of each coordinate axis. The position error vector is obtained by combining the deviation values ​​of each coordinate axis into a vector.

6. The adaptive coordinate calibration method for samples inside a liquid nitrogen tank according to claim 2, characterized in that, The step of applying the initial coordinate estimate to the correction vector, performing iterative correction until convergence, and generating calibration coordinates includes: The initial coordinate estimate is added to the correction vector to obtain the updated coordinate estimate. The updated coordinate estimates are repeatedly subjected to vector correction and convergence judgment to generate calibration coordinates.

7. An adaptive coordinate calibration system for samples inside a liquid nitrogen tank, applied to the adaptive coordinate calibration method for samples inside a liquid nitrogen tank as described in any one of claims 1-6, characterized in that, The system includes: A multi-source environmental sensing module is used to acquire multi-source environmental data inside the liquid nitrogen tank and process the multi-source environmental data to generate multi-source sensor data, including: acquiring infrared ranging data and image feature data, which are combined to form original position data; acquiring tank mechanical vibration data and real-time temperature inside the tank; and performing temperature shielding processing on the original position data and tank mechanical vibration data based on the real-time temperature inside the tank to generate multi-source sensor data. The dynamic weight generation module is used to generate dynamic weight allocation parameters based on the environmental state features in the multi-source environmental data, including: extracting fog density parameters and vibration frequency parameters from the multi-source environmental data; and combining the fog density parameters and vibration frequency parameters as environmental state features to generate the dynamic weight allocation parameters. The coordinate calibration module is used to perform fusion compensation calculations on the multi-source sensor data according to the dynamic weight allocation parameters to generate calibration coordinates. This includes: weighting and fusing the infrared ranging data and image feature data according to the sensor weight allocation ratio in the dynamic weight allocation parameters to generate preliminary fused coordinates; performing vibration displacement compensation on the preliminary fused coordinates using vibration compensation data from the multi-source sensor data according to the vibration compensation intensity in the dynamic weight allocation parameters to generate dynamically compensated coordinates; and iteratively correcting the dynamically compensated coordinates to generate calibration coordinates. The deviation feedback analysis module is used to obtain the actual grasping position of the sample storage, and calculate the deviation value by comparing the actual grasping position with the calibration coordinates; The weight adaptive update module is used to update the weight calculation rules of the dynamic weight allocation parameters according to the deviation value.

Images