Sensor batch calibration method and system based on multi-vision fusion

Abstract

The application belongs to the technical field of sensor calibration, and provides a sensor batch calibration method and system based on multi-vision fusion. First, all-around vision information of the sensor and the detection table is acquired. Pose estimation is performed by fusing the all-around vision information. Then, on the basis of the pose estimation, the sensor barycenter position, the grabbing point intensity and the avoidance of sensitive areas are considered, the grabbing point selection is optimized, the impedance control algorithm is used for flexible grabbing, the grabbing force size is automatically adjusted according to the sensor material and the surface friction coefficient, and the sensor lofting is realized. The problem of sensor deformation or damage caused by excessive grabbing force is avoided, the calibration demand of different specifications of sensors is realized, and the batch sensor calibration precision and efficiency are improved.

Description

Technical Field

[0001] This invention belongs to the field of sensor calibration technology, and in particular relates to a method and system for batch calibration of sensors based on multi-vision fusion. Background Technology

[0002] As core sensing components in modern industrial production, intelligent manufacturing, and the Internet of Things (IoT) systems, the measurement accuracy of sensors directly affects the reliability and accuracy of the entire system. With the development of intelligent agriculture, the demand for soil parameter sensors has increased dramatically. These sensors must undergo precise calibration during mass production to ensure measurement accuracy.

[0003] Existing automated calibration systems mostly employ fixed workstations and fixtures, lacking flexible control strategies and unable to adapt to the calibration needs of sensors of different specifications. Robots only perform simple loading and unloading actions, lacking intelligent perception and adaptive capabilities. Especially when the gripping force is too large, it can easily cause sensor deformation or damage, affecting the accuracy and efficiency of batch sensor calibration. Furthermore, most calibration systems lack or only have simple vision systems, failing to achieve automatic sensor model identification, precise positioning, and attitude adjustment. They rely on manual intervention or fixed fixture positioning, resulting in weak visual perception capabilities. They only use simple linear correction, with insufficient compensation for nonlinear errors and a lack of intelligent optimization based on historical data, making it difficult to continuously improve calibration parameters and limiting calibration accuracy. The calibration systems are sensitive to environmental changes and lack adaptive compensation mechanisms, leading to poor consistency of calibration results and poor environmental adaptability, further impacting the accuracy of batch sensor calibration and the efficiency of intelligent calibration. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a method and system for batch sensor calibration based on multi-vision fusion. This invention performs pose estimation by fusing omnidirectional visual information. Based on this, it optimizes the selection of gripping points by considering factors such as the sensor's center of gravity, gripping point strength, and avoidance of sensitive areas. An impedance control algorithm is employed for flexible gripping, and the gripping force is automatically adjusted according to the sensor material and surface friction coefficient to achieve sensor placement. This avoids the problem of sensor deformation or damage caused by excessive gripping force, fulfills the calibration requirements of sensors of different specifications, and improves the accuracy and efficiency of batch sensor calibration.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: In a first aspect, the present invention provides a method for batch calibration of sensors based on multi-vision fusion, comprising: Acquire comprehensive visual information from sensors and the detection station; Pose estimation is performed by fusing comprehensive visual information. Based on pose estimation, the selection of gripping points is optimized by considering the sensor's center of gravity position, gripping point strength, and avoiding sensitive areas. An impedance control algorithm is used for flexible gripping, and the gripping force is automatically adjusted according to the sensor material and surface friction coefficient to achieve sensor lofting. Select the corresponding algorithm from the preset algorithm library for calibration.

[0006] Furthermore, the omnidirectional visual information includes the sensor's position and orientation in three-dimensional space acquired through a binocular vision system, the sensor model, QR code, and sensor surface identified by an industrial camera, and the sensor's fit to the test stage surface and the alignment of the sensor electrodes with the test stage pins detected by orthogonal dual cameras.

[0007] Furthermore, the omnidirectional visual information fusion employs a Kalman filter algorithm, and the Kalman filter state update formula is: ; ; in, For posterior state estimation; For prior state estimation; Kalman gain; These are the observed values; The observation matrix; Let be the posterior error covariance.

[0008] Furthermore, the impedance control algorithm is as follows: ; in, Joint torque; It is a Jacobian matrix; For the power of expectation; For positional stiffness; This is for positional deviation; For velocity damping; For speed.

[0009] Furthermore, when optimizing the selection of the gripping point, let the gripping posture parameters be... G The optimization objective is to minimize the following comprehensive cost function: ; ; ; ; in, The overall cost of capturing the posture; The center of gravity balance function; To capture the coordinates of the point; The coordinates of the centroid; This is the maximum allowed distance; For the capture point intensity function; The coefficient of friction; For normal grasping force; This is a sensitive area avoidance function; This is the minimum distance from the capture point to the sensitive area; For scale parameters; , and These are the weighting coefficients.

[0010] Furthermore, when automatically adjusting the gripping force based on the sensor material and surface friction coefficient, the gripping force should meet the following requirements: ; ; ; in: Minimum gripping force to prevent slipping; For sensor quality; It is the acceleration due to gravity; The inertial force caused by dynamic acceleration; The static friction coefficient between the sensor surface and the gripper material; Maximum gripping force for material safety; The yield strength of the sensor housing material; This represents the contact area between the gripper and the sensor. This is for the safety factor.

[0011] Furthermore, dynamic weight adjustment is introduced: ; in: This is the safety margin factor; Safety is a priority factor.

[0012] Secondly, the present invention also provides a batch calibration system for sensors based on multi-vision fusion, comprising: The data acquisition module is configured to acquire omnidirectional visual information from the sensors and the detection station; The pose estimation module is configured to perform pose estimation by fusing omnidirectional visual information. The lofting module is configured to: based on pose estimation, consider factors such as the sensor's center of gravity position, gripping point strength, and avoidance of sensitive areas, optimize the selection of gripping points, use an impedance control algorithm for flexible gripping, and automatically adjust the gripping force according to the sensor material and surface friction coefficient to achieve sensor lofting; The calibration module is configured to select the corresponding algorithm from a preset algorithm library for calibration.

[0013] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-vision fusion-based sensor batch calibration method described in the first aspect.

[0014] Fourthly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the steps of the multi-vision fusion-based sensor batch calibration method described in the first aspect.

[0015] Fifthly, the present invention also provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the steps of the multi-vision fusion-based sensor batch calibration method described in the first aspect.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention estimates pose by fusing omnidirectional visual information. Based on this, it optimizes the selection of gripping points by considering factors such as the sensor's center of gravity, gripping point strength, and avoiding sensitive areas. It employs an impedance control algorithm for flexible gripping and automatically adjusts the gripping force according to the sensor material and surface friction coefficient to achieve sensor placement. This avoids the problem of sensor deformation or damage caused by excessive gripping force, meets the calibration requirements of sensors of different specifications, and improves the calibration accuracy and efficiency of batch sensors.

[0017] 2. This invention takes into account factors such as the sensor's center of gravity position, gripping point strength, and avoidance of sensitive areas, and optimizes the selection of gripping points to ensure torque balance during gripping, avoid tilting or slipping, and achieve avoidance of sensitive areas such as electrodes and vulnerable parts, so as to avoid damage caused by gripping or affect calibration accuracy. Attached Figure Description

[0018] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0019] Figure 1 This is a flowchart of the method in Embodiment 1 of the present invention. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0022] Example 1: As core sensing components in modern industrial production, intelligent manufacturing, and the Internet of Things (IoT) systems, sensors directly impact the reliability and accuracy of the entire system through their measurement accuracy. With the development of intelligent agriculture, the demand for soil parameter sensors (such as those for soil moisture, temperature, nutrients, and pH) has increased dramatically. These sensors must undergo precise calibration during mass production to ensure measurement accuracy.

[0023] Traditional calibration methods rely on manual, single-piece approaches, which are cumbersome, inefficient, and unsuitable for mass production. Manual loading, positioning, testing, and recording are time-consuming, with a typical calibration cycle of 10-15 minutes per unit, resulting in low efficiency. Existing automated calibration systems often use fixed workstations and fixtures, lacking flexibility and unable to adapt to the calibration needs of sensors of different specifications. Robots only perform simple loading and unloading actions, lacking intelligent perception and adaptive capabilities, resulting in insufficient automation. Most calibration systems lack or have only simple vision systems, failing to achieve automatic sensor model identification, precise positioning, and attitude adjustment, relying on manual intervention or fixed fixtures for positioning, resulting in weak visual perception capabilities. Simple linear correction is used, with insufficient compensation for nonlinear errors; the lack of intelligent optimization based on historical data makes it difficult to continuously improve calibration parameters, limiting calibration accuracy. Calibration systems are sensitive to environmental changes (temperature, humidity, light, etc.) and lack adaptive compensation mechanisms, leading to poor consistency and environmental adaptability in calibration results. Lacking embodied intelligence, it cannot autonomously adjust its calibration strategy based on visual perception results, nor can it learn and optimize algorithm parameters from historical calibration data, resulting in a low level of intelligence.

[0024] To address at least one of the aforementioned problems, this embodiment provides a batch sensor calibration method based on multi-vision fusion, which solves the technical problems of low efficiency, insufficient automation, low intelligence, and poor environmental adaptability in existing sensor calibration technologies. Figure 1 As shown, the method in this embodiment includes: S1. Multi-vision collaborative perception and sensor recognition: A multi-channel vision system is used to acquire comprehensive information from sensors and the inspection station. The inspection station and robot can be implemented using conventional techniques.

[0025] S1.1, Robot end-effector binocular vision system: A binocular camera is mounted on the end effector of an industrial robot, moving with the robot to acquire the precise position and orientation of the sensor in three-dimensional space using stereo vision technology. This allows for real-time determination of the sensor's pose matrix relative to the robot's end effector coordinate system, providing precise guidance for grasping. The baseline distance of the binocular camera is also considered. ,focal length Field of view Working distance range .

[0026] S1.2, Robot end-effector industrial camera system: A high-resolution industrial camera (5 megapixels, 30fps) is mounted on the robot's end effector to identify sensor models, read QR codes / barcodes, and detect surface defects on sensors. The industrial camera supports OCR (Optical Character Recognition) to recognize text markings on sensors, automatically obtaining information such as sensor model, batch number, and specifications; it can be used with a ring LED light source to ensure stable imaging under different lighting conditions.

[0027] S1.3, Orthogonal Dual-Camera System for Inspection Platform: Two industrial cameras are placed orthogonally on either side of the inspection stage, located along the X-axis and Y-axis respectively, to monitor the sensor's placement and electrode contact on the stage in real time. Optionally, the X-axis camera detects whether the sensor is fully in contact with the stage surface, and the Y-axis camera detects the alignment of the sensor electrodes with the stage pins. The camera resolution is optionally 2 megapixels, and the working distance is optional. The field of view can be selected as .

[0028] S1.4, Multi-visual information fusion: The Kalman filter algorithm is used to fuse multiple visual data streams to improve pose estimation accuracy. The Kalman filter state update formula is as follows: ; ; in, For posterior state estimation; For prior state estimation; Kalman gain; These are the observed values; The observation matrix; Let be the posterior error covariance.

[0029] Establish a unified transformation relationship between the world coordinate system, robot base coordinate system, camera coordinate system, and inspection table coordinate system. Obtain the transformation matrix between each coordinate system through visual calibration to achieve accurate registration of the coordinate systems.

[0030] S2, Embodied Intelligent Decision-Making and Adaptive Control: Based on visual perception results and historical calibration data, intelligent decision-making and adaptive control are performed. The decision-making process is as follows: S2.1 Adaptive adjustment of grasping posture: Based on the 3D pose of the sensors acquired through binocular vision, the robot calculates the optimal grasping trajectory in real time. Considering factors such as sensor center of gravity position, grasping point strength, and avoiding sensitive areas, the selection of grasping points is optimized. An impedance control algorithm is employed to achieve flexible grasping and avoid sensor damage. The impedance control equation is as follows: ; in, Joint torque; It is a Jacobian matrix; For the power of expectation; For positional stiffness; This is for positional deviation; For velocity damping; For speed.

[0031] When optimizing the selection of gripping points, let the gripping posture parameters be: G The optimization objective is to minimize the following comprehensive cost function: ; in, The smaller the value, the better the gripping point selection; The center of gravity balance function reflects the degree of matching between the gripping point and the sensor's center of gravity; it can be defined as the normalized value of the Euclidean distance between the gripping point and the center of gravity. ,in, To capture the coordinates of the point, Using the coordinates of the centroid, This is the maximum allowed distance; The gripping point strength function reflects the structural strength or gripping stability of the gripping point. It can be modeled based on factors such as the material strength and surface friction coefficient at the gripping point, for example: ,in, The coefficient of friction, For normal grasping force; The sensitive area avoidance function reflects the distance between the gripping point and the sensor's sensitive area (such as electrodes or vulnerable parts). It can be defined as the negative exponential form of the distance from the gripping point to the nearest edge of the sensitive area: ,in, To capture the minimum distance from the point to the sensitive area, For scale parameters; , and These are weighting coefficients, used to adjust the importance of center of gravity position, gripping point strength, and sensitive area avoidance in the optimization process. These coefficients can be adaptively adjusted according to sensor type, material, and calibration requirements.

[0032] In this embodiment, factors such as the sensor's center of gravity position, gripping point strength, and avoidance of sensitive areas are considered to optimize the selection of gripping points, ensuring torque balance during gripping, avoiding tilting or slipping, and avoiding sensitive areas such as electrodes and vulnerable parts, thus preventing damage or impact on calibration accuracy caused by gripping.

[0033] The gripping force is adaptively adjusted based on the sensor material and surface friction coefficient, automatically adjusting the gripping force magnitude (range). Optionally, to ensure that the sensor does not slip (static friction failure) or become crushed (material yielding failure) during the gripping process, the gripping force should meet the following requirements: ; in: To determine the minimum gripping force required to prevent slippage and overcome the sliding tendency caused by gravity and inertia, the minimum normal gripping force is determined by static friction equilibrium. ; in, For sensor quality; It is the acceleration due to gravity; For inertial force caused by dynamic acceleration, such as when a robotic arm accelerates, it can be set to 0.1~0.3×m·g; The static friction coefficient between the sensor surface and the gripper material is given by various parameters, such as: plastic–rubber μ≈0.6~0.8, metal–silicone μ≈0.4~0.7.

[0034] To ensure material safety and prevent deformation, cracking, or damage to sensitive structures (such as electrodes and circuit boards) due to excessive clamping force, a maximum permissible normal force is set: ; in, The yield strength of the sensor housing material; This represents the contact area between the gripper and the sensor. For safety factors, a value of 0.3 to 0.7 is usually used, with the lower limit used for brittle materials, such as ceramics k=0.3 and tough plastics k=0.6.

[0035] Based on the material identification results, the system calls upon the materials mechanics database to calculate... To ensure a firm grip and prevent damage.

[0036] To avoid extreme values ​​(such as...) μ Extremely low (Too large), introduce dynamic weight correction: ; in: For safety margin factor, =0.1~0.3, to deal with surface oil stains and vibration disturbance; As a safety priority factor, .

[0037] S2.2 Automatic selection of calibration scheme: The system has a built-in library of various calibration algorithms, including linear regression, polynomial fitting, BP neural network, and Gaussian process regression algorithms.

[0038] Based on the sensor model, specifications, and historical calibration data, the system automatically selects the optimal calibration scheme; for new sensors, the system automatically performs preliminary calibration, evaluates the performance of each algorithm, and selects the best algorithm.

[0039] Optionally, an algorithm can be selected based on the nonlinearity and sensor model, if the nonlinearity... Linear regression is chosen because it has small errors, fast computation, and strong interpretability, making it suitable for highly linear sensors, such as some temperature sensors; however, if the nonlinearity is... If the sample size is greater than 50, then the polynomial fitting algorithm is selected, which can effectively compensate for mild nonlinearity and avoid overfitting, making it suitable for most NPK or humidity sensors; if the nonlinearity is... If the sample size is between 20 and 50, then a backpropagation (BP) neural network is chosen because it can model strong nonlinear relationships, has strong generalization ability, and is suitable for environmentally sensitive sensors; if the nonlinearity is... If the sample size is less than 20, the Gaussian process regression algorithm is selected to provide uncertainty estimation, suitable for small-sample, high-precision calibration scenarios, such as research-grade soil pH sensors. Nonlinearity Calculations should be based on actual calibration experimental data: ; in, The sensor at the input value The actual output can be obtained by averaging multiple measurements; To find the ideal straight line fitted by the endpoint method, connect the minimum and maximum inputs to find the corresponding outputs. This represents the full-scale output range of the sensor.

[0040] S2.3 Automatic compensation for the position of the testing station: An orthogonal camera monitors the sensor's positional deviation on the detection stage in real time and determines the positional deviation. , and (Angle deviation); Automatic position compensation is achieved by fine-tuning the detection platform or adjusting the robot's placement, with a position compensation accuracy of [missing information]. The angle compensation accuracy is A PID control algorithm is used for position compensation. ; in, To control the output; This refers to the positional error; , , These are PID parameters.

[0041] S2.4 Environmental Adaptive Compensation: Real-time monitoring of environmental parameters such as temperature, humidity, and light intensity; establishment of a model to assess the impact of environmental parameters on calibration results. ; in, For temperature change; For humidity changes; For changes in light intensity; This is the influence coefficient.

[0042] The calibration parameters are automatically adjusted based on environmental parameters to compensate for environmental impacts. The temperature compensation range is [range missing]. The compensation accuracy is .

[0043] S3. Self-learning of algorithm parameters based on historical data: Establish a calibration history database to continuously optimize algorithm parameters, including: S3.1 Establishment of historical calibration database: The database stores the raw data, calibration parameters, verification results, and environmental parameters for each calibration. It contains no fewer than 100,000 calibration records and can be stored using a time-series database, supporting efficient querying and analysis.

[0044] S3.2 Adaptive optimization of algorithm parameters: A reinforcement learning approach is employed, using calibration accuracy and efficiency as reward functions to optimize algorithm parameters. The reward function is defined as follows: ; in, As a reward value; This is for calibration error; For calibration time; Maximum allowable calibration time; A success indicator function; Weighting coefficients ( ).

[0045] The Bayesian optimization algorithm is used to efficiently search for the optimal parameter combination. The Bayesian optimization acquisition function is as follows: ; in, For acquisition functions (such as Expected Improvement); For historical datasets; The parameter vector is used; the parameter update strategy is as follows: ; in, These are algorithm parameters; The learning rate; These are the updated algorithm parameters; These are the algorithm parameters before the update.

[0046] S3.3, Dynamic adjustment of the calibration model: Based on statistical analysis of historical calibration data, sensor performance drift trends are identified. When a systematic deviation in calibration error is detected, the calibration model is automatically updated. Optionally, an online learning algorithm is used to update model parameters in real time. When the average calibration error of 100 consecutive sensors exceeds a set threshold, the sensor model changes, and / or environmental conditions change significantly, the calibration model is automatically updated.

[0047] S4. Automated calibration execution: S4.1 Automatic feeding: The industrial robot picks up the sensor to be calibrated from the tray. Binocular vision guides the robot in real time to accurately align with the sensor's gripping point. The gripping force is adaptively adjusted to ensure stable gripping without damaging the sensor.

[0048] S4.2 Precise Placement: The robot, carrying sensors, moves above the inspection platform. The platform's orthogonal camera provides real-time feedback on the sensor's pose relative to the platform. Based on this visual feedback, the robot makes fine adjustments to ensure accurate sensor placement. The placement accuracy is [insert accuracy here]. , angle is .

[0049] S4.3 Electrode contact verification: The orthogonal camera detects the contact between the sensor electrodes and the detection stage pins. Once a complete fit is confirmed, the electrical connection is automatically initiated. If poor contact is detected, the position is automatically fine-tuned.

[0050] S4.4 Standard Quantity Application and Data Acquisition: Standard excitation signals (temperature, humidity, nutrient concentration, etc.) are automatically applied through a standard sample library, enabling high-speed data acquisition via multiple channels with a sampling rate of no less than [missing information]. The collected data is transmitted to the data processing system in real time, and abnormal data points are automatically removed. Optionally, the following methods can be used: Criteria: Data Points satisfy Then it is considered an outlier.

[0051] in, This represents the data mean. The standard deviation is denoted as .

[0052] S4.5 Calibration parameter calculation: Based on the collected data, the selected calibration algorithm is used to calculate the calibration parameters, supporting online parameter optimization to ensure calibration accuracy.

[0053] S4.6 Parameter Writing and Verification: The calibration parameters are written to the sensor's internal storage unit (EEPROM) and protected by encryption using the AES-128 encryption algorithm. The parameters are read and verified immediately after writing to ensure that they are written correctly. If the verification fails, the parameters are rewritten, with a maximum of 3 retries.

[0054] S4.7, Material cutting and marking: The robot picks up the calibrated sensors and places them on the unloading tray. The laser marking machine marks the calibration date and verification results on the sensor housing. Qualified sensors and unqualified sensors are stored separately.

[0055] S5. Calibration Quality Traceability and Analysis: S5.1 Calibration report is automatically generated: Each sensor generates a detailed calibration report, including: raw data, calibration parameters, verification results, environmental parameters, operator information, etc.; the report is stored in PDF format and supports QR code scanning for querying.

[0056] S5.2 Quality Statistical Analysis: The system provides real-time statistics on key indicators such as calibration pass rate, calibration error distribution, and calibration efficiency. It employs the SPC (Statistical Process Control) method to monitor the quality of the calibration process. When an abnormality is detected, it automatically alarms and indicates possible causes.

[0057] S5.3 Data Traceability: It supports tracing the complete calibration history through sensor serial numbers and allows querying calibration data by time, batch, model, and other conditions; providing data support for quality traceability and problem analysis.

[0058] The fully automated calibration process described in this embodiment eliminates the need for manual intervention, thus improving calibration efficiency. More than double. The calibration time for a single sensor can be reduced to 2-3 minutes, with 24-hour continuous operation capability. Employing multi-vision fusion for precise positioning, placement accuracy is significantly improved. Advanced algorithms such as neural networks are employed to improve the accuracy of nonlinear error compensation. The above indicates that the standard deviation of the calibration error has decreased. The robot's end effector uses binoculars and an industrial camera to achieve precise 3D spatial perception; the inspection platform has orthogonal dual cameras to accurately monitor placement and electrode contact; and multi-channel visual information fusion improves pose estimation accuracy. It supports calibration of various types of sensors (soil moisture, temperature, nutrients, pH, etc.) and different sensor sizes (size range). It automatically adapts to different calibration schemes without requiring hardware replacement. Complete calibration history records are available, each sensor is traceable, and real-time quality statistical analysis enables timely detection of quality issues. Continuous optimization based on historical data improves overall quality levels.

[0059] In some embodiments, a linear regression algorithm is employed for... calibration point ,in This is the standard value. This is the sensor output. A straight line is fitted using the least squares method with regularization: The objective function is: ; The formula for calculating the coefficient is: ; ; in, This is a regularization parameter to prevent overfitting.

[0060] Using a polynomial fitting algorithm, for The order polynomial is: ; Construct the matrix equation: .in: ; Solve using regularized least squares: ; in, It is the identity matrix. This is the regularization parameter.

[0061] The BP neural network algorithm is used, including the input layer ( Hidden layer () ) and output layer ( Forward propagation: ; ; ; ; in, , This is the weight matrix; , It is the bias vector; For the activation function (ReLU: ); The output activation function is linear.

[0062] The error function (mean square error) is: ; Backpropagation is: ; ; ; ; ; ; in, The target value; The learning rate; Momentum factor; Element-wise multiplication; parameter update: ; ; The Levenberg-Marquardt algorithm is used for neural network training, and the parameter update formula is as follows: ; in, These are the network parameters (weights and biases); It is a Jacobian matrix; This is the error vector; is the damping factor.

[0063] When using an adaptive strategy, if the new error is less than the old error, then μ = μ / β, where β > 1, which is usually taken as 1; otherwise, μ = μβ.

[0064] An environmental adaptive compensation algorithm is used to establish a model of the influence of environmental parameters on the calibration results: The linear compensation model is as follows: ; The secondary compensation model is as follows: ; Using historical calibration data, model parameters are determined through regression analysis or neural network training, and compensation applications are implemented. ; The objective function of the grasping posture optimization algorithm in the embodied intelligent decision-making algorithm is: ; in, To capture the gripping posture parameters (grip point position, gripping angle, gripping force); This is a stability function (degree of balance of the center of gravity); For security functions (avoiding sensitive areas); The efficiency function (shortest fetching time); , , These are the weighting coefficients.

[0065] The position compensation control algorithm uses PID control: ; in, This represents the position error (target position - actual position). For control output (robot movement commands); These are the PID parameters (adaptively adjusted based on sensor type). Feedforward compensation is: ; ; in, This is a feedforward compensation term used for rapid response to large errors.

[0066] In reinforcement learning algorithms, the state space is: ; The motion space is: ; The reward function is: ; in, This is for calibration error; For calibration time; Maximum allowable calibration time; A success indicator function; These are the weighting coefficients.

[0067] Using either Deep Q-Network (DQN) or Policy Gradient (PPO) algorithms, the DQN loss function is: ; in, For Q-networks; For network parameters; For target network parameters; This is the discount factor.

[0068] During traceability, forward traceability includes scanning the sensor's QR code, querying the database to obtain complete calibration records, and displaying calibration parameters, verification results, environmental parameters, etc. Reverse traceability includes querying calibration records by time, batch, and model, statistically analyzing calibration quality, identifying quality problems, and tracing related sensors.

[0069] When quality traceability is performed, if an abnormality is found in a certain batch of sensors, the calibration records of all sensors in that batch are queried to analyze the cause of the abnormality and trace the scope of impact.

[0070] During quality statistical analysis, calibration pass rates are statistically analyzed by model, batch, and time to identify trends in quality problems; calibration error distribution analysis is performed, an error distribution histogram is plotted, and the mean, standard deviation, and extreme values ​​are calculated to identify systematic errors. The error statistics are as follows: ; ; ; SPC (Statistical Process Control) is used to draw control charts (X-bar charts, R charts) to monitor and calibrate process stability and promptly detect process anomalies. The X-bar control limits are: ; ; in, The mean of the sample; The mean of the sample standard deviation; For control limit coefficients.

[0071] Example 2: Based on Example 1, this embodiment focuses on the mass production and calibration of soil moisture sensors used in agricultural intelligent monitoring systems. The sensor type is a capacitive soil moisture sensor, with dimensions of [missing information]. The electrode interface is Pin pins, spacing The measurement range is Volumetric moisture content, accuracy requirement: .

[0072] Optionally, the baseline of the robot's end-effector binocular camera is... Working distance is The robot's end-effector industrial camera is 10,000 pixels, equipped with a ring LED light source; the detection table has two orthogonal cameras. 10 megapixels, working distance A six-axis industrial robot is used, with a working radius of [missing information]. The load is The repeatability accuracy is The flexible gripper uses a three-finger gripper, adaptable to... Size. The standard soil sample library is set to have moisture contents of [specific values]. , , , , and Standard soil samples, the testing station uses Workstation, temperature controlled at The data acquisition system is Bit resolution Sampling rate. The calibration process includes: S1. Sensor identification and model confirmation: The robot picks up a sensor from the tray, the industrial camera at the robot's end reads the QR code on the sensor, the OCR identifies the sensor model, batch and production date, and queries the database to obtain the calibration specifications for that sensor model.

[0073] S2. Adaptive adjustment of grasping posture: The binocular vision system acquires the sensor's 3D pose, calculates the sensor's center of gravity and the optimal grasping point, adjusts the robot's grasping posture to ensure stable grasping, and sets the grasping force to... .

[0074] S3. Place precisely on the testing station: The robot, carrying sensors, moves above the inspection platform. An orthogonal camera on the platform monitors the sensor's position in real time. The robot makes fine adjustments based on visual feedback. The sensor's placement accuracy at position and angle are as follows: ,and .

[0075] S4. Electrode contact verification: The camera detection sensor pins in the Y-axis direction are aligned with the electrodes of the detection stage. Once a perfect fit is confirmed, the electrical connection is automatically initiated. If a deviation is detected, automatic position fine-tuning is performed (accuracy). ).

[0076] S5. Automatic selection of calibration scheme: System analysis of sensor type (capacitive), range ( ), nonlinearity (historical data approximately ), Automatic selection of polynomial fitting algorithm ( (Step) is used as the calibration scheme.

[0077] S6. Automatically apply standard values ​​and collect data: The standard soil sample library provides samples sequentially. Each moisture content sample was stable. After a few seconds, the sensor outputs data, at a sampling frequency of [frequency missing]. Data collected at each calibration point Each data point automatically removes outlier data. (Guidelines). S7. Calibration parameter calculation: use Polynomial fitting algorithm of order: ; in, y represents the actual moisture content; x The sensor output voltage; the polynomial coefficients are calculated using the least squares method. , and .for n calibration point Construct the matrix equation: ; in: ; Solve using regularized least squares: ; in, It is the identity matrix; Regularization parameter; calculate goodness of fit ,Require : ; in, This is a predicted value; This is the mean.

[0078] S8. Parameter writing and verification: The calibration parameters are written into the sensor's internal EEPROM, and the parameters are encrypted using AES-484. The algorithm reads and verifies the written parameters. Once the verification is successful, the laser marking machine marks the calibration date.

[0079] S9. Verification Test: The selected verification points had the following moisture contents: , , , and Perform verification tests and calculate the calibration error; the error requirement is... The verification pass rate was .

[0080] The calibration time for a single sensor is minutes, traditional methods require Minutes; calibration pass rate can reach The standard deviation of batch calibration consistency is: The maximum error is The average error is .

[0081] Example 3: Based on Example 1, several different types of soil sensors, including those for moisture, temperature, pH, and conductivity, were simultaneously calibrated. The moisture sensor was... Capacitive Pin pins; temperature sensor is PT100 Needle pins; pH sensor is Glass electrode, BNC interface; conductivity sensor is Four electrodes Pin pin.

[0082] The testing station is Workstation, each type of sensor Each workstation features a standard sample library containing standards for various sensors. The flexible grippers have replaceable gripper heads to accommodate sensors of different sizes. The calibration process includes: S1. Automatic sensor type identification: The robot picks up sensors from the tray, and the vision system identifies the sensor's shape, size, and interface type; the OCR reads the sensor's model label; and the robot automatically determines the sensor type.

[0083] S2. Automatic selection of calibration scheme: The moisture sensor uses polynomial fitting ( (Step); the temperature sensor uses a BP neural network, with resistance value and ambient temperature as inputs and actual temperature as output; the pH sensor uses polynomial fitting (Step). (Step); the conductivity sensor uses linear regression.

[0084] S3, Flexible Grabbing and Placement: Based on the sensor size, the appropriate gripper head is automatically selected; the binocular vision guides the robot to grasp precisely, and it is automatically assigned to the corresponding inspection station according to the sensor type.

[0085] S4. Parallel calibration: Each workstation operates simultaneously, for each type of sensor. Each workstation has its own standard sample library that automatically provides the corresponding standard, and the data acquisition system simultaneously collects data from all workstations.

[0086] S5. Parameter writing and verification: Different parameter formats are used to write data according to the sensor type, with a unified verification process.

[0087] In this embodiment, the average calibration time for a single sensor is minute, Simultaneous calibration of multiple sensors improves overall efficiency. The calibration pass rate is: moisture content ,temperature pH value Electrical conductivity .

[0088] Example 4: Building upon Example 1, this example incorporates self-learning of algorithm parameters based on historical data. By continuously calibrating historical data over a long period, the algorithm parameters are continuously optimized. The calibration period is [duration missing]. Months; calibrated quantity approximately For sensors only, data types include raw data, calibration parameters, verification results, and environmental parameters. The self-learning optimization process includes: S1. Data Collection and Preprocessing: collect Monthly calibration history data, data cleaning: removing abnormal and missing data, data annotation: annotating calibration accuracy, calibration time, environmental conditions, etc.

[0089] S2, Feature Engineering: Extract features such as sensor model, ambient temperature, ambient humidity, calibration time, and calibration accuracy, and normalize the features of different dimensions to... For intervals, correlation analysis is used to select key features.

[0090] S3. Establish a prediction model: A calibration accuracy prediction model is built using Gradient Boosting Tree (XGBoost). Input features include sensor model, environmental parameters, and algorithm parameters; the output is the predicted calibration accuracy. During model training, the previous... Training on data for one month, then Validation using data from one month; the XGBoost objective function is: ; in, The loss function; For regularization terms; For the first A tree.

[0091] S4. Algorithm parameter optimization: The Bayesian optimization algorithm is used to search for the optimal algorithm parameters, and the objective function is to maximize the calibration accuracy (weights). ) and minimize calibration time (weight) ).

[0092] Optimization parameters include the number of hidden layer nodes in the neural network, the learning rate, and the momentum factor; the order of the polynomial in the polynomial fitting; and the regularization parameters for linear regression. The Bayesian optimization Gaussian process model is as follows: ; in, It is a mean function; Let be the covariance function.

[0093] S5. Online learning and updates: Employing online learning algorithms, the model parameters are updated in real time, with each calibration... Each sensor triggers a parameter update once, and the parameter update uses an incremental learning method to avoid full retraining.

[0094] Through this implementation, the average error is reduced from... Reduce to The calibration time is shortened: from minutes reduced to Minutes, calibration pass rate improved: From Upgraded to .

[0095] Example 5: Based on Example 1, environmental adaptive compensation is performed to automatically compensate for the calibration results under different environmental conditions (temperature and humidity changes). Specifically, the diurnal temperature range is 15℃ to 35℃, and the seasonal humidity variation is 40%. ~80 The lighting changes from natural daylight to artificial nighttime lighting. The steps for implementing environmental adaptive compensation are as follows: S1. Real-time monitoring of environmental parameters: The temperature and humidity sensor collects environmental parameters in real time, with a sampling frequency of [missing information]. Using moving average filtering (window size) The moving average filter is: ; in, The value after smoothing; Historical data; This refers to the window size.

[0096] S2. Environmental Impact Model Establishment: Collect calibration data under different environmental conditions, analyze the impact of environmental parameters on the calibration results, and establish an environmental impact model. ; Coefficients were determined through regression analysis. .

[0097] Multiple linear regression is: ; in, For the response vector; For designing the matrix; These are the regression coefficients; This represents the error term. The parameter estimate is: ; S3, Real-time compensation calculation: Calculate the compensation value and compensation calibration parameters based on real-time environmental parameters. for: ; in, These are calibration parameters under reference environmental conditions.

[0098] S4. Verification of compensation effect: The calibration accuracy before and after compensation was compared to verify the effectiveness of the compensation model.

[0099] Through this embodiment, the influence of temperature change on calibration results is demonstrated from... Reduce to The effect of humidity changes on calibration results is as follows: Reduce to Overall environmental adaptability is improved, and the standard deviation of calibration error is reduced. .

[0100] Example 6: This embodiment provides a batch calibration system for sensors based on multi-vision fusion, including: The data acquisition module is configured to acquire omnidirectional visual information from the sensors and the detection station; The pose estimation module is configured to perform pose estimation by fusing omnidirectional visual information. The lofting module is configured to: based on pose estimation, consider factors such as the sensor's center of gravity position, gripping point strength, and avoidance of sensitive areas, optimize the selection of gripping points, use an impedance control algorithm for flexible gripping, and automatically adjust the gripping force according to the sensor material and surface friction coefficient to achieve sensor lofting; The calibration module is configured to select the corresponding algorithm from a preset algorithm library for calibration.

[0101] The working method of the system is the same as that of the multi-vision fusion-based sensor batch calibration method in Embodiment 1, and will not be repeated here.

[0102] Example 7: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the multi-vision fusion-based sensor batch calibration method described in Embodiment 1.

[0103] Example 8: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, it implements the steps of the multi-vision fusion-based sensor batch calibration method described in Embodiment 1.

[0104] Example 9: This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the multi-vision fusion-based sensor batch calibration method described in Embodiment 1.

[0105] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A batch calibration method for sensors based on multi-vision fusion, characterized in that, include: Acquire comprehensive visual information from sensors and the detection station; Pose estimation is performed by fusing comprehensive visual information. Based on pose estimation, the selection of gripping points is optimized by considering factors such as the sensor's center of gravity position, gripping point strength, and avoidance of sensitive areas. An impedance control algorithm is used for flexible gripping, and the gripping force is automatically adjusted according to the sensor material and surface friction coefficient to achieve sensor placement. Select the corresponding algorithm from the preset algorithm library for calibration.

2. The sensor batch calibration method based on multi-vision fusion as described in claim 1, characterized in that, The comprehensive visual information includes the sensor's position and orientation in three-dimensional space acquired through a binocular vision system, the sensor model, QR code, and sensor surface identified by an industrial camera, and the sensor's fit to the test stage surface and the alignment of the sensor electrodes with the test stage pins detected by orthogonal dual cameras.

3. The sensor batch calibration method based on multi-vision fusion as described in claim 2, characterized in that, The omnidirectional visual information fusion employs the Kalman filter algorithm, and the Kalman filter state update formula is as follows: ； ； in, For posterior state estimation; For prior state estimation; Kalman gain; These are the observed values; The observation matrix; Let be the posterior error covariance.

4. The sensor batch calibration method based on multi-vision fusion as described in claim 1, characterized in that, The impedance control algorithm is as follows: ； in, Joint torque; It is a Jacobian matrix; For the power of expectation; For positional stiffness; This is for positional deviation; For velocity damping; For speed.

5. The sensor batch calibration method based on multi-vision fusion as described in claim 4, characterized in that, When optimizing the selection of the grasping point, let the grasping posture parameter be... G The optimization objective is to minimize the following comprehensive cost function: ； ； ； ； in, The overall cost of capturing the posture; The center of gravity balance function; To capture the coordinates of the point; The coordinates of the centroid; This is the maximum allowed distance; The capture point intensity function; The coefficient of friction; For normal grasping force; This is a sensitive area avoidance function; This is the minimum distance from the capture point to the sensitive area; For scale parameters; , and These are the weighting coefficients.

6. The sensor batch calibration method based on multi-vision fusion as described in claim 1, characterized in that, When automatically adjusting the gripping force based on the sensor material and surface friction coefficient, the gripping force should meet the following requirements: ； ； ； in: Minimum gripping force to prevent slipping; For sensor quality; It is the acceleration due to gravity; The inertial force caused by dynamic acceleration; The static friction coefficient between the sensor surface and the gripper material; Maximum gripping force for material safety; The yield strength of the sensor housing material; This represents the contact area between the gripper and the sensor. This is for the safety factor.

7. The sensor batch calibration method based on multi-vision fusion as described in claim 6, characterized in that, Introducing dynamic weight adjustment: ； in: This is the safety margin factor; Safety is a priority factor.

8. A batch calibration system for sensors based on multi-vision fusion, characterized in that, include: The data acquisition module is configured to acquire omnidirectional visual information from the sensors and the detection station; The pose estimation module is configured to perform pose estimation by fusing omnidirectional visual information. The lofting module is configured to: based on pose estimation, consider factors such as the sensor's center of gravity position, gripping point strength, and avoidance of sensitive areas, optimize the selection of gripping points, use an impedance control algorithm for flexible gripping, and automatically adjust the gripping force according to the sensor material and surface friction coefficient to achieve sensor lofting; The calibration module is configured to select the corresponding algorithm from a preset algorithm library for calibration.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the program, it implements the steps of the multi-vision fusion-based sensor batch calibration method as described in any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the multi-vision fusion-based sensor batch calibration method as described in any one of claims 1-7.

Images