A sterile rapid transfer device for cell culture flasks

Abstract

This invention provides a rapid aseptic transfer device for cell culture flasks, belonging to the interdisciplinary field of bioengineering equipment and automation control. The device includes a culture flask status and aseptic environment sensing module, a surface fitting algorithm module, a transfer condition determination algorithm module, a single pendulum wave algorithm module, a robotic arm transfer path planning algorithm module, a robotic arm transfer execution module, and a transfer process control module. This invention, through a design approach of "hardware modularization + deep algorithm integration + full-process collaboration," constructs a closed-loop system of "sensing-preprocessing-decision-motion modeling-path planning-execution-monitoring," fundamentally solving the asepticity, efficiency, and accuracy problems of existing technologies.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of bioengineering equipment and automation control, and in particular to a sterile rapid transfer device for cell culture flasks. Background Technology

[0002] In the field of cell culture, the transfer of culture flasks is a crucial step connecting cell expansion with subsequent processing (such as harvesting and passage). Its core requirement is to achieve both speed and precision while ensuring sterility. However, existing technologies suffer from the following significant drawbacks, making it difficult to meet practical application needs:

[0003] 1. Manual transfer is the primary method, making it difficult to balance sterility and efficiency.

[0004] Currently, most laboratories or small-to-medium-sized production settings still use a manual transfer mode of "manual operation + biosafety cabinet": operators need to wear sterile protective clothing and hold the culture bottles inside the biosafety cabinet to complete the transfer and docking. This method has two major problems: First, it has a high risk of sterility. The operator's hand movements and breathing airflow can easily disturb the sterile environment inside the biosafety cabinet, and repeated operation can easily lead to fatigue, increasing the risk of contamination caused by operational errors (such as collision of bottle mouths or hand contact with the bottle body); Second, it is inefficient. Manual transfer requires completing multiple steps such as "opening the cap - aligning - transferring - closing the cap", which takes a long time for a single bottle transfer and is difficult to adapt to the batch transfer needs of large-scale culture.

[0005] 2. Traditional automated transfer equipment has technical shortcomings.

[0006] Automated transfer equipment has been introduced in some large-scale production settings, but existing equipment still has significant technical shortcomings:

[0007] Insufficient sterility assurance: Traditional control of robotic arms often uses "rigid motion trajectory". Rapid acceleration and sharp turns can easily generate airflow disturbances, causing the sterile airflow field in the biosafety cabinet to become turbulent and bringing suspended microorganisms into the culture bottle. At the same time, the disinfection methods of the end effector of the equipment (such as ultraviolet irradiation) are difficult to cover all contact surfaces, posing a risk of microbial residue.

[0008] Poor accuracy and coordination: Existing equipment mostly uses "single-point sampling" to sense the status of culture bottles. The discrete data is easily affected by measurement noise, resulting in large deviations in the docking of the robotic arm (such as a bottle mouth alignment deviation exceeding 0.1mm). Moreover, each link of "sensing-judgment-execution" operates independently, lacking data coordination. For example, sterility parameters such as environmental temperature and humidity and microbial concentration are not linked with the movement of the robotic arm. Even if environmental parameters exceed the standard, the equipment may still continue to perform the transfer action, causing contamination.

[0009] Poor adaptability: Existing equipment is mostly designed for culture flasks of fixed specifications, making it difficult to adapt to culture flasks of different sizes such as 100mL, 250mL, and 500mL. When changing the flask type, the parameters of the robotic arm need to be adjusted manually, which is complicated and prone to introducing errors.

[0010] 3. Insufficient coordination between algorithms and hardware makes it difficult to balance "speed" and "smoothness".

[0011] Current automated equipment motion control often employs algorithms for "uniform linear motion" or "simple curvilinear motion," failing to consider the smoothness requirements of a sterile environment. Increasing motion speed for rapid transfer can easily lead to robotic arm vibration and airflow disturbance; conversely, reducing speed to ensure smoothness sacrifices transfer efficiency. Furthermore, the algorithms are not integrated with the actual geometry of the culture flasks (e.g., flask opening orientation, flask surface curvature), resulting in a disconnect between path planning and actual docking requirements, increasing the risk of operational errors. Summary of the Invention

[0012] This invention provides a sterile rapid transfer device for cell culture flasks. When using a traditional robotic arm, a new control algorithm is designed to control the robotic arm to perform the transfer. The overall solution adopts the design concept of "hardware modularization + deep algorithm integration + full-process collaboration" to build a closed-loop system of "perception-preprocessing-decision-motion modeling-path planning-execution-monitoring", which fundamentally solves the problems of sterility, efficiency and accuracy of existing technologies.

[0013] To achieve the above objectives, the present invention adopts the following technical solution:

[0014] A sterile rapid transfer device for cell culture flasks includes:

[0015] The culture bottle status and aseptic environment sensing module is used to collect discrete data on the position of the outer surface of the culture bottle, the direction of the bottle opening, the ambient temperature and humidity, the microbial concentration, and the pressure inside the culture bottle through laser displacement sensor, angle sensor, temperature and humidity sensor, microbial sensor, and pressure sensor.

[0016] The surface fitting algorithm module is connected to the culture bottle state and sterile environment sensing module. It is used to receive discrete points of position and orientation, convert them into continuous surface models of the outer surface of the culture bottle and the orientation of the bottle mouth, and calculate the deviation of each discrete point from the corresponding surface and the total fitting error.

[0017] The transfer condition determination algorithm module is connected to the surface fitting algorithm module and the culture bottle status and sterile environment sensing module, respectively. It is used to receive position surface model, orientation surface model, deviation of each discrete point, total fitting error and environmental temperature, environmental humidity, microbial concentration and pressure data inside the culture bottle, determine whether the current state meets the transfer conditions, and output the determination result and abnormal parameter set.

[0018] The pendulum wave algorithm module, connected to the transition condition determination algorithm module, is used to receive the position surface model and orientation surface model when the determination result is that the transition condition is met, and to simulate the pendulum motion to generate motion data of the robotic arm (curves of angular displacement, linear velocity and acceleration changing with time).

[0019] The robotic arm transition path planning algorithm module is connected to the single pendulum wave algorithm module. It is used to receive curves of angular displacement, linear velocity and acceleration changing over time, as well as position surface models and orientation surface models, and generate control commands for the robotic arm (starting point coordinates, ending point coordinates, actual movement speed, actual turning angle and total movement time).

[0020] The robotic arm transfer execution module is connected to the robotic arm transfer path planning algorithm module. It is used to receive the starting point coordinates, ending point coordinates, actual movement speed, actual turning angle and total movement time, drive the robotic arm to perform the transfer action, and provide real-time feedback on the movement completion signal, real-time position and key component status.

[0021] The overall control module for the transfer process is connected to the culture bottle status and sterile environment sensing module, the transfer condition determination algorithm module, and the robotic arm transfer execution module, respectively. It is used to receive information from each module, output control commands, and coordinate the entire process.

[0022] In this specification, the deviations of discrete points at each position and the deviations of discrete points at each orientation calculated by the surface fitting algorithm module are transmitted in real time to the transition condition determination algorithm module. During the determination process, if any discrete point deviation at any position exceeds a preset threshold for a single position deviation or any discrete point deviation at any orientation exceeds a preset threshold for a single orientation deviation, the transition condition determination algorithm module directly marks the discrete point corresponding to the deviation as an anomaly point and includes the discrete point deviation exceeding the standard into the set of abnormal parameters. The determination of this type of anomaly is not based on the total fitting error alone.

[0023] In this specification, the surface fitting algorithm module uses a bicubic B-spline surface model, and the node spacing of its basis functions is the optimal value determined during the model training phase—the node spacing in the position dimension is 8 mm and the node spacing in the orientation dimension is 3 mm. The node spacing parameters are stored in the parameter library of the surface fitting algorithm module and are automatically loaded each time the fitting calculation is started without manual adjustment. Furthermore, if a change in the size of the culture bottle is detected during the fitting process, the pre-trained node spacing parameters of the corresponding size culture bottle are automatically called.

[0024] In this specification, the judgment logic of the transition condition judgment algorithm module is as follows: when the total fitting error, the deviation of discrete points at each position, and the deviation of discrete points at each orientation do not exceed the preset threshold, and the ambient temperature, ambient humidity, microbial concentration, and pressure inside the culture bottle are all within the preset range, it is determined that the transition condition is met; otherwise, the corresponding abnormal parameters are output.

[0025] In this specification, the pendulum motion simulated by the pendulum wave algorithm module is characterized by the pendulum amplitude being the straight-line distance from the starting point to the ending point of the robotic arm, the pendulum period being determined by the equivalent pendulum length of the robotic arm and the gravitational acceleration, and the maximum pendulum angle not exceeding 0.3 rad to ensure smooth motion.

[0026] In this specification, when the robotic arm transition path planning algorithm module generates the actual motion speed, it multiplies the pendulum speed by a preset speed safety factor and corrects it according to the angle between the speed direction and the bottle opening direction; when generating the actual turning angle, it fuses the pendulum angular displacement with the bottle opening orientation curved surface model through a preset angle adjustment coefficient.

[0027] In this specification, after the single pendulum wave algorithm module generates the linear velocity curve of the robotic arm's motion, it calculates the first derivative of the linear velocity to obtain the acceleration curve, which is then synchronously transmitted to the robotic arm transition path planning algorithm module. When generating the actual motion speed, the robotic arm transition path planning algorithm module sets an upper limit for acceleration based on the acceleration curve to ensure that the actual motion acceleration of the robotic arm does not exceed the preset maximum allowable acceleration of the robotic arm, thus avoiding vibration disturbance to the sterile environment caused by excessive acceleration of the robotic arm.

[0028] In this specification, if the transfer condition determination algorithm module determines that the transfer conditions are not met and outputs an abnormal parameter set, it synchronously feeds back the abnormal parameter set to the surface fitting algorithm module and the culture bottle status and sterile environment sensing module. If the abnormal parameter is that the total fitting error exceeds the standard, the surface fitting algorithm module reloads the discrete point data and adjusts the node interval of the B-spline basis function, and re-executes the fitting calculation. If the abnormal parameter is that the environmental parameter exceeds the standard, the culture bottle status and sterile environment sensing module re-collects environmental data at intervals of 5 to 10 seconds until the environmental parameter meets the preset range or the transfer process control module outputs a stop transfer command.

[0029] In this specification, when the single pendulum wave algorithm module determines the single pendulum amplitude, it first extracts the maximum curvature point on the outer surface of the culture bottle from the position surface model, calculates the straight-line distance from the starting point of the robotic arm to the maximum curvature point, and then reduces the straight-line distance by 10% to 15% as the final single pendulum amplitude to ensure that the robotic arm's motion trajectory avoids the protruding area on the outer surface of the culture bottle. Moreover, the single pendulum amplitude adjustment process is based on the orientation surface model verification to ensure that the adjusted amplitude does not change the relative angle between the end of the robotic arm and the bottle opening.

[0030] In this specification, the robotic arm transition path planning algorithm module, when generating the actual motion speed and actual turning angle, discretizes the continuous pendulum motion curve into parameter values ​​for multiple time nodes with a time step of 0.02 seconds. The actual motion speed at each time node must match the actual turning angle of that node. When the actual turning angle is greater than a preset angle threshold, the actual motion speed of the corresponding node is automatically reduced. The reduction magnitude is positively correlated with the turning angle to ensure smooth movement of the robotic arm when turning.

[0031] In summary, the present invention has at least the following beneficial effects:

[0032] 1. Significantly improved sterility assurance capabilities

[0033] The three-layer aseptic protection mechanism fundamentally reduces the risk of contamination: First, the smooth motion trajectory generated by the single pendulum wave algorithm avoids airflow disturbances caused by the movement of the robotic arm, ensuring the stability of the flow field in the aseptic environment; second, the end effector of the execution module adopts a combination of high-pressure steam sterilization and ethanol disinfection to ensure that there are no microbial residues on the contact surface; third, the transfer condition judgment algorithm monitors environmental microbial concentration, temperature, humidity and other parameters in real time. If the parameters exceed the standard, the transfer is terminated immediately, blocking the possibility of contamination from the decision-making level and completely solving the aseptic problems caused by "motion disturbance" and "lack of environmental monitoring" in the existing technology.

[0034] 2. Transfer efficiency and ease of operation have been greatly improved.

[0035] The device eliminates the need for manual operation and achieves fully automated workflow: from sensor sampling and algorithm processing to robotic arm execution, no human intervention is required, significantly reducing operation steps and time consumption. At the same time, the device has the ability to adapt to different sizes of culture bottles. When changing bottle types, there is no need to manually adjust the core parameters. You only need to select the bottle type through the human-machine interface of the central control module, which greatly reduces the complexity of operation and solves the defects of existing equipment in "low efficiency" and "weak adaptability".

[0036] 3. Improved switching accuracy and operational reliability

[0037] Through data preprocessing and path adaptation, the transfer action is ensured to be precise and controllable: the surface fitting algorithm transforms discrete sensing data into a continuous surface model, eliminating errors caused by measurement noise and providing high-precision basic data for subsequent judgment and path planning; the path planning algorithm combines the actual geometric state of the culture bottle (position surface, orientation surface) with the single pendulum wave motion curve, and the generated path parameters are highly matched with the execution requirements of the robotic arm, avoiding docking deviation; the force sensor and position feedback encoder of the execution module monitor the contact force and position deviation in real time, and stop the action immediately if an abnormality occurs, ensuring that the culture bottle and equipment are undamaged, solving the problems of "poor accuracy" and "low reliability" in existing technologies.

[0038] 4. The system's coordination and intelligence levels have been significantly improved.

[0039] A closed-loop data chain of "sensing-processing-decision-execution-monitoring" is constructed, with each module transmitting data in real time through a standardized communication protocol to achieve deep collaboration. For example, the surface fitting results are directly used as input to the judgment algorithm, the single pendulum wave motion curve directly determines the path planning parameters, and the execution status is fed back to the central control module in real time, ensuring that there is no data redundancy or delay in each link. At the same time, the central control module has historical data storage and fault statistics functions, which can trace the key parameters of each transfer, facilitating later maintenance and troubleshooting, realizing the upgrade from "automation" to "intelligence", and meeting the stringent requirements of "traceability" and "stability" of equipment in the fields of biopharmaceuticals and cell therapy. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the aseptic rapid transfer device for cell culture flasks involved in the present invention.

[0041] Figure 2 This is a schematic diagram of the data acquisition-preprocessing optimization process involved in this invention.

[0042] Figure 3 This is a schematic diagram of the motion parameter optimization process involved in this invention.

[0043] Figure 4 This is a schematic diagram of the feedback iterative closed-loop process involved in this invention. Detailed Implementation

[0044] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0045] like Figure 1 As shown, this embodiment provides a sterile rapid transfer device for cell culture flasks, including:

[0046] The culture bottle status and aseptic environment sensing module is used to collect discrete data on the position of the outer surface of the culture bottle, the direction of the bottle opening, the ambient temperature, the ambient humidity, the microbial concentration, and the pressure inside the culture bottle through laser displacement sensor, angle sensor, temperature and humidity sensor, microbial sensor, and pressure sensor.

[0047] The surface fitting algorithm module is connected to the culture bottle state and sterile environment sensing module. It is used to receive discrete points of position and orientation, convert them into continuous surface models of the outer surface of the culture bottle and the orientation of the bottle mouth, and calculate the deviation of each discrete point from the corresponding surface and the total fitting error.

[0048] The transfer condition determination algorithm module is connected to the surface fitting algorithm module and the culture bottle status and sterile environment sensing module, respectively. It is used to receive position surface model, orientation surface model, deviation of each discrete point, total fitting error and environmental temperature, environmental humidity, microbial concentration and pressure data inside the culture bottle, determine whether the current state meets the transfer conditions, and output the determination result and abnormal parameter set.

[0049] The pendulum wave algorithm module, connected to the transition condition determination algorithm module, is used to receive the position surface model and orientation surface model when the determination result is that the transition condition is met, and to simulate the pendulum motion to generate curves of the angular displacement, linear velocity and acceleration of the robotic arm motion over time.

[0050] The robotic arm transition path planning algorithm module, connected to the single pendulum wave algorithm module, is used to receive curves of angular displacement, linear velocity and acceleration changing over time, as well as position surface models and orientation surface models, to generate the robotic arm's starting coordinates, ending coordinates, actual movement speed, actual turning angle and total movement time.

[0051] The robotic arm transfer execution module is connected to the robotic arm transfer path planning algorithm module. It is used to receive the starting point coordinates, ending point coordinates, actual movement speed, actual turning angle and total movement time, drive the robotic arm to perform the transfer action, and provide real-time feedback on the movement completion signal, real-time position and key component status.

[0052] The overall control module for the transfer process is connected to the culture bottle status and sterile environment sensing module, the transfer condition determination algorithm module, and the robotic arm transfer execution module, respectively. It is used to receive information from each module, output control commands, and coordinate the entire process.

[0053] In some embodiments, the transfer condition determination algorithm module adopts a "priority sorting" logic when determining environmental parameters: whether the microbial concentration exceeds the standard is the highest priority, followed by the pressure inside the culture bottle, and finally the ambient temperature and humidity; if multiple types of environmental parameters are abnormal at the same time, the determination algorithm first outputs the abnormal parameter with the highest priority, and the central control module processes the abnormality first. After the highest priority abnormality is resolved, the remaining low priority parameters are then determined to be normal.

[0054] In some embodiments, when calculating the total fitting error, if the surface fitting algorithm module detects that the deviation of a certain discrete point is much greater than that of other discrete points (the deviation is 3 times or more of the average deviation of other discrete points), it will automatically reduce the weight of the discrete point to 50% of the original weight and recalculate the total fitting error. This avoids the excessive influence of a single abnormal discrete point on the overall surface model accuracy. The weight adjustment process needs to be recorded in the log of the overall control module to facilitate the later tracking of the discrete point acquisition quality.

[0055] In some embodiments, when the pendulum wave algorithm module calculates the pendulum period, it needs to adjust the period in conjunction with the mechanical arm load information fed back by the mechanical arm transfer execution module (collected by the force sensor of the execution module, transmitted to the main control module and then forwarded to the pendulum wave algorithm module): when the load increases (such as transferring a 500mL culture bottle), the pendulum period is extended by 5% to 8% to ensure that the linear velocity does not decrease with the increase of the load, maintain the transfer efficiency, and the period adjustment range does not exceed the preset maximum period adjustment threshold (10%).

[0056] In some embodiments, after the robotic arm transition path planning algorithm module generates the endpoint coordinates, it needs to perform "collision verification" based on the position surface model and the orientation surface model: along the path from the robotic arm's starting point to the endpoint, a verification point is taken every 1mm to determine whether the verification point is within the range of the outer surface surface model of the culture bottle. If any verification point is outside the range, the x / y / z axis values ​​of the endpoint coordinates are adjusted (the adjustment range does not exceed 0.5mm) until all verification points do not collide with the culture bottle surface. Only after the verification is passed can the path parameters be output to the execution module.

[0057] The device of this invention adopts a collaborative architecture of 7 modules:

[0058] The sensing module, serving as the foundation for data input, collects multi-dimensional data on the physical state (position, orientation) of the culture flasks and environmental aseptic parameters (temperature, humidity, microbial concentration, and internal pressure) through laser displacement sensors, angle sensors, temperature and humidity sensors, microbial concentration sensors, and pressure sensors, ensuring comprehensive and reliable raw data. Four algorithm modules form the core processing chain: a surface fitting algorithm transforms discrete sensing data into a continuous, smooth surface model, eliminating noise and errors; a transfer condition determination algorithm integrates geometric accuracy (surface fitting results) with environmental parameters to scientifically determine transfer feasibility; a single pendulum wave algorithm simulates smooth motion curves, avoiding airflow disturbances; and a path planning algorithm transforms theoretical curves into executable path parameters for the robotic arm, achieving a "theory-practice" fit. The execution module, as the action output end, employs a six-axis collaborative robotic arm and a sterile end effector to complete precise transfers according to the planned path, providing real-time feedback on the execution status. The central control module, as the global coordination center, receives data from each module, outputs control commands, and achieves fully automated monitoring and anomaly handling throughout the entire process.

[0059] The core logic of the entire solution is: taking "sterility" as the core constraint, ensuring smoothness of movement and scientific judgment through algorithm fusion, ensuring the accuracy of data acquisition and action execution through hardware modularization, and ultimately achieving the transfer goal of "sterility, speed and accuracy", while having the flexibility to adapt to different sizes of culture bottles.

[0060] I. Culture Bottle Status and Aseptic Environment Sensing Module: High-Precision Multi-Dimensional Data Acquisition

[0061] 1. Module hardware composition and functional division

[0062] To ensure the accuracy and comprehensiveness of the collected data, the module adopts a "multi-sensor collaborative acquisition" scheme, with the hardware configuration and corresponding functions as follows:

[0063] Laser displacement sensor (model: Keyence IL-600): used to acquire position data of the outer surface of culture flasks, with a measurement accuracy of ±0.01mm, a sampling frequency of 100Hz, and can simultaneously acquire x / y / z three-dimensional coordinates (i.e., A total of three samples were deployed (collected from the front, side, and top respectively to ensure full coverage of the culture flask surface);

[0064] Three-axis angle sensor (model: Bosch BNO055): used to acquire bottle opening orientation data, measurement range ±180°, accuracy ±0.5°, outputting α / β / γ three-dimensional angles (i.e., ( ), installed on the vision positioning component at the end of the robotic arm, and moves synchronously with the robotic arm to sample;

[0065] Temperature and humidity sensor (model: Sensirion SHT31): used to collect the temperature and humidity of the internal environment of the adapter. The temperature measurement range is 0-65℃ (accuracy ±0.2℃), and the humidity measurement range is 0%~100%RH (accuracy ±2%RH). The output parameters are T (temperature) and H (humidity). It is deployed at the top of the device (away from heat sources and dead air areas).

[0066] Laser particle counter (model: Lighthouse 3016): used to detect the concentration of environmental microorganisms (indirectly reflecting the microbial content by counting particles ≥0.5μm), measurement range 0-50000CFU / m³, accuracy ±10%, output parameter C (microbial concentration), deployed on the side of the culture bottle placement area, sampling interval 1s;

[0067] Pressure sensor (model: Freescale MPX5700): Used to acquire pressure inside culture flasks, measurement range 0-200 kPa, accuracy ±0.1 kPa, output parameters (Intra-bottle pressure), the culture bottle is connected through a sterile puncture needle (the needle is sterilized with ethylene oxide to avoid contamination).

[0068] 2. Data Acquisition Process and Quality Control

[0069] To avoid data anomalies caused by sensor interference or operational errors during the acquisition process, the module is equipped with a strict acquisition process and quality control mechanism:

[0070] 1. Preparation before data acquisition: Preheat all sensors (temperature and humidity sensors for 30 minutes, pressure sensors for 10 minutes), and perform zero-point calibration using standard calibration pieces (e.g., laser displacement sensors are calibrated using standard blocks, and angle sensors are calibrated using a water platform).

[0071] 2. Regional sampling: Divide the culture bottle into 3 sampling regions according to "bottle body - bottle mouth - bottle bottom", and collect 2-4 discrete points in each region (determine the total number of sampling points m=6, 8, 10 according to the size of the culture bottle) to ensure that the sampling points are evenly distributed (e.g., 100mL culture bottle: 2 points on the bottle body, 2 points on the bottle mouth, and 2 points on the bottle bottom, for a total of m=6).

[0072] 3. Data validity verification: Calculate the standard deviation of three repeated measurements for each sampling point. If the standard deviation is >0.05mm (location) or >0.2° (orientation), the point should be re-sampled. For environmental parameters (T, H, C, ... The data was collected five times consecutively, and the average value was taken as the final data to avoid the influence of instantaneous fluctuations.

[0073] 4. Data format conversion: Convert the collected raw data (such as voltage signals output by sensors) into physical quantity units (mm, °, ℃, %RH, CFU / m³, kPa) that the algorithm module can recognize, and encapsulate them in the format of "sampling point number-parameter type-value-collection time" to ensure data traceability.

[0074] 3. Module Output and Data Transfer

[0075] Output: Discrete set of locations of the culture flask: (k=1,...,m) (including the 3D coordinates of each point and the sampling area identifier); Discrete set of points with the bottle opening facing: (k=1,...,m) (including the three-dimensional angle and sampling area identifier for each point); Environmental sterility parameters: T (average temperature), H (average humidity), C (average microbial concentration). (Mean bottle pressure); Data quality indicators: Standard deviation of each parameter (e.g., ...) Location point standard deviation (where is the standard deviation of temperature), used in subsequent algorithms to determine data reliability.

[0076] Data transmission method: Real-time transmission to the surface fitting algorithm module and the transition condition judgment algorithm module via industrial Ethernet (Profinet protocol) (environmental parameters need to be transmitted to the judgment module synchronously), with a transmission rate of 100Mbps and a latency of ≤10ms to ensure data timeliness.

[0077] 4. Core Function of the Module

[0078] Provides "source data assurance" for the entire system: Through multi-sensor collaboration and quality control mechanisms, ensures that the collected location, orientation, and environmental parameters are accurate and reliable, avoiding subsequent algorithm misjudgments or execution deviations due to errors in the original data; Achieves "synchronous monitoring of sterility and geometric state": Not only collects physical location data of the culture bottles, but also monitors key sterility indicators such as environmental temperature and humidity and microbial concentration in real time, covering the core needs of the transfer process from a data perspective.

[0079] II. Surface Fitting Algorithm Module (Algorithm 1): Continuous Modeling of Discrete Data

[0080] 1. Model Building Details

[0081] The core issue is that the position and orientation data of the culture flasks collected by the sensing module are discrete (e.g., 6, 8, or 10 sampling points) and contain measurement noise (e.g., sensor accuracy error ±0.1mm). Directly using this data for subsequent judgment or path planning will lead to error accumulation. Therefore, it is necessary to transform the discrete points into a continuous and smooth surface model to improve data reliability.

[0082] Model selection: Bicubic B-spline surface model. This model has the following advantages: ① Local support: the error of a single sampling point will not affect the entire surface; ② Smoothness: the surface and its first and second derivatives are continuous, which conforms to the physical properties of the culture bottle surface; ③ Flexibility: it can be adapted to culture bottles of different shapes by adjusting the distribution of basis function nodes.

[0083] : Sampling point on the outer surface of the kth culture flask (k=1,2,...,m, m is the number of sampling points, 6, 8, 10, selected according to the size of the culture flask, m=6 for 100mL culture flask, m=10 for 500mL culture flask); (These are three-dimensional coordinates, in mm). The k-th bottle opening faces the sampling point ( Let be the rotation angle about the x-axis. Let be the rotation angle about the y-axis. (The angle of rotation about the z-axis, in degrees). The weight of the kth sampling point ( =1 / m, ensuring the total weight is 1, so that the contribution of each sampling point to the surface is balanced. Equation of the continuous surface of the outer surface of the culture flask (expression: ,in For spline coefficients, (These are cubic B-spline basis functions). The equation for the continuous surface facing the bottle opening (same form) , The distribution of basis function nodes and match). The deviation between the sampling point at the k-th position and the fitted surface ( (Unit: mm, reflects the fitting accuracy of a single point). : The deviation between the k-th orientation sampling point and the fitted surface ( (Unit: °). Total fitting error ( ,unit: (Comprehensive evaluation of the overall surface fitting quality).

[0084] Derivation of core formula:

[0085] To ensure the surface fits the discrete sampling points as closely as possible, a weighted least squares method is used to minimize the total fitting error.

[0086] (1)

[0087] (2)

[0088] Will and Substituting the basis function expressions into the above formula, we can transform the problem into solving a system of linear equations for the spline coefficients. Finally, the continuous surface model was determined.

[0089] 2. Model Training Process

[0090] Training data preparation: Collect high-precision data from 500 sets of standard culture bottles (the actual position and orientation were obtained by a coordinate measuring machine, with an error ≤0.01mm), covering three commonly used culture bottle types: 100mL culture bottles: 200 sets (diameter 50mm, height 100mm); 250mL culture bottles: 200 sets (diameter 70mm, height 120mm); 500mL culture bottles: 100 sets (diameter 90mm, height 150mm).

[0091] Training objective: To adjust the node distribution of the B-spline basis functions to improve the average total fitting error on the training set. This ensures that the surface model accurately reflects the true shape of the culture flask.

[0092] Training steps:

[0093] 1. Initialization: Set the basis function order to 3 (cubic spline), the initial value of the position node spacing to 10mm (same for x / y / z axes), and the initial value of the orientation node spacing to 5° (same for α / β / γ axes).

[0094] 2. Error Calculation: Substitute the training data into formulas (1) and (2) to calculate the initial error. (Initial value approximately 0.4-0.5).

[0095] 3. Parameter optimization: The node spacing is adjusted using a grid search method (position spacing range 5-20mm, step size 1mm; orientation spacing range 2-10°, step size 0.5°), and recalculated after each adjustment. .

[0096] 4. Convergence Judgment: When the position interval = 8mm and the orientation interval = 3°, the average Stop training and save the basis function parameters at this point.

[0097] 3. Model Application Process

[0098] Input: Output of the perception module (k=1,...,m) and (k=1,...,m).

[0099] Calculation process:

[0100] 1. Substitute the discrete sampling points into the trained B-spline basis function, and obtain the spline coefficients by solving the linear equations of formulas (1) and (2). ( and ),Sure and .

[0101] 2. Calculate the deviation between each sampling point and the surface: , .

[0102] 3. Calculate the total fitting error: .

[0103] Output: , , (k=1,...,m) (k=1,...,m) .

[0104] Core function:

[0105] By eliminating noise and errors in discrete sampling, "point data" is transformed into "surface models," providing continuous and smooth basic data for subsequent decision-making and path planning.

[0106] pass and , Quantify the reliability of sensing data to avoid overall judgment errors caused by abnormal measurements of a single point.

[0107] III. Transfer Condition Determination Algorithm Module (Algorithm 2): Asepticity Determination Based on Multi-Parameter Fusion

[0108] 1. Model Building Details

[0109] The core issue is that the transfer process must simultaneously meet both "geometric accuracy" (accurate position and orientation of the culture bottle) and "environmental sterility" (temperature, humidity, microbial concentration, etc. must meet standards). Judging based on a single parameter may lead to misjudgment (e.g., considering only position while ignoring microbial concentration can cause contamination). Therefore, it is necessary to construct a multi-parameter fusion judgment model.

[0110] Parameters from Algorithm 1: , , (k=1,...,m) (k=1,...,m) Environmental parameters from the sensing module: T, H, C, .

[0111] Judgment threshold (experimentally validated aseptic transfer cutoff value): (Total fitting error threshold; exceeding this value indicates that the surface model is unreliable). (Single position point deviation threshold to ensure the docking accuracy of the robotic arm); =1.0° (Single orientation point deviation threshold to prevent leakage when the bottle neck is connected); =36.0℃, =38.0℃ (based on the optimal temperature range for cell culture (36-38℃), deviations will affect cell viability); =40%RH =60%RH (humidity range; too high a humidity level can easily promote the growth of microorganisms, while too low a humidity level will cause the culture medium to evaporate). (Upper limit of microbial concentration; exceeding this value greatly increases the risk of contamination.) =99.5kPa, =100.5kPa (pressure range inside the bottle; excessive pressure difference with the environment can lead to airflow exchange contamination).

[0112] Output parameters: D: Judgment result (enumerated values: D=1 (meets transfer conditions), D=0 (does not meet transfer conditions)); Ab: Set of abnormal parameters (7 possible abnormal types in total). : > (Total fitting error exceeds the standard); : > (Position deviation exceeds standard); : > (Orientation deviation exceeds standard); :T< or T> (Temperature exceeded the standard); H< or H> (Humidity exceeds the standard); :C> (Microbial concentration exceeds the standard); : < or > (Abnormal pressure inside the bottle)

[0113] Core formula logic:

[0114] The judgment model adopts a "full satisfaction" logic (that is, all conditions must be met before it is judged as compliant), and is carried out in two steps:

[0115] 1. Geometric accuracy determination: ;

[0116] 2. Environmental parameter determination:

[0117] ;

[0118] 3. Overall Judgment Result: (D=1 only when F=1 and En=1) (3)

[0119] 2. Model Training Process

[0120] Training data preparation: 1000 sets of historical transfer records (including successful and failed cases), each set of data includes: Input parameters: , , T, H, C Actual results: =1 (Transfer successful, no contamination and accurate connection) or =0 (transfer failure, including contamination, misalignment, etc.).

[0121] Training objective: Optimize the decision threshold so that the model's decision accuracy (number of correct decisions / total number of samples) is ≥99% and the false negative rate (the proportion of actual failures but judged as successes) is ≤0.5% (to avoid contamination risk).

[0122] Training steps:

[0123] 1. Initialize the threshold: Set the initial threshold based on literature and experience (e.g., =0.5), calculate D for each sample group according to formula (3). 2. Performance evaluation: Compare D with Statistical accuracy and false negative rate (initial accuracy approximately 92%, false negative rate approximately 3%). 3. Threshold optimization: Fine-tuning the threshold using gradient descent (adjustment step size: It is 0.01. (e.g., 0.05mm), prioritize reducing the false negative rate. 4. Convergence judgment: When the threshold is adjusted to the above defined value, the accuracy = 99.2% and the false negative rate = 0.3%, which meets the requirements, and training is stopped.

[0124] 3. Model Application Process

[0125] Input: All parameters output by Algorithm 1 and environmental parameters of the sensing module.

[0126] Calculation process: 1. Calculate the geometric accuracy judgment result F: Check Is it ≤ ,all Is it ≤ ,all Is it ≤ 2. Calculate and determine the environmental parameter result En: Check T, H, C, 3. Calculate the total judgment result D according to formula (3): If F=1 and En=1, then D=1; otherwise, D=0. 4. List the abnormal parameters Ab: If D=0, add all parameter types that cause F=0 or En=0 to Ab (e.g., add T=39℃ to Ab). ).

[0127] Output: D, Ab (when D=1, Ab is an empty set and is output to the single pendulum wave algorithm module; when D=0, it is output to the main control module).

[0128] Core Function: As the "decision-making center" of the device, it integrates geometric and environmental parameters to scientifically determine whether a transfer can be performed, thus preventing contamination and operational errors at the source. By listing abnormal parameters Ab, it provides the main control module with specific reasons for the anomalies, facilitating rapid problem troubleshooting (e.g., ...). (The environment needs to be disinfected).

[0129] IV. Single Pendulum Wave Algorithm Module (Algorithm 3): Smooth Motion Modeling in a Sterile Environment

[0130] 1. Model Building Details

[0131] The core issue is that rapid acceleration and sharp turns in robotic arm movements can cause airflow disturbances within the device, potentially introducing microorganisms from the environment into the culture flask (disrupting sterility). The motion of a simple pendulum, characterized by "continuous velocity change and no abrupt acceleration," can be simulated to achieve smooth motion.

[0132] Parameters from Algorithm 2: D=1 (meets the switching condition) (Position surface) (Towards the curved surface). Robotic arm inherent parameters: L (equivalent swing length of the robotic arm, unit: mm, e.g., 250, 300, 350, selected according to the device model; standard model uses 300mm). (Gravity acceleration, fixed value). Motion parameters: The amplitude of a simple pendulum (unit: mm, equal to the straight-line distance from the starting point to the ending point of the robotic arm, calculated by...) (Calculated) Period of a simple pendulum (unit: s) (reflects the speed of movement); Maximum swing angle (unit: rad) (Small angles ensure smooth movement) : Function of angular displacement of a simple pendulum as a function of time (unit: rad, t is the time variable) ); : Cycloidal velocity as a function of time (unit: mm / s); : Function of acceleration of a simple pendulum as a function of time (unit: mm / s²).

[0133] Derivation of core formula:

[0134] Simple pendulum motion at small angles ( The following can be approximated as simple harmonic motion, as shown in the following formula:

[0135] 1. Angular displacement function: (4)

[0136] When t=0 =0, corresponding to the starting point of the robotic arm; t= 4 o'clock = This corresponds to the endpoint of the robotic arm, ensuring a single direction of movement.

[0137] 2. Linear velocity function: for Differentiate and multiply by the pendulum length L to obtain the linear velocity:

[0138] (5)

[0139] The speed increases from 0, reaches its maximum value at the midpoint, and then decreases back to 0 without any sudden changes.

[0140] 3. Acceleration function: for Taking the derivative, we obtain the acceleration:

[0141] (6)

[0142] The acceleration changes continuously without impact, avoiding airflow disturbance.

[0143] 2. Model Training Process

[0144] Training data preparation: 300 sets of robotic arm motion test data, covering 3 pendulum lengths (250, 300, 350 mm), with 100 sets for each pendulum length. Each set of data includes:

[0145] Input parameters: L, ;

[0146] Output metrics: Motion stability score (1-10 points, measured by a particle image velocimeter to measure airflow disturbance intensity; the smaller the disturbance, the higher the score), motion time (time from start to finish). ).

[0147] Training objective: Determine The optimal range is to achieve a stability score of ≥9 (minimal airflow disturbance) and a motion time of ≤6s (meeting the requirements for rapid transfer).

[0148] Training steps:

[0149] 1. For each set of data, calculate according to formulas (4)-(6). , , 1. Simulate the movement of a robotic arm. 2. Measure and score the intensity of airflow disturbance, and record the movement time. 3. Analyze the results. Relationship with performance: When Too small (<60mm), too long movement time (>8s); when When it is too large (>120mm). >0.3 rad, stationarity score <8 points. 4. Determine the optimal range: (When L=300mm, =3-5s, rating = 9.2 points).

[0150] 3. Model Application Process

[0151] Input: D=1 from the output of Algorithm 2. , , default L.

[0152] Calculation process:

[0153] 1. From Extraction robotic arm starting point (Fixed to (0,0,0)) and endpoint (Coordinates of the center of the bottle mouth) ), calculate distance (like If the value exceeds [60, 120] mm, automatically adjust L. (Falling within the range). 2. Calculate the period of the simple pendulum: (e.g., when L=300mm, ≈1.1s). 3. Calculate the maximum swing angle: (make sure ≤0.3rad). 4. Substitute into formulas (4)-(6) to generate Within range , , .

[0154] Output: , , , .

[0155] Core function: Providing a motion reference curve with "continuous velocity and no abrupt acceleration" avoids airflow disturbance from a dynamic perspective, which is crucial for ensuring the sterility of the transfer process. By controlling the motion time through a single pendulum cycle, a balance is struck between "smoothness" and "speed" (e.g., the transfer time for a 500mL culture flask can be controlled within 4-5 seconds).

[0156] V. Robotic Arm Transfer Path Planning Algorithm Module (Algorithm 4): Conversion from Theoretical Model to Execution Parameters

[0157] 1. Model Building Details

[0158] The core issue is that the single pendulum wave algorithm generates an ideal motion curve, which needs to be combined with the actual position and orientation of the culture flask. , The parameters are adapted and converted into specific execution parameters for the robotic arm (such as three-dimensional coordinates, actual speed and angle) to ensure precise docking between the end effector and the bottle opening.

[0159] Parameters from Algorithm 3: , , , Parameters from Algorithm 1: (Position surface) (Towards the curved surface). Path parameters: : Coordinates of the robotic arm's starting point (unit: mm, fixed at (0,0,0)); : Coordinates of the robotic arm's endpoint (unit: mm); =0.9: Speed ​​safety factor (reduces speed by 10% to avoid overshoot caused by robotic arm response delay); =0.8: Angle adjustment coefficient (reduces the angle by 20% to ensure the end orientation matches the bottle opening); Δθ(t): Angle between the velocity direction and the bottle opening orientation (unit: rad). ); : Actual movement speed of the robotic arm (unit: mm / s); : Actual turning angle of the robotic arm (unit: rad); Total motion time (unit: s) ).

[0160] Derivation of core formula:

[0161] 1. Calculation of endpoint coordinates: from the position surface Extract the coordinates of the bottle mouth center:

[0162] (7)

[0163] The preset coordinates of the bottle neck center on the curved surface.

[0164] 2. Speed ​​adaptation: Adjust the speed of the pendulum. Multiply by a safety factor and correct according to the angle between the velocity direction and the direction the bottle opening is facing (the larger the angle, the greater the velocity discount):

[0165] (8)

[0166] Ensure that the direction of the robotic arm's speed is aligned with the direction the bottle opening is facing when it approaches the bottle opening to avoid collision.

[0167] 3. Angle adaptation: Adjust the angular displacement of the pendulum. With the bottle opening facing the curved surface Fusion, by adjusting the coefficients to adapt to the actual rotation range of the robotic arm:

[0168] (9)

[0169] Ensure that the orientation of the robotic arm's end effector is perfectly matched with the bottle opening to achieve seamless docking.

[0170] 2. Model Training Process

[0171] Training data preparation: 400 sets of path planning test data, covering different culture flask types (100mL, 250mL, 500mL) and positions (center, edge, tilted placement). Each set of data includes:

[0172] Input parameters: , , , ;

[0173] Output metrics: Path execution success rate (number of successful dockings / total number of attempts), docking deviation (distance between actual docking location and ideal location, unit: mm).

[0174] Training objective: Optimize , This ensures a path execution success rate of ≥99.5% and a docking deviation of ≤0.2mm.

[0175] Training steps:

[0176] 1. Initialization =0.5、 =0.5, generate path parameters according to formulas (7)-(9). , 2. Control the robotic arm to move along the generated path, and statistically analyze the success rate and docking deviation (initial success rate approximately 90%, deviation approximately 0.5mm). 3. Adjust using a grid search method. (0.6-1.0) (0.6-1.0), retest after each adjustment. 4. Convergence criterion: When =0.9、 When the coefficient of performance is 0.8, the success rate is 99.7% and the deviation is 0.15mm, which meets the requirements, so training should be stopped.

[0177] 3. Model Application Process

[0178] Input: All parameters output by Algorithm 3 and , .

[0179] Calculation process:

[0180] 1. Determine the endpoint of the robotic arm according to formula (7). 2. Real-time calculation (t is the motion time). 3. Calculate the actual speed according to formula (8). Calculate the actual turning angle according to formula (9). 4. Generate a discrete set of path points (including coordinates, velocity, and angle at each time step) with a time step of Δt = 0.02s.

[0181] Output: , , , , (All data is transferred to the robotic arm's transfer execution module).

[0182] Core function: Serving as a bridge between the "theoretical model" and "actual execution," it combines the smooth characteristics of a single pendulum wave with the actual shape of the culture bottle to generate directly executable path parameters. Through dynamic adaptation of speed and angle, it ensures that the robotic arm can move quickly and smoothly while achieving high-precision docking with the bottle mouth (deviation ≤0.2mm).

[0183] Integration and interaction relationships between algorithm modules

[0184] (a) Interaction between surface fitting algorithm (algorithm 1) and transition condition determination algorithm (algorithm 2)

[0185] Interaction logic: Algorithm 1 is the "data preprocessing layer" of Algorithm 2, providing it with high-quality geometric parameters; Algorithm 2 is the "quality verification layer" of Algorithm 1, providing feedback on the reliability of the surface model through the judgment results.

[0186] Specific interaction process:

[0187] 1. Algorithm 1 will , , , , Output to Algorithm 2, where , , It is directly used as the input to F (geometric accuracy determination) in Algorithm 2. , Then, it serves as the basic data for subsequent path planning and is temporarily stored in Algorithm 2. 2. Algorithm 2 uses F in formula (3) to calculate whether the surface fitting quality meets the standard: If > If F=0 and D=0, then we can directly determine that F=0 and D=0, without considering environmental parameters. In this case, the output of Algorithm 1 becomes the key to the analysis of the cause of the anomaly (e.g. (A prompt indicates that sensing data needs to be re-collected). 3. If D=1, Algorithm 2 will... , The model is passed to Algorithm 3 to achieve the connection between the "geometric model" and the "motion model".

[0188] The interactive formula demonstrates that the geometric accuracy determination F of Algorithm 2 depends on the output of Algorithm 1. , The decision function in formula (3) is the output of algorithm 1, which directly determines the value of F.

[0189] (II) Interaction between the single pendulum wave algorithm (Algorithm 3) and the path planning algorithm (Algorithm 4)

[0190] Interaction logic: Algorithm 3 is the "motion reference layer" of Algorithm 4, providing the theoretical curve of smooth motion; Algorithm 4 is the "execution adaptation layer" of Algorithm 3, converting the theoretical curve into parameters that the robotic arm can execute.

[0191] Specific interaction process:

[0192] 1. Algorithm 3 will , , , All outputs are sent to Algorithm 4, where , As in Algorithm 4 , The baseline parameters. 2. Algorithm 4 uses formulas (8) and (9) to combine the pendulum wave parameters with... , Fusion: Velocity Parameters Adapt to bottle neck orientation (corrected by Δθ(t)), angle parameters The orientation of the bottle opening itself (by) (Revised). 3. Algorithm 3 Directly determines the total motion time of Algorithm 4 This ensures that the rhythm of the movement is consistent with the characteristics of a pendulum.

[0193] Interactive formula manifestation:

[0194] The core parameters of Algorithm 4 are directly calculated from the output of Algorithm 3:

[0195] , ;

[0196] Algorithm 3 , This is the theoretical source of the algorithm's four parameters.

[0197] Through the fusion design of the above algorithms, the device can realize the fully automated processing of "continuous sensing data → multi-parameter sterility determination → smooth motion modeling → precise path planning", ultimately achieving sterile, rapid and precise transfer of cell culture flasks.

[0198] VI. Robotic Arm Transfer Execution Module: High-Precision Motion Execution and Status Feedback

[0199] 1. Module hardware configuration and functional adaptation

[0200] The module adopts a hardware architecture of "multi-joint robotic arm + end effector", and the hardware selection and functions are strictly matched to the requirements of sterile, fast and accurate transfer:

[0201] Six-axis collaborative robotic arm (model: ABB YuMi IRB 14000): As the core execution component, it has a repeatability of ±0.02mm, a maximum load of 5kg (meeting the weight requirements of culture bottles), a movement speed range of 0.1~1m / s (which can be dynamically adjusted according to path planning parameters), joint drives using servo motors (with braking function to prevent displacement during power outages), and a surface made of stainless steel (which can withstand high-temperature sterilization).

[0202] Aseptic end effector: Features a dual-mode design of "vacuum adsorption + flexible clamping," adaptable to different sizes of culture flasks (100mL, 250mL, 500mL).

[0203] Vacuum adsorption: The adsorption force is adjusted from 0 to 50 N by a silicone suction cup (food-grade silicone, heat-resistant for sterilization). The adsorption force is set according to the thickness of the bottle, such as 20 N for a 100 mL culture bottle.

[0204] Flexible clamping: The bottle body is clamped from both sides by polyurethane claws (hardness 50 Shore A), with a clamping stroke of 0-80mm (covering the diameter range of culture bottle 50-90mm), avoiding rigid contact that could damage the bottle body;

[0205] Position feedback encoder: Each joint is equipped with a 17-bit absolute encoder to acquire joint angles in real time (resolution 0.001°), and calculate the real-time position of the end effector of the robotic arm (i.e., the output (x(t),y(t),z(t)))) through inverse kinematics, with a sampling frequency of 500Hz;

[0206] Force sensor (model: ATI Nano17): Installed between the end effector of the robotic arm and the actuator, with a measurement range of ±100N (force) and ±5Nm (torque), used to detect the contact force during docking (if the force exceeds 5N when docking the bottle mouth, the protection is triggered to avoid damaging the bottle mouth).

[0207] 2. Action execution process and aseptic assurance

[0208] To ensure that the sterile environment and culture flasks are not damaged during execution, the module is equipped with standardized execution procedures and multiple protection mechanisms:

[0209] 1. Preparations before execution:

[0210] Aseptic treatment: The end effector (suction cup, gripper) is sterilized by high-pressure steam at 121℃ for 20 minutes, and the joints and surfaces of the robotic arm are wiped and disinfected with 75% ethanol to ensure no microbial residue;

[0211] Parameter loading: Receives the output of the path planning algorithm module. (starting point), (end), (Actual speed) (From a practical perspective) (Total motion time) The parameters are parsed (this parsing method is a conventional existing technology, which this invention will not improve; refer to the conventional existing technology or the example below) into robotic arm joint motion commands (such as the angle change curve of each joint). After receiving the commands, parameter verification is performed first, such as determining whether the start / end point is within the robotic arm's motion range (e.g., if the arm length is 350mm, the end point distance should not exceed 350mm) and whether the speed exceeds the robotic arm's maximum safe speed (e.g., ≤50mm / s, to avoid impact). If the parameters are abnormal, feedback is sent to the main control module to trigger a "stop transfer command". Parameter parsing: From "spatial parameters" to "joint commands", the core is to convert "Cartesian space motion parameters" (position, speed, angle) into "robotic arm joint space commands" (angle change of each joint). Taking a common 6-axis robotic arm as an example:

[0212] First, calculate using inverse kinematics: based on the robotic arm's joint structure (such as joint length, joint type, rotary / translational joints), determine the starting point... ,end The corresponding spatial pose is broken down into the initial angles of 6 joints. and target angle ; and then combine and Generate "Joint Angle Change Curve": based on total motion time Divide the time into time steps (e.g., 0.02s / step), and calculate the angle values ​​that the six joints need to reach at each time point (e.g., from the initial angle). To the target angle According to speed (corresponding to uniform rate changes), ultimately forming a continuous angle change curve for each joint.

[0213] After parsing, the "joint angle change curve" is converted into a format recognizable by the robotic arm controller (such as pulse signals or Modbus protocol commands), and compared with the total motion time. Synchronization—ensuring all joints move at the same tempo (e.g., joint 1 moves from 0° to 30°, joint 2 moves from 10° to 45°, all within the same time frame). =Completed within 3 seconds), avoiding deviation of the robotic arm end from the target path, and providing a basis for subsequent precise docking of culture bottles.

[0214] Zero-point reset: The robotic arm automatically returns to its starting point. The end position is calibrated by the encoder to ensure that the starting point accuracy is ≤0.01mm;

[0215] 2. Execution of the handover action (in three stages):

[0216] Approaching phase ( ): Robotic arm press From the starting point Move toward the culture flask, gradually increasing the speed from 0 to the maximum value, while the force sensor monitors the contact force in real time (at which point the contact force should be 0).

[0217] docking phase ( When the end of the robotic arm approaches the culture flask, the actuator switches to the "vacuum adsorption + flexible clamping" mode. The position of the end is finely adjusted by the position feedback encoder (deviation ≤ 0.02mm) to ensure that the bottle mouth and the adapter interface are precisely aligned. The force sensor monitors the contact force (controlled between 1 and 3N to avoid excessive force).

[0218] Transition phase ( ): Robotic arm press Transfer the culture flask from its initial position to the target position (such as a culture rack inside a biosafety cabinet) at a speed that is gradually reduced from its maximum value to 0 to avoid shaking of the culture medium due to the impact of stopping the machine.

[0219] 3. Post-execution confirmation: The robotic arm has reached the endpoint. Afterwards, the actuator releases the culture flask (closes the vacuum, releases the grippers), and the position feedback encoder confirms the end-effector position. If the deviation is ≤0.02mm, the action is considered complete. The force sensor confirms that there is no residual clamping force (if it is 0, the release is considered successful).

[0220] 3. Status Feedback and Exception Handling

[0221] The module not only executes actions, but also needs to report the execution status to the central control module in real time and respond quickly to abnormal situations.

[0222] Real-time status feedback content:

[0223] Motion completion signal (complete, incomplete, abnormal interruption): The motion execution has reached... If the positional deviation is ≤0.02mm, output "Completed"; otherwise, output "Not Completed". If the position deviation is greater than 0.02mm, the output will be "Incomplete"; if the contact force exceeds 5N or the encoder malfunctions, the output will be "Abnormal Interruption".

[0224] Real-time position feedback: Outputs the three-dimensional coordinates (x(t), y(t), z(t)) (unit: mm) of the robotic arm's end effector at a frequency of 500Hz, compares them with the theoretical position planned by the path, and calculates the deviation. , , Synchronous feedback of deviation value;

[0225] Key component status: Output actuator suction force, clamping force, joint temperature and other parameters (e.g., suction force <15N indicates "insufficient suction force");

[0226] Abnormal handling mechanism: Contact force exceeds threshold (>5N): Immediately stop the movement of the robotic arm, output "abnormal interruption" signal, and trigger an alarm in the main control module; Position deviation exceeds threshold (>0.05mm): Automatically start deviation correction (compensate for deviation through joint fine-tuning). If the deviation still exceeds the threshold after 3 corrections, output "abnormal interruption"; Sensor failure (e.g., no encoder signal): Immediately cut off the power to the robotic arm, lock the joint to avoid loss of control, output "abnormal interruption" and indicate the faulty component.

[0227] 4. Core Function of the Module

[0228] As a "physical action executor," it transforms the theoretical parameters of the path planning algorithm into actual transfer actions, ensuring the precise transfer of culture bottles through a high-precision robotic arm and end effector; as a "sterility guardian," it avoids microbial contamination and damage to culture bottles during transfer through mechanisms such as aseptic processing, flexible execution, and contact force control; and as a "status feedback node," it transmits the execution status to the central control module in real time, providing a basis for global monitoring and anomaly handling.

[0229] VII. Transfer Process Control Module: Full-Process Coordination and Monitoring

[0230] 1. Module Hardware and Software Composition

[0231] The module adopts an architecture of "industrial controller + human-machine interface", and has powerful logic operation, data storage and instruction output capabilities:

[0232] Hardware components:

[0233] Industrial controller (model: Siemens S7-1500 PLC): CPU clock speed 1.5GHz, memory 1GB, supports multi-protocol communication (Profinet, Modbus, EtherNet / IP), can establish data connection with sensing, algorithm and execution modules at the same time, and the operation cycle is ≤1ms;

[0234] Human-Machine Interface (HMI, model: Weinview MT8150iE): 15-inch touchscreen with a resolution of 1920×1080, supports Chinese / English switching, can display the status of each module, parameter curves, and alarm information in real time, and supports manual operation (such as emergency stop and parameter setting).

[0235] Alarm device: Includes an audible and visual alarm (red LED light + buzzer, alarm volume adjustable from 80-120dB) and fault indicator lights (classified by abnormality type: environmental abnormality, algorithm abnormality, execution abnormality), installed in a conspicuous position on the top of the device;

[0236] Data storage module: SD card (capacity greater than or equal to 32GB), used to store historical transfer data (including parameters, judgment results, and execution status of each transfer), with a storage period of ≥1 year, and supports data export (USB interface).

[0237] Software components:

[0238] Data acquisition and parsing module: Receives output data from each module in real time, parses it into a unified format (such as JSON format), and stores it in the database;

[0239] Logic control module: Based on the result of the transfer condition judgment and the execution status feedback, output control commands (start alarm, stop transfer, complete transfer) according to preset logic.

[0240] Status display module: Dynamically displays the operating status of each module on the HMI (such as "sampling in" and "sampling completed" for the sensing module, and "standby" and "moving" for the execution module), as well as key parameter curves (such as temperature change curve over time and robotic arm position deviation curve).

[0241] Alarm and Log Module: Records abnormal events (time, abnormality type, and handling measures), and displays the cause of the abnormality and the solution when an alarm is triggered (such as prompting "start ultraviolet disinfection" when "microbial concentration exceeds the standard").

[0242] 2. End-to-end coordination and control logic

[0243] The modules coordinate their operation in four stages: "preprocessing - decision-making - execution - cleanup," forming a closed-loop control.

[0244] 1. Pretreatment stage (0-10s):

[0245] The central control module sends a "start sampling" command to the sensing module, and the sensing module starts the sensor to collect data; the central control module sends a "data reception preparation" command to the surface fitting algorithm module, and the algorithm module enters the data waiting state; the HMI displays "preprocessing in progress" and updates the sensor warm-up progress and the number of sampling points in real time;

[0246] 2. Judgment Phase (10-15 seconds):

[0247] After the sensing module completes sampling, it sends a "sampling complete" signal to the central control module, which then forwards the data to the surface fitting algorithm module. After completing the fitting, the surface fitting algorithm module... , , The parameters are sent to the central control module, which then forwards them to the transfer condition judgment algorithm module. The judgment algorithm module outputs D (judgment result) and Ab (abnormal parameters). Upon receiving these, the central control module: if D=1 (condition met): sends a "start motion modeling" command to the single-pendulum wave algorithm module, and the HMI displays "Judgment passed, prepare path planning"; if D=0 (condition not met): triggers an audible and visual alarm, and the HMI displays abnormal parameters (such as "..."). If the microbial concentration exceeds the standard (20 CFU / m³), send a "Stop Transfer" command to all modules.

[0248] 3. Implementation phase (15~15+) s):

[0249] After the single pendulum wave algorithm module completes motion modeling, it will... , The data is sent to the central control module, which then forwards it to the path planning algorithm module; the path planning algorithm module generates the path. , , , The data is sent to the central control module, which then forwards it to the execution module. The execution module initiates the transfer action and provides real-time feedback of "motion completion signal" and "real-time position feedback" to the central control module. The central control module displays the robotic arm's motion trajectory and position deviation (e.g., "current deviation: 0.01mm") on the HMI. If the execution module reports an "abnormal interruption," the central control module immediately triggers an alarm, sends a "stop transfer" command, and records the abnormality log.

[0250] 4. Final stage (15+) -20+ s):

[0251] After the execution module reports "Motion Completed," the central control module sends a "Release Culture Bottle" command, and the execution module completes the release. The central control module sends a "Resample" command to the sensing module, which collects the status of the culture bottle at the target location (confirming accurate location) and sends a "Position Normal" signal. The central control module outputs a "Transfer Completed" command, the HMI displays "Transfer Successful," and records the transfer data (time, culture bottle type, parameters of each module) to the SD card. All modules return to standby mode, waiting for the next transfer command.

[0252] 3. Human-computer interaction and data management functions

[0253] To improve the ease of use and traceability of the device, the module enhances human-computer interaction and data management capabilities:

[0254] Manual operation functions: Emergency stop: A red "emergency stop" button is set on the HMI. Pressing it immediately cuts off power to all modules, suitable for sudden failures; Parameter settings: Supports manual adjustment of the number of sensor sampling points m, the robotic arm swing length L, and the judgment threshold (e.g., However, an administrator password is required (to avoid accidental operation); Manual control: Supports manual control of the robotic arm movement via the HMI joystick (only enabled in "Debug Mode" to avoid interfering with normal transfer);

[0255] Data Management Functions: Historical Data Query: Supports querying historical data by time (year, month, day), transfer result (success / failure), and culture flask type, displaying detailed parameters (e.g., transfer at 10:00 AM on May 10, 20XX, culture flask type 100mL, T=37.0℃, D=1); Data Export: Supports exporting historical data to Excel format (including parameter name, value, unit, and acquisition time), and transferring it to a computer via USB interface; Fault Statistics: Automatically counts the occurrence frequency of each type of anomaly (e.g., ... (Temperature exceeding the limit) occurred 5 times. (If the pressure abnormality occurs 3 times), a monthly statistical report will be generated for easy maintenance.

[0256] 4. Core Function of the Module

[0257] As a "global coordinator," it unifies the operation rhythm of the sensing, algorithm, and execution modules, ensuring that each module works in sequence and avoiding process chaos; as an "anomaly handling center," it receives anomaly feedback from each module in real time, and quickly controls risks and reduces losses through alarms and stop commands; as a "human-machine interaction window," it provides operators with intuitive status display, convenient manual operation, and complete data traceability, improving the ease of use and reliability of the device.

[0258] Full-module collaboration and data chain closed loop

[0259] (a) Physical connection and communication protocol between modules

[0260] To ensure the real-time performance and reliability of data transmission between modules, the system adopts standardized physical connections and communication protocols: Sensing module → Surface fitting algorithm module: Industrial Ethernet (Profinet protocol), transmission rate 100Mbps, latency ≤10ms, transmitting discrete sampling points and environmental parameters; Surface fitting algorithm module → Switching condition determination algorithm module: Industrial Ethernet (Profinet protocol), transmission rate 100Mbps, latency ≤10ms, transmitting the surface model and fitting error; Switching condition determination algorithm module → Single pendulum wave algorithm module / central control module: Dual-link communication (Profinet + Modbus), Profinet transmits determination results and the surface model (to the single pendulum wave module), Modbus transmission is interrupted. Parameters (to the main control module), delay ≤10ms; Single pendulum wave algorithm module ↔ Path planning algorithm module: Industrial Ethernet (Profinet protocol), transmission rate 100Mbps, delay ≤5ms (requires fast response to motion parameters), transmission content is single pendulum wave motion curve; Path planning algorithm module ↔ Execution module: Industrial Ethernet (EtherNet / IP protocol, adapted for robotic arm communication), transmission rate 100Mbps, delay ≤5ms, transmission content is path parameters; Execution module / judgment module ↔ Main control module: Industrial Ethernet (Profinet protocol) + hardwire connection (emergency stop signal), Profinet transmits status feedback and control commands, hardwire connection ensures no delay in emergency stop signal, transmission rate 100Mbps, delay ≤1ms.

[0261] (II) Examples of Data Chain Closed Loop and Collaboration

[0262] Taking "transfer of 100mL culture flask" as an example, the process of full-module collaboration and data chain closed loop is demonstrated:

[0263] 1. Sensing Phase: The sensing module activates 3 laser displacement sensors (sampling points m=6), 1 angle sensor, 1 temperature and humidity sensor, 1 particle counter, and 1 pressure sensor to collect data. (mm)... (mm) (°)... (°), T=37.0℃, H=50%RH, C=8CFU / m³, =100.0 kPa, the data is transmitted to the surface fitting algorithm module via Profinet;

[0264] 2. Preprocessing stage: The surface fitting algorithm module inputs the bicubic B-spline model to calculate... (Position surface) (Towards the curved surface) (≤) =0.3), Maximum value 0.3mm (≤ =0.5), Maximum value 0.8° (≤ =1.0), the data is transmitted to the judgment module;

[0265] 3. Judgment Phase: The judgment module calculates F=1 (geometric accuracy meets standards), En=1 (environmental parameters meet standards), D=1, and there is no abnormal parameter Ab. Then, D=1, , The data is transmitted to the single-pendulum wave module and the main control module, and the main control module HMI displays "Judgment passed";

[0266] 4. Motion modeling stage: The single pendulum wave module is preset with L=300mm, from... extract , ,calculate =80mm, ≈1.1s, (rad), (mm / s), data is transmitted to the path planning module;

[0267] 5. Path planning stage: The path planning module calculates... (Approximately 0.1 rad) (mm / s) ,generate , , =0.275s, data is transmitted to the execution module;

[0268] 6. Execution Phase: The execution module parses the path parameters and drives the robotic arm to... The system provides real-time feedback on movement, including "In Motion," "Current Position (55, 105, 162.5) mm," and "Deviation 0.01 mm." The central control module (HMI) displays the trajectory.

[0269] 7. Finalization Phase: The execution module arrives. The system displays "Motion complete" and releases the culture flask. The sensing module resamples to confirm the position is normal, the main control module outputs "Transfer complete", the data is recorded to the SD card, and the HMI displays "Transfer successful".

[0270] System core advantages and application scenarios

[0271] (I) Core Advantages

[0272] 1. Asepticity Guarantee: A three-tiered aseptic guarantee system of "prevention-monitoring-control" is constructed through smooth motion of the single-pendulum wave algorithm (avoiding airflow disturbance), aseptic processing of the execution module (autoclaving, ethanol disinfection), and microbial concentration monitoring of the sensing module; 2. Improved Speed: Total transfer time (including pretreatment, judgment, and execution) ≤ 20s (e.g., for a 100mL culture flask). =0.275s (18s for the entire process), much faster than manual transfer (approximately 60s); 3. Precision control: the robotic arm's repeatability is ±0.02mm, and the bottle mouth docking deviation is ≤0.02mm, avoiding contamination or damage caused by misalignment; 4. Intelligent collaboration: each module collaborates deeply through the data chain, automatically completing sampling, judgment, and execution without manual intervention, reducing human error; 5. Strong traceability: the central control module stores all transfer data, supports querying and exporting, facilitating quality traceability and fault analysis.

[0273] (II) Application Scenarios

[0274] 1. Biopharmaceutical field: Used for transferring cell culture flasks during vaccine and antibody production to ensure the sterile environment is not compromised and improve production efficiency; 2. Cell therapy field: Used for transferring flasks during CAR-T cell and stem cell culture to avoid cell contamination and ensure cell viability; 3. Scientific research field: Used for automated transfer of laboratory cell cultures to reduce repetitive operations by researchers and improve experimental reproducibility.

[0275] Through optimized design and deep collaboration across all modules, this device, a sterile rapid transfer device for cell culture flasks, achieves the core requirements of "sterility, speed, precision, and intelligence," providing reliable technical theoretical support for the automation upgrade of the cell culture process.

[0276] In some implementations, to improve the system's adaptability to complex culture bottle shapes and dynamic environments, the Gradient Boosting Decision Tree (GBDT) algorithm is introduced as the core of collaborative optimization, forming a closed-loop fusion mechanism of data-correction-feedback with surface fitting and pendulum wave algorithms. That is, the device of this invention also includes a gradient boosting tree algorithm module. The data acquisition-preprocessing optimization process (sensing → surface fitting → GBDT) is as follows: Figure 2 As shown, the motion parameter optimization process (surface fitting → pendulum wave → GBDT) is as follows: Figure 3 As shown, the feedback iteration closed-loop process (execution → GBDT → next optimization) is as follows: Figure 4 As shown.

[0277] The core interaction relationship among the three is as follows:

[0278] 1. GBDT → Surface Fitting: By learning the mapping relationship between historical fitting errors and culture bottle features, the dynamic weights are output to correct the surface fitting model, reducing fitting bias in edge regions; 2. Surface Fitting → Pendulum Wave: The optimized surface model provides accurate geometric parameters of the culture bottle (curvature, bottle mouth orientation) as constraints on the pendulum motion trajectory; 3. Pendulum Wave → GBDT: The actual deviation data after the robotic arm performs the pendulum motion is fed back to GBDT for model iterative training, improving the accuracy of subsequent optimization; 4. GBDT → Pendulum Wave: Based on historical motion smoothness data, the optimal adjustment coefficient of the pendulum wave parameters is predicted to balance motion speed and aseptic environment stability.

[0279] Gradient boosting tree algorithm module:

[0280] I. Gradient Boosting Tree (GBDT) Model Construction and Training

[0281] 1. Definition of Model Input Features and Output Target

[0282] Input feature set : Culture flask specifications (discrete values: 100mL=1, 250mL=2, 500mL=3); Total number of sampling points m (6 / 8 / 10); Mean position deviation ( (where the deviation between the k-th discrete point and the initial surface is denoted as ). Maximum positional deviation ; Average orientation deviation ( (where k is the deviation between the k-th discrete point and the initial surface). Maximum orientation deviation ; Ambient temperature T; Ambient humidity (H); Microbial concentration C; Initial value of pendulum amplitude ; Initial value of the pendulum period ; Actual motion deviation of the robotic arm (Historical data);

[0283] Output target set : Surface fitting weight correction coefficient (Used to adjust the weights of discrete points in the fitting) Adjustment coefficient for single pendulum wave parameters (Used to correct amplitude and period).

[0284] 2. Model Training Process

[0285] 1. Training Data Collection: Collect 1000 sets of historical data under different culture flask specifications and environmental conditions. Each set includes input features. Initial fitting error Initial motion deviation and the optimal correction factor for manual annotation , ;

[0286] 2. Loss function definition: The mean squared error (MSE) is used as the loss function.

[0287]

[0288] Where n is the number of samples. , The predicted value of GBDT for the i-th sample (corresponding to) , );

[0289] 3. Model Training:

[0290] Initialize weak learner (Constant model); Iteratively generate 100 decision trees: Calculate the negative gradient (residual): Fitting residuals using decision trees , obtain the m-th tree Calculate the learning rate Update the model: Final model: ;

[0291] 4. Model Validation: Validate using 20% ​​of the test set, requiring a minimum prediction error. and Otherwise, increase the number of trees and retrain.

[0292] II. Integration and Optimization with Surface Fitting Algorithms (GBDT → Surface Fitting)

[0293] 1. Initial surface fitting model (bicubic B-spline)

[0294] Let the original discrete points be... (k=1,2,...,m), the initial surface model is:

[0295]

[0296] For parameter coordinates; The basis functions are cubic B-spline functions. To control the vertex matrix;

[0297] Initial fitting error calculation: (Sum of squares of positional deviation and orientation deviation);

[0298] 2. GBDT-corrected surface fitting model

[0299] GBDT Output Weight Correction Coefficient (Range 0.8-1.2), dynamically adjust the fitting weights for each discrete point:

[0300] The corrected weight of the k-th point:

[0301] The historical standard deviation is 0.2 mm² + °².

[0302] Corrected surface model:

[0303]

[0304] Corrected fitting error: ;

[0305] 3. Interaction Process Description

[0306] 1. The surface fitting algorithm first calculates the initial surface based on equal weights. and initial error 2. Combine initial error and input features Input GBDT, get 3. Surface fitting algorithm Recalculate the weights of each point Generate optimized surfaces 4. Optimized and The data is transmitted to the switching condition determination module and the single pendulum wave algorithm module.

[0307] Core function: By dynamically adjusting the weights of GBDT, the impact of high deviation points on the overall surface (such as sampling errors caused by reflections at the edge of the culture bottle) is reduced, making the surface model more closely resemble the actual shape and reducing the average fitting error.

[0308] III. Integration and Optimization with Single Pendulum Wave Algorithm (Surface Fitting → Single Pendulum Wave → GBDT)

[0309] 1. Initial Calculation of Simple Pendulum Wave Parameters Based on Optimized Surface

[0310] The single pendulum wave algorithm optimizes the surface Extract key parameters:

[0311] Starting coordinates and endpoint coordinates Straight-line distance: ;

[0312] Curvature values ​​at the point of maximum curvature on the path: ( , (First and second derivatives of the surface); initial amplitude: (The greater the curvature, the greater the amplitude reduction, thus avoiding collisions); Initial period: (L is the equivalent pendulum length of the robotic arm, and g = 9.8 m / s² is the acceleration due to gravity).

[0313] 2. GBDT-corrected parameters of a single pendulum wave

[0314] GBDT output parameter adjustment coefficient (Range 0.9-1.1), corrected amplitude and period:

[0315] Corrected amplitude: (C represents the microbial concentration,) (The threshold value is used; the higher the concentration, the smaller the amplitude to reduce airflow). Corrected period: (Opposite to amplitude correction, to avoid velocity fluctuations); Corrected angular displacement curve: Corrected linear velocity curve: ;

[0316] 3. Feedback Iteration Mechanism (Single Pendulum Wave → GBDT)

[0317] After the robotic arm performs the movement, the path planning algorithm calculates the actual movement deviation:

[0318]

[0319] Let x(t), y(t), and z(t) represent the total motion time, and x(t), y(t), and z(t) represent the actual positions.

[0320] Will The input set of GBDT is supplemented as a new feature. Retrain the model:

[0321]

[0322] If the actual deviation is positive, increase If negative, decrease. ;

[0323] 4. Interaction Process Description

[0324] 1. The single pendulum wave algorithm is based on optimized surfaces. Calculate initial parameters , 2. Set initial parameters and input features Input GBDT, get 3. Single pendulum wave algorithm Adjust parameters to generate the final motion curve. , 4. After the motion is executed, calculate This information is then fed back to GBDT to update the model parameters.

[0325] Core function: By constraining the safety of the pendulum trajectory through surface parameters, and utilizing the speed and smoothness of GBDT dynamic balancing, it automatically reduces motion disturbances and decreases actual motion deviations in scenarios with high microbial concentrations. When the system continuously meets the three conditions of error convergence, anomaly stability, and parameter invariance through closed-loop iteration, it signifies that the optimization closed loop can end and the system enters a stable operating state.

[0326] IV. Collaborative Workflow and Core Contributions of the Three Algorithms

[0327] 1. Complete Collaboration Process

[0328] 1. Data Acquisition: The sensing module outputs discrete points. , and environmental parameters; 2. Initial fitting: generated by surface fitting algorithm , 3. GBDT First Optimization: Input and Output 4. Optimized Fitting: Surface Fitting Algorithm Generation , 5. Initial pendulum parameters: The pendulum wave algorithm is based on... calculate , 6. GBDT Second Optimization: Input and Output 7. Optimize motion curve: Generated by single pendulum wave algorithm , 8. Execution and Feedback: The robotic arm executes movements and calculates... And update the GBDT model.

[0329] 2. Core Technology Contributions

[0330] 1. Dynamic Adaptability: GBDT learns the optimal parameter patterns for different scenarios through historical data, enabling the system to adapt to various sizes (e.g., 100-500mL) of culture flasks and complex environments without manual adjustment; 2. Error Chain Closed-Loop Control: A complete feedback chain is formed from fitting error to motion deviation. The model automatically iterates after each transfer, improving accuracy over long-term use; 3. Sterility-Efficiency Balance: Through the coupling correction of microbial concentration and curvature, the transfer time is controlled within the optimal range while ensuring a sterile environment, resolving the contradiction of traditional equipment where speed leads to contamination and stability leads to inefficiency.

[0331] Through the deep integration of the three algorithms, the system achieves full-link intelligent control of perception accuracy, model optimization, and motion control, providing core algorithm support that is both reliable and adaptable for cell culture flask transfer.

Claims

1. A sterile rapid transfer device for cell culture flasks, characterized in that, include: The culture bottle status and aseptic environment sensing module is used to collect discrete data on the outer surface of the culture bottle, the direction of the bottle opening, the ambient temperature and humidity, the concentration of microorganisms, and the pressure inside the culture bottle. The surface fitting algorithm module receives discrete points of position and orientation, transforms them into continuous surface models of the outer surface of the culture bottle and the orientation of the bottle opening, and calculates the deviation between each discrete point and the corresponding surface and the total fitting error. The transition condition determination algorithm module is used to receive position surface model, orientation surface model, deviation of each discrete point, total fitting error, and environmental temperature and humidity, microbial concentration, and pressure data in culture flask, and determine whether the current state meets the transition conditions. The pendulum wave algorithm module is used to receive the position surface model and the orientation surface model when the judgment result meets the transfer conditions, and simulate the pendulum motion to generate motion data of the robotic arm. The robotic arm transition path planning algorithm module is used to receive motion data and position surface model and orientation surface model, and generate control commands for the robotic arm. The robotic arm transfer execution module is used to receive control commands, drive the robotic arm to perform transfer actions, and provide real-time feedback on motion completion signals, real-time position, and the status of key components. The overall control module for the transfer process is used to receive information from each module, output control commands, and coordinate the operation of the entire process.

2. The aseptic rapid transfer device for cell culture flasks according to claim 1, characterized in that, The deviations of discrete points at each position and at each orientation calculated by the surface fitting algorithm module are transmitted in real time to the transition condition determination algorithm module. During the determination process, if any discrete point deviation at any position exceeds a preset threshold for a single position deviation or any discrete point deviation at any orientation exceeds a preset threshold for a single orientation deviation, the discrete point corresponding to the deviation is directly marked as an anomaly point, and the discrete point deviation exceeding the standard is included in the set of abnormal parameters. The determination of this type of anomaly is not based on the total fitting error alone.

3. The aseptic rapid transfer device for cell culture flasks according to claim 1, characterized in that, The surface fitting algorithm module uses a bicubic B-spline surface model, and the node spacing of its basis functions is the optimal value determined during the model training phase—the node spacing in the position dimension is 8 mm and the node spacing in the orientation dimension is 3 mm. The node spacing parameters are stored in the parameter library of the surface fitting algorithm module and are automatically loaded each time the fitting calculation is started without manual adjustment. Furthermore, if a change in the size of the culture bottle is detected during the fitting process, the pre-trained node spacing parameters of the corresponding size culture bottle are automatically called.

4. The aseptic rapid transfer device for cell culture flasks according to claim 1, characterized in that, The judgment logic of the transition condition judgment algorithm module is as follows: when the total fitting error, the deviation of discrete points at each position, and the deviation of discrete points at each orientation do not exceed the preset threshold, and the ambient temperature and humidity, microbial concentration, and pressure inside the culture bottle are all within the preset range, it is determined that the transition condition is met; otherwise, the corresponding abnormal parameters are output.

5. The aseptic rapid transfer device for cell culture flasks according to claim 1, characterized in that, In the pendulum motion simulated by the pendulum wave algorithm module, the pendulum amplitude is the straight-line distance from the starting point to the ending point of the robotic arm, the pendulum period is determined by the equivalent pendulum length of the robotic arm and the gravitational acceleration, and the maximum pendulum angle does not exceed 0.3 rad to ensure smooth motion.

6. The aseptic rapid transfer device for cell culture flasks according to claim 1, characterized in that, When the robotic arm transfer path planning algorithm module generates the actual movement speed, it multiplies the cycloidal speed by a preset speed safety factor and corrects it according to the angle between the speed direction and the bottle opening direction. When generating the actual turning angle, the pendulum angular displacement is fused with the bottle opening facing the curved surface model using a preset angle adjustment coefficient.

7. The aseptic rapid transfer device for cell culture flasks according to claim 1, characterized in that, After generating the linear velocity curve of the robotic arm's motion using the single pendulum wave algorithm module, the first derivative of the linear velocity is calculated to obtain the acceleration curve, which is then synchronously transmitted to the robotic arm transition path planning algorithm module. When generating the actual motion speed, the robotic arm transition path planning algorithm module sets an upper limit for acceleration based on the acceleration curve to ensure that the actual motion acceleration of the robotic arm does not exceed the preset maximum allowable acceleration of the robotic arm, thus avoiding vibration disturbance to the sterile environment caused by excessive acceleration of the robotic arm.

8. The aseptic rapid transfer device for cell culture flasks according to claim 4, characterized in that, If the transfer condition determination algorithm module determines that the transfer conditions are not met and outputs an abnormal parameter set, it synchronously feeds back the abnormal parameter set to the surface fitting algorithm module and the culture bottle status and sterile environment sensing module. If the abnormal parameter is that the total fitting error exceeds the standard, the surface fitting algorithm module reloads the discrete point data and adjusts the node interval of the B-spline basis function, and re-executes the fitting calculation. If the abnormal parameter is that the environmental parameter exceeds the standard, the culture bottle status and sterile environment sensing module re-collects environmental data at intervals of 5 to 10 seconds until the environmental parameter meets the preset range or the transfer process control module outputs a stop transfer command.

9. The aseptic rapid transfer device for cell culture flasks according to claim 5, characterized in that, When determining the pendulum amplitude, the single pendulum wave algorithm module first extracts the maximum curvature point on the outer surface of the culture bottle from the position surface model, calculates the straight-line distance from the starting point of the robotic arm to the maximum curvature point, and then reduces the straight-line distance by 10% to 15% as the final single pendulum amplitude. This ensures that the robotic arm's motion trajectory avoids the protruding area on the outer surface of the culture bottle. Furthermore, the single pendulum amplitude adjustment process is based on the orientation surface model verification to ensure that the adjusted amplitude does not change the relative angle between the end of the robotic arm and the bottle opening.

10. The aseptic rapid transfer device for cell culture flasks according to claim 6, characterized in that, When generating the actual motion speed and actual turning angle, the robotic arm transition path planning algorithm module discretizes the continuous pendulum motion curve into parameter values ​​for multiple time nodes with a time step of 0.02 seconds. The actual motion speed at each time node must match the actual turning angle of that node. When the actual turning angle is greater than a preset angle threshold, the actual motion speed of the corresponding node is automatically reduced. The reduction is positively correlated with the turning angle to ensure smooth movement of the robotic arm when turning.

Images