An Adaptive Control Method for Eye Drop Processing Equipment Based on Digital Twin

Abstract

This invention relates to the field of industrial control system technology, specifically to an adaptive control method for eye drop processing equipment based on digital twins. The method includes acquiring the transient pressure decay sequence of the eye drop filling pipeline, the braking displacement sequence of the filling pump rotor, and real-time temperature data; extracting the rheological relaxation time constant based on the pressure sequence, analyzing the transient shear rate matrix inside the pipeline based on the displacement sequence, and determining the dynamic viscosity reference parameter based on the temperature data; inputting the above parameters into a digital twin model, mapping and outputting the dynamic evolution time sequence of the first normal stress difference, thereby predicting the liquid bridge fracture time and analyzing the residual normal tensile stress at the needle tip; analyzing the reverse suction momentum compensation parameter based on the residual normal tensile stress and the predicted fracture time, generating an asymmetric suction control pulse sequence, and finally converting it into servo drive commands and sending them to the control unit. This invention achieves adaptive control through digital twin modeling.

Description

Technical Field

[0001] This invention relates to the field of industrial control system technology, specifically to an adaptive control method for eye drop processing equipment based on digital twins. Background Technology

[0002] In the field of precision motion control, high-precision adjustment of actuators during transient processes is one of the core tasks of automated control systems. Current control schemes typically employ high-performance servo drives in conjunction with position / velocity closed-loop feedback to drive actuators to complete complex physical tasks through preset motion trajectory planning.

[0003] As control systems develop towards ultra-precision and high dynamic response, the nonlinear coupling relationship between the physical properties of the controlled object and its motion state has become a bottleneck limiting further improvements in control accuracy. During the transient phase when the actuator switches from high-speed operation to braking and stopping, there is often a lag in the evolution of the physical state within the controlled object (such as stress release, energy dissipation, or inertial fluctuations). This lag phenomenon is not completely synchronized with the mechanical execution command in the time domain.

[0004] Traditional closed-loop control systems primarily rely on mechanical displacement or velocity feedback for regulation, making it difficult to detect complex transient physical parameter shifts within the controlled object. When the external environment or the physical characteristics of the controlled object drift, the lack of deep coupling regulation of the internal physical evolution process means that the pulse commands generated by the control system cannot accurately cancel the nonlinear energy residue in the transient process. Consequently, it becomes difficult to maintain control consistency and dynamic stability under different operating conditions.

[0005] The technical problem to be solved by this application is how to reduce end-efficiency deviation caused by mismatch between control commands and physical state during the dynamic switching process of the actuator.

[0006] To address this, an adaptive control method for eye drop processing equipment based on digital twins is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide an adaptive control method for eye drop processing equipment based on digital twins. Adaptive control is achieved through digital twin modeling. This includes acquiring the transient pressure decay sequence of the eye drop filling pipeline, the braking displacement sequence of the filling pump rotor, and real-time temperature data; extracting the rheological relaxation time constant from the pressure sequence, analyzing the transient shear rate matrix inside the pipeline based on the displacement sequence, and determining the dynamic viscosity reference parameter based on the temperature data; inputting the above parameters into the digital twin model, mapping and outputting the dynamic evolution time sequence of the first normal stress difference, thereby predicting the liquid bridge fracture time and analyzing the residual normal tensile stress at the needle tip; analyzing the reverse suction momentum compensation parameter based on the residual normal tensile stress and the predicted fracture time, generating an asymmetric suction control pulse sequence, and finally converting it into servo drive commands and sending them to the control unit.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] An adaptive control method for an eye drop processing device based on digital twins, comprising:

[0010] The transient pressure decay time series of the eye drop filling pipeline was obtained, as well as the braking displacement sequence of the filling pump rotor and the real-time temperature data of the medicine solution.

[0011] Based on the transient pressure decay time series, the rheological relaxation time constant is extracted; combined with the braking displacement sequence, the transient shear rate matrix inside the pipeline during the shutdown phase is analyzed, and the dynamic viscosity reference parameter of the current batch of drug solution is determined based on real-time temperature data.

[0012] The rheological relaxation time constant, transient shear rate matrix and dynamic viscosity reference parameters are input into the digital twin model, and the dynamic evolution time sequence of the first normal stress difference is mapped and output. The corresponding predicted liquid bridge fracture time is extracted based on the dynamic evolution time sequence of the first normal stress difference, and the residual normal tensile stress at the tip of the needle is analyzed.

[0013] Based on the residual normal tensile stress and the predicted liquid bridge fracture time, the reverse backflow momentum compensation parameters are analyzed; based on the reverse backflow momentum compensation parameters, an asymmetric backflow control pulse sequence is generated, which includes trigger delay time, starting acceleration scalar and braking pulse width.

[0014] The asymmetric backflow control pulse sequence is converted into servo drive commands and sent to the control unit for adaptive control.

[0015] Preferably, the process of acquiring data from the eye drop filling tubing includes: using a sterile, isolated pressure sensor located at the end of the eye drop filling tubing, after receiving a shutdown trigger signal from the filling pump, collecting transient fluid dynamics fluctuation characteristics within the tubing at a preset high-frequency sampling rate to construct a transient pressure decay time series; using an absolute servo encoder coupled to the filling pump drive shaft, synchronously collecting angular displacement decay pulses from the moment the shutdown trigger signal is issued until the drive shaft is completely locked and stationary to construct a braking displacement sequence; and using a non-contact infrared temperature measurement component attached to the outside of the eye drop filling tubing, collecting the thermal radiation parameters of the tubing wall in real time, and mapping the thermal radiation parameters to the real-time temperature data of the liquid.

[0016] Preferably, the process of extracting the rheological relaxation time constant includes: extracting the initial pressure peak value at the moment the filling pump stops and the steady-state pressure value of the substrate after the fluid has completely stopped in the transient pressure decay time series; calculating the transient pressure difference between the initial pressure peak value and the steady-state pressure value of the substrate; locating the target time node corresponding to when the actual pressure decay reaches the preset proportional threshold of the transient pressure difference along the time axis of the transient pressure decay time series, wherein the preset proportional threshold is set based on the natural exponential decay characteristics of stress relaxation of non-Newtonian fluids; calculating the time span from the moment of shutdown to the target time node, and determining the time span as the rheological relaxation time constant characterizing the elastic memory release rate of the drug solution.

[0017] Preferably, the process of obtaining the transient shear rate matrix and dynamic viscosity reference parameter includes: performing time differentiation on the braking displacement sequence to obtain the transient angular velocity sequence of the filling pump rotor during the shutdown deceleration phase; obtaining a preset peristaltic pump single-cycle volumetric displacement parameter, and mapping the transient angular velocity sequence to a transient volumetric flow rate sequence inside the pipeline; obtaining a preset filling pipeline inner diameter parameter, and calculating the stratified velocity gradient generated by the transient volumetric flow rate sequence on the pipeline path section based on the laminar flow profile distribution law of the fluid in the circular pipe, and constructing the stratified velocity gradient as the transient shear rate matrix characterizing the relative sliding strength of the fluid; extracting a pre-stored drug liquid viscosity-temperature characteristic mapping relationship table, inputting the real-time temperature data as a query index into the relationship table, obtaining the initial fluid viscosity under no-shear force state through interpolation matching, and using the initial fluid viscosity as the dynamic viscosity reference parameter.

[0018] Preferably, the process of obtaining the residual normal tensile stress includes: using the rheological relaxation time constant and the dynamic viscosity reference parameter as physical property constraints, and combining the transient shear rate matrix, driving the viscoelastic constitutive algorithm in the digital twin model to perform time-domain iterative calculation, and outputting the dynamic evolution time sequence of the first normal stress difference of the fluid micro-particle at the tip of the needle during the shutdown and re-suction process; monitoring the numerical change of the dynamic evolution time sequence of the first normal stress difference in real time, locating the time axis coordinate when the value drops to the critical point of fluid surface tension equilibrium, and determining the time axis coordinate as the predicted liquid bridge breakage time; wherein, the critical point of fluid surface tension equilibrium is obtained in real time through a preset viscosity-surface tension mapping function based on the dynamic viscosity reference parameter; extracting the instantaneous peak value of the dynamic evolution time sequence of the first normal stress difference before the predicted liquid bridge breakage time, and multiplying the instantaneous peak value with the flow cross-sectional area parameter at the tip of the needle to obtain the residual normal tensile stress characterizing the traction resistance at the moment of droplet detachment.

[0019] Preferably, the process of generating the asymmetric backflow control pulse sequence based on the reverse backflow momentum compensation parameter includes: performing time-domain integration of the residual normal tensile stress and the predicted liquid bridge fracture time to obtain a target impulse value, and determining the target impulse value as the reverse backflow momentum compensation parameter; extracting the mechanical response dead time of the filling pump actuator, subtracting the mechanical response dead time from the predicted liquid bridge fracture time to obtain the trigger delay time of the asymmetric backflow control pulse sequence; combining the reverse backflow momentum compensation parameter and the braking pulse width to calculate the peak velocity scalar of the asymmetric backflow control pulse sequence; retrieving the torque constant of the filling pump motor, and calculating the starting acceleration scalar of the asymmetric backflow control pulse sequence based on the peak velocity scalar, the reverse backflow momentum compensation parameter, and the current limit of the driver; combining the liquid viscosity corresponding to the real-time temperature data to calculate the liquid surface fluctuation recovery time caused by the reverse backflow action, and using the recovery time as the braking pulse width of the asymmetric backflow control pulse sequence.

[0020] Preferably, the process of sending the data to the control unit for adaptive control includes: mapping the trigger delay time to the timer interrupt offset of the servo controller and synchronizing it with the filling pump stop trigger signal; constructing a transient motion curve from zero speed to the target suction speed using an S-shaped acceleration / deceleration algorithm based on the starting acceleration scalar and the peak speed scalar as the upper speed limit constraint, and limiting the slope of the curve according to the rated torque of the servo motor; converting the asymmetric suction control pulse sequence into a pulse train of the corresponding frequency and sending it to the driver via real-time Ethernet; simultaneously, acquiring the encoder feedback data of the servo motor in real time and performing closed-loop deviation compensation between the actual suction displacement and the target pulse sequence; controlling the stationary time after suction stops according to the braking pulse width, and adjusting the holding torque of the driver in conjunction with the real-time temperature data to suppress secondary oscillations of the liquid surface at the tip of the needle.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0022] 1. By extracting the rheological relaxation time constant from the transient pressure decay sequence, the system can quantify the release rate of the "elastic memory" of the controlled medium at the moment of shutdown. This process utilizes the natural exponential decay characteristics of stress relaxation in non-Newtonian fluids. By calculating the time span between the initial pressure difference and the target node, it provides a key characteristic parameter for the control system to characterize the evolution of the physical state, and provides high-confidence physical property constraints for the digital twin model.

[0023] 2. By performing time-domain iterative calculations using a viscoelastic constitutive algorithm within the digital twin model, the system achieves a deep mapping of the dynamic evolution process of the first normal stress difference of the fluid micro-particle at the tip of the needle. It can accurately capture the critical point at which the fluid surface tension and internal stress reach equilibrium, thereby predicting the precise moment of liquid bridge fracture and resolving the residual normal tensile stress at the moment of droplet detachment. This logic transforms the originally invisible and complex physical evolution process into a quantifiable control criterion.

[0024] 3. By analyzing the reverse backflow momentum compensation parameters and generating an asymmetric backflow control pulse sequence, this control strategy comprehensively considers the mechanical response dead zone of the actuator, the motor torque constant, and the fluctuation recovery time caused by the viscosity of the liquid. It transforms the complex physical prediction results into specific execution commands with trigger delay, starting acceleration, and braking pulse width, and can actively intervene in the nonlinear energy residue in the transient process. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of an adaptive control method for an eye drop processing equipment based on digital twins according to the present invention.

[0026] Figure 2 This is a schematic diagram illustrating the process of obtaining the residual normal tensile stress according to the present invention.

[0027] Figure 3 This is a schematic diagram of the process for generating an asymmetric backflow control pulse sequence based on the reverse backflow momentum compensation parameters according to the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Please see Figures 1 to 3 This invention provides an adaptive control method for eye drop processing equipment based on digital twins, the technical solution of which is as follows:

[0030] Example 1:

[0031] An adaptive control method for eye drop processing equipment based on digital twins, the specific process of which is as follows: Figure 1 As shown, it includes:

[0032] The transient pressure decay time series of the eye drop filling pipeline was obtained, as well as the braking displacement sequence of the filling pump rotor and the real-time temperature data of the medicine solution.

[0033] Based on the transient pressure decay time series, the rheological relaxation time constant is extracted; combined with the braking displacement sequence, the transient shear rate matrix inside the pipeline during the shutdown phase is analyzed, and the dynamic viscosity reference parameter of the current batch of drug solution is determined based on real-time temperature data.

[0034] The rheological relaxation time constant, transient shear rate matrix and dynamic viscosity reference parameters are input into the digital twin model, and the dynamic evolution time sequence of the first normal stress difference is mapped and output. The corresponding predicted liquid bridge fracture time is extracted based on the dynamic evolution time sequence of the first normal stress difference, and the residual normal tensile stress at the tip of the needle is analyzed.

[0035] Based on the residual normal tensile stress and the predicted liquid bridge fracture time, the reverse backflow momentum compensation parameters are analyzed; based on the reverse backflow momentum compensation parameters, an asymmetric backflow control pulse sequence is generated, which includes trigger delay time, starting acceleration scalar and braking pulse width.

[0036] The asymmetric backflow control pulse sequence is converted into servo drive commands and sent to the control unit for adaptive control.

[0037] Furthermore, the process of acquiring data from the eye drop filling tubing includes: using a sterile, isolated pressure sensor located at the end of the eye drop filling tubing, after receiving a stop trigger signal from the filling pump, collecting transient fluid dynamics fluctuation characteristics within the tubing at a preset high-frequency sampling rate to construct a transient pressure decay time series; using an absolute servo encoder coupled to the filling pump drive shaft, synchronously collecting angular displacement decay pulses from the moment the stop trigger signal is issued until the drive shaft is completely locked and stationary to construct a braking displacement sequence; and using a non-contact infrared temperature measurement component attached to the outside of the eye drop filling tubing, collecting the thermal radiation parameters of the tubing wall in real time, and mapping the thermal radiation parameters to the real-time temperature data of the liquid.

[0038] During the acquisition of transient pressure data, a sterile, isolated piezoresistive pressure sensor with a range of 0 to 0.6 MPa and an accuracy of 0.1 is installed at the end of the eye drop filling tubing (50 mm to 100 mm from the filling needle outlet). The industrial host computer of the control system monitors the operating status of the filling pump in real time through a high-speed data acquisition card (such as a sampling rate of 20 kHz or higher). When the controller sends a stop command signal to the filling pump driver, the hardware interrupt of the acquisition card is triggered simultaneously. The system continuously records the internal pressure values ​​of the tubing at no less than 200 sampling points at a time step of 50 microseconds. These data are arranged in the order of collection. After eliminating the zero-point drift of the sensors, a transient pressure decay time series reflecting the evolution of pressure from steady state to static state is constructed. Specifically, during the 100ms static period before the filling pump starts, the system pre-collects 2000 pressure points in 50-microsecond steps and calculates the arithmetic mean, which is defined as the zero-pressure reference under the current environment. When constructing the sequence, all original sampled values ​​are subtracted from this reference value to eliminate the dynamic bias of the sensors caused by the temperature rise due to continuous operation.

[0039] Considering the spatial displacement between the pressure sensor installation location and the liquid bridge breakage point at the needle tip, gravity compensation calculation is performed after acquiring the original pressure signal. The specific logic is as follows: the system calculates the displacement pressure difference between the sensor and the needle tip based on the physical length of the tubing, the real-time density of the drug solution, and the installation angle of the tubing. By superimposing or subtracting this gravity pressure head value caused by the height difference from the measured value, the pressure data measured at the installation point can be accurately converted and mapped to the real-time pressure state at the needle outlet. To address the high-frequency signal attenuation problem that may occur due to the elastic diaphragm of the sterile isolation sensor, the system introduces a dynamic response correction mechanism. The dynamic response characteristic function of the sensor diaphragm is pre-calibrated to describe the gain loss of the diaphragm under high-frequency fluctuations at the 10 kHz level. During the data processing stage, the system uses this characteristic function to perform reverse compensation processing on the acquired transient pressure sequence, thereby eliminating the physical interference of the diaphragm's own elasticity on the viscoelastic relaxation signal of the drug solution and ensuring that the captured pressure attenuation curve can truly restore the transient hydrodynamic characteristics inside the drug solution.

[0040] The high-frequency sampling rate is determined based on the highest frequency component to be captured. Considering that the characteristic frequency of viscoelastic relaxation of the drug solution is generally in the range of 100Hz to 1000Hz, according to the Nyquist sampling theorem, the sampling rate should be at least twice that frequency. The present invention preferably uses a sampling rate of 1kHz to 50kHz, more preferably 10kHz to 20kHz, to ensure accurate capture of transient pressure fluctuations.

[0041] For acquiring the braking displacement sequence, a 23-bit absolute servo encoder coaxially connected to the drive shaft of the filling pump servo motor is used. When the PLC control unit issues a stop trigger signal, the system immediately records the current absolute angular position pulse count as the starting point. As the motor decelerates until the servo driver feedbacks "motor completely stopped" or the speed monitoring value returns to zero, the system continuously reads the encoder's feedback pulse increment with a sampling period of 1 millisecond. By dividing the cumulative pulse count by the encoder's single-turn resolution, the data points of angular displacement changing with time during the stopping process are obtained. These data points are integrated in order of time step to form a complete braking displacement sequence, which is used to characterize the physical movement of the mechanical actuator during the stopping transient. For the motion trajectory, since the sampling frequency of the displacement sequence (1kHz) is lower than that of the pressure sequence (20kHz), the displacement sequence needs to be upsampled using linear interpolation before being input into the digital twin model. The data points with a step size of 1ms are expanded into an equally spaced sequence with a step size of 50us to ensure that the mechanical motion characteristics and the fluid pressure fluctuation characteristics are aligned at the microsecond level on the time axis. Furthermore, combined with the preset peristaltic pump single-cycle displacement (e.g., 0.5ml / cycle) and the filling tube inner diameter (e.g., 3mm), the angular displacement change rate is mapped to the axial velocity of the fluid in the tube, and the instantaneous velocity gradient of the tube path to different positions is analyzed based on the laminar flow model, thereby generating a transient shear rate matrix characterizing the micro-deformation intensity of the fluid.

[0042] In practice, the mapping and parsing process includes the following logical steps:

[0043] The system first performs a first-order time derivative on the acquired angular displacement sequence to calculate the instantaneous angular velocity of the filling pump rotor at each sampling moment. Then, it multiplies this angular velocity by a preset actuator proportionality coefficient (which is calculated by dividing the peristaltic pump's single-revolution displacement, such as 0.5 ml per revolution, by the total radian of a single revolution) to obtain the instantaneous volumetric flow rate of the liquid medicine in the pipeline.

[0044] Based on the preset inner diameter parameters of the filling pipeline (e.g., 3 mm), the instantaneous volumetric flow rate is converted into the cross-sectional average velocity. Considering the non-Newtonian fluid characteristics of the eye drops, the system invokes the built-in power-law fluid velocity distribution model. This model presets the fluidity index of the liquid to describe the velocity decrease law of the fluid from the center of the pipeline to the pipe wall, that is, the velocity is the largest at the center of the pipe and the velocity tends to be zero near the pipe wall, forming a parabolic velocity profile that varies with the radius.

[0045] To digitize the aforementioned physical evolution, the system divides the pipeline cross-section into several concentric annular regions at equal intervals along the radial direction (e.g., 10 layers, numbered sequentially from the center to the pipe wall). For each time sampling step (e.g., 50 microseconds), the system calculates the radial velocity change rate of these 10 layers under the current velocity profile, i.e., the instantaneous shear rate at that location.

[0046] These shear rate values ​​at different levels and at different times are arranged in a structure with spatial hierarchy as rows and time step as columns, and finally a two-dimensional data table reflecting the interlayer sliding intensity of the fluid at the moment of shutdown is constructed, namely the transient shear rate matrix.

[0047] In the real-time temperature data acquisition process, a non-contact infrared temperature measurement module is fixed to the outer wall of the eye drop filling tubing, ensuring that the sensor detection window is perpendicular to the tubing axis. The module senses the 8-14 micrometer long-wave infrared radiation energy emitted by the outer wall of the tubing and converts it into a millivolt-level voltage signal. The signal processing circuit performs nonlinear compensation correction based on the pre-set emissivity of the tubing material (e.g., 0.93 for silicone tubing). The system inputs the processed radiation energy parameters into the built-in conversion model to calculate the corresponding Celsius temperature value in real time, which is then stored as the real-time temperature data of the current batch of eye drops. This temperature acquisition is synchronized with the filling cycle to ensure that the temperature status of each bottle of eye drops at the end of filling is verifiable. This temperature value needs to be corrected by thermal conduction compensation: based on the thermal conductivity of the tubing material (e.g., medical-grade silicone) and the wall thickness (e.g., 1.5 mm), combined with the preset ambient reference temperature, the built-in one-dimensional steady-state thermal conductivity model performs gradient compensation on the infrared-sensed wall temperature, thereby obtaining the true real-time temperature data of the core of the eye drops and ensuring the accuracy of the viscosity reference parameter calculation.

[0048] By simultaneously acquiring pressure decay, displacement, and temperature data from the filling pipeline, this scheme transforms the complex viscoelastic characteristics of fluids into a quantifiable physical sequence. Utilizing the transient shear rate matrix derived from the laminar flow model, multidimensional property constraints are provided for the digital twin model, effectively linking mechanical motion commands with the microscopic deformation characteristics of the drug solution. This lays the data foundation for subsequent accurate prediction of the liquid bridge fracture moment and residual stress, enabling the system to perform quantitative analysis of the stress release process of the drug solution.

[0049] Further, the process of extracting the rheological relaxation time constant includes: extracting the initial pressure peak value at the moment the filling pump stops and the steady-state pressure value of the substrate after the fluid is completely still in the transient pressure decay time series; calculating the transient pressure difference between the initial pressure peak value and the steady-state pressure value of the substrate; locating the target time node corresponding to when the actual pressure decay reaches the preset proportional threshold of the transient pressure difference along the time axis of the transient pressure decay time series, wherein the preset proportional threshold is set based on the natural exponential decay characteristics of stress relaxation of non-Newtonian fluids; calculating the time span from the moment of shutdown to the target time node, and determining the time span as the rheological relaxation time constant characterizing the elastic memory release rate of the drug solution.

[0050] In the data preprocessing stage, the system first retrieves the constructed transient pressure decay time series from the cache. This series is a data set containing at least 200 pressure sampling points with a step size of 50 microseconds. The processor first retrieves the pressure values ​​in the time interval from 0 milliseconds to 10 milliseconds in the series, and uses a maximum value search algorithm to lock the highest value point, which is determined as the initial pressure peak at the moment the filling pump stops. Before the search, the processor first performs a five-point moving average filter or a digital low-pass filter with a cutoff frequency of 500Hz on the original series within 10 milliseconds to eliminate high-frequency random noise pulses. Then, it finds the highest value point in the filtered smoothed series. Subsequently, the processor retrieves the data of the last 50 milliseconds (i.e., the last 1000 sampling points) of the series and calculates its arithmetic mean. This average value is used as the base steady-state pressure value after the fluid is completely still, in order to eliminate the interference of single-point noise on the reference pressure.

[0051] In the differential pressure calculation logic, the processor executes a subtraction instruction, subtracting the base steady-state pressure value from the initial pressure peak to obtain the transient differential pressure characterizing the total system pressure drop. Based on the natural exponential decay characteristic of stress relaxation in non-Newtonian fluids, a proportional threshold constant of 0.632 (i.e., the value of 1 - 1 / e) is pre-stored in the controller's non-volatile memory. In this embodiment, this proportional threshold of 0.632 corresponds to the main pressure relaxation period of a typical non-Newtonian fluid. However, in practical applications, those skilled in the art can determine the appropriate threshold based on the viscoelastic curves of eye drops with different components. Within the range of 0.5 to 0.8, the proportional threshold is empirically fine-tuned. The processor calculates the product of the transient pressure difference and the proportional threshold to obtain the target pressure attenuation. Then, the target pressure determination threshold is calculated by subtracting the target pressure attenuation from the initial pressure peak. For drug solutions containing polymeric components such as sodium hyaluronate, the threshold is moved closer to 0.8 to capture a longer relaxation period; for drug solutions that are approximately Newtonian fluids, the threshold is moved closer to 0.5. This dynamic adjustment based on the drug solution composition ensures the prediction accuracy of the digital twin model under different batch operating conditions.

[0052] In the time node positioning stage, the system starts a linear scanning program, starting from the timestamp corresponding to the initial pressure peak, and reads the pressure value point by point along the time axis of the time series. The program logically compares the real-time pressure value of each point with the target pressure judgment threshold mentioned above. When the pressure value of a certain sampling point is detected to be less than or equal to the target pressure judgment threshold for the first time, the program immediately triggers an interrupt and records the current timestamp. This point is defined as the target time node. In order to prevent false triggering caused by signal fluctuations, the program adds an anti-jitter mechanism when the first value is less than or equal to the target pressure judgment threshold: that is, at least 3 sampling points (i.e., for 150 microseconds) must be continuously detected to remain below the target pressure judgment threshold before the first sampling point that meets the condition is locked as the target time node.

[0053] During the constant determination phase, the processor retrieves the timestamp value of the target time node and subtracts the timestamp value at the moment of shutdown (i.e., the initial pressure peak). Here, the initial pressure peak is defined as the calculation origin. If the system detects that the pressure peak occurs more than 2 milliseconds after the pump shutdown command trigger signal, the starting timestamp needs to be advanced and compensated according to the inherent physical response time constant of the pressure sensor to ensure that the time span truly reflects the rheological relaxation process of the liquid. The resulting difference is the time span during which the liquid decays from a high-pressure stress state to a specific proportion. The system formally marks this as the rheological relaxation time constant. This constant is stored in the characteristic register of the control unit as a floating-point number and serves as a core physical property parameter characterizing the elastic memory release rate of the liquid, directly participating in the subsequent time-domain iterative calculation of the digital twin model.

[0054] By extracting the rheological relaxation time constant from the transient pressure decay sequence, the elastic memory release rate of the liquid at the moment of shutdown can be quantified. This process utilizes the fluid stress relaxation characteristics to provide high-confidence physical property constraints for the digital twin model. This transforms the invisible physical evolution process of filling into a quantifiable control criterion, laying the foundation for subsequent accurate prediction of the liquid bridge fracture moment and analysis of residual tensile stress, and enhancing the adaptive adjustment capability of the equipment during transient switching.

[0055] Furthermore, the process of obtaining the transient shear rate matrix and dynamic viscosity reference parameters includes: performing time differentiation on the braking displacement sequence to obtain the transient angular velocity sequence of the pump rotor during the shutdown deceleration phase; obtaining a preset peristaltic pump single-cycle volumetric displacement parameter, and mapping the transient angular velocity sequence to a transient volumetric flow rate sequence inside the pipeline; obtaining a preset filling pipeline inner diameter parameter, and based on the laminar flow profile distribution law of the fluid in the circular pipe, calculating the stratified velocity gradient generated by the transient volumetric flow rate sequence on the pipeline path, and constructing the stratified velocity gradient as the transient velocity gradient characterizing the relative sliding strength of the fluid. Specifically, the laminar flow profile distribution follows a power-law fluid model. The system corrects the parabolic velocity profile based on the fluidity index of the drug solution (e.g., between 0.6 and 0.8 for solutions containing sodium hyaluronate), thereby accurately capturing the drastic velocity gradient changes caused by shear thinning near the wall. A pre-stored viscosity-temperature characteristic mapping table of the drug solution is extracted, and the real-time temperature data is used as a query index in the table. The initial fluid viscosity under no-shear conditions is obtained through interpolation matching, and this initial fluid viscosity is used as the dynamic viscosity reference parameter.

[0056] In terms of extracting the transient angular velocity sequence, the system processor retrieves a pre-stored braking displacement sequence, which records the angular displacement points of the motor from receiving the stop signal to coming to a complete stop. The processor uses first-order backward differential logic to divide the angular displacement increment between two adjacent sampling points by a fixed sampling time interval to calculate the instantaneous rotational speed at each moment, thereby constructing a transient angular velocity sequence that reflects the deceleration characteristics of the actuator. Before performing the differential operation, the original displacement sequence needs to be processed by a sliding arithmetic mean based on 5 or 7 points; or after differentiation, a Butterworth low-pass filter with a cutoff frequency of 100Hz is applied to the generated transient angular velocity sequence to filter out the interference of encoder quantization noise on the transient motion characteristics. Subsequently, the system obtains the preset peristaltic pump single-cycle volumetric displacement parameters and converts each value in the above angular velocity sequence with the displacement constant proportionally, thereby mapping the motion information of the mechanical end into a transient volumetric flow rate sequence of the fluid inside the pipeline.

[0057] In the construction of the transient shear rate matrix, the system first obtains the preset inner diameter parameters of the filling pipeline. The processor, based on the laminar flow profile distribution law of the fluid in the circular pipe, specifically, describes the laminar flow profile distribution law using a power-law fluid model. That is, the steepness of the velocity distribution curve is corrected according to the fluidity index n of the liquid (e.g., between 0.6 and 0.8) to ensure that the stratified velocity gradient can accurately reflect the shear thinning behavior of non-Newtonian fluids near the pipe wall. The pipeline cross-section is divided into several concentric annular regions at equal intervals from the central axis to the pipe wall as spatial discrete layers. The spatial discrete layers are preferably set to 8 to 15 layers, and the closer to the pipe... In the wall region, the denser the layer spacing (using non-equidistant spacing), the higher resolution transient shear rate matrix is ​​constructed in the near-wall region where shear stress changes drastically. For each time point in the transient volumetric flow rate sequence, the system calculates the flow velocity corresponding to each annulus according to the flow velocity distribution model, and further calculates the ratio of the flow velocity change between adjacent layers to the radial distance change, thus obtaining the layered flow velocity gradient of each layer. Finally, the system integrates these flow velocity gradient values ​​covering different radial positions (spatial dimension) and different sampling times (time dimension) into a two-dimensional data table, which generates the transient shear rate matrix characterizing the microscopic relative sliding intensity of the fluid.

[0058] To determine the dynamic viscosity reference parameter, the system first retrieves a pre-stored viscosity-temperature characteristic mapping table of the drug solution from non-volatile memory. This table details the physical characteristics of a specific drug solution at different calibration temperatures. The mapping table records viscosity reference values ​​within the range of 5 to 45 degrees Celsius, with a step size of 0.5 degrees Celsius. A bidirectional lookup algorithm is used for the query index. If the real-time temperature deviation exceeds the calibration point, a linear interpolation algorithm is used to calculate the dynamic viscosity reference parameter at equilibrium under the current ambient temperature. The processor inputs the previously collected real-time temperature data of the drug solution as the query index into the table for matching. If the real-time temperature value lies between two adjacent calibration temperature points in the table, the processor initiates the linear interpolation algorithm to perform mathematical fitting within the corresponding viscosity range, thereby obtaining the initial fluid viscosity at that specific temperature under a shear-free state. This initial viscosity is then formally confirmed as the dynamic viscosity reference parameter and stored in the physical feature register for subsequent use by the digital twin model.

[0059] When calculating the velocity distribution within the pipe, the relationship between velocity and radial position is set according to the power-law fluid characteristics. Specifically, the velocity at the central axis of the pipe is set to its maximum value, while the velocity near the pipe wall decays nonlinearly according to the flow index. The system calculates the instantaneous velocity of the liquid at different cross-sectional levels by calculating the ratio of the current radial coordinate to the total pipe radius and combining it with the power factor determined by the flow index. This calculation method can realistically reflect the physical characteristics of non-Newtonian fluids flowing in pipes, which are "faster in the middle and slower at the edges" and have a flatter cross-sectional shape than Newtonian fluids.

[0060] By analyzing the radial rate of change of the aforementioned velocity profile, the transient shear rate characterizing the microscopic deformation of the fluid is obtained. Specifically, the ratio of the velocity difference between adjacent annular layers to the radial distance difference is calculated, and combined with a proportional correction coefficient determined by the flowability index, the local shear rate of that layer is derived. By integrating the data from each layer at each time step, the system ultimately constructs a transient shear rate matrix reflecting the relative sliding intensity between layers within the fluid, providing accurate mechanical input for the digital twin model.

[0061] The flow index is a core parameter characterizing the degree of shear thinning of the drug solution. Its specific value is pre-calibrated based on the composition of the drug solution. For eye drops containing high molecular weight components such as sodium hyaluronate and exhibiting significant shear thinning characteristics, the index is set in the range of 0.6 to 0.8. For physiological saline solutions with physical properties close to Newtonian fluids, the index is set to be close to one. The system locks the parameter value based on the real-time retrieved drug solution type information, ensuring physical consistency when converting flow data to strain data.

[0062] By mapping the braking displacement in the mechanical dimension to the transient shear rate matrix in the fluid dimension and coupling it with a temperature-corrected dynamic viscosity benchmark, this process achieves a digital characterization of the relative sliding strength of the fluid during the filling shutdown phase. This provides a high-confidence physical property constraint for the digital twin model, transforming the originally invisible complex physical evolution into a quantifiable control criterion.

[0063] Furthermore, the process of obtaining the residual normal tensile stress includes: using the rheological relaxation time constant and the dynamic viscosity reference parameter as physical property constraints, combined with the transient shear rate matrix, driving the viscoelastic constitutive algorithm in the digital twin model to perform time-domain iterative calculation, outputting the dynamic evolution time sequence of the first normal stress difference of the fluid micro-particle at the needle tip during the shutdown and backflow process; before starting the iterative calculation, the system directly maps the extracted rheological relaxation time constant to the relaxation term parameter in the constitutive equation, and sets the initial stress tensor to the steady flow field solution at the shutdown trigger moment; at the same time, a nonlinear mobility factor between 0.1 and 0.5 is preset according to the drug properties to establish the physical boundary of the model's evolution from static to dynamic; the numerical change of the dynamic evolution time sequence of the first normal stress difference is monitored in real time, and the point at which the value drops to the critical point of fluid surface tension equilibrium is located. The time axis coordinates are determined as the predicted liquid bridge breakage time. The fluid surface tension equilibrium critical point is obtained in real-time through a preset viscosity-surface tension mapping function based on the dynamic viscosity reference parameter. Before judgment, the system needs to simultaneously extract the liquid bridge characteristic radius at the needle outlet, divide the analyzed fluid surface tension value by this characteristic radius to convert it into surface tension additional pressure with pressure dimensions, and then compare the dynamic value of the first normal stress difference with this additional pressure to ensure the dimensional consistency of the mechanical judgment logic. The instantaneous peak value of the dynamic evolution sequence of the first normal stress difference before the predicted liquid bridge breakage time is extracted, and the instantaneous peak value is multiplied by the flow cross-sectional area parameter at the needle tip to obtain the residual normal tensile stress characterizing the traction resistance at the moment of droplet detachment. The specific process is as follows: Figure 2 As shown.

[0064] First, the computational engine in the digital twin model is activated. The rheological relaxation time constant and dynamic viscosity reference parameters obtained in the previous steps are used as fluid property constraint boundaries. The computational engine calls the built-in viscoelastic constitutive algorithm (such as the step-by-step iterative form of the Giesekus model or the Oldroyd-B model), and combines the spatial layered data in the transient shear rate matrix to perform time-domain rolling calculations with a time step of 50 microseconds. In each time step, the algorithm iteratively calculates the normal stress component generated by the fluid micro-element at the tip of the needle according to the shear rate of the current layer and the stress state of the previous moment. Finally, it maps and outputs the dynamic evolution time sequence of the first normal stress difference, which reflects the elastic energy storage and release process inside the fluid. Before starting the iteration, the system needs to initialize the viscoelastic tensor and set its initial value to the steady shear stress state of the filling pump during the stable operation phase. At the same time, according to the solvent properties of the drug solution, the nonlinear mobility parameter (range 0.1 to 0.5) and solvent viscosity ratio in the Giesekus model are preset as physical constraint constants in the iteration process.

[0065] While the stress is iterating, the system retrieves the preset viscosity-surface tension mapping function (this function is constructed based on experimental calibration data and records the correspondence between viscosity and surface tension with a resolution of 0.01 mN / m). The processor takes the dynamic viscosity reference parameter as the input value and, through piecewise linear interpolation logic, analyzes the fluid surface tension value of the drug solution under the current temperature and concentration conditions in real time, and sets it as the fluid surface tension equilibrium critical point for determining the liquid bridge fracture.

[0066] The system initiates real-time monitoring, continuously comparing the dynamically evolving timeline of the generated first normal stress difference with the aforementioned critical point for fluid surface tension equilibrium. When the value of the first normal stress difference continuously decreases and reaches or falls below the critical point for the first time, the system determines that the viscoelastic restoring force within the fluid is insufficient to overcome the surface tension, and the liquid bridge is about to break. At this point, the system immediately captures and records the time axis coordinates corresponding to the current sampling point, officially determining it as the predicted moment of liquid bridge breakage. Before comparison, the system needs to incorporate the characteristic radius of the liquid bridge at the needle outlet. Specifically, the real-time analyzed fluid surface tension is divided by the current characteristic radius of the liquid bridge, converting it into surface tension additional pressure with pressure dimensions. Then, the system compares the value of the first normal stress difference with this additional pressure to ensure dimensional consistency in the mechanical determination.

[0067] After locking the fracture moment, the processor traces back the dynamic evolution timeline of the first normal stress difference prior to that moment and extracts the instantaneous peak value in the timeline using an extreme value retrieval algorithm (this peak value represents the maximum normal tensile effect accumulated by the fluid just before fracture). Subsequently, the system retrieves the preset needle tip flow cross-sectional area parameter (for example, the cross-sectional area corresponding to a needle with an inner diameter of 0.3 mm is approximately 0.07 square millimeters), and multiplies the instantaneous peak value with the cross-sectional area parameter. The result of this calculation is determined as the residual normal tensile stress characterizing the traction resistance at the moment of droplet detachment and is stored in the system's execution control register as the core criterion for reverse backflow compensation. Finally, the system combines the predicted liquid bridge fracture moment and performs time integration on the residual normal tensile stress within the evolution period at the moment of fracture, thereby converting it into a compensation parameter with momentum properties. This parameter is used to determine the total energy output of the reverse backflow control pulse to achieve precise physical offsetting of the residual tensile energy at the moment of fluid fracture.

[0068] The system pre-stores surface tension calibration data of the medication at different temperatures. For example, for a typical eye drop solution (water-based solution), the surface tension is approximately 50 mN / m at 20°C and approximately 48 mN / m at 30°C. Based on real-time temperature data, the system uses linear interpolation to retrieve the surface tension corresponding to the current temperature from this calibration table. In this invention, the liquid bridge characteristic radius is defined as the minimum cross-sectional radius of the droplet neck at the needle exit. During the initial stage of filling shutdown (before the liquid bridge significantly shrinks), the liquid bridge characteristic radius is approximately taken as half the needle's inner diameter.

[0069] When calculating the characteristic radius, the system dynamically updates the volume by monitoring the volume change generated by the back suction action in real time. The specific logic is as follows: Based on the real-time pulse count fed back by the absolute value servo encoder, the physical displacement of the back suction piston or pump shaft is calculated, and the displacement is multiplied by the cross-sectional area inside the needle to obtain the real-time back suction volume. The initial liquid bridge volume (estimated by the needle inner diameter and the initial length of the liquid bridge) is subtracted from the real-time back suction volume to obtain the current residual volume of the liquid bridge. Assuming that the liquid bridge maintains a cylindrical deformation before breakage, the system uses the residual volume to divide the liquid bridge height and perform a square root operation to dynamically lock the current liquid bridge characteristic radius.

[0070] To eliminate dimensional differences, the surface tension value obtained in the previous steps is multiplied by 2 and then divided by the real-time updated liquid bridge characteristic radius. This transforms the mechanical quantity that originally described the surface properties into surface tension additional pressure with the dimension of pressure (Pascal). During the calculation, the system automatically retrieves the physical inner diameter of the needle (e.g., 0.3 mm) as the reference value of the initial characteristic radius. When the viscosity of the drug solution increases, causing the liquid bridge to stretch and become thinner, the characteristic radius decreases, and the corresponding surface tension additional pressure will increase non-linearly. This dynamic conversion process ensures that the reference point for judgment can track the physical contraction state of the liquid bridge neck in real time.

[0071] In the final determination stage, the system employs a three-point stability verification algorithm. The digital twin model outputs the first normal stress difference with a period of 50 microseconds. The real-time comparison module performs a real-time subtraction operation between this value and the surface tension and additional pressure at the current moment. When the calculation result is less than or equal to zero for the first time, and the difference between the next two consecutive sampling points (i.e., within a total time window of 150 microseconds) remains negative or zero, it is determined that the liquid bridge has reached the critical state of physical fracture. The system automatically captures the timestamp of the first of the three sampling points, officially marking it as the predicted time of liquid bridge fracture, and uses it as the zero-point coordinate to trigger subsequent momentum compensation.

[0072] The digital twin model is constructed based on viscoelastic fluid dynamics theory. The core of the model adopts a viscoelastic constitutive algorithm. Specifically, it introduces a rheological relaxation time constant that reflects the fluid memory effect as a relaxation term and combines it with dynamic viscosity reference parameters to define the internal friction characteristics of the fluid. In order to improve the real-time performance of the calculation, the model simplifies the three-dimensional needle flow field into an axial one-dimensional tensile flow model, thereby significantly reducing the computational cost while ensuring physical accuracy.

[0073] In terms of data mapping mechanism, the system uses the acquired transient shear rate matrix as the dynamic driving source of the model. Specifically, the model extracts the spatial layered data near the pipe wall (i.e. the region where the shear stress changes most drastically) from the matrix and converts it into a strain rate input that varies with time, thereby simulating the intense deformation of the drug microparticles at the moment of shutdown. Through this mapping method, the braking displacement characteristics of the mechanical end are effectively converted into the mechanical excitation signal of the fluid end, realizing the deep coupling between the physical state and the digital model.

[0074] The execution of the digital twin model employs a time-domain rolling iterative algorithm. The system strictly sets the time step to 50 microseconds to ensure that the extremely short relaxation process of the eye drops can be captured. Within each calculation step, the model uses numerical methods to discretize the nonlinear constitutive equations and seeks the stress equilibrium solution at the current moment through iterative methods. The iterative process sets a strict convergence criterion, namely, the relative deviation between two consecutive calculation results must be less than one part per million to ensure the physical reliability of the output timing.

[0075] Before the calculation begins, the model performs a physical initialization of the internal stress state. The initial state is set as the analytical solution of the steady flow field at the instant the pump stops, ensuring that the simulation starts from the real physical boundary. During the iteration process, the model calculates and outputs the first normal stress difference at the end of the needle in real time (i.e., the numerical difference between the axial normal stress and the radial normal stress). The curve of the change of this value over time constitutes the dynamic evolution time sequence of the first normal stress difference, which intuitively reflects the release trajectory of the residual elastic energy inside the drug solution.

[0076] Finally, the model sets clear calculation termination conditions. When the total duration of iterative calculation reaches 200 milliseconds, or when the output first normal stress difference value is detected to drop below the preset surface tension balance critical point, the digital twin engine will automatically stop running. This automated calculation management ensures that the system can provide accurate data support for the generation of subsequent back suction control commands at the first moment of predicting the liquid bridge fracture.

[0077] By employing a viscoelastic constitutive algorithm within a digital twin model, the system achieves a deep mapping of the fluid stress evolution process at the needle tip, transforming the invisible and complex physical evolution into quantifiable control criteria. This process accurately captures the equilibrium critical point between surface tension and internal stress, thereby predicting the moment of fluid bridge fracture and analyzing residual normal tensile stress. This provides a core physical basis for offsetting the residual energy at the moment of fracture.

[0078] Although this invention uses a viscoelastic model of shear flow, considering that the fluid still retains residual stress from the shear history during the shutdown and backflow process, the first normal stress difference can still be used as a quantitative indicator of the residual viscoelastic effect. In order to further improve the accuracy, a tensile correction factor is introduced when calculating the first normal stress difference. The calculation results are compensated and corrected according to the axial tensile rate, thereby accurately characterizing the residual viscoelastic effect under tensile flow conditions.

[0079] The stretching correction factor is a constant based on the drug components calibrated in advance through stretching rheological experiments. It is used to quantify the stress enhancement effect of the fluid under uniaxial stretching compared to shearing. In a preferred embodiment of the present invention, for eye drops containing polymeric thickeners, the correction factor is set between 0.1 and 0.3. The system automatically retrieves the corresponding correction value according to the physical property parameters of the currently filled drug solution as the proportional benchmark for stress compensation calculation.

[0080] The axial tensile rate reflects the degree of deformation of the liquid bridge in the length direction during the back suction process. The calculation process is as follows: the system uses the back suction speed fed back by the absolute encoder, combined with the predicted liquid bridge fracture time, to analyze the rate of change of the liquid bridge length with time in real time. The axial tensile rate is obtained by dividing the rate of change of the length by the current real-time length of the liquid bridge. This parameter dynamically captures the tensile strength under the transient back suction during shutdown, providing real-time data support for stress correction.

[0081] By multiplying the tensile correction factor by the axial tensile rate, a dynamic compensation coefficient is obtained. Then, the first normal stress difference initially output by the digital twin model based on the shear algorithm is multiplied by the weighted sum of "the above dynamic compensation coefficient". This step can be described by textual logic as: on the basis of the original shear stress, an enhancement component proportional to the tensile rate is superimposed to obtain the final corrected first normal stress difference. This result truly reflects the complex mechanical action of the fluid on the eve of fracture, ensuring the accuracy of subsequent liquid bridge fracture prediction.

[0082] Furthermore, the process of generating an asymmetric backflow control pulse sequence based on the reverse backflow momentum compensation parameter includes: performing time-domain integration of the residual normal tensile stress and the predicted liquid bridge fracture time to obtain a target impulse value, and determining the target impulse value as the reverse backflow momentum compensation parameter. By applying a reverse backflow action at the predicted time before liquid bridge fracture, the liquid in the needle is actively retracted, shortening the tensile length of the liquid bridge in space, thereby effectively reducing the volume of liquid remaining at the needle tip when the liquid bridge fractures. This process is not a simple energy cancellation, but rather changes the geometry of the liquid bridge, making it more responsive to physical fracture. Before the liquid bridge breaks, it actively retracts into the needle tip. Specifically, the magnitude of the backflow momentum is quantified based on the residual normal tensile stress analyzed earlier. The generated reverse motion command creates a transient negative pressure gradient inside the needle tip. This negative pressure gradient is sufficient to overcome the viscoelastic tensile force of the liquid itself. Through this transformation of mechanical balance, the liquid bridge neck retracts into the needle tip outlet at the moment of breakage, fundamentally avoiding droplet adhesion and subsequent liquid surface oscillation. The mechanical response dead time of the filling pump actuator is extracted, and the predicted liquid bridge breakage time is subtracted from the mechanical response dead time to obtain the trigger delay of the asymmetric backflow control pulse sequence. In the meantime, the mechanical response dead time is dynamically maintained by the system through online monitoring logic. Specifically, it calculates the difference between the timestamp of the controller issuing the stop command and the timestamp of the first change in the encoder feedback pulse, and dynamically updates the preset dead time parameter using the multi-cycle average of this difference to compensate for response offset caused by mechanical wear. Combining the reverse suction momentum compensation parameter and the braking pulse width, the peak speed scalar of the asymmetric suction control pulse sequence is calculated. The torque constant of the filling pump motor is retrieved, and based on the peak speed scalar, the reverse suction momentum compensation parameter, and the current limit of the driver, the non-... The starting acceleration scalar of the symmetrical back-suction control pulse sequence is pre-stored in the system, which contains the total equivalent moment of inertia J of the actuator (including the motor rotor, coupling, and pump body rotating parts). The processor, according to Newton's second law, divides the electromagnetic torque corresponding to the driver current limit by the total equivalent moment of inertia J, and compares this value with the quotient of the target speed / acceleration time, taking the minimum value as the final starting acceleration scalar. Combining this with the real-time temperature data and the corresponding drug viscosity, the liquid surface fluctuation recovery time caused by the reverse back-suction action is calculated, and this recovery time is used as the braking pulse width of the asymmetric back-suction control pulse sequence. The specific process is as follows: Figure 3 As shown.

[0083] The processor retrieves the residual normal tensile stress and the predicted time of fluid bridge fracture determined in the previous stage. The system employs a numerical integration method (such as trapezoidal integral logic) within the time interval from the onset of instability to the predicted fracture of the fluid bridge. The onset of instability is defined as the moment when the value of the dynamic evolution sequence of the first normal stress difference drops to 90% of the peak value after passing through the instantaneous peak. Using this moment as the lower limit of integration and the predicted time of fluid bridge fracture as the upper limit of integration, time-domain integration is performed on the tensile stress. The result of this calculation is the target impulse value used to offset the residual tensile energy at the moment of fluid fracture. The system defines this as the reverse suction momentum compensation parameter and stores it in the dynamic parameter table.

[0084] The processor retrieves preset hardware parameters, namely the mechanical response dead time of the filling pump actuator (this time is usually determined by the driver communication delay, the time required for the motor to overcome static friction, and mechanical clearance, for example, set to 5 to 10 milliseconds). The processor performs a subtraction operation, subtracting the mechanical response dead time from the predicted liquid bridge breakage time, thereby calculating the trigger delay time of the asymmetric back suction control pulse sequence. This time point ensures that the physical output of the back suction action coincides exactly with the moment of fluid breakage on the time axis.

[0085] The system combines the determined reverse suction momentum compensation parameters with the preset or subsequently calculated braking pulse width. Based on the transformation relationship of the momentum theorem, the processor distributes the target impulse value within the pulse width period and calculates the peak velocity scalar required for the suction action. This scalar limits the highest point of the suction pulse curve, ensuring that the suction kinetic energy is sufficient to neutralize the residual normal tensile stress. Specifically, the asymmetric suction control pulse sequence adopts a triangular velocity profile. The processor's distribution logic is: multiply the displacement corresponding to the target impulse value by 2, and then divide by the braking pulse width to obtain the peak velocity scalar of the triangular wave.

[0086] The processor retrieves the torque constant of the filling pump motor (such as the torque value generated per unit current), and combines it with the aforementioned peak speed scalar, reverse suction momentum compensation parameters, and the current limit preset by the servo driver (to prevent overload protection from triggering). The system calculates the starting acceleration scalar of the asymmetric suction control pulse sequence through the torque balance equation. This acceleration value determines the slope of the suction pulse leading edge, ensuring that the motor reaches the required momentum state for compensation at the fastest response speed within the current safety range.

[0087] Finally, the system utilizes the viscosity characteristics of the drug solution corresponding to the acquired real-time temperature data. The processor queries the built-in viscosity-level calming time experimental data mapping relationship. This mapping relationship table is pre-calibrated by conducting back-suction experiments on standard samples of different viscosities and measuring the time for oscillation decay to steady state using a laser liquid level displacement sensor. For example, when the real-time viscosity is 50 mPa·s, the corresponding calming time is set to 15 ms. The system calculates the calming time required for the liquid level fluctuations that may be caused by this reverse back-suction action to completely disappear. The system confirms this calming time as the braking pulse width of the asymmetric back-suction control pulse sequence. This parameter defines the duration for which the motor maintains stillness or holds torque after the back-suction command is executed, in order to physically suppress secondary oscillations of the liquid surface at the tip of the needle. The specific adjustment logic is as follows: the system establishes a positive correlation mapping between viscosity and current loop gain. When the real-time temperature increases, causing the drug solution viscosity to decrease compared to the reference viscosity, the system proportionally increases the position holding current of the servo driver. For example, for every 10% decrease in viscosity, the system increases the current gain corresponding to the holding torque by 3% to 5% to enhance the axial rigidity of the motor at the moment of back-suction stop.

[0088] The asymmetry in this invention refers to the asymmetrical distribution of the execution duration and acceleration amplitude of the back-suction pulse in the acceleration and deceleration phases. This is to adapt to the nonlinear physical characteristics of fluid viscoelastic release, which involves rapid stretching and slow recovery. Specifically, an asymmetric profile of "fast acceleration and slow deceleration" is adopted, with the acceleration phase accounting for 40% of the total back-suction time and the deceleration phase accounting for 60%. At the same time, the acceleration amplitude of the deceleration phase is set to 0.6 to 0.8 times that of the acceleration phase. Through this dual asymmetric design of time and mechanical amplitude, it is possible to ensure that the negative pressure gradient generated by the back-suction accurately offsets the residual stress and does not introduce additional mechanical oscillations due to excessively rapid deceleration.

[0089] To eliminate logical circular dependencies between control parameters, the analysis is performed in a strict sequence. First, based on the obtained reverse suction momentum compensation parameters and the principle of energy equivalence, the total displacement required for this suction action is estimated. Then, instead of directly calculating the velocity, the system first looks up the preset "viscosity-liquid surface recovery time" mapping table based on the viscosity of the liquid corresponding to the current real-time temperature. This directly locks the braking pulse width (i.e., the total suction duration) of the asymmetric pulse sequence. For example, when the system detects a viscosity of 50 mPa·s, it automatically sets the braking pulse width to 15 milliseconds. This logic of first determining the total displacement and total duration provides a unique and definite benchmark for the subsequent calculation of motion parameters.

[0090] After determining the total displacement and total duration, the specific motion command is analyzed through a geometric model. In this embodiment, the recoil velocity curve is set as a triangular profile. Twice the total displacement is divided by the braking pulse width to calculate the peak velocity scalar of this action. Then, based on the peak velocity and the previously determined 40% acceleration time ratio, the starting acceleration scalar is calculated. Before the command is issued, the system automatically retrieves the motor's torque constant and current limit to check whether the acceleration exceeds the hardware load capacity. If it does, the maximum acceleration allowed by the hardware is used as the starting value, and the acceleration time is extended accordingly to ensure the high reliability of recoil momentum compensation at the physical execution level.

[0091] To further improve control smoothness, the system incorporates an S-shaped transition algorithm at the start and end points of both the acceleration and deceleration phases. By limiting the rate of change of acceleration (i.e., jerk), mechanical shock is avoided. Based on the magnitude of the initial acceleration and a preset proportion of 10% to 20% of the total acceleration phase duration, a jerk limit threshold is set. This microscopic smoothing process ensures that the back suction action maintains high dynamic response and accurately offsets residual stress while effectively filtering out secondary vibrations of the liquid surface at the needle tip caused by sudden acceleration changes, thereby achieving higher precision filling control.

[0092] By analyzing the reverse backflow momentum compensation parameters and generating an asymmetric backflow pulse sequence, this strategy comprehensively considers the dead zone of the actuator, the motor torque constant, and the fluctuation recovery time caused by the viscosity of the liquid. It transforms the complex physical prediction results into specific execution instructions with trigger delay, acceleration, and pulse width. This not only allows for proactive intervention in the nonlinear energy residue during transient processes but also effectively suppresses secondary oscillations of the liquid surface at the tip of the needle.

[0093] Furthermore, the process of sending the data to the control unit for adaptive control includes: mapping the trigger delay time to the timer interrupt offset of the servo controller and synchronizing it with the filling pump stop trigger signal; constructing a transient motion curve from zero speed to the target suction speed using an S-shaped acceleration / deceleration algorithm based on the starting acceleration scalar and the peak speed scalar as the upper speed limit constraint, and limiting the slope of the curve according to the rated torque of the servo motor; converting the asymmetric suction control pulse sequence into a pulse train of the corresponding frequency and sending it to the driver via real-time Ethernet; simultaneously, acquiring the encoder feedback data of the servo motor in real time and performing closed-loop deviation compensation for the actual suction displacement and the target pulse sequence; controlling the stationary time after suction stops according to the braking pulse width, and adjusting the holding torque of the driver in conjunction with the real-time temperature data to suppress secondary oscillations of the liquid surface at the tip of the needle.

[0094] During the trigger and clock synchronization phase, the controller converts the calculated trigger delay time into a counter value based on the system clock frequency (for example, when the system clock is 100MHz, 1 millisecond corresponds to 100,000 counting pulses) and writes it into the timer interrupt offset register of the servo controller. This timer is associated with the hardware interrupt of the filling pump stop trigger signal. When the falling edge of the stop signal is captured, the timer starts counting down to ensure that the back suction task logic is triggered after the precise delay time is reached.

[0095] In the curve construction and limiting process, the system calls the built-in S-curve acceleration / deceleration algorithm (such as the seven-segment acceleration / deceleration logic) to obtain the starting acceleration scalar as the initial acceleration parameter, and uses the peak speed scalar as the upper speed limit constraint during operation. The processor verifies the rated torque parameter of the servo motor in real time. If the instantaneous torque demand of the S-curve at the acceleration / deceleration transition point (calculated based on angular acceleration and system inertia) exceeds 1.2 times the rated torque, the acceleration parameter is automatically reduced to limit the slope of the curve, thereby constructing a transient motion curve that conforms to physical characteristics. The specific adjustment logic is as follows: the acceleration parameter is reduced proportionally in steps of 10% of the current acceleration parameter. The instantaneous torque demand is recalculated each time it is reduced until the demand value drops to within 1.2 times the rated torque limit value, thereby determining the final S-curve acceleration configuration.

[0096] In the command issuance and communication process, the motion controller spatially discretizes the generated transient motion curve and converts it into a pulse train command of the corresponding frequency. Through the periodic data object (PDO) mapping of real-time Ethernet (such as EtherCAT or Profinet), the position command is sent to the servo driver with a synchronization period of 125 microseconds or 250 microseconds.

[0097] In the closed-loop compensation stage, the controller synchronously transmits the encoder feedback data of the servo motor via real-time Ethernet. The system calculates the difference between the target pulse sequence and the actual feedback displacement in real time, and uses feedforward control logic to superimpose the deviation into the instruction of the next execution cycle, thereby realizing closed-loop deviation compensation for the suction trajectory and ensuring that the actual suction momentum is consistent with the compensation momentum predicted by the digital twin model. The feedforward control logic adopts a speed feedforward superposition mode. The system calculates the position deviation change rate of the current sampling cycle and multiplies it by a preset feedforward gain coefficient (with a value range of 0.5 to 0.8). The resulting value is directly accumulated into the speed instruction register of the next cycle as a compensation component.

[0098] During the oscillation suppression phase, the system sets a forced still timer after the backflow stops based on the braking pulse width. Simultaneously, the processor adjusts the motor's holding torque via the driver control word based on the currently recorded real-time temperature data and the corresponding drug viscosity characteristics. (In low-viscosity conditions, the current loop gain is appropriately increased. The adjustment coefficient of the holding torque is positively correlated with the drug viscosity. For example, when the viscosity corresponding to the real-time temperature is 30% lower than the reference viscosity, the system increases the proportional gain of the current loop by 15% to 20% via the driver control word to enhance the rigidity of the motor at zero speed and counteract the strong liquid surface sloshing energy of the low-viscosity fluid.) During the stillness period after the backflow ends, the motor rotor position is locked to physically suppress the secondary oscillation of the liquid surface at the tip of the needle.

[0099] This process achieves closed-loop control from physical algorithms to mechanical execution by deeply coupling predicted parameters with underlying drive commands. The S-curve algorithm and torque limiting ensure the motor's dynamic stability under extreme response conditions, while clock synchronization technology guarantees microsecond-level timeliness of compensation actions. Combined with holding torque based on viscosity dynamic adjustment, residual stress is precisely offset, suppressing liquid surface oscillations.

[0100] By constructing a digital twin model and fusing pressure decay sequence, braking displacement, and liquid temperature data in real time, a deep perception of the transient physical process of eye drop filling shutdown is achieved. Utilizing the rheological relaxation time constant and transient shear rate matrix, the system can quantify the release of elastic memory and stress evolution within the fluid, transforming the originally invisible physical hysteresis into a high-confidence control criterion. This adaptive control method generates asymmetric backflow control pulses by analyzing the reverse backflow momentum compensation parameters, which can accurately offset the residual tensile energy at the moment of fluid fracture and suppress secondary oscillations of the liquid surface. This effectively reduces the mismatch between mechanical execution commands and the physical state of the controlled object.

[0101] Example 2:

[0102] In this embodiment, for the filling scenario of high-viscosity liquid medicine, the control system first reads the analog signal of the pressure sensor through the high-frequency acquisition card and converts it into a transient pressure decay time series. At the same time, the servo controller synchronously latches the pulse value of the absolute encoder through the real-time bus to construct the braking displacement sequence. The infrared temperature measurement module continuously monitors the radiation energy of the pipe wall and analyzes it into the real-time temperature data of the liquid medicine through the heat conduction compensation model, which serves as the benchmark for subsequent viscosity correction.

[0103] The processor calls a preset filtering algorithm to smooth the pressure sequence, automatically identifies the maximum value point in the sequence and defines it as the initial pressure peak. The system calculates the target decay pressure point according to the preset natural exponential decay characteristic ratio (such as 1-1 / e), and starts a time axis scanning program to find the corresponding target time node. By calculating the time difference between the shutdown time and the node, the rheological relaxation time constant characterizing the elastic release rate of the high-viscosity drug solution is analyzed.

[0104] The system performs derivative calculations on the braking displacement, mapping the mechanical rotation speed to a transient volumetric flow rate sequence of the fluid inside the pipe. Based on a power-law fluid model, the processor corrects the radial velocity distribution according to the fluidity index of the liquid and spatially discretizes the pipe cross-section into layers, thereby calculating the stratified velocity gradient of each layer of fluid at different times. Finally, it encapsulates this into a transient shear rate matrix. Simultaneously, using real-time temperature as an index, the system matches the dynamic viscosity reference parameter of the liquid from a viscosity-temperature table.

[0105] The processor loads the extracted physical constants and matrices into the digital twin engine, drives the viscoelastic constitutive algorithm (such as the Giesekus model) to perform rolling calculations, and the system simultaneously analyzes the surface tension equilibrium critical point (the pressure value after conversion via the characteristic radius of the liquid bridge) at the current viscosity, and performs a real-time logical comparison with the first normal stress difference N_1 output by the simulation. When the two reach equilibrium, the system automatically locks and records the predicted liquid bridge fracture time corresponding to the sampling point, and captures the stress peak before fracture to analyze the residual normal tensile stress.

[0106] The system performs time-domain integration on the residual normal tensile stress, converting the residual energy into the target impulse value (reverse pull-back momentum compensation parameter). The processor retrieves the mechanical response dead time calibrated online, removes this delay from the predicted fracture moment, calculates the trigger delay time of the compensation pulse, and, based on the triangular velocity profile logic and combined with the system's moment of inertia J, analyzes the starting acceleration scalar and peak velocity scalar required for the pull-back action.

[0107] The controller maps the calculated parameters into servo commands and sends them to the driver via real-time Ethernet. During the execution of the back suction pulse, the system compares the actual displacement fed back by the encoder in real time and compensates for the trajectory deviation in real time through speed feedforward logic. After the back suction task is completed, the system controls the motor to enter the stationary timing stage according to the calming time determined by viscosity, and dynamically increases the current loop gain to enhance the holding torque, thereby suppressing the secondary oscillation of the liquid surface at the tip of the needle at the physical level.

[0108] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A digital-twin-based adaptive control method for eye drop processing equipment, characterized in that, include: The transient pressure decay time series of the eye drop filling pipeline was obtained, as well as the braking displacement sequence of the filling pump rotor and the real-time temperature data of the medicine solution. Based on the transient pressure decay time series, the rheological relaxation time constant is extracted; combined with the braking displacement sequence, the transient shear rate matrix inside the pipeline during the shutdown phase is analyzed, and the dynamic viscosity reference parameter of the current batch of drug solution is determined based on real-time temperature data. The rheological relaxation time constant, transient shear rate matrix and dynamic viscosity reference parameters are input into the digital twin model, and the dynamic evolution time sequence of the first normal stress difference is mapped and output. The corresponding predicted liquid bridge fracture time is extracted based on the dynamic evolution time sequence of the first normal stress difference, and the residual normal tensile stress at the tip of the needle is analyzed. Based on the residual normal tensile stress and the predicted liquid bridge fracture time, the reverse backflow momentum compensation parameters are analyzed; based on the reverse backflow momentum compensation parameters, an asymmetric backflow control pulse sequence is generated, which includes trigger delay time, starting acceleration scalar and braking pulse width. The process of generating an asymmetric backflow control pulse sequence based on the reverse backflow momentum compensation parameter includes: performing time-domain integration of the residual normal tensile stress and the predicted liquid bridge fracture time to obtain a target impulse value, and determining the target impulse value as the reverse backflow momentum compensation parameter; extracting the mechanical response dead time of the filling pump actuator; subtracting the mechanical response dead time from the predicted liquid bridge fracture time to obtain the trigger delay time of the asymmetric backflow control pulse sequence; combining the reverse backflow momentum compensation parameter and the braking pulse width to calculate the peak velocity scalar of the asymmetric backflow control pulse sequence; retrieving the torque constant of the filling pump motor; calculating the starting acceleration scalar of the asymmetric backflow control pulse sequence based on the peak velocity scalar, the reverse backflow momentum compensation parameter, and the current limit of the driver; and calculating the liquid surface fluctuation recovery time caused by the reverse backflow action based on the liquid viscosity corresponding to the real-time temperature data, and using the recovery time as the braking pulse width of the asymmetric backflow control pulse sequence. The asymmetric backflow control pulse sequence is converted into servo drive commands and sent to the control unit for adaptive control.

2. The digital-twin-based adaptive control method for eye drop processing equipment according to claim 1, wherein The process of acquiring data from the eye drop filling tubing includes: using a sterile, isolated pressure sensor located at the end of the eye drop filling tubing, after receiving a stop trigger signal from the filling pump, collecting transient fluid dynamic fluctuation characteristics within the tubing at a preset high-frequency sampling rate to construct a transient pressure decay time series; using an absolute servo encoder coupled to the filling pump drive shaft, synchronously collecting angular displacement decay pulses from the moment the stop trigger signal is issued until the drive shaft is completely locked and stationary to construct a braking displacement sequence; and using a non-contact infrared temperature measurement component attached to the outside of the eye drop filling tubing, collecting the thermal radiation parameters of the tubing wall in real time, and mapping the thermal radiation parameters to the real-time temperature data of the liquid.

3. The digital-twin-based adaptive control method for eye drop processing equipment according to claim 1, wherein The process of extracting the rheological relaxation time constant is as follows: In the transient pressure decay time series, the initial pressure peak value at the moment the filling pump stops and the steady-state pressure value of the substrate after the fluid is completely still are extracted; the transient pressure difference between the initial pressure peak value and the steady-state pressure value of the substrate are calculated; along the time axis of the transient pressure decay time series, the target time node corresponding to when the actual pressure decay reaches the preset proportional threshold of the transient pressure difference is located, the preset proportional threshold is set based on the natural exponential decay characteristics of stress relaxation of non-Newtonian fluids; the time span from the moment of shutdown to the target time node is calculated, and the time span is determined as the rheological relaxation time constant characterizing the elastic memory release rate of the drug solution.

4. The digital-twin-based adaptive control method for eye drop processing equipment according to claim 1, wherein The process of obtaining the transient shear rate matrix and dynamic viscosity reference parameter includes: performing time differentiation on the braking displacement sequence to obtain the transient angular velocity sequence of the filling pump rotor during the shutdown deceleration phase; obtaining the preset peristaltic pump single-cycle volumetric displacement parameter, and mapping the transient angular velocity sequence to the transient volumetric flow rate sequence inside the pipeline; obtaining the preset filling pipeline inner diameter parameter, and calculating the stratified velocity gradient generated by the transient volumetric flow rate sequence on the pipeline path section based on the laminar flow profile distribution law of the fluid in the circular pipe, and constructing the stratified velocity gradient as the transient shear rate matrix characterizing the relative sliding strength of the fluid; extracting the pre-stored drug liquid viscosity-temperature characteristic mapping relationship table, inputting the real-time temperature data as the query index into the relationship table, obtaining the initial fluid viscosity under the no-shear force state through interpolation matching, and using the initial fluid viscosity as the dynamic viscosity reference parameter.

5. The digital-twin-based adaptive control method for eye drop processing equipment according to claim 1, wherein, The process of obtaining the residual normal tensile stress includes: using the rheological relaxation time constant and the dynamic viscosity reference parameter as physical property constraints, combined with the transient shear rate matrix, driving the viscoelastic constitutive algorithm in the digital twin model to perform time-domain iterative calculation, and outputting the dynamic evolution time sequence of the first normal stress difference of the fluid micro-particle at the tip of the needle during the shutdown and re-suction process; monitoring the numerical change of the dynamic evolution time sequence of the first normal stress difference in real time, locating the time axis coordinate when the value drops to the critical point of fluid surface tension equilibrium, and determining the time axis coordinate as the predicted liquid bridge breakage time; wherein, the critical point of fluid surface tension equilibrium is obtained in real time through a preset viscosity-surface tension mapping function based on the dynamic viscosity reference parameter; extracting the instantaneous peak value of the dynamic evolution time sequence of the first normal stress difference before the predicted liquid bridge breakage time, multiplying the instantaneous peak value with the flow cross-sectional area parameter at the tip of the needle, and obtaining the residual normal tensile stress characterizing the traction resistance at the moment of droplet detachment.

6. The adaptive control method for an eye drop processing equipment based on digital twins according to claim 1, characterized in that, The process of sending the data to the control unit for adaptive control includes: mapping the trigger delay time to the timer interrupt offset of the servo controller and synchronizing it with the filling pump stop trigger signal; constructing a transient motion curve from zero speed to the target suction speed using an S-shaped acceleration / deceleration algorithm based on the starting acceleration scalar and the peak speed scalar as the upper speed limit constraint, and limiting the slope of the curve according to the rated torque of the servo motor; converting the asymmetric suction control pulse sequence into a pulse train of the corresponding frequency and sending it to the driver via real-time Ethernet; simultaneously acquiring the encoder feedback data of the servo motor in real time and performing closed-loop deviation compensation between the actual suction displacement and the target pulse sequence; controlling the stationary time after suction stops according to the braking pulse width, and adjusting the holding torque of the driver in conjunction with the real-time temperature data to suppress secondary oscillations of the liquid surface at the tip of the needle.

Images