A melt electrospinning intelligent monitoring and regulation system and method based on orthogonal binocular vision

Abstract

The application provides a melt electrospinning intelligent monitoring and regulating system and method based on orthogonal double vision, and belongs to the cross field of machine vision and melt electrospinning process control. The system adopts two cameras with mutually orthogonal optical axes and perpendicular intersection with the nozzle axis to synchronously collect orthogonal projection images of Taylor cones and jets; based on double-view segmentation masks, an elliptical cross-sectional area integration method is used to reconstruct the three-dimensional volume of the Taylor cone, and the real-time value thereof is used as a feedback signal to adjust the spinning voltage through a closed-loop controller; meanwhile, a complete two-dimensional jet lag vector is synthesized from the double views, the dynamic response thereof is represented by an inertial differential equation, and a two-stage strategy combining steady-state global offset and corner dynamic transition is used to pre-compensate the numerical control track instruction offline. The voltage regulation and track compensation respectively act on two independent physical channels of a high-voltage power supply and a motion controller, and do not interfere with each other in time sequence, so that the synchronous control of fiber diameter uniformity and deposition track precision is realized.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of machine vision and melt electrospinning process control, specifically relating to an intelligent monitoring and control system and method for melt electrospinning based on orthogonal dual vision. Background Technology

[0002] Melt electrospinning is a processing technology that uses a high-voltage electrostatic field to stretch molten polymers to form micro- and nanofibers. In near-field direct writing mode, by shortening the distance between the nozzle and the collecting plate and controlling the nozzle's movement path, the jet can be deposited along a controllable trajectory, enabling the orderly arrangement and precise patterning of micro- and nanofibers. This technology has broad application prospects in fields such as tissue engineering scaffolds, flexible electronic devices, and functional fiber membranes.

[0003] A Taylor cone is a conical structure formed by the melt at the nozzle tip under the combined action of electric field force and surface tension. Its shape and volume directly affect the stability of the jet and the uniformity of fiber diameter. Online monitoring of the Taylor cone state is fundamental to the control of the spinning process. Chinese patent CN110852998B discloses a Taylor cone detection method and system based on deep learning. It uses a deep learning model to perform binary classification detection of the presence or absence of Taylor cones in images acquired by a single camera, achieving automatic discrimination of the existence of Taylor cones. This method provides a framework for automated monitoring of the spinning process, but it only achieves qualitative detection of Taylor cones and does not involve quantitative estimation of the three-dimensional volume of Taylor cones, making it difficult to provide continuous feedback signals for precise control of spinning voltage. Chinese patent CN211367826U discloses a near-field direct-write electrospinning fiber trajectory and morphology control system. It uses multiple sensors such as cameras, piezoelectric lenses, and photoresistors to acquire fiber image, morphology, and position information, and adjusts the solution spray volume of the nozzle through a control device. The system achieves multi-dimensional information perception of the fiber deposition process, but its monitoring object is the deposited fiber rather than the Taylor cone at the spinning source, which belongs to the post-event detection mode and the response lags behind the actual changes in the spinning process.

[0004] In estimating the three-dimensional volume of a Taylor cone, a single-camera approach is typically used to acquire its two-dimensional profile. Based on the body-of-revolution assumption, this profile is rotated around its central axis to approximate the three-dimensional volume. This method provides reasonable estimation results under ideal conditions where the Taylor cone's shape is approximately axisymmetric. However, in actual spinning processes, factors such as air disturbances introduced by platform motion, electric field fluctuations, and viscoelastic stretching of the melt can cause asymmetric deformation of the Taylor cone, deviating its horizontal cross-section from the ideal circle. In this case, a difference exists between the circular cross-section in the body-of-revolution assumption and the actual cross-sectional shape, potentially introducing volume estimation errors. Furthermore, the single-view approach can only acquire profile information in one projection direction and cannot detect deformation perpendicular to the observation direction, lacking the ability to detect asymmetric states of the Taylor cone.

[0005] Fiber diameter uniformity is a crucial indicator of spun product quality. Spinning voltage affects the feed and consumption balance of the Taylor cone by altering the electric field strength, thus influencing fiber diameter. Chinese patent CN216550835U discloses a near-field direct-writing micro / nano 3D electrospinning device, integrating a precision sample feeding mechanism and a temperature and humidity control mechanism, enabling control of spinning solution supply and environmental conditions. However, it still employs an open-loop mode with a preset constant value for the spinning voltage. Chinese patent CN110264487A discloses a detection method for electrospun products, determining fiber diameter and distribution parameters through grayscale processing and contour extraction of electron microscopy images of the spun products. This provides an offline detection method for evaluating spinning quality, but it lacks the ability to sense and adjust the Taylor cone state in real time during spinning. Currently, existing work on spinning voltage control mainly focuses on optimizing feed and environmental parameters; adaptive voltage adjustment based on Taylor cone volume feedback has not yet been reported.

[0006] In the near-field direct writing process of melt electrospinning, when the nozzle moves horizontally relative to the collecting plate, the jet cannot instantly reach the ideal position directly below the nozzle due to its own mass inertia and viscoelastic properties. This causes the jet landing point to deviate horizontally along the direction of motion, forming a jet hysteresis effect. During the uniform motion of the straight trajectory segment, the hysteresis length remains relatively stable; however, when the nozzle's motion direction changes at a corner, the hysteresis vector needs to transition from the original direction to the new direction. This transition process has dynamic characteristics, resulting in a deviation between the actual fiber deposition trajectory and the target trajectory. To address this problem, He et al. (Journal of Physics D: Applied Physics, 2016, 49(5): 055504) disclosed a corner deceleration method, which reduces the hysteresis length by reducing the collecting speed before the corner. However, speed changes cause fiber diameter fluctuations, and the buffer parameters need to be calibrated for specific process conditions. Cao et al. (Journal of Manufacturing Processes, 2024, 125: 265-279) proposed a matching equation between nozzle position and jet landing point based on vector analysis and differential geometry, providing a theoretical basis for toolpath planning of curved trajectories. This method is based on the steady-state assumption and is mainly applicable to uniform curved trajectory conditions. Duan et al. (Virtual and Physical Prototyping, 2022, 17(4): 1048-1066) disclosed a method for detecting jet deviation using a single-camera vision system, providing feedback signals for online control, but there is a measurement blind zone when the motion direction is parallel to the camera optical axis.

[0007] Based on the above analysis, the existing technology has the following three main shortcomings:

[0008] First, the existing methods lack the ability to accurately estimate the three-dimensional volume of the Taylor cone. Current solutions rely on a combination of a single camera and the assumption of a rotating body, providing reasonable estimates when the Taylor cone's shape is approximately axisymmetric. However, they cannot address asymmetric deformations caused by external disturbances during actual spinning. A single viewpoint can only acquire contour information in one projection direction. When the Taylor cone's cross-section deviates from a circle, a discrepancy arises between the cross-sectional area calculated from the width in a single direction and the actual cross-sectional area. Furthermore, the magnitude and direction of this discrepancy vary with the observation angle, leading to inconsistent volume estimation results. Simultaneously, existing methods lack the ability to quantitatively assess and provide early warnings for the asymmetric deformation state of the Taylor cone, failing to promptly alert the operator or automatically implement protective measures when deformation reaches a certain level.

[0009] Secondly, the spinning voltage lacks a closed-loop control mechanism based on real-time feedback of the Taylor cone volume. Existing spinning equipment generally employs an open-loop voltage control mode. During spinning, various factors such as nozzle temperature drift, changes in polymer viscosity with thermal history, fluctuations in feed gas pressure, and changes in ambient temperature and humidity can cause a shift in the Taylor cone's feed-consumption equilibrium point, resulting in a slow drift in the Taylor cone volume, which is reflected in the fiber diameter along the spinning path. Due to the lack of a technical approach to incorporate Taylor cone volume information into the voltage control loop, the uniformity of fiber diameter highly depends on the accuracy of manually preset process parameters and the stability of external conditions. This limits the ability to actively control the process drift during long-duration printing and multi-batch preparation.

[0010] Third, there is a lack of effective dynamic compensation methods for deposition trajectory deviations at corners. The aforementioned corner deceleration method, geometric model compensation method, and single-camera measurement method each have their own limitations in terms of fiber uniformity, dynamic transition description, and measurement direction coverage. From a common perspective, none of these methods have established a mathematical model capable of describing the dynamic transition characteristics of the hysteresis vector from the old equilibrium state to the new equilibrium state at corners, thus failing to quantitatively predict and accurately compensate for the trajectory deviations that evolve over time within the transition zone. Furthermore, as a two-dimensional physical quantity, the complete measurement of the hysteresis vector requires simultaneous acquisition of component information in two independent directions. However, existing single-camera configurations have projection blind spots in specific motion directions, limiting the omnidirectional measurement of the hysteresis vector and the implementation of compensation methods based on this limitation. Summary of the Invention

[0011] The purpose of this invention is to propose an intelligent monitoring and control system and method for melt electrospinning based on orthogonal dual vision, so as to overcome the technical problems in the existing melt electrospinning process, such as the Taylor cone volume estimation being constrained by the single-view rotating body assumption, the difficulty of the open-loop control of the spinning voltage in dealing with process drift, and the lack of dynamic compensation for corner trajectory deviation, so as to achieve the coordinated work of fiber diameter uniformity control and deposition trajectory accuracy control.

[0012] To achieve the above objectives, the technical solution of the present invention is as follows:

[0013] A smart monitoring and control system for melt electrospinning based on orthogonal dual vision, the system using an orthogonal dual-camera vision monitoring module as a common visual sensing basis, includes:

[0014] The orthogonal dual-camera vision monitoring module includes an X-axis camera and a Y-axis camera whose optical axes are orthogonal to each other and whose respective optical axes are perpendicular to the nozzle axis and intersect at the horizontal plane where the nozzle tip is located. These cameras are used to simultaneously acquire orthogonal projection images of the Taylor cone and the jet during the electrospinning process of the melt.

[0015] The visual feature extraction module processes the dual-view images acquired by the orthogonal dual-camera visual monitoring module and outputs the instance segmentation mask of the Taylor cone region under the dual view, as well as the pixel coordinates of the nozzle tip position and the jet landing point position.

[0016] The Taylor cone volume estimation module, based on the instance segmentation mask of the Taylor cone region under the dual-view perspective, discretizes the Taylor cone along the axis into multiple horizontal slices. Each slice cross section is regarded as an ellipse. The contour width of the two orthogonal views is used to determine the two orthogonal diameters of each elliptical cross section. The cross-sectional area of ​​each slice is accumulated along the axis to obtain the real-time three-dimensional volume of the Taylor cone.

[0017] The voltage closed-loop control module uses the real-time three-dimensional volume as a feedback signal and compares it with a preset volume target value. The closed-loop controller adjusts the spinning voltage output by the high-voltage power supply in real time during the printing process so that the real-time three-dimensional volume tracks the volume target value.

[0018] The hysteresis vector measurement module extracts the components of the two-dimensional hysteresis vector in the two orthogonal coordinate axes from the perspectives of the X-direction camera and the Y-direction camera, respectively, based on the pixel coordinates of the nozzle tip position and the jet landing point position under dual perspectives, and synthesizes them into a complete two-dimensional hysteresis vector for hysteresis monitoring and modeling of the dynamic response model of the inertial differential equation.

[0019] The trajectory pre-compensation module stores a pre-identified dynamic response model of the inertial differential equation, which describes the dynamic process of the two-dimensional hysteresis vector transitioning from the original steady state to the new steady state when the relative motion direction between the nozzle and the collection plate changes. Before the three-axis motion platform executes printing, the trajectory pre-compensation module performs offline correction on all trajectory points of the original CNC trajectory command based on the dynamic response model of the inertial differential equation, and outputs the corrected CNC trajectory command to the motion controller.

[0020] The voltage closed-loop control module acts on the high-voltage power supply in real time during the printing process, while the trajectory pre-compensation module acts on the motion controller offline before printing. The two do not interfere with each other in terms of timing.

[0021] Preferably, the Taylor cone volume estimation module reconstructs the real-time three-dimensional volume in the following manner:

[0022] Using the nozzle tip position as the common reference zero point for both perspectives, the pixel coordinates are converted into physical coordinates using a pre-calibrated pixel equivalent coefficient, so that the instance segmentation mask of the Taylor cone region under both perspectives is aligned in the physical height space.

[0023] The aligned mask region is uniformly divided into n horizontal slices with a thickness of Δh along the jet axis from the reference zero point.

[0024] For the i-th slice, extract its horizontal widths d(x,i) and d(y,i) from the instance segmentation mask corresponding to the camera in the X direction and the instance segmentation mask corresponding to the camera in the Y direction, respectively, as the two orthogonal diameters of the current slice's elliptical cross-section; calculate the cross-sectional area of ​​each slice according to the elliptical area formula S(i)=π·d(x,i)·d(y,i) / 4, and accumulate the product of each slice's cross-sectional area and slice thickness Δh along the axial direction to obtain the real-time three-dimensional volume V of the Taylor cone:

[0025] V = Σ π·d(x,i)·d(y,i)·Δh / 4 i=1,2,…,n.

[0026] Preferably, the Taylor cone volume estimation module further includes a morphological asymmetry early warning unit;

[0027] For each frame of image, the morphological asymmetry early warning unit calculates the ratio of the absolute value of the difference between the two orthogonal diameters of each slice layer to the larger value, and takes the average value of the corresponding ratios of all slice layers as the cross-sectional average eccentricity exponent E of the frame.

[0028] When the detected E values ​​in M ​​consecutive frames all exceed the abnormal threshold E th When an asymmetric deformation of the Taylor cone is detected, the morphological asymmetry early warning unit triggers the voltage closed-loop control module to pause voltage regulation output and issue an early warning signal; when the E value recovers to E for M consecutive frames... th In the following cases, the voltage closed-loop control will be automatically restored.

[0029] Preferably, the abnormal threshold E th The value range is 0.10~0.25, and the value range of consecutive frame number M is 3~10 frames.

[0030] Preferably, the dynamic response model of the inertial differential equation is in the following form:

[0031] τ·(dL / dt) + L = K·v

[0032] Where L is the two-dimensional hysteresis vector, v is the velocity vector of the nozzle relative to the collection plate, τ is the time constant, K is the steady-state gain coefficient, and both τ and K are functions of the relative velocity magnitude |v|.

[0033] The steady-state solution of the dynamic response model of the inertial differential equation when v is constant is L=K·v; when v changes, the dynamic response model of the inertial differential equation describes the dynamic process of the two-dimensional hysteresis vector L transitioning from the original steady state to the new steady state according to the first-order exponential law.

[0034] Preferably, the time constant τ and the steady-state gain coefficient K are pre-identified and stored in the trajectory pre-compensation module in the following manner:

[0035] Within the system's operating speed range, select at least three different relative motion speed levels and execute test trajectories that include sudden changes in direction at corners.

[0036] The orthogonal dual-camera visual monitoring module and the hysteresis vector measurement module are used to synchronously collect the temporal change data of the two-dimensional hysteresis vector before and after the corner;

[0037] The least squares method is used to fit τ and K at each relative motion speed level for the collected data, and the mapping relationship between τ and K with respect to relative motion speed is established by linear interpolation. The mapping relationship is used as the lookup table basis for trajectory pre-compensation.

[0038] Preferably, the trajectory pre-compensation module uses a two-level compensation strategy to perform offline correction of the original CNC trajectory command:

[0039] The first level is global offset compensation, which offsets each trajectory point in the original CNC trajectory command by a distance of |K·v| in the opposite direction of the motion direction of the trajectory segment to eliminate steady-state lag deviation, where v is the motion velocity vector corresponding to the trajectory segment;

[0040] The second level is dynamic transition compensation for corners. For each trajectory point in the original CNC trajectory command, the angle between the direction of the previous trajectory segment and the direction of the next trajectory segment is greater than the corner recognition threshold θ. th The trajectory points are determined as corner points; for each corner point, the corresponding time constant τ and steady-state gain coefficient K are determined according to the relative velocity |v| at the corner point, and the difference vector ΔL between the steady-state lag vectors before and after the corner point is calculated; M′ compensation points are generated in the compensation interval of length N·τ·|v| after the corner point, where N is a multiplier coefficient, and the position offset of each compensation point decays from |ΔL| to zero along the direction of motion according to a first-order exponential decay law;

[0041] The compensation points are sequentially inserted into the trajectory sequence after the corner point, and the corrected trajectory sequence is re-encoded into CNC trajectory instructions for output.

[0042] Preferably, the corner recognition threshold θ th The value range is 3° to 10°, the multiplier N ranges from 2 to 5, and the number of compensation points M′ ranges from 5 to 20.

[0043] The system employs a dual-channel decoupled control architecture during the printing process: the trajectory compensation channel is completed offline by the trajectory pre-compensation module before printing begins, and its target is the motion controller, used to correct the original CNC trajectory commands received by the motion controller offline; the voltage regulation channel is run in real time by the voltage closed-loop regulation module during printing, and its target is the high-voltage power supply, used to adjust the spinning voltage output by the high-voltage power supply in real time; the trajectory compensation channel and the voltage regulation channel act on two independent physical channels, motion control and voltage control, respectively, and do not interfere with each other in timing.

[0044] A method for intelligent monitoring and control of melt electrospinning based on orthogonal dual vision, wherein the method is implemented using any of the above-mentioned intelligent monitoring and control systems for melt electrospinning, and includes an offline preprocessing stage and an online printing stage;

[0045] The offline preprocessing stage includes: based on the pre-identified dynamic response model of the inertial differential equation, performing two-level corrections on all trajectory points of the original CNC trajectory command, namely global offset compensation and corner dynamic transition compensation, to generate the compensated CNC trajectory command; this step is applied to the CNC trajectory command received by the motion controller and is executed offline before printing;

[0046] The online printing phase includes the following steps, which are executed concurrently during the printing process:

[0047] The compensated CNC trajectory command is loaded into the motion controller to drive the three-axis motion platform to perform printing.

[0048] The orthogonal projection images of the Taylor cone and the jet during the electrospinning process of melt are acquired simultaneously using two cameras with mutually orthogonal optical axes.

[0049] The Taylor cone region of the acquired dual-view images is segmented into instances to obtain the instance segmentation mask of the Taylor cone region under dual-view conditions, and the pixel coordinates of the nozzle tip position and the jet landing point position under dual-view conditions are detected respectively.

[0050] Based on the instance segmentation mask of the Taylor cone region under the dual-view perspective, the real-time three-dimensional volume of the Taylor cone is reconstructed using the dual-view elliptical section integration method. The dual-view elliptical section integration method discretizes the Taylor cone into multiple horizontal slices along the axial direction. Each slice is regarded as an ellipse. The two orthogonal diameters of each elliptical slice are determined by the contour width of the two orthogonal perspectives. The real-time three-dimensional volume of the Taylor cone is obtained by accumulating the product of the elliptical area and the slice thickness of each slice along the axial direction.

[0051] Using the real-time three-dimensional volume as a feedback signal, the spinning voltage is adjusted in real time by a closed-loop controller so that the real-time three-dimensional volume tracks the preset volume target value; this step is applied to the spinning voltage output by the high-voltage power supply and is executed in real time during the printing process;

[0052] Based on the coordinates of the nozzle tip position and jet landing point position of the dual-view nozzle, the components of the two-dimensional hysteresis vector in the two orthogonal coordinate axes are extracted respectively, and a complete two-dimensional hysteresis vector is synthesized for hysteresis monitoring and modeling of the dynamic response model of the inertial differential equation.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) Two orthogonally arranged dual cameras are used to synchronously acquire the two orthogonal projected contours of the Taylor cone. The cross-sections at each height are approximated as ellipses and discretely integrated along the axis to reconstruct the three-dimensional volume. Compared with the traditional single-view rotating body assumption method, this method can reflect the anisotropic changes in shape through the cross-sectional dimensions in two independent directions when the Taylor cone undergoes asymmetric deformation. This significantly improves the consistency problem of volume estimation caused by different observation angles and enhances the reliability of volume estimation under dynamic conditions. At the same time, an asymmetric quantitative assessment and early warning mechanism is established using dual-view cross-sectional information. This mechanism can automatically pause closed-loop control and trigger an early warning when the Taylor cone deformation exceeds the threshold, thereby enhancing the process safety of the system under abnormal conditions.

[0055] (2) Using the real-time estimated Taylor cone volume as a feedback signal, the spinning voltage is automatically adjusted through a closed-loop controller, establishing a closed-loop control channel between the Taylor cone volume and the spinning voltage. Compared with the open-loop constant voltage mode, this mechanism can compensate for Taylor cone volume drift caused by factors such as nozzle temperature drift, material viscosity changes, and environmental condition fluctuations during the spinning process in real time, keeping the fiber diameter fluctuation along the process at a low level and reducing the dependence of fiber diameter uniformity on the accuracy of manually preset process parameters. It is particularly suitable for scenarios where process drift accumulation is significant, such as long-term continuous printing and multi-batch preparation.

[0056] (3) By using orthogonal dual cameras, the blind zone of projection measurement under specific motion directions of a single camera is overcome, and the complete measurement of the two independent components of the hysteresis vector is realized. On this basis, a mathematical model describing the dynamic transition characteristics of the hysteresis vector is established, and offline pre-compensation is performed on the CNC trajectory command accordingly. This method does not require changing the collection speed before the corner, avoiding fiber diameter fluctuations caused by speed changes. At the same time, it breaks through the limitations of the steady-state assumption and can quantitatively predict the dynamic transition process of the hysteresis vector from the old equilibrium state to the new equilibrium state at the corner, providing effective compensation for trajectory deviation in the transition zone. Experimental verification shows that this method has achieved good compensation results under various corner angles and collection speed conditions.

[0057] (4) By applying voltage closed-loop regulation and trajectory pre-compensation to two independent physical channels—the high-voltage power supply and the motion controller—without temporal interference, this invention achieves system-level synergy between fiber diameter uniformity control and deposition trajectory accuracy control. In the fabrication of functional micro / nano structures such as the Janus unidirectional water transport structure, which require simultaneous assurance of fiber diameter uniformity and trajectory accuracy, the synergistic effect of the two control channels can reliably achieve functional goals that are difficult to achieve with a single control channel. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0059] Figure 2 This is a schematic diagram of the spatial layout of the orthogonal dual-camera visual monitoring module;

[0060] Figure 3 A schematic diagram of the data flow of the main control computer software module;

[0061] Figure 4 This is a schematic diagram illustrating the principle of Taylor's cone elliptical section integration method.

[0062] Figure 5 A schematic diagram of the hysteresis effect and micro-element force analysis of melt electrospinning jet;

[0063] Figure 6 A schematic diagram illustrating hysteresis vector measurement and orientation angle definition;

[0064] Figure 7 The fitted curves of the hysteresis vector dynamic response are shown for typical collection velocities (where: (a) collection velocity 20 mm / s, (b) collection velocity 30 mm / s, (c) collection velocity 45 mm / s).

[0065] Figure 8 This is a schematic diagram illustrating the principle of dynamic transition compensation at corners.

[0066] Figure 9 Here is a flowchart of the trajectory pre-compensation algorithm;

[0067] Figure 10 The graphs show a comparison of the Taylor cone volume and spinning voltage response curves in open-loop and closed-loop modes (where: (a) Taylor cone volume response curve, (b) spinning voltage response curve).

[0068] Figure 11 Comparison of fiber deposition trajectories before and after corner compensation at different angles (where: (a) 45° corner, (b) 90° corner, (c) 135° corner).

[0069] In the diagram: 1-Nozzle; 2-Nozzle tip; 3-Taylor cone; 4-Jet; 5-Fiber; 6-Collection plate; 7-Triaxial motion platform; 8-Temperature-controlled feeding system; 9-Heating device; 10-Pneumatic feeding device; 11-High-voltage power supply; 12-X-direction camera; 13-Y-direction camera; 14-X-direction backlight; 15-Y-direction backlight; 16-Motion controller; 17-Main control computer; 18-Orthogonal dual-camera visual monitoring module; 19-Visual feature extraction module ; 20-Taylor cone volume estimation module; 21-Voltage closed-loop control module; 22-Hysteresis vector measurement module; 23-Trajectory pre-compensation module; 24-Horizontal slice; 25-Elliptical cross section; 26-Orthogonal diameter in the X direction; 27-Orthogonal diameter in the Y direction; 28-Actual landing point; 29-Target landing point; 30-Micro element; 31-Gravity; 32-Electric force; 33-Air resistance; 34-Hysteresis vector X-direction component; 35-Hysteresis vector Y-direction component; 36-Direction angle. Detailed Implementation

[0070] The following is in conjunction with the appendix Figure 1-11 The technical solution of the present invention will be described in detail below. The following embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0071] like Figure 1 As shown in the figure, the structural layout of the intelligent monitoring and control system for melt electrospinning based on orthogonal dual vision provided in this embodiment is as follows.

[0072] The nozzle 1 is fixedly mounted on the frame, and the nozzle tip 2 is provided at the end of the nozzle 1. The melt below the nozzle tip 2 is stretched to form a Taylor cone 3, and the end of the Taylor cone 3 is stretched to form a jet 4, which is deposited on the surface of the collecting plate 6 to form fibers 5. The collecting plate 6 is fixed on the XY worktable of the three-axis motion platform 7 and moves in two dimensions in the horizontal plane with the three-axis motion platform 7. The X / Y axes of the three-axis motion platform 7 are used to drive the collecting plate 6 to move in the horizontal plane according to a preset trajectory to achieve patterned deposition, and its Z axis is used to support the collecting plate 6 and adjust the collection distance between the nozzle tip 2 and the collecting plate 6.

[0073] The temperature-controlled feeding system 8 includes a heating device 9 and a pneumatic feeding device 10. The output end of the temperature-controlled feeding system 8 is connected to the nozzle 1 via a feeding pipeline. The heating device 9 is used to heat the spinning material to a molten state, and the pneumatic feeding device 10 is used to push the molten spinning material to the nozzle 1 at a set pressure.

[0074] A high-voltage power supply 11 is used to establish a high-voltage electrostatic field between the nozzle tip 2 and the collecting plate 6. The positive terminal of the high-voltage power supply 11 is connected to the nozzle 1, and the negative terminal of the high-voltage power supply 11 is connected to the collecting plate 6 via a grounding terminal. Under the combined action of the high-voltage electrostatic field and the surface tension of the melt, the melt below the nozzle tip 2 is stretched to form the Taylor cone 3. The end of the Taylor cone 3 is further stretched into the jet 4 under the action of the electric field force.

[0075] like Figure 2 As shown, the X-direction camera 12, Y-direction camera 13, X-direction backlight 14, and Y-direction backlight 15 are all fixedly mounted on the bracket, together constituting the orthogonal dual-camera visual monitoring module 18. The optical axis of the X-direction camera 12 is arranged along the Y-direction, used to acquire orthogonal projection images of the Taylor cone 3 and the jet 4 along the Y-direction; the optical axis of the Y-direction camera 13 is arranged along the X-direction, used to acquire orthogonal projection images of the Taylor cone 3 and the jet 4 along the X-direction. The X-direction backlight 14 is arranged opposite to the nozzle 1 along the optical axis of the X-direction camera 12 on the other side, and the Y-direction backlight 15 is arranged opposite to the nozzle 1 along the optical axis of the Y-direction camera 13 on the other side. The two backlights provide uniform and stable backlight illumination for their respective cameras, so that the Taylor cone 3 and the jet 4 present a high-contrast silhouette effect in the image, which facilitates the instance segmentation and key point detection of the visual feature extraction module 19 in the subsequent process. The optical axes of the X-axis camera 12 and the Y-axis camera 13 are orthogonal to each other and both intersect perpendicularly with the axis of the nozzle 1 at the horizontal plane where the nozzle tip 2 is located. The technical significance of this orthogonal arrangement is that when the collecting plate 6 moves along the X-axis, the Y-axis camera 13 can clearly observe the offset component of the jet 4 in the Y-axis, while the X-axis camera 12 can measure the offset component of the jet 4 in the X-axis; the situation is similar when the collecting plate 6 moves along the Y-axis. Therefore, regardless of the direction in which the collecting plate 6 moves, at least one camera can effectively measure the hysteresis component in the corresponding direction, eliminating the projection measurement blind spot present in the single-camera scheme when the direction of movement is parallel to the optical axis, and realizing the complete measurement of the two-dimensional hysteresis vector under any direction of movement.

[0076] The output of the motion controller 16 is connected to the three-axis motion platform 7, and is used to receive and execute CNC trajectory commands to drive the three-axis motion platform 7 to move according to the commands. The output of the main control computer 17 is connected to the motion controller 16 and the high-voltage power supply 11 respectively. The main control computer 17 is equipped with a visual feature extraction module 19, a Taylor cone volume estimation module 20, a voltage closed-loop control module 21, a hysteresis vector measurement module 22, and a trajectory pre-compensation module 23. The visual feature extraction module 19, the Taylor cone volume estimation module 20, the voltage closed-loop control module 21, the hysteresis vector measurement module 22, and the trajectory pre-compensation module 23 can all be implemented by software running on the main control computer 17. Those skilled in the art can use other software implementations with the same functions according to actual needs.

[0077] like Figure 3 As shown, the signal flow relationships between the components of the system are as follows: the dual-view images acquired by the X-direction camera 12 and the Y-direction camera 13 are input to the visual feature extraction module 19; the visual feature extraction module 19 processes the dual-view images respectively, and its output dual-view segmentation mask is input to the Taylor cone volume estimation module 20, and its output pixel coordinates of the nozzle tip position and jet landing point position are input to the hysteresis vector measurement module 22; the real-time three-dimensional volume value of the Taylor cone 3 output by the Taylor cone volume estimation module 20 is input to the electrical... The voltage closed-loop control module 21 outputs voltage adjustment commands to the high-voltage power supply 11 to adjust the spinning voltage in real time; the hysteresis vector measurement module 22 outputs hysteresis vector timing data to pre-identify the dynamic response model parameters in the trajectory pre-compensation module 23; before the three-axis motion platform 7 starts printing, the trajectory pre-compensation module 23 performs a one-time offline correction of the original CNC trajectory command based on the dynamic response model, and the corrected CNC trajectory command is sent to the three-axis motion platform 7 for execution via the motion controller 16.

[0078] The system employs a dual-channel decoupled control architecture during the printing process: the voltage regulation channel, operated in real-time by the voltage closed-loop regulation module 21, acts on the high-voltage power supply 11; the trajectory compensation channel, completed offline once before printing begins by the trajectory pre-compensation module 23, acts on the numerical control trajectory commands received by the motion controller 16. The voltage regulation channel and the trajectory compensation channel operate on two independent physical channels—voltage control and motion control—and do not interfere with each other in timing.

[0079] In this embodiment, the X / Y axis travel of the three-axis motion platform 7 is 150mm, and the repeatability is ±2μm. The X-axis camera 12 and the Y-axis camera 13 are both industrial-grade area array cameras with a resolution of 1440×1080 and a frame rate of 150fps. The maximum output voltage of the high-voltage power supply 11 is 30kV, with an output accuracy of ±10V, and the spinning voltage setting is limited to a safe operating range of 4.8kV to 5.4kV. The motion controller 16 supports standard G-code input. The spinning material is polycaprolactone (PCL, number average molecular weight 50000), and the basic process parameters are set as follows: heating temperature 165℃, feeding air pressure 30kPa, nozzle inner diameter 0.51mm, collection distance 9mm, and initial spinning voltage setting 5.0kV. Those skilled in the art will understand that the above hardware configuration and process parameters are merely exemplary values ​​given for illustrative purposes and are not intended to limit the invention.

[0080] The following provides a detailed description of the specific implementation methods of each software module within the main control computer 17.

[0081] The visual feature extraction module 19 processes the dual-view images acquired by the orthogonal dual-camera visual monitoring module 18, including two functions: instance segmentation and keypoint detection. The instance segmentation function performs pixel-level segmentation of the Taylor cone 3 region in each of the dual-view images, outputting a binarized segmentation mask for the Taylor cone 3. The keypoint detection function detects the position coordinates of the nozzle tip 2 and the landing point of the jet 4 in each of the dual-view images. These two functions independently perform inference processing on the images acquired by the X-direction camera 12 and the Y-direction camera 13. In this embodiment, the instance segmentation uses the YOLO11n-seg model, and the keypoint detection uses the YOLO11n-pose model. Those skilled in the art can use other neural network models with the same functions according to actual needs.

[0082] like Figure 4 As shown, the Taylor cone volume estimation module 20 receives the dual-view segmentation mask output by the visual feature extraction module 19 and uses the elliptical section integration method to estimate the real-time three-dimensional volume of the Taylor cone 3. The method discretizes the Taylor cone 3 along the axial direction into multiple horizontal slices 24, each horizontal slice 24 having a cross-section approximately equal to an elliptical cross-section 25. The diameters of the two orthogonal directions of each elliptical cross-section 25 are determined using the contour widths of the two orthogonal viewpoints, and the real-time three-dimensional volume of the Taylor cone 3 is reconstructed through discrete integration along the jet axis.

[0083] The specific calculation process is as follows. First, mask preprocessing and height space alignment are performed: using the position point of the nozzle tip 2 output by the visual feature extraction module 19 as the common reference zero point for both views, the vertical coordinates of the two mask images are respectively zeroed, and the pixel coordinates are converted into physical coordinates using their respective calibrated pixel equivalent coefficients, so that the two masks are aligned layer by layer in the physical height space. In this embodiment, the pixel equivalent coefficients calibrated by the X-direction camera 12 and the Y-direction camera 13 are 29.1 μm / pixel and 29.3 μm / pixel, respectively. The position coordinates of the nozzle tip 2 also serve as the common reference zero point for subsequent Taylor cone volume estimation and hysteresis vector measurement.

[0084] Then, orthogonal contour extraction and slicing are performed: the aligned mask area is uniformly divided into n horizontal slices 24 with a thickness of Δh along the jet axis from the reference zero point. The horizontal width of the i-th slice (i=1,2,…,n) in the mask of the X-direction camera 12 and the mask of the Y-direction camera 13 is extracted and converted into physical dimensions d(x,i) and d(y,i), which are used as the X-direction orthogonal diameter 26 and Y-direction orthogonal diameter 27 of the elliptical section 25 at that height.

[0085] The cross-sectional area of ​​the i-th slice is calculated using the ellipse area formula S(i)=π·d(x,i)·d(y,i) / 4. The product of the cross-sectional area of ​​each slice and its thickness Δh is accumulated along the axial direction to obtain the real-time three-dimensional volume of the Taylor cone 3.

[0086] V = Σ π·d(x,i)·d(y,i)·Δh / 4 (i=1,2,…,n).

[0087] Where Δh is the physical thickness of the slice. Under the steady-state condition where the Taylor cone 3 is approximately axisymmetric, the two orthogonal diameters d(x,i) and d(y,i) are approximately equal, and the elliptical section 25 degenerates into a circular section. The results of the elliptical section integration method are consistent with those of the traditional circular section rotational volume integral method. Under dynamic conditions, the elliptical section integration method can capture the anisotropic changes in the cross-sectional shape, thereby reflecting the asymmetric deformation of the Taylor cone 3.

[0088] The Taylor cone volume estimation module 20 also includes a morphological asymmetry early warning unit. For each frame of the image, the early warning unit calculates the ratio of the absolute value of the difference between the two orthogonal diameters on each horizontal slice 24 to the larger of the two values, and takes the average of this ratio across all slice layers as the cross-sectional average eccentricity exponent E for that frame.

[0089] E = (1 / n) · Σ |d(x,i)-d(y,i)| / max(d(x,i),d(y,i)), i=1,2,…,n.

[0090] The eccentricity index E is used to quantify the degree of asymmetry of the Taylor cone 3. In this embodiment, based on statistical analysis under steady-state spinning conditions, the anomaly threshold E is set. th Set to 0.15 (range 0.10~0.25), and the number of consecutive frames M is set to 5 frames (range 3~10 frames). When the detected E value exceeds E for M consecutive frames... th When the Taylor cone 3 is determined to have undergone asymmetric deformation, the voltage regulation output of the voltage closed-loop control module 21 is automatically paused and a warning signal is triggered; when the E value recovers to E for M consecutive frames... th In the following case, the closed-loop control of the voltage closed-loop control module 21 is automatically restored. In actual operation, the E... th Both M and M can be adjusted within the above range according to the process conditions.

[0091] The voltage closed-loop control module 21 receives the real-time three-dimensional volume value output by the Taylor cone volume estimation module 20 as a feedback signal. After comparing it with the preset volume target value, it automatically adjusts the output voltage of the high-voltage power supply 11 through the closed-loop controller, so that the real-time three-dimensional volume tracks the volume target value, thereby maintaining the volume stability of the Taylor cone 3. In this embodiment, the closed-loop controller adopts an incremental PID control algorithm, and the control parameter is configured as K. p =0.5, K i =0.10, K d =0.05, the spinning voltage operating range is set to 4.8kV to 5.4kV, and a limiting protection is triggered when the voltage exceeds this operating range. Those skilled in the art can use other closed-loop control algorithms with the same function, such as fuzzy control or model predictive control, according to actual needs. The PID parameters and the operating range should be adjusted according to the actual process conditions.

[0092] like Figure 5 As shown, during the near-field direct writing process of melt electrospinning, when the collecting plate 6 moves in two dimensions along the horizontal direction, the jet 4 cannot reach directly below the nozzle 1 instantaneously due to its own mass inertia and viscoelastic properties. This causes the actual landing point 28 of the jet 4 to be horizontally offset relative to the target landing point 29 directly below the nozzle 1 along the direction of movement of the collecting plate 6 relative to the nozzle 1, forming a jet hysteresis effect. The horizontal offset between the actual landing point 28 and the target landing point 29 is the jet hysteresis, and its representation as a vector is the two-dimensional hysteresis vector.

[0093] The physical cause of the jet hysteresis effect can be explained by the force balance of the jet element: such as Figure 5 As shown, any infinitesimal mass element 30 (dm) on the jet 4 is subjected to three main forces: gravity 31 (dm·g), electric field force 32 (dF), and gravitational force 31 (dm·g). e) and air resistance 33 (dF a The gravity 31 acts vertically downward on the micro-element 30; the electric field 32, generated by the high-voltage electrostatic field between the nozzle tip 2 and the collecting plate 6, acts on the charge carried by the micro-element 30, pointing from the micro-element 30 to the collecting plate 6, and is the main driving force for stretching the jet 4 and depositing it on the collecting plate 6; the air resistance 33 is generated by the motion of the jet 4 relative to the surrounding air, and its direction is opposite to the motion direction of the micro-element 30 relative to the air. When the collecting plate 6 is stationary, the three forces are symmetrically distributed along the central axis of the jet 4, and the jet 4 extends vertically directly below the nozzle 1; however, when there is relative motion between the nozzle 1 and the collecting plate 6, the stretching direction of the upper section of the jet 4 deflects relative to the vertical direction, and the resultant force of the three forces is no longer vertical, causing the jet 4 to bend as a whole, and the actual landing point 28 shifts horizontally relative to the target landing point 29 along the direction of relative motion.

[0094] When the nozzle 1 moves at a constant speed relative to the collecting plate 6 along a straight line and the system reaches dynamic equilibrium, the magnitude and direction of the two-dimensional hysteresis vector remain stable. However, when the direction of motion of the nozzle 1 relative to the collecting plate 6 changes abruptly at a corner, the two-dimensional hysteresis vector needs to transition from the original direction to a new direction. This transition process has dynamic characteristics and is the main reason why the actual fiber deposition trajectory at the corner deviates from the target trajectory.

[0095] like Figure 6 As shown, in the measurement coordinate system of the two-dimensional hysteresis vector, with the position point of the nozzle tip 2 as the origin, and the positive directions of the X-axis and Y-axis of the three-axis motion platform 7 as the positive directions of the horizontal and vertical coordinate axes respectively, the two-dimensional hysteresis vector L is expressed in the form of two-dimensional coordinate components: L = (L x ,L y ).

[0096] The hysteresis vector measurement module 22 receives the dual-view keypoint coordinates output by the visual feature extraction module 19, extracts the components of the two-dimensional hysteresis vector in the corresponding coordinate axis directions from the views of the X-direction camera 12 and the Y-direction camera 13, and synthesizes them into a complete two-dimensional hysteresis vector. The specific calculation method is as follows:

[0097] The hysteresis vector X-direction component 34 (L) x The following is calculated from the image of the X-direction camera 12:

[0098] L x = (u drop - u tip ) × k x

[0099] Where u drop and u tip k represents the horizontal pixel coordinates of the landing point of the jet 4 and the nozzle tip 2 in the image acquired by the X-direction camera 12, respectively. x is the pixel equivalent coefficient of the X-direction camera 12.

[0100] The hysteresis vector Y-direction component 35 (L) y The following is calculated from the image of the Y-direction camera 13:

[0101] L y = (p drop - p tip ) × k y

[0102] Where p drop and p tip k represents the horizontal pixel coordinates of the jet 4 landing point and the nozzle tip 2 in the image acquired by the Y-direction camera 13, respectively. y The pixel equivalent coefficient is the value of the Y-direction camera 13. In this embodiment, the pixel equivalent coefficients of the X-direction camera 12 and the Y-direction camera 13 obtained by the camera calibration program are 29.1 μm / pixel and 29.3 μm / pixel, respectively.

[0103] The magnitude and direction angle 36 of the two-dimensional hysteresis vector L are further expressed as:

[0104] |L| = √(L x 2 + L y 2 )

[0105] θ L = arctan(L y / L x )

[0106] Where |L| is the magnitude of the two-dimensional hysteresis vector, θ L The direction angle 36° is the two-dimensional hysteresis vector relative to the positive X-axis. Under the conditions of the calibration accuracy of the pixel equivalent coefficient and the key point detection accuracy of the visual feature extraction module 19, the measurement accuracy of the two-dimensional hysteresis vector is conservatively estimated to be ±0.1 mm.

[0107] The trajectory pre-compensation module 23 uses a first-order inertial differential equation dynamic response model to describe the dynamic response characteristics of the two-dimensional hysteresis vector:

[0108] τ·(dL / dt) + L = K·v

[0109] Where L is the two-dimensional hysteresis vector, v is the velocity vector of the nozzle 1 relative to the collection plate 6, τ is the time constant of the dynamic response model of the first-order inertial differential equation, K is the steady-state gain coefficient, and both τ and K are functions of the relative velocity magnitude |v|.

[0110] The first-order inertial vector equation holds independently in the X and Y directions, and expands into two component equations:

[0111] τ·(dL x / dt) + L x = K·v x

[0112] τ·(dL y / dt) + L y = K·v y

[0113] Where v x and v y These are the components of the velocity vector v in the X and Y directions, respectively. The component equations in both directions have the same time constant τ and steady-state gain coefficient K.

[0114] The steady-state solution of the first-order inertial differential equation dynamic response model when v is constant is L = K·v; when v undergoes a step change, the dynamic process of the two-dimensional hysteresis vector L transitioning from the original steady state to the new steady state according to the first-order exponential law is described. The transition time of this transition process is determined by the time constant τ. After v undergoes a step change, the steady state can be basically restored after about 3 times τ.

[0115] The parameters τ and K of the dynamic response model of the first-order inertial differential equation are pre-identified through a corner direction change experiment. The specific experimental scheme is as follows: an L-shaped test trajectory is designed, consisting of two straight line segments, each 50 mm long and orthogonal to each other, with a 90° direction change at the corner. The collecting plate 6 moves along the L-shaped test trajectory at six relative motion speed levels: 20 mm / s, 25 mm / s, 30 mm / s, 35 mm / s, 40 mm / s, and 45 mm / s, with each speed level repeated five times. During the movement of the collecting plate 6, the orthogonal dual-camera visual monitoring module 18 and the hysteresis vector measurement module 22 are used to synchronously collect the time-series change data of the two-dimensional hysteresis vector before and after the corner. The least squares method is used to perform curve fitting on the time-series data collected at each speed level. That is, by solving for the parameter value that minimizes the sum of squared errors between the predicted and measured values ​​of the dynamic response model of the first-order inertial differential equation, the corresponding τ and K at each relative motion speed are obtained.

[0116] Furthermore, based on the τ and K identified at the six speed levels, a mapping relationship between τ and K with respect to the relative motion speed |v| is established using a linear interpolation method. This mapping relationship is stored in the trajectory pre-compensation module 23 for subsequent trajectory compensation as a lookup reference. A typical fitting curve of the parameter identification results is shown below. Figure 7 As shown.

[0117] Specifically, Figure 7 (a), (b), and (c) show the measured dynamic response data of the two-dimensional hysteresis vector of the collecting plate 6 relative to the nozzle 1 at three typical velocity levels of 20 mm / s, 30 mm / s, and 45 mm / s, respectively, and the fitting curves of the dynamic response model of the first-order inertial differential equation. In the identification results at the six velocity levels, the goodness of fit R for each working condition is... 2 All values ​​were above 0.98, verifying the good descriptive ability of the first-order inertial differential equation dynamic response model for the dynamic response of the two-dimensional hysteresis vector in the relative motion velocity range of 20 mm / s to 45 mm / s.

[0118] The identification results further indicate that the time constant τ decreases monotonically with the increase of the velocity of the nozzle 1 relative to the collecting plate 6, decreasing from 284.8 ms at 20 mm / s to 43.9 ms at 45 mm / s; the steady-state gain K remains stable between 0.0715 s and 0.0907 s within the measured velocity range. The trends of τ and K reflect the physical characteristics of the viscoelastic response of the jet 4 as a function of relative velocity. The greater the relative velocity, the stronger the drag effect of the deposited fiber 5 on the jet 4, and the faster the jet 4 is pulled to a new equilibrium position, resulting in a corresponding decrease in the time constant τ.

[0119] In this embodiment, the mapping relationship between τ and K with respect to relative motion speed is established by linear interpolation of the parameter values ​​identified at the six speed levels. When the speed of the nozzle 1 relative to the collection plate 6 exceeds the range of 20 mm / s to 45 mm / s covered by the identification experiment, the parameter identification experiment of the first-order inertial differential equation dynamic response model needs to be repeated to obtain the τ and K parameters at the corresponding speed range.

[0120] like Figure 8 As shown, the trajectory pre-compensation module 23 adopts a two-level compensation strategy that combines steady-state global offset and corner dynamic transition compensation to perform offline correction of the original CNC trajectory command.

[0121] The first level is steady-state global offset compensation: for a straight line segment moving at a constant velocity, the steady-state value of the two-dimensional hysteresis vector is L. ss= K·v·d, where d is the unit vector of the trajectory segment's motion direction, and v is the magnitude of the relative motion velocity corresponding to the trajectory segment. The trajectory pre-compensation module 23 offsets each trajectory point in the original CNC trajectory command by a distance |K·v| in the opposite direction of the trajectory segment's motion direction d, so that the landing point of the jet 4 after compensation returns exactly to the target trajectory, thereby eliminating the constant steady-state hysteresis deviation of the straight segment.

[0122] The second level is corner dynamic transition compensation: Assume the nozzle 1 moves relative to the collection plate 6 along direction d1, and after passing the corner point O at time t=0, the relative motion direction abruptly changes to d2, where d1 and d2 are both unit vectors. From the step response solution of the first-order inertial differential equation dynamic response model, the evolution of the two-dimensional hysteresis vector over time after the corner is as follows:

[0123] L(t) = L2 + (L1 - L2)·e -t / τ

[0124] Where L1 = K·v·d1 is the steady-state lag vector before the corner, and L2 = K·v·d2 is the steady-state lag vector after the corner. The difference between the steady-state lag vectors before and after the corner is defined as follows:

[0125] ΔL = K·v·(d2 - d1)

[0126] Assuming the steady-state hysteresis has been eliminated by the first-level global offset, the dynamic deviation that still needs additional compensation at the corner is ΔL·e. -t / τ Therefore, the compensation trajectory of nozzle 1 within the transition zone after the corner can be obtained as follows:

[0127] P c (t) = O + v·t·d2 - ΔL·e -t / τ

[0128] In the formula, O represents the corner point. The above formula indicates that the compensated trajectory deviates by |ΔL| relative to the original straight segment at the corner point along the -ΔL direction. The deviation then decays exponentially. When t≈3τ, the deviation decays to less than 5% of the initial value, and the transition process is essentially complete. In the above compensation formula, τ and K are parameters related to relative motion velocity. In actual compensation calculations, the corresponding parameter values ​​are obtained by linear interpolation based on the relative motion velocity |v| of the trajectory segment where the corner is located, using the mapping relationship between τ and K with respect to relative motion velocity.

[0129] Furthermore, the continuous compensation trajectory needs to be discretized to convert it into the CNC trajectory command. The effective compensation length of the transition zone after the corner is set to L. comp= N·τ·|v|, where N is the multiplier coefficient of the coverage transition time (ranging from 2 to 5, N=3 in this embodiment), and the effective compensation length is divided into M′ segments (ranging from 5 to 20, M′=10 in this embodiment). Let t be the time corresponding to the end of the kth segment. k = k·N·τ / M', then the coordinates of the kth compensation point are:

[0130] P c,k = O + (k / M')·L comp ·d2 - ΔL·e -N·k / M' k=1,2,…,M′

[0131] The calculated M' compensation points are sequentially inserted into the numerical control trajectory sequence following the corner point O, thus completing the dynamic transition compensation for the corner. Those skilled in the art can adjust the specific values ​​of N and M' within the above range according to actual process accuracy requirements, or adopt other discretization strategies.

[0132] The synergistic effect of the two-level compensation strategy covers the complete compensation requirements of the straight section before the corner, the corner point, and the transition zone after the corner: the first-level global offset makes the nozzle 1 always "ahead" of the target trajectory, naturally forming an overshoot effect at the corner; the second-level dynamic compensation precisely controls the transition process after the corner, so that the landing trajectory of the jet 4 gradually converges to the target trajectory.

[0133] Based on the above two-level compensation strategy, the complete processing flow of the trajectory pre-compensation module 23 is as follows: Figure 9 As shown. The processing flow includes the following steps:

[0134] First, the original CNC trajectory instructions are parsed, converting the instruction sequence into an ordered sequence of trajectory points {P}. i} and the corresponding feed rate sequence {F i For circular interpolation instructions (such as G02 and G03) in CNC trajectory instructions, the circular arc is discretized into several short straight lines by a fixed angle increment Δθ=5°, and then uniformly converted into a linear interpolation instruction (G01) for processing.

[0135] Secondly, perform the first-level steady-state global offset compensation: for each trajectory point in the sequence of trajectory points obtained from the analysis, offset by a distance |K·v| in the opposite direction of the motion direction d of the trajectory segment to eliminate the steady-state lag deviation of the straight line segment. K and v are the steady-state gain coefficient and relative motion velocity vector corresponding to the current trajectory segment, respectively.

[0136] Then, the corner points are identified by traversing the trajectory point sequence one by one, and a second-level dynamic transition compensation for corners is performed. For the i-th trajectory point P... iCalculate the angle θ between its incident direction vector and its exit direction vector. i When the included angle is strictly greater than the preset corner recognition threshold θ th (The value range is 3°~10°, and in this embodiment θ) th When the angle is 5°, determine the trajectory point P. i The corner point is identified, and its geometric features, such as the incident unit vector d1 and the outgoing unit vector d2, are extracted. Based on the relative velocity magnitude |v| at the corner, the corresponding τ and K are obtained through linear interpolation within the mapping relationship between τ and K and relative velocity. Then, the difference in steady-state hysteresis vectors before and after the corner, ΔL = K·v·(d2 - d1), is calculated, and the coordinates P of M′ compensation points are calculated using the aforementioned discretization formula. c,k (k=1,2,…,M′), the M′ compensation points are sequentially inserted into the trajectory sequence following the corner point. The corner recognition threshold θ th The increment Δθ of the discretized arc angle is taken to be the same value of 5°, and the judgment condition is strictly greater than 5°, so as to ensure that the 5° directional change generated by the discretization of the arc is not misjudged as a corner point.

[0137] Furthermore, when there are adjacent corners in the trajectory point sequence and the distance between the two corners is less than the effective compensation interval length L... comp When this is done, a truncation strategy is adopted: the end of the compensation interval of the previous corner is cut off to the position of the next corner point, and the number of compensation points in the compensation interval of the previous corner is reduced accordingly, so as to ensure the independence of each corner compensation interval and avoid the unexpected superposition of compensation amount.

[0138] Finally, the corrected trajectory point sequence after two-level compensation is re-encoded into a standard CNC trajectory instruction format for output. The corner points and inserted compensation points are represented by linear interpolation instructions (G01). The feed rate of each instruction is inherited from the feed rate setting of the corresponding trajectory segment in the original CNC trajectory instruction. The corrected CNC trajectory instruction can be directly loaded into the motion controller 16 for execution.

[0139] like Figure 10 As shown, the control effect of the voltage closed-loop control module 21 was experimentally verified. In the experiment, the target volume value of the Taylor cone 3 was increased from approximately 0.34 mm in its natural steady state. 3 Set to 0.22mm 3 Under the same basic process parameters, the spinning process is run in open-loop mode (the spinning voltage is constant at the initial set value of 5.0kV) and closed-loop mode (the voltage closed-loop control module 21 automatically adjusts the spinning voltage according to the preset PID control algorithm), and the volume response curve of the Taylor cone 3 and the response curve of the spinning voltage are recorded. Figure 10In Figure (a), the volume response curve of the Taylor cone 3 is shown. Figure 10 Figure (b) shows the response curve of the spinning voltage. Experimental results show that the volume of the Taylor cone 3 in open-loop mode is approximately 0.339 mm. 3 Fluctuations were observed, with a volume variation coefficient of approximately 1.0%. In closed-loop mode, the system entered a steady state after approximately 35 seconds of adjustment, with a steady-state average volume of 0.249 mm². 3 The steady-state volume variation coefficient decreased to 0.8%, a reduction of 26.5% compared to the open-loop mode. Further comparative experiments on fiber diameter showed that the diameter variation coefficient of the fiber 5 decreased from 7.92% in the open-loop mode to 2.21% in the closed-loop mode, an improvement rate of 72.10%, verifying the effectiveness of the voltage closed-loop control module 21 in suppressing the volume drift of the Taylor cone 3 and improving the diameter uniformity of the fiber 5.

[0140] like Figure 11 As shown, the compensation effect of the trajectory pre-compensation module 23 was experimentally verified. Figure 11 (a), (b), and (c) show the deposition trajectory of the fiber 5 before and after compensation when the collecting plate 6 moves at a relative speed of 30 mm / s with test trajectories at 45°, 90°, and 135° corners, respectively. The experiment used normalized trajectory error as the evaluation index. The results showed that the trajectory error improvement rate was 91.77% for the 45° corner, 90.16% for the 90° corner, and 83.90% for the 135° corner. Furthermore, the 90° corner compensation experiment conducted at six relative speed levels (20 mm / s, 25 mm / s, 30 mm / s, 35 mm / s, 40 mm / s, and 45 mm / s) showed that the trajectory error improvement rate was above 84% at each speed level, verifying the engineering applicability of the trajectory pre-compensation module 23 within the tested corner angle range and relative speed range.

[0141] The working process of each module of the system described above corresponds to the steps of the method of the present invention: Step S1 is executed offline by the trajectory pre-compensation module 23 before printing begins, performing two-level pre-compensation processing on the original CNC trajectory command based on the pre-identified dynamic response model parameters of the first-order inertial differential equation and generating the compensated CNC trajectory command; Step S2 loads the compensated CNC trajectory command into the motion controller 16 to drive the three-axis motion platform 7 to perform printing; Step S3 is executed by the orthogonal dual-camera visual monitoring module 18, using the X-direction camera 12 and the Y-direction camera 13 to simultaneously acquire the orthogonal projection images of the Taylor cone 3 and the jet 4; Step S4 is executed by the visual feature extraction module 19, performing instance segmentation and key point detection on the dual-view images respectively, and outputting the dual-view segmentation mask as well as the position of the nozzle tip 2 and the landing point of the jet 4. The pixel coordinates of the position; Step S5 is executed by the Taylor cone volume estimation module 20, which obtains the real-time three-dimensional volume of the Taylor cone 3 based on the dual-view segmentation mask using the dual-view elliptical section integration method; Step S6 is executed by the voltage closed-loop control module 21, which automatically adjusts the spinning voltage output by the high-voltage power supply 11 using the real-time three-dimensional volume value of the Taylor cone 3 as a feedback signal; Step S7 is executed by the hysteresis vector measurement module 22, which extracts and synthesizes a complete two-dimensional hysteresis vector from the dual-view coordinates; Steps S3 to S7 run concurrently during the printing process of the three-axis motion platform 7, and the offline trajectory pre-compensation of Step S1 and the online voltage closed-loop control of Step S6 act on two independent physical channels, namely the numerical control trajectory command received by the motion controller 16 and the spinning voltage output by the high-voltage power supply 11, respectively, without interfering with each other in timing.

[0142] The overall performance of the system was verified using the fabrication of a Janus unidirectional water transport structure as an application example. The Janus unidirectional water transport structure consists of eight orthogonal meshes with dimensions of 25mm × 25mm, and four levels of fiber spacing gradients along the stacking direction: 600μm for the bottom two layers (layers 1 and 2), 500μm for layers 3 and 4, 400μm for layers 5 and 6, and 300μm for the top two layers (layers 7 and 8), thus forming a gradient porosity structure decreasing from the bottom to the top. During the printing process, the voltage closed-loop control module 21 and the trajectory pre-compensation module 23 were simultaneously activated. Experimental results show that the coefficient of variation of the fiber diameter 5 in the fabricated structure is 2.46%, and the layer spacing error is controlled within 5%. The test results of the unidirectional water transport function show that the experimental group (with voltage closed-loop control and trajectory pre-compensation enabled simultaneously) achieved a forward penetration time of approximately 2 seconds and failed to penetrate in the reverse direction after 60 seconds, exhibiting significant unidirectional water transport characteristics. In contrast, the control group (with the voltage closed-loop control module 21 and trajectory pre-compensation module 23 disabled) also failed to penetrate in the forward direction under the same structural design. These comparative results demonstrate that the synergistic effect of the voltage closed-loop control module 21 and the trajectory pre-compensation module 23 is a necessary condition for the reliable realization of the Janus unidirectional water transport function, verifying the comprehensive performance of the system of this invention in the fabrication of functional micro / nano structures.

[0143] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Various changes, modifications, equivalent substitutions, and improvements made by those skilled in the art based on the technical solutions of the present invention should be included within the scope of protection of the present invention. For example:

[0144] (1) The spinning material is not limited to polycaprolactone used in this embodiment. Polylactic acid, polycaprolactone-polylactic acid blends and other thermoplastic polymers that can be melt-spun can also be used. However, after changing the spinning material, the time constant τ and steady-state gain K of the dynamic response model of the first-order inertial differential equation should be re-identified.

[0145] (2) The dynamic response model of the first-order inertial differential equation is not limited to the form described in this embodiment. Those skilled in the art can adopt a second-order inertial model or other higher-order dynamic response models according to actual accuracy requirements.

[0146] (3) The closed-loop controller in the voltage closed-loop control module 21 is not limited to the incremental PID control algorithm used in this embodiment. Those skilled in the art can use other closed-loop control algorithms with the same function, such as fuzzy control, model predictive control, and sliding mode control, according to actual needs.

[0147] (4) The abnormal threshold E of the morphological asymmetry early warning unit in the Taylor cone volume estimation module 20 thAnd the number of consecutive frames M, the effective compensation interval length coefficient N, the number of compensation points M', and the corner recognition threshold θ of the trajectory pre-compensation module 23 th And parameters such as the discretized arc angle increment Δθ can be adjusted by those skilled in the art within the range of values ​​given in the claims according to actual process conditions and accuracy requirements;

[0148] (5) The instance segmentation model and key point detection model of the visual feature extraction module 19 are not limited to the YOLO11n-seg and YOLO11n-pose models used in this embodiment. Those skilled in the art can use other neural network models with the same function according to actual needs.

[0149] (6) The specific hardware configuration of the orthogonal dual-camera visual monitoring module 18, the specific model of the high-voltage power supply 11, and the specific model of the motion controller 16 do not constitute a substantial limitation on the technical solution of the present invention.

[0150] Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention shall be included within the scope of protection of this invention.

Claims

1. A smart monitoring and control system for melt electrospinning based on orthogonal dual vision, characterized in that, include: The orthogonal dual-camera vision monitoring module includes an X-axis camera and a Y-axis camera whose optical axes are orthogonal to each other and whose respective optical axes are perpendicular to the nozzle axis and intersect at the horizontal plane where the nozzle tip is located. These cameras are used to simultaneously acquire orthogonal projection images of the Taylor cone and the jet during the electrospinning process of the melt. The visual feature extraction module processes the dual-view images acquired by the orthogonal dual-camera visual monitoring module and outputs the instance segmentation mask of the Taylor cone region under the dual view, as well as the pixel coordinates of the nozzle tip position and the jet landing point position. The Taylor cone volume estimation module, based on the instance segmentation mask of the Taylor cone region under the dual-view perspective, discretizes the Taylor cone along the axis into multiple horizontal slices. Each slice cross section is regarded as an ellipse. The contour width of the two orthogonal views is used to determine the two orthogonal diameters of each elliptical cross section. The cross-sectional area of ​​each slice is accumulated along the axis to obtain the real-time three-dimensional volume of the Taylor cone. The voltage closed-loop control module uses the real-time three-dimensional volume as a feedback signal and compares it with a preset volume target value. The closed-loop controller adjusts the spinning voltage output by the high-voltage power supply in real time during the printing process so that the real-time three-dimensional volume tracks the volume target value. The hysteresis vector measurement module extracts the components of the two-dimensional hysteresis vector in the directions of two orthogonal coordinate axes based on the pixel coordinates of the nozzle tip position and the jet landing point position under dual perspectives, and synthesizes them into a complete two-dimensional hysteresis vector for use in hysteresis monitoring and modeling of the dynamic response model of the inertial differential equation. The trajectory pre-compensation module stores a pre-identified dynamic response model of the inertial differential equation, which describes the dynamic process of the two-dimensional hysteresis vector transitioning from the original steady state to the new steady state when the relative motion direction between the nozzle and the collection plate changes. Before the three-axis motion platform executes printing, the trajectory pre-compensation module performs offline correction on all trajectory points of the original CNC trajectory command based on the dynamic response model of the inertial differential equation, and outputs the corrected CNC trajectory command to the motion controller.

2. The intelligent monitoring and control system for melt electrospinning based on orthogonal dual vision as described in claim 1, characterized in that, The Taylor cone volume estimation module reconstructs the real-time three-dimensional volume as follows: Using the nozzle tip position as the common reference zero point for both perspectives, the pixel coordinates are converted into physical coordinates using a pre-calibrated pixel equivalent coefficient, so that the instance segmentation mask of the Taylor cone region under both perspectives is aligned in the physical height space. The aligned mask region is uniformly divided into n horizontal slices with a thickness of Δh along the jet axis from the reference zero point. For the i-th slice, extract its horizontal widths d(x,i) and d(y,i) from the instance segmentation mask corresponding to the camera in the X direction and the instance segmentation mask corresponding to the camera in the Y direction, respectively, as the two orthogonal diameters of the current slice's elliptical cross-section; calculate the cross-sectional area of ​​each slice according to the elliptical area formula, and accumulate the product of each slice's cross-sectional area and slice thickness Δh along the axial direction to obtain the real-time three-dimensional volume V of the Taylor cone: V = Σ π·d(x,i)·d(y,i)·Δh / 4 i=1,2,…,n.

3. The intelligent monitoring and control system for melt electrospinning based on orthogonal dual vision as described in claim 2, characterized in that, The Taylor cone volume estimation module also includes a morphological asymmetry early warning unit; For each frame of image, the morphological asymmetry early warning unit calculates the ratio of the absolute value of the difference between the two orthogonal diameters of each slice layer to the larger value, and takes the average value of the corresponding ratios of all slice layers as the cross-sectional average eccentricity exponent E of the frame. When the detected E values ​​in M ​​consecutive frames all exceed the abnormal threshold E th When an asymmetric deformation of the Taylor cone is detected, the morphological asymmetry early warning unit triggers the voltage closed-loop control module to pause voltage regulation output and issue an early warning signal; when the E value recovers to E for M consecutive frames... th In the following cases, the voltage closed-loop control will be automatically restored.

4. The intelligent monitoring and control system for melt electrospinning based on orthogonal dual vision as described in claim 3, characterized in that, Abnormal threshold E th The value range is 0.10~0.25, and the value range of consecutive frame number M is 3~10 frames.

5. The intelligent monitoring and control system for melt electrospinning based on orthogonal dual vision as described in claim 1, characterized in that, The dynamic response model of the inertial differential equation is in the following form: τ·(dL / dt) + L = K·v Where L is the two-dimensional hysteresis vector, v is the velocity vector of the nozzle relative to the collection plate, τ is the time constant, K is the steady-state gain coefficient, and both τ and K are functions of the relative velocity magnitude |v|. The steady-state solution of the dynamic response model of the inertial differential equation when v is constant is L=K·v; when v changes, the dynamic response model of the inertial differential equation describes the dynamic process of the two-dimensional hysteresis vector L transitioning from the original steady state to the new steady state according to the first-order exponential law.

6. The intelligent monitoring and control system for melt electrospinning based on orthogonal dual vision as described in claim 5, characterized in that, The time constant τ and the steady-state gain coefficient K are pre-identified and stored in the trajectory pre-compensation module in the following manner: Within the system's operating speed range, select at least three different relative motion speed levels and execute test trajectories that include sudden changes in direction at corners. The orthogonal dual-camera visual monitoring module and the hysteresis vector measurement module are used to synchronously collect the temporal change data of the two-dimensional hysteresis vector before and after the corner; The least squares method was used to fit τ and K at each relative motion speed level to the collected data, and the mapping relationship between τ and K with respect to relative motion speed was established by linear interpolation.

7. The intelligent monitoring and control system for melt electrospinning based on orthogonal dual vision as described in claim 5, characterized in that, The trajectory pre-compensation module uses a two-level compensation strategy to perform offline correction of the original CNC trajectory command: The first level is global offset compensation, which offsets each trajectory point in the original CNC trajectory command by a distance of |K·v| in the opposite direction of the motion direction of the trajectory segment, where v is the motion velocity vector corresponding to the trajectory segment; The second level is dynamic transition compensation for corners. For each trajectory point in the original CNC trajectory command, the angle between the direction of the previous trajectory segment and the direction of the next trajectory segment is greater than the corner recognition threshold θ. th The trajectory points are determined as corner points; for each corner point, the corresponding time constant τ and steady-state gain coefficient K are determined according to the relative velocity |v| at the corner point, and the difference vector ΔL between the steady-state lag vectors before and after the corner point is calculated; M′ compensation points are generated in the compensation interval of length N·τ·|v| after the corner point, where N is a multiplier coefficient, and the position offset of each compensation point decays from |ΔL| to zero along the direction of motion according to a first-order exponential decay law; The compensation points are sequentially inserted into the trajectory sequence after the corner point, and the corrected trajectory sequence is re-encoded into CNC trajectory instructions for output.

8. The intelligent monitoring and control system for melt electrospinning based on orthogonal dual vision as described in claim 7, characterized in that, Corner recognition threshold θ th The value range is 3° to 10°, the multiplier N ranges from 2 to 5, and the number of compensation points M′ ranges from 5 to 20.

9. A method for intelligent monitoring and control of melt electrospinning based on orthogonal dual vision, characterized in that, The method is implemented using the intelligent monitoring and control system for melt electrospinning as described in any one of claims 1-8, and includes an offline pretreatment stage and an online printing stage. The offline preprocessing stage includes: based on the pre-identified dynamic response model of the inertial differential equation, performing two-level corrections on all trajectory points of the original CNC trajectory command, namely global offset compensation and corner dynamic transition compensation, to generate the compensated CNC trajectory command. The online printing stage includes the following steps: The compensated CNC trajectory command is loaded into the motion controller to drive the three-axis motion platform to perform printing. The orthogonal projection images of the Taylor cone and the jet are acquired simultaneously using two cameras with mutually orthogonal optical axes; The Taylor cone region of the acquired dual-view images is segmented into instances to obtain the instance segmentation mask of the Taylor cone region under dual-view conditions, and the pixel coordinates of the nozzle tip position and the jet landing point position under dual-view conditions are detected. Based on the instance segmentation mask of the Taylor cone region under the dual-view perspective, the real-time three-dimensional volume of the Taylor cone is reconstructed using the dual-view elliptical section integration method. The dual-view elliptical section integration method discretizes the Taylor cone into multiple horizontal slices along the axial direction. Each slice is regarded as an ellipse. The two orthogonal diameters of each elliptical slice are determined by the contour width of the two orthogonal perspectives. The real-time three-dimensional volume of the Taylor cone is obtained by accumulating the product of the elliptical area and the slice thickness of each slice along the axial direction. Using the real-time three-dimensional volume as a feedback signal, the spinning voltage is adjusted in real time by a closed-loop controller so that the real-time three-dimensional volume tracks the preset volume target value. Based on the coordinates of the nozzle tip position and jet landing point position of the dual-view nozzle, a complete two-dimensional hysteresis vector is synthesized for use in hysteresis monitoring and modeling of the dynamic response model of the inertial differential equation.

Images