A precision alignment mechanical transmission mechanism and control system for anti-counterfeiting screen printing
By using four sets of independent electric telescopic rods in parallel drive and CCD vision full positioning closed loop, combined with deep learning adaptive control, high-precision alignment of anti-counterfeiting screen printing is achieved, solving the problems of registration accuracy and equipment stability, and possessing adaptive and predictive maintenance capabilities.
Patent Information
- Application Number
- CN202611177882.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-05
- Publication Date
- 2026-11-10
AI Technical Summary
In existing anti-counterfeiting screen printing technologies, the registration accuracy is difficult to achieve within ±0.01mm, the equipment cost is high, and there is a lack of real-time monitoring and predictive maintenance capabilities for equipment operation status, making it difficult to meet the needs of long-term high-precision production.
It adopts four sets of independent electric telescopic rods in parallel drive, combined with CCD vision full positioning closed loop and deep learning adaptive control algorithm to achieve precise alignment of eight motion modes in the plane. It is equipped with an embedded motion controller for three-level closed loop control and has equipment health monitoring and predictive maintenance functions.
It significantly improves alignment accuracy and operational stability, has adaptive capabilities, reduces unplanned downtime, and meets the needs of long-term high-precision production.
Smart Images

Figure CN123061009A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of screen printing technology, and in particular to a precision alignment mechanical transmission mechanism and control system for anti-counterfeiting screen printing. Background Technology
[0002] As a core process for secure printing of banknotes, certificates, and high-end anti-counterfeiting labels, anti-counterfeiting screen printing has extremely high requirements for registration accuracy, typically requiring registration error to be controlled within ±0.01mm.
[0003] A search revealed that, in the prior art, such as the Chinese utility model patent with authorization announcement number CN205615162U, a sheet screen printing machine with automatic screen alignment is disclosed. It uses four sets of XYθ alignment units to drive the screen alignment. Although the structure of three sets of active units plus one set of follower units achieves automatic alignment, the multi-axis series stacking arrangement causes the error to be transmitted layer by layer, resulting in significant Abbe error. The moving parts have large mass and insufficient rigidity, and are prone to vibration during high-speed alignment. Ultimately, the positioning accuracy is limited and the equipment cost is high.
[0004] For example, Chinese invention patent application CN117419749A discloses a visual alignment calibration method and related device for a UVW alignment platform. Although the UVW alignment platform adopts a coplanar parallel structure to improve rigidity, its kinematic calculation is complex, and the control system generally adopts the traditional PID algorithm with fixed parameters. It is difficult to maintain optimal performance under different loads and different speed conditions, resulting in insufficient alignment accuracy and stability. At the same time, the position feedback of this type of platform mostly relies on the encoder semi-closed loop or the grating ruler in a limited direction, which cannot fully cover the detection of motion deviation in any direction in the plane.
[0005] For example, Chinese utility model patent CN205439549U discloses a screen printing system for automatic alignment of roll materials. It achieves positioning through a four-axis automatic alignment platform combined with CCD camera, but it only has basic alignment control functions and lacks the ability to perceive and predict the operating status of the equipment. Faults such as wear of transmission pairs, insufficient lubrication, and vacuum leakage are often only discovered after the fact, affecting the continuity of production. Moreover, the thermal deformation and residual vibration generated during long-term operation also lack effective compensation and suppression methods, making it difficult to meet the requirements of long-term continuous high-precision production of anti-counterfeiting printing.
[0006] Therefore, this application proposes a precision alignment mechanical transmission mechanism and its control system for anti-counterfeiting screen printing, which is flexible in movement, more precise, and has intelligent adaptive capabilities and health management functions. Summary of the Invention
[0007] One objective of this invention is to propose a precision alignment mechanical transmission mechanism and its control system for anti-counterfeiting screen printing. This invention can achieve three-degree-of-freedom precision alignment of eight motion modes in a plane through the parallel drive of four independent electric telescopic rods. Combined with CCD vision full positioning closed loop and deep learning adaptive control algorithm, it significantly improves alignment accuracy and operational stability while simplifying the mechanical structure. It also has equipment health monitoring and predictive maintenance functions to meet the needs of long-term continuous high-precision production of anti-counterfeiting screen printing.
[0008] According to an embodiment of the present invention, a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing includes a support mechanism, a drive assembly, a platform mechanism, a motion mechanism, a sensor detection assembly, and an embedded motion controller.
[0009] The support mechanism includes a base, and four shafts are vertically installed at the four corners of the top of the base;
[0010] The drive assembly includes four sets of high-precision electric telescopic rods, which are respectively arranged at the four corners of the top of the base. Each set of high-precision electric telescopic rods has a bushing installed at its fixed end. The bushing is movably fitted onto a shaft, allowing the high-precision electric telescopic rod to swing around the shaft in the horizontal plane. Each set of high-precision electric telescopic rods has a vertically upward connecting shaft fixedly connected to its extended end. Each high-precision electric telescopic rod has a pull rod type displacement sensor integrated inside.
[0011] The motion mechanism includes a movable plate, which is horizontally positioned above four sets of high-precision electric telescopic rods. Bearings are embedded at the four corners of the movable plate, and the connecting shafts are inserted into the inner rings of the bearings and fixedly connected to achieve the hinge connection between the high-precision electric telescopic rods and the movable plate. A servo motor is fixedly installed at the bottom center of the movable plate, and the output shaft of the servo motor is connected upward to a drive shaft. The drive shaft passes vertically through the central through hole of the movable plate, and a high-precision circular grating encoder is mounted on the drive shaft.
[0012] The platform mechanism includes a motion platform, which is horizontally fixedly installed at the top of the transmission shaft, and the upper surface of the motion platform is machined with an array of vacuum adsorption grooves.
[0013] The sensor detection assembly includes a CCD vision alignment module, a photoelectric origin sensor, and a vacuum pressure sensor. The CCD vision alignment module is mounted directly above the motion platform and directly detects the actual position and orientation of the motion platform in any direction in the plane by photographing the alignment marks on the substrate, including X-axis displacement, Y-axis displacement, and rotation angle. The vacuum pressure sensor is connected in series in the vacuum adsorption gas path of the motion platform to monitor the adsorption pressure.
[0014] The embedded motion controller is fixedly installed in the side compartment of the base. The signal input terminal of the controller is connected to each sensor, and the control output terminal is connected to the drivers of four sets of high-precision electric telescopic rods and the drivers of servo motors.
[0015] Furthermore, a fixing hole is provided around the base and centered between adjacent shafts, and mounting holes for mounting the printing worktable are provided at the four corners of the motion platform. An origin baffle is fixedly installed on one side edge of the motion platform, and a photoelectric origin sensor is fixedly installed at the corresponding position on the base.
[0016] Furthermore, a ball bearing seat is fixedly installed at the bottom of the high-precision electric telescopic rod and directly below the connecting shaft. A ball bearing is movably embedded in the ball bearing seat, and the bottom of the ball bearing makes rolling contact with the upper surface of the base.
[0017] Furthermore, the four sets of high-precision electric telescopic rods are arranged symmetrically at a diagonal angle on the top of the base. In the standby state, each set of high-precision electric telescopic rods is at half of its total stroke. The four sets of telescopic rods can synthesize translational motion in any direction in the plane by independently controlling the extension and retraction. This includes at least eight basic motion modes: left front group extension and retraction alone, right front group extension and retraction alone, left rear group extension and retraction alone, right rear group extension and retraction alone, left front and right rear synchronous extension and retraction, left front and left rear synchronous extension and retraction, and right front and right rear synchronous extension and retraction.
[0018] Furthermore, the CCD vision alignment module includes at least two global shutter CCD industrial cameras, equipped with dual telecentric lenses and a coaxial light source, which are respectively aligned with two diagonal alignment marks on the substrate. The camera resolution is no less than 10,000 pixels, and the acquisition frame rate is no less than fps. With the help of a sub-pixel edge extraction algorithm, the alignment detection resolution can reach the sub-micron level. The CCD vision alignment module can detect translation and rotation in any direction in the plane, covering pose feedback for all eight motion modes.
[0019] Furthermore, the high-precision circular grating encoder is fixed in the middle of the drive shaft. The stator of the high-precision circular grating encoder is fixedly mounted on the upper surface of the movable plate by a bracket. The rotor rotates synchronously with the drive shaft. The high-precision circular grating encoder has no less than [number] grating lines and is used to directly detect the actual rotation angle of the motion platform.
[0020] Furthermore, the embedded motion controller adopts an ARM+FPGA heterogeneous architecture control board, which is installed in the electrical control compartment on the side of the base. The ARM processor runs kinematic calculation algorithms, intelligent control algorithms and visual communication programs, while the FPGA runs five-axis synchronous pulse generation, encoder high-speed counting and multi-sensor synchronous acquisition programs. The controller has a built-in three-level closed-loop control logic, which consists of a visual full positioning closed loop, a telescopic rod semi-closed loop and a current inner loop, from the outside to the inside.
[0021] A control system for a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing is provided. The control system is mounted in an embedded motion controller and includes four levels: a three-level closed-loop alignment control subsystem, a deep learning adaptive parameter optimization subsystem, a multi-mode motion planning subsystem, and an equipment health management and predictive maintenance subsystem.
[0022] The three-level closed-loop alignment control subsystem adopts a nested three-loop structure. The outermost layer is a vision-based full positioning loop, which directly calculates the pose deviation in any direction in the plane based on the alignment mark image acquired by the CCD vision alignment module and outputs the target position command, covering all eight motion modes. The middle layer is a telescopic rod semi-closed loop, which uses feedback from the built-in rod-type displacement sensors of each telescopic rod to achieve independent and precise position control of each telescopic rod through an improved PID algorithm combined with a disturbance observer (DOB). The PID control law satisfies the following formula:
[0023]
[0024] in For the first The output volume is controlled by the beat. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients. For the first Shooting position deviation, This is the feedforward compensation amount;
[0025] The innermost layer is the current inner loop, which enables rapid tracking control of motor torque;
[0026] The deep learning adaptive parameter optimization subsystem incorporates a deep reinforcement learning (DQN) network. It uses the current alignment deviation, motion speed, load estimation, and motion pattern as state inputs (s), and the adjustments to the PID parameters and feedforward gain as action outputs (a). The reward function R(s,a) is a weighted value of alignment accuracy, settling time, and vibration amplitude, and is implemented using the Bellman equation. Iteratively update the Q-value table, where For learning rate, As a discount factor, For instant rewards, online self-optimization control parameters are implemented;
[0027] The multi-mode motion planning subsystem has built-in kinematic models of eight basic motion modes, which can automatically select the optimal motion combination based on the target displacement and plan the coordinated motion trajectory of each telescopic rod.
[0028] The equipment health management and predictive maintenance subsystem collects multi-dimensional operating data such as motor current, displacement deviation, and vacuum pressure, performs anomaly detection using the isolated forest algorithm, and predicts the remaining service life of key components using the Weibull degradation model.
[0029] Furthermore, the three-level closed-loop alignment control subsystem also integrates an alignment trajectory planning module, installed on the ARM terminal of the embedded motion controller. It uses a seven-segment S-shaped acceleration / deceleration curve to generate the motion trajectory, ensuring continuous displacement, velocity, acceleration, and jerk, thus avoiding mechanical impact. It also incorporates ZV input shaping technology to pre-filter the command signal, ensuring the shaper pulse sequence meets the requirements. and ,in To ensure the system's inherent vibration period and suppress residual oscillations after motion stops, the disturbance observer DOB configured in the semi-closed loop of the telescopic rod satisfies the following formula:
[0030]
[0031] in For the nominal model transfer function, To control the input, To output speed, the estimated disturbance amount Feedforward compensation is sent to the control input.
[0032] Furthermore, the deep learning adaptive parameter optimization subsystem adopts a two-stage operation mode of offline pre-training plus online fine-tuning. In the offline stage, deep Q-network parameters are pre-trained using working condition data simulated under various loads, strokes, and eight motion modes in a simulation environment. The loss function of the Q-network is:
[0033]
[0034] in For the current network parameters, For the target network parameters, during the actual operation of the device in the online phase, after each alignment is completed, a four-tuple of state, action, reward, and next state is generated. Store the data in the experience replay pool, and periodically randomly sample small batches of data to update the network weights using gradient descent. The specific formula for the reward function is:
[0035]
[0036] in ~ These are the weighting coefficients. , , The target accuracy threshold, To stabilize the time, The amplitude of the vibration;
[0037] The control system also establishes a positioning knowledge base in the controller's local storage, recording the motion mode, working condition data and final accuracy of each positioning. When encountering similar working conditions, it directly calls the historical optimal parameters to achieve continuous self-improvement of positioning performance.
[0038] The beneficial effects of this invention are:
[0039] In this invention, four sets of independently arranged high-precision electric telescopic rods are used to drive the movable plate in parallel. Through different telescopic combinations, eight basic motion modes can be realized, covering positive and negative translation in the X, Y and two diagonal directions in the plane. The structure is simple, the motion is highly flexible, and the structure is symmetrical and the force is balanced.
[0040] Combined with a CCD vision alignment module to form a full vision positioning loop, it can directly detect positional deviations in any direction within the plane, breaking through the limitation of traditional grating rulers that can only detect displacements in orthogonal directions. Together with the telescopic rod semi-closed loop and the current inner loop, it forms a three-level nested control architecture, balancing alignment speed and final accuracy.
[0041] The control system incorporates a deep learning adaptive parameter optimization subsystem based on a deep Q-network. Through a two-stage mode of offline pre-training and online fine-tuning, it uses alignment accuracy, stability time, and vibration amplitude as comprehensive rewards to online self-tune PID parameters and feedforward gain. This enables the equipment to have a self-learning ability that becomes more accurate with use and can adapt to different working conditions. Combined with a disturbance observer to compensate for friction and load disturbances in real time, and ZV input shaping technology to suppress residual vibration, the system further improves alignment robustness and motion smoothness.
[0042] The equipment health management and predictive maintenance subsystem uses the isolated forest algorithm for multi-dimensional anomaly detection and combines it with the Weibull degradation model to predict the remaining service life of key components, thereby enabling graded alarms and predictive maintenance recommendations, effectively reducing unplanned downtime. Attached Figure Description
[0043] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0044] Figure 1 This is a schematic diagram of the overall structure of a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing proposed in this invention.
[0045] Figure 2 This is a schematic diagram showing the disassembled structure of a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing proposed in this invention.
[0046] Figure 3 This is a front view schematic diagram of a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing proposed in this invention.
[0047] Figure 4 This is a schematic diagram of the bottom structure of the movable plate of a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing proposed in this invention.
[0048] Figure 5 This is a schematic diagram of a high-precision electric telescopic rod structure for a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing proposed in this invention.
[0049] Figure 6 This is a control system framework diagram of a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing proposed in this invention.
[0050] In the diagram: 1. Support mechanism; 11. Base; 12. Shaft; 13. Fixing hole; 2. Drive assembly; 21. High-precision electric telescopic rod; 22. Bushing; 23. Connecting shaft; 24. Ball bearing seat; 25. Ball bearing; 3. Platform mechanism; 31. Motion platform; 32. Mounting hole; 33. Vacuum adsorption groove array; 34. Origin baffle; 4. Motion mechanism; 41. Movable plate; 42. Servo motor; 43. Drive shaft; 44. Bearing; 45. High-precision circular grating encoder; 5. Sensor detection assembly; 51. CCD vision alignment module; 52. Photoelectric origin sensor; 6. Embedded motion controller. Detailed Implementation
[0051] To make the technical means and objectives and effects of the present invention easier to understand, the embodiments of the present invention will be described in detail below with reference to specific illustrations.
[0052] Example 1.
[0053] like Figure 1-5 As shown, the present invention discloses a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing, including a support mechanism 1, a drive component 2, a platform mechanism 3, a motion mechanism 4, a sensor detection component 5, and an embedded motion controller 6.
[0054] The support mechanism 1 includes a base 11, which is integrally cast from HT300 gray cast iron. It has a box-shaped closed structure with internal grid-like reinforcing ribs to ensure high rigidity and damping characteristics. The upper surface of the base 11 is precision ground to a flatness of no more than 0.02mm / 300mm, serving as the installation reference for the entire machine.
[0055] Four shafts 12 are vertically installed at the four corners of the top of the base 11. The shafts 12 are made of GCr15 bearing steel, which has been quenched and ground. The cylindricity is no more than 0.003mm. They serve as the fulcrum for the rotation of the electric telescopic rod.
[0056] Secondly, fixing holes 13 are provided on the four sides of the base 11 and at the center position between two adjacent shafts 12, for fixing the alignment mechanism to the printing machine main frame by bolts.
[0057] The drive assembly 2 includes four sets of high-precision electric telescopic rods 21, which are independently arranged at the four corners of the top of the base 11, and are symmetrically distributed in a rectangular shape.
[0058] Each high-precision electric telescopic rod 21 is a servo electric cylinder with a built-in ball screw transmission pair, achieving a repeatability of ±0.005mm. Each high-precision electric telescopic rod 21 has a bushing 22 installed at its fixed end, which is movably fitted onto a shaft 12 with a clearance controlled between 0.008 and 0.012mm, allowing the high-precision electric telescopic rod 21 to swing freely around the shaft 12 in the horizontal plane. Each high-precision electric telescopic rod 21 has a vertically upward connecting shaft 23 fixedly connected to its extended end. The connecting shaft 23 is made of stainless steel with a chrome-plated surface.
[0059] Secondly, each high-precision electric telescopic rod 21 is equipped with a rod-type displacement sensor, which adopts a differential transformer type LVDT structure with a resolution of up to 0.1μm. This sensor is used to detect the actual extension and retraction of the telescopic rod in real time, forming a semi-closed-loop feedback.
[0060] At the bottom of each high-precision electric telescopic rod 21, and directly below the connecting shaft 23, a ball bearing seat 24 is fixedly installed. A ball bearing 25 is movably embedded in the ball bearing seat 24. The ball bearing 25 is made of silicon carbide ceramic ball material. The bottom spherical surface of the ball bearing 25 rolls in contact with the upper surface of the base 11, which plays a role in auxiliary support and reducing friction. A miniature thin-film pressure sensor is also embedded inside the ball bearing seat 24 to monitor the pressure change of the support point in real time and indirectly reflect the load distribution status.
[0061] It should be noted that all four sets of high-precision electric telescopic rods 21 are at half of their total stroke in standby mode. By independently controlling the extension and retraction of each telescopic rod, translational motion in any direction in the plane can be synthesized, which includes at least eight basic motion modes: left front group extension and retraction alone, right front group extension and retraction alone, left rear group extension and retraction alone, right rear group extension and retraction alone, left front and right rear synchronous extension and retraction, left rear and right front synchronous extension and retraction, left front and left rear synchronous extension and retraction, and right front and right rear synchronous extension and retraction. These correspond to eight basic motion directions: positive X-axis translation, negative X-axis translation, positive Y-axis translation, negative Y-axis translation, and positive and negative translation along the two diagonals.
[0062] The motion mechanism 4 includes a movable plate 41, which is made of 7075 aluminum alloy with hard anodizing treatment. The lightweight design ensures sufficient rigidity. The movable plate 41 is horizontally set above four sets of high-precision electric telescopic rods 21.
[0063] Among them, bearings 44 are embedded in the four corners of the movable plate 41. High-precision angular contact ball bearings are used. The connecting shaft 23 passes through the inner ring of the bearing 44 and is fixedly connected, so as to realize the independent hinge connection between each set of high-precision electric telescopic rods 21 and the corresponding corner of the movable plate 41.
[0064] Secondly, a servo motor 42 is fixedly installed at the bottom center of the movable plate 41. It is a low-inertia AC servo motor with a rated power of 400W and an encoder resolution of 23 bits.
[0065] In addition, the output shaft of the servo motor 42 is connected to the drive shaft 43, which passes vertically through the center through hole of the movable plate 41. A high-precision circular grating encoder 45 is mounted on the drive shaft 43.
[0066] It should be noted that the stator of the high-precision circular encoder 45 is fixedly mounted on the upper surface of the movable plate 41 by a bracket, and the rotor rotates synchronously with the transmission shaft 43. The circular encoder has no less than 36,000 grid lines, and the angular resolution is better than 1 arcsecond after electronic subdivision. It is used to directly detect the actual rotation angle of the motion platform and form a high-precision closed-loop control of the independent rotating shaft.
[0067] The platform mechanism 3 includes a motion platform 31, which is made of microcrystalline granite or 7075 aluminum alloy. The upper surface is precision ground and the flatness is no more than 0.005mm / length. The motion platform 31 is horizontally fixedly installed at the top of the transmission shaft 43 and is directly driven to rotate by the servo motor 42 through the transmission shaft 43.
[0068] The motion platform 31 has multiple sets of mounting holes 32 at its four corners, which are used to install corresponding jigs or suction trays according to different substrate specifications.
[0069] Secondly, the upper surface of the motion platform 31 is processed with a vacuum adsorption groove array 33, which is 1mm wide and 0.5mm deep and is distributed in a grid pattern. It is connected to the negative pressure system through the vacuum air circuit interface at the bottom to firmly adsorb the substrate and prevent displacement during registration and printing.
[0070] In addition, an origin baffle 34 is fixedly installed on one side edge of the motion platform 31, and a photoelectric origin sensor 52 is fixedly installed at the corresponding position on the base 11. A slotted photoelectric switch is used. When the origin baffle 34 moves with the motion platform, it can trigger the photoelectric origin sensor 52 to achieve mechanical return to zero and establish an absolute coordinate reference.
[0071] The sensor detection component 5 includes a CCD vision alignment module 51, a photoelectric origin sensor 52, and a vacuum pressure sensor.
[0072] The CCD vision alignment module 51 is mounted directly above the motion platform 31. It includes at least two global shutter CCD industrial cameras, which, together with dual telecentric lenses and coaxial light sources, are aligned with two diagonal alignment marks on the substrate. The camera resolution is no less than 2 million pixels, the acquisition frame rate is no less than 30fps, and with the sub-pixel edge extraction algorithm, the alignment detection resolution can reach the sub-micron level.
[0073] Secondly, the CCD vision alignment module 51 can detect translation and rotation in any direction within the plane, covering pose feedback for all eight motion modes. It can directly calculate the actual pose of the motion platform by photographing the alignment marks on the substrate, including X-axis displacement, Y-axis displacement, and rotation angle.
[0074] In addition, a vacuum pressure sensor is connected in series to the vacuum adsorption gas path of the motion platform 31. It is a high-precision diffused silicon pressure sensor with a range of -100kPa to 0kPa and an accuracy of 0.25%FS. It is used to monitor the adsorption pressure in real time to ensure reliable adsorption of the substrate and trigger an alarm when the pressure is insufficient.
[0075] The embedded motion controller 6 is fixedly installed in the electrical control compartment on the side of the base 11. It adopts a control board with an ARM+FPGA heterogeneous architecture. The ARM processor runs kinematics calculation algorithm, intelligent control algorithm and visual communication program, while the FPGA runs five-axis synchronous pulse generation, encoder high-speed counting and multi-sensor synchronous acquisition program.
[0076] Specifically, the signal input terminals of the controller are connected to the CCD vision alignment module 51, the photoelectric origin sensor 52, the vacuum pressure sensor, the rod-type displacement sensor built into each telescopic rod, and the high-precision circular grating encoder 45, respectively. The control output terminals are connected to the servo drivers of the four sets of high-precision electric telescopic rods 21 and the servo drivers of the servo motors 42, respectively.
[0077] The controller has a built-in three-level closed-loop control logic, consisting of a visual full positioning closed loop, a telescopic rod semi-closed loop, and a current inner loop, from the outside to the inside, to achieve closed-loop control for precise three-degree-of-freedom alignment.
[0078] Example 2
[0079] like Figure 6 As shown, the control system in this embodiment is mounted in the embedded motion controller 6 in embodiment 1. It includes four levels: a three-level closed-loop alignment control subsystem, a deep learning adaptive parameter optimization subsystem, a multi-mode motion planning subsystem, and an equipment health management and predictive maintenance subsystem. Each level works together to achieve high-precision and intelligent alignment control.
[0080] The three-level closed-loop positioning control subsystem adopts a nested three-loop structure, consisting of a visual full positioning loop, a telescopic rod semi-closed loop, and a current inner loop, from the outside to the inside.
[0081] The outermost visual full positioning ring is based on the alignment mark image acquired by the CCD visual alignment module 51. It directly calculates the pose deviation in any direction in the plane through template matching and sub-pixel positioning algorithm, and outputs the target position command. It can cover all eight motion modes. The control cycle of the visual ring is about 50ms. Its function is to quickly guide the platform to the vicinity of the target position. It has a large convergence range and fast response speed.
[0082] The middle layer is a semi-closed loop for telescopic rods. Based on the feedback from the built-in rod-type displacement sensors 26 of each telescopic rod, the independent and precise position control of each telescopic rod is achieved through an improved PID algorithm combined with a disturbance observer (DOB). The semi-closed loop control cycle is 1ms, which is the core link to ensure accuracy.
[0083] The PID control law satisfies the following formula:
[0084]
[0085] in For the first The output volume is controlled by the beat. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients. For the first Shooting position deviation, This is the feedforward compensation amount.
[0086] The improved PID adopts integral separation and derivative-first structure. When the position deviation is greater than the set threshold, the integral action is temporarily canceled to avoid overshoot caused by integral saturation. The derivative-first structure only differentiates the measured value and not the set value, reducing the impact when the set value changes abruptly.
[0087] Feedforward compensation Calculation based on the velocity and acceleration of the target trajectory, i.e. Used to improve trajectory tracking accuracy, among which For velocity feedforward coefficient, For acceleration feedforward coefficients, For the target speed, Accelerate towards the target.
[0088] The innermost layer is the current inner loop, with a control cycle of 125μs. It achieves rapid tracking of motor torque through PI regulation, ensuring the speed and stability of motor output.
[0089] The three-level closed-loop alignment control subsystem also integrates an alignment trajectory planning module, which is installed on the ARM terminal of the embedded motion controller 6. It uses a seven-segment S-shaped acceleration and deceleration curve to generate the motion trajectory, dividing the motion process into seven stages: acceleration, uniform acceleration, deceleration, uniform speed, acceleration and deceleration, uniform deceleration, and deceleration and deceleration. Displacement, velocity, acceleration, and jerk all change continuously to avoid mechanical impact and vibration.
[0090] The trajectory planning module also incorporates ZV input shaping technology to pre-filter the command signal. The shaper consists of two pulses, and the pulse sequence satisfies... and ,in , These are the amplitudes of the two pulses, , These are the times when the two pulses act, This is the system's natural vibration period.
[0091] The two pulses generate vibrations with equal amplitude and opposite phase, which cancel each other out, thus effectively suppressing residual oscillations after the motion stops and shortening the settling time.
[0092] The disturbance observer DOB configured in the semi-closed loop of the telescopic pole satisfies the following formula:
[0093]
[0094] in For the nominal model transfer function, To control the input, For output speed, To estimate the amount of disturbance.
[0095] The disturbance observer estimates external disturbances such as friction and load changes by observing the difference between the motor current and the actual speed, and then converts the estimated disturbance amount into a quantity. Feedforward compensation is sent to the control input to effectively counteract the effects of external disturbances and improve the system's robustness to load changes and friction disturbances.
[0096] The deep learning adaptive parameter optimization subsystem incorporates a deep reinforcement learning DQN network, also known as a deep Q-network, to achieve online self-optimization of control parameters.
[0097] The subsystem uses the current alignment deviation, movement speed, load estimation, and movement mode as state inputs s. The state vector includes dimensions such as X-direction deviation, Y-direction deviation, rotation angle deviation, current movement speed, load estimation value calculated from motor current, and current movement mode number.
[0098] The adjustment of PID parameters and feedforward gain is used as the action output 'a'. The action space includes position loop proportional coefficient adjustment, integral coefficient adjustment, feedforward gain adjustment, S-curve jerk adjustment, etc., with each parameter increasing or decreasing by a fixed step size.
[0099] Using the weighted values of alignment accuracy, settling time, and vibration amplitude as the reward function R(s,a), the Q-value table is updated iteratively through the Bellman equation, which is:
[0100]
[0101] in The learning rate, ranging from 0 to 1, determines the degree to which new information overwrites old information. This is a discount factor, ranging from 0 to 1, which determines the weight of future rewards. The closer it is to 1, the more emphasis is placed on long-term returns. The immediate reward obtained after performing an action.
[0102] The physical meaning of value update is that the new estimate equals the old estimate plus the learning rate multiplied by the time-series difference error, which is the difference between the actual return and the original estimate.
[0103] The deep learning adaptive parameter optimization subsystem adopts a two-stage operation mode of offline pre-training plus online fine-tuning.
[0104] In the offline phase, the system simulates various loads, strokes, and eight motion modes within the MATLAB / Simulink simulation environment. This allows for the pre-training of deep Q-network parameters, enabling basic parameter tuning capabilities and preventing poor performance during the initial deployment phase. The loss function of the Q-network is:
[0105]
[0106] in These are the weight parameters of the current main network. These are the weight parameters of the target network, which are copied and updated from the main network every fixed number of steps to improve training stability.
[0107] The loss function measures the current The mean squared error between the network's predicted and target values is minimized during training using gradient descent. To update network parameters .
[0108] During the online phase of device operation, after each alignment is completed, a four-tuple of state, action, reward, and next state is generated. The data is stored in the experience replay pool, and small batches of data are randomly sampled from the replay pool periodically to update the network weights using gradient descent. By utilizing the experience replay mechanism, the temporal correlation between samples is broken, thereby improving training stability.
[0109] The specific formula for the reward function is:
[0110]
[0111] in to These are the weighting coefficients for each item, all of which are positive numbers and sum to 1. Recommended values are 0.3, 0.3, 0.2, 0.1, and 0.1, which can be adjusted according to the process emphasis.
[0112] , , These are the target accuracy thresholds in the X, Y, and θ directions, respectively, which represent the maximum allowable deviation. The alignment settling time is the time from the start of the motion to the position entering the accuracy zone and remaining stable. The maximum allowed alignment time; This represents the residual vibration amplitude after the motion ends.
[0113] The comprehensive reward function considers three dimensions simultaneously: alignment accuracy, stability time, and vibration magnitude, guiding the agent to minimize alignment time and reduce vibration while ensuring accuracy.
[0114] The control system also establishes a positioning knowledge base in the controller's local storage, recording the motion mode, working condition data and final accuracy of each positioning. When encountering similar working conditions, it can directly call the historical optimal parameters to speed up the convergence speed and achieve continuous self-improvement of positioning performance.
[0115] The multi-mode motion planning subsystem has built-in kinematic models for eight basic motion modes and stores the mapping relationship between the extension and retraction of the four telescopic rods and the platform displacement in each mode.
[0116] After receiving the target displacement command, the subsystem automatically determines which basic motion mode the target displacement direction is closest to, selects the optimal motion combination, and plans the coordinated motion trajectory of each telescopic rod to make the motion path shortest, energy consumption lowest, and accuracy highest.
[0117] For any target displacement in any direction, the subsystem calculates the target extension amount that each of the four telescopic rods needs to achieve through the inverse kinematics algorithm, and then distributes the motion velocity curves of each axis to ensure that the platform moves smoothly during the four-axis coordinated motion.
[0118] The equipment health management and predictive maintenance subsystem collects multi-dimensional operating data such as motor current, displacement deviation, and vacuum pressure in real time, and performs anomaly detection through the isolated forest algorithm.
[0119] The Isolation Forest algorithm constructs multiple isolated trees and recursively divides the sample space by randomly selecting features and split points. Abnormal samples are more likely to be isolated because they are few in number and have large feature differences, i.e., the path depth is shorter.
[0120] The formula for calculating the anomaly score of sample x is:
[0121]
[0122] in For the sample The average path depth across all isolated trees. for The average path length normalization factor for each sample is used to normalize the path depth.
[0123] The abnormal score ranges from 0 to 1. A score greater than 0.6 is considered an anomaly.
[0124] The subsystem can identify abnormal states including increased friction manifested as abnormal current rise, increased transmission clearance manifested as abnormal position tracking error, and vacuum system leakage manifested as abnormal pressure fluctuation.
[0125] Combining the Weibull degradation model to predict the remaining service life of critical components, the reliability function of the Weibull distribution is:
[0126]
[0127] in For shape parameters, Characteristic lifetime.
[0128] Based on data such as cumulative operating mileage, load spectrum, and temperature history, the Weibull distribution parameters are estimated, and then the remaining service life (RUL) of key components such as lead screws, guide rails, and bearings is calculated.
[0129] The subsystem performs graded alarms based on anomaly detection results and lifespan predictions, divided into three levels: early warning, alarm, and shutdown. It also provides predictive maintenance suggestions, such as the estimated number of hours after which the guide rails need to be lubricated, to help users schedule maintenance in advance, reduce unplanned downtime, and improve equipment uptime.
[0130] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing, characterized in that, It includes a support mechanism (1), a drive assembly (2), a platform mechanism (3), a motion mechanism (4), a sensor detection assembly (5), and an embedded motion controller (6). The support mechanism (1) includes a base (11), and four shafts (12) are vertically installed at the four corners of the top of the base (11). The drive assembly (2) includes four sets of high-precision electric telescopic rods (21), which are respectively arranged at the four corners of the top of the base (11). Each set of high-precision electric telescopic rods (21) has a bushing (22) installed at its fixed end. The bushing (22) is correspondingly and movably fitted on a shaft (12), so that the high-precision electric telescopic rod (21) can swing around the shaft (12) in the horizontal plane. The extended end of each set of high-precision electric telescopic rods (21) is fixedly connected to a vertically upward connecting shaft (23). Each high-precision electric telescopic rod (21) has a pull rod type displacement sensor integrated inside. The motion mechanism (4) includes a movable plate (41), which is horizontally positioned above four sets of high-precision electric telescopic rods (21). Bearings (44) are embedded at the four corners of the movable plate (41). The connecting shaft (23) passes through the inner ring of the bearing (44) and is fixedly connected to achieve the hinge between the high-precision electric telescopic rod (21) and the movable plate (41). A servo motor (42) is fixedly installed at the bottom center of the movable plate (41). The output shaft of the servo motor (42) is connected upward to a transmission shaft (43). The transmission shaft (43) passes vertically through the central through hole of the movable plate (41). A high-precision circular grating encoder (45) is mounted on the transmission shaft (43). The platform mechanism (3) includes a motion platform (31), which is horizontally fixed at the top of the transmission shaft (43). The upper surface of the motion platform (31) is machined with a vacuum adsorption groove array (33). The sensor detection component (5) includes a CCD vision alignment module (51), a photoelectric origin sensor (52), and a vacuum pressure sensor. The CCD vision alignment module (51) is mounted directly above the motion platform (31) and directly detects the actual position and posture of the motion platform in any direction in the plane by photographing the alignment marks on the substrate, including X-direction displacement, Y-direction displacement, and rotation angle. The vacuum pressure sensor is connected in series to the vacuum adsorption gas path of the motion platform (31) to monitor the adsorption pressure. The embedded motion controller (6) is fixedly installed in the side compartment of the base (11). The signal input terminal of the controller is connected to each sensor, and the control output terminal is connected to the driver of the four sets of high-precision electric telescopic rods (21) and the driver of the servo motor (42).
2. The anti-counterfeiting screen printing precision alignment mechanical transmission mechanism according to claim 1, characterized in that, The base (11) has a fixing hole (13) centered between adjacent shafts (12) around its perimeter. The motion platform (31) has mounting holes (32) at its four corners for mounting the printing workbench. An origin baffle (34) is fixedly installed on one side edge of the motion platform (31). A photoelectric origin sensor (52) is fixedly installed on the base (11) at the corresponding position.
3. The anti-counterfeiting screen printing precision alignment mechanical transmission mechanism according to claim 1, characterized in that, A ball bearing seat (24) is fixedly installed at the bottom of the high-precision electric telescopic rod (21) and directly below the connecting shaft (23). A ball bearing (25) is movably embedded in the ball bearing seat (24), and the bottom of the ball bearing (25) rolls in contact with the upper surface of the base (11).
4. The anti-counterfeiting screen printing precision alignment mechanical transmission mechanism according to claim 1, characterized in that, Four sets of high-precision electric telescopic rods (21) are arranged symmetrically at a 45° angle on the top of the base (11). In the standby state, each set of high-precision electric telescopic rods (21) is at half of the total stroke. The four sets of telescopic rods can synthesize translational motion in any direction in the plane by independently controlling the telescopic amount. Among them, there are at least eight basic motion modes: left front group telescopic alone, right front group telescopic alone, left rear group telescopic alone, right rear group telescopic alone, left front and right rear synchronous telescopic, left rear and right front synchronous telescopic, left front and left rear synchronous telescopic, and right front and right rear synchronous telescopic.
5. The anti-counterfeiting screen printing precision alignment mechanical transmission mechanism according to claim 1, characterized in that, The CCD vision alignment module (51) includes at least two global shutter CCD industrial cameras, which are equipped with dual telecentric lenses and coaxial light sources to align with two diagonal alignment marks on the substrate. The camera resolution is not less than 2 million pixels and the acquisition frame rate is not less than 30fps. With the sub-pixel edge extraction algorithm, the alignment detection resolution can reach the sub-micron level. The CCD vision alignment module (51) can detect translation and rotation in any direction in the plane, covering the pose feedback of all eight motion modes.
6. The anti-counterfeiting screen printing precision alignment mechanical transmission mechanism according to claim 1, characterized in that, The high-precision circular encoder (45) is fixed in the middle of the transmission shaft (43). The stator of the high-precision circular encoder (45) is fixed on the upper surface of the movable plate (41) by a bracket. The rotor rotates synchronously with the transmission shaft (43). The high-precision circular encoder (45) has no less than 36,000 grid lines and is used to directly detect the actual rotation angle of the motion platform.
7. The anti-counterfeiting screen printing precision alignment mechanical transmission mechanism according to claim 1, characterized in that, The embedded motion controller (6) adopts an ARM+FPGA heterogeneous architecture control board, which is installed in the electrical control compartment on the side of the base (11). The ARM processor runs kinematics calculation algorithm, intelligent control algorithm and visual communication program, while the FPGA runs five-axis synchronous pulse generation, encoder high-speed counting and multi-sensor synchronous acquisition program. The controller has a built-in three-level closed-loop control logic, which consists of a visual full positioning closed loop, a telescopic rod semi-closed loop and a current inner loop from the outside to the inside.
8. The control system for a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing according to any one of claims 1-7, characterized in that, The control system is mounted in an embedded motion controller (6) and includes four levels: a three-level closed-loop alignment control subsystem, a deep learning adaptive parameter optimization subsystem, a multi-mode motion planning subsystem, and an equipment health management and predictive maintenance subsystem. The three-level closed-loop alignment control subsystem adopts a nested three-loop structure. The outermost layer is a visual full positioning loop, which directly calculates the pose deviation in any direction in the plane and outputs the target position command based on the alignment mark image collected by the CCD visual alignment module (51), which can cover all eight motion modes. The middle layer is a telescopic rod semi-closed loop, which uses the feedback of the rod-type displacement sensor built into each telescopic rod to achieve independent and precise position control of each telescopic rod through an improved PID algorithm combined with a disturbance observer (DOB). The PID control law satisfies the following formula: in For the first The output volume is controlled by the beat. This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients. For the first Shooting position deviation, This is the feedforward compensation amount; The innermost layer is the current inner loop, which enables rapid tracking control of motor torque; The deep learning adaptive parameter optimization subsystem incorporates a deep reinforcement learning (DQN) network. It uses the current alignment deviation, motion speed, load estimation, and motion pattern as state inputs (s), and the adjustments to the PID parameters and feedforward gain as action outputs (a). The reward function R(s,a) is a weighted value of alignment accuracy, settling time, and vibration amplitude, and is implemented using the Bellman equation. Iteratively update the Q-value table, where For learning rate, As a discount factor, For instant rewards, online self-optimization control parameters are implemented. The multi-mode motion planning subsystem has built-in kinematic models of eight basic motion modes, which can automatically select the optimal motion combination based on the target displacement and plan the coordinated motion trajectory of each telescopic rod. The equipment health management and predictive maintenance subsystem collects multi-dimensional operating data such as motor current, displacement deviation, and vacuum pressure, performs anomaly detection using the isolated forest algorithm, and predicts the remaining service life of key components using the Weibull degradation model.
9. The control system for a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing according to claim 8, characterized in that, The three-level closed-loop alignment control subsystem also integrates an alignment trajectory planning module, which is installed on the ARM terminal of the embedded motion controller (6). It uses a seven-segment S-shaped acceleration and deceleration curve to generate the motion trajectory. The displacement, velocity, acceleration, and jerk are all continuous to avoid mechanical impact. It also combines ZV input shaping technology to pre-filter the command signal, and the shaper pulse sequence satisfies and ,in To ensure the system's inherent vibration period and suppress residual oscillations after motion stops, the disturbance observer DOB configured in the semi-closed loop of the telescopic rod satisfies the following formula: in For the nominal model transfer function, To control the input, To output speed, the estimated disturbance amount Feedforward compensation is sent to the control input.
10. The control system for a precision alignment mechanical transmission mechanism for anti-counterfeiting screen printing according to claim 8, characterized in that, The deep learning adaptive parameter optimization subsystem adopts a two-stage operation mode of offline pre-training plus online fine-tuning. In the offline stage, the parameters of the deep Q-network are pre-trained in a simulation environment by simulating working condition data under various loads, various strokes, and eight motion modes. The loss function of the Q-network is: in For the current network parameters, For the target network parameters, during the actual operation of the device in the online phase, after each alignment is completed, a four-tuple of state, action, reward, and next state is generated. Store the data in the experience replay pool, and periodically randomly sample small batches of data to update the network weights using gradient descent. The specific formula for the reward function is: in ~ These are the weighting coefficients. , , The target accuracy threshold, To stabilize the time, The amplitude of the vibration; The control system also establishes a positioning knowledge base in the controller's local storage, recording the motion mode, working condition data and final accuracy of each positioning. When encountering similar working conditions, it directly calls the historical optimal parameters to achieve continuous self-improvement of positioning performance.
Citation Information
Patent Citations
Visual alignment calibration method of UVW alignment platform and related device
CN117419749A
Coil stock automatic alignment's screen printing system
CN205439549U
Half tone automatic alignment's sheet stock screen printing machine
CN205615162U
