Large UV printer nozzle three-dimensional space automatic calibration system

Abstract

The invention relates to the technical field of printing equipment manufacturing, and particularly discloses a large UV printer nozzle three-dimensional space automatic calibration system which comprises a dynamic reference reconstruction module, a non-contact three-dimensional pose measurement module, a dynamic error modeling and compensation module and a closed-loop feedback control module. According to the invention, the normal form of nozzle calibration is changed, and static and off-line reference point measurement is converted into dynamic and on-line space reference reconstruction and real-time pose feedback. Through the dynamic reference reconstruction module, the system can perceive and calculate six-degree-of-freedom micro-motion, generated by mechanical vibration, of the nozzle mounting seat in real time, so that a reference coordinate system dynamically changing along with actual working conditions is established. In this way, a real dynamic reference system is provided for follow-up nozzle tip pose measurement, the core contradiction that the static calibration benchmark and the dynamic operation state are disjointed is solved, and a correct foundation is laid for high-precision dynamic compensation.

Description

Technical Field

[0001] This invention belongs to the field of printing equipment manufacturing technology, specifically relating to a three-dimensional automatic calibration system for the printhead of a large UV printer. Background Technology

[0002] In the field of industrial large-format printing, UV curing printing technology has become one of the mainstream technologies due to its high precision, high efficiency, and wide adaptability to various substrates. This technology precisely deposits UV ink onto the surface of the printing medium by controlling a printhead array and then instantly cures it using ultraviolet light to achieve high-quality image output. Maintaining a precise and constant relative spatial position between the printhead and the printing medium is a core prerequisite for ensuring print quality and avoiding printhead collision damage.

[0003] Large-scale UV printing equipment typically employs a gantry or cantilever mechanical structure, with its printhead mounted on a beam driven by a multi-axis motion system to achieve a wide range of three-dimensional spatial movement. To ensure printing accuracy, the printhead's spatial position must be calibrated after equipment installation or maintenance to establish a precise coordinate relationship between the printhead tip and the printing platform or reference plane.

[0004] Static calibration methods are employed, where the printhead's relative position to a preset reference point is measured using laser ranging, visual positioning, or contact probes while the equipment is stationary. However, during high-speed operation of large printing equipment, the mechanical vibrations generated by its moving parts (such as guide rails and lead screws) are transmitted through the structure to the printhead mounting base, causing a significant drift in the actual dynamic position of the printhead relative to the static calibration reference. This positional deviation under dynamic operating conditions cannot be captured and compensated for by static calibration, resulting in a severe disconnect between the calibration results and the actual operating conditions.

[0005] Over long-term operation, the accumulated vibration effects may further exacerbate minute changes in printhead position, affecting not only the edge sharpness and registration accuracy of the printed pattern, but also potentially causing collisions between the printhead and the media or platform due to misalignment during high-speed printing or platform lifting, resulting in costly printhead damage and production interruptions. Summary of the Invention

[0006] The purpose of this invention is to provide a three-dimensional automatic calibration system for the printhead of a large UV printer, in order to solve the problem that existing static calibration methods cannot capture and compensate for the actual position drift of the printhead caused by mechanical vibration during high-speed dynamic operation, thereby causing a decline in print quality and the risk of collision.

[0007] This invention provides a three-dimensional automatic calibration system for the printhead of a large UV printer. This system is integrated into the motion control architecture of a large UV printing device and includes: The dynamic reference reconstruction module is used to collect and fuse data from multiple high-precision vibration sensors in real time during the operation of the printing equipment, so as to calculate and generate a dynamically changing printhead mount space reference coordinate system. The non-contact 3D pose measurement module is used to measure the 3D coordinates and attitude angles of the nozzle tip in the dynamic reference coordinate system generated by the dynamic reference reconstruction module in a non-contact manner during the operation of the printing equipment. The dynamic error modeling and compensation module is used to construct a spatiotemporal evolution model of the dynamic position error of the nozzle based on the nozzle position data measured in real time by the non-contact three-dimensional position measurement module and the preset ideal position model of the nozzle. Based on the model, the multi-axis linkage compensation amount used to correct the motion control command is calculated in real time. The closed-loop feedback control module receives the multi-axis linkage compensation amount output by the dynamic error modeling and compensation module, and superimposes it with the original printing path planning instruction in real time to generate the final motion control instruction, which drives the multi-axis motion system to execute, thereby realizing closed-loop correction of the actual operating posture of the printhead.

[0008] Furthermore, the dynamic reference reconstruction module includes a vibration sensing array and a reference calculation unit.

[0009] The vibration sensing array consists of at least three triaxial accelerometers and at least three triaxial gyroscopes. These sensors are rigidly connected and fixedly mounted on the rigid main structure of the nozzle mounting base. Their installation positions are pre-calibrated to determine the precise geometric relationship between them.

[0010] The reference calculation unit performs the following processes: real-time synchronous acquisition of raw data from all sensors in the vibration sensor array; gravity component stripping and coordinate transformation processing of accelerometer data, and attitude calculation and drift correction processing of gyroscope data; based on the fixed geometric relationship between sensors, using the principle of spatial rigid body kinematics, fusing and processing the data from all sensors, and calculating the real-time six-degree-of-freedom pose change of the nozzle mount in the global coordinate system; continuously updating the preset static mount reference coordinate system with this real-time pose change, thereby outputting a dynamic reference coordinate system that can characterize the actual dynamic position and attitude of the nozzle mount.

[0011] Furthermore, the non-contact three-dimensional pose measurement module includes a laser displacement sensing array and a pose calculation unit.

[0012] The laser displacement sensor array consists of four or more single-point laser displacement sensors, which are rigidly mounted on the printer frame in a non-coplanar manner, and their measurement beams converge in a preset measurement space near the nozzle tip.

[0013] The pose calculation unit performs the following process: controlling each sensor in the laser displacement sensor array to synchronously emit a measurement beam to a pre-set, highly reflective spherical target surface on the nozzle tip; receiving the precise distance values ​​to the target surface returned by each sensor; based on the pre-calibrated spatial installation position and beam direction of each sensor, and the multiple measured distance values, using the principle of spatial multi-point intersection measurement, calculating the three-dimensional coordinates of the center point of the spherical target in the global coordinate system; further, by analyzing the positional change of the reflected light spot on the target surface on the sensor receiver, calculating the small attitude angle of the target, i.e., the nozzle tip, relative to the measurement beam; finally, converting the calculated nozzle tip pose data to the dynamic reference coordinate system provided by the dynamic reference reconstruction module, and outputting the real-time three-dimensional coordinates and attitude angle of the nozzle tip relative to its mounting base.

[0014] Furthermore, the dynamic error modeling and compensation module includes an error extraction unit, a model building unit, and a compensation calculation unit.

[0015] The error extraction unit is used to receive the real-time nozzle pose data output by the non-contact 3D pose measurement module, and to retrieve the ideal pose model data of the nozzle corresponding to the current printing path point from the preset storage unit; the real-time measured pose is compared with the ideal model pose item by item, and the real-time error values ​​of the nozzle in a total of 6 degrees of freedom in the three translational directions of X, Y, and Z and the three rotational directions around the X, Y, and Z axes are calculated.

[0016] The model building unit is used to continuously collect the time-series error data output by the error extraction unit during a continuous equipment operation period, and at the same time, collect the operating state parameters of the equipment at this moment. The operating state parameters include the motion velocity and acceleration of each axis of the multi-axis motion system, as well as the vibration spectrum characteristics collected by the vibration sensor array. Using the system identification method, a dynamic error spatiotemporal evolution model is established with the operating state parameters as input and the six-degree-of-freedom error value as output. This model characterizes the mapping relationship between factors such as mechanical vibration and motion inertia and nozzle pose error.

[0017] The compensation calculation unit is used to receive the current motion state parameters in real time during the subsequent operation of the equipment and input them into the established dynamic error spatiotemporal evolution model to predict the pose error that the nozzle will generate in future control cycles in advance. Based on the predicted error, combined with the inverse kinematics model of the multi-axis motion system, the compensation displacement and compensation angle that need to be applied to each motion axis driver are solved, i.e., the multi-axis linkage compensation amount.

[0018] Furthermore, the closed-loop feedback control module is integrated into the motion controller of the device, and the module includes an instruction overlay unit and a motion control unit.

[0019] The instruction overlay unit is used to receive the original path planning instruction from the upper-level print management system. This instruction contains a series of discrete printhead target pose point sequences calculated based on the ideal device model. At the same time, it receives the multi-axis linkage compensation amount from the dynamic error modeling and compensation module in real time. It then performs vector overlay of the multi-axis linkage compensation amount corresponding to the current moment with the original target pose point to generate the actual target pose instruction after error compensation.

[0020] The motion control unit is used to calculate and output drive signals to the servo drivers of each axis in real time based on the actual target pose command through the three-loop control algorithm of the motion controller's position loop, velocity loop and current loop. The servo drivers drive the motor to drive the motion mechanism, so that the nozzle actually moves to the compensated target pose, thereby forming a closed-loop control system with non-contact three-dimensional pose measurement as feedback and dynamic error model prediction compensation as feedforward.

[0021] Furthermore, the system operates within a phased calibration workflow, which includes an offline modeling phase and an online compensation phase.

[0022] The offline modeling stage is performed after the equipment is installed and debugged or during regular maintenance. During this stage, the printing equipment is driven to run unloaded in the entire workspace of the multi-axis motion system with a variety of typical speed and acceleration combinations. In this process, the dynamic error modeling and compensation module simultaneously collects a large amount of error data and operating status parameters, completes the global training and parameter calibration of the dynamic error spatiotemporal evolution model, and stores the final model parameters.

[0023] The online compensation phase is continuously executed during the daily printing operations of the equipment. During this phase, the dynamic error model trained in the offline modeling phase is loaded. The system simultaneously activates the dynamic reference reconstruction module, the non-contact 3D pose measurement module, the dynamic error modeling and compensation module, and the closed-loop feedback control module during the printing process to achieve real-time measurement, error prediction, and motion compensation of the printhead pose, ensuring the consistency between the actual pose and the ideal pose of the printhead throughout the entire printing process.

[0024] Furthermore, the pose calculation unit in the non-contact three-dimensional pose measurement module is also used to perform target integrity verification. This process analyzes the echo signal intensity and waveform characteristics received by each laser displacement sensor to determine whether the measurement beam completely illuminates the effective reflection area of ​​the spherical target. When the signal characteristics of any sensor indicate that the beam is blocked by the nozzle body or ink, the pose calculation unit immediately marks the data at that moment as invalid and triggers the data interpolation mechanism. The data interpolation mechanism estimates the most likely nozzle pose at the current moment using a Kalman filter algorithm based on historical valid pose data and the current device motion state. This is then used as temporary data input to the dynamic error modeling and compensation module to ensure the continuity of the control closed loop.

[0025] Furthermore, the model building unit in the dynamic error modeling and compensation module uses a recursive least squares method based on the forgetting factor to fine-tune the model parameters online. During the online compensation phase, the system continuously compares the error value predicted by the dynamic error model with the error value actually measured by the non-contact 3D pose measurement module to calculate the residual. Using this residual and the current operating state parameters, the key parameters of the model are updated online in a small and gradual manner through the recursive least squares method, so that the dynamic error model can adapt to the slow drift of the characteristics of the equipment caused by long-term operation, wear or temperature changes, and maintain long-term calibration accuracy.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention revolutionizes nozzle calibration, shifting from static, offline reference point measurement to dynamic, online spatial reference reconstruction and real-time pose feedback. Through the dynamic reference reconstruction module, the system can sense and calculate the six-degree-of-freedom micro-motions of the nozzle mounting base caused by mechanical vibration in real time, thereby establishing a reference coordinate system that dynamically changes with actual operating conditions. This provides a true dynamic reference system for subsequent nozzle tip pose measurements, resolving the core contradiction of the disconnect between static calibration reference and dynamic operating state, and laying a correct foundation for high-precision dynamic compensation.

[0027] 2. This invention achieves a closed-loop process of "measurement-modeling-prediction-compensation" for printhead pose error through the synergy of a non-contact 3D pose measurement module and a dynamic error modeling and compensation module. The system not only measures the error in real time but also reveals the intrinsic relationship between the error and the motion state through modeling, and uses the model for feedforward prediction and compensation. This composite control strategy, combining real-time feedback and model feedforward, can significantly suppress or even eliminate dynamic pose deviations caused by mechanical vibration, ensuring that the printhead maintains a precise and constant relative position with the printing medium even at high speeds. This greatly improves the edge sharpness and registration accuracy of the printed pattern and fundamentally eliminates the risk of collisions caused by dynamic pose inaccuracies.

[0028] 3. This invention designs a phased workflow that combines offline modeling and online compensation, along with an online model fine-tuning mechanism, enabling the system to possess both high accuracy and strong robustness. In the offline modeling phase, a high-precision basic error model is obtained through global data training; in the online compensation phase, this model is used to achieve real-time, high-performance compensation.

[0029] 4. The online fine-tuning function enables the system to adapt to slow changes in equipment status, resist performance degradation and external interference, ensure the long-term stability and reliability of calibration results, reduce the equipment's reliance on periodic manual recalibration, and improve the intelligence level and continuous production capacity of large-scale UV printing equipment. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the overall technical solution architecture of the large UV printer printhead three-dimensional spatial automatic calibration system proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of dynamic reference reconstruction and real-time pose measurement in this invention; Figure 3 This is a logical flowchart of the dynamic error modeling and prediction compensation in this invention; Figure 4 This is a logical framework diagram of the two-stage workflow of offline modeling and online compensation in this invention; Figure 5 This is a schematic diagram of the data flow and control flow of closed-loop feedback control and multi-module collaboration in this invention. Detailed Implementation

[0031] This embodiment details a specific implementation scheme for a three-dimensional spatial automatic calibration system for the printhead of a large UV printer. Please refer to the appendix. Figures 1 to 5 This system is integrated into the motion control architecture of large-scale UV printing equipment. Its core function is to sense, model, and compensate for the three-dimensional spatial pose deviation of the printhead caused by mechanical vibration and other factors during high-speed dynamic operation in real time. The system is composed of a dynamic reference reconstruction module, a non-contact three-dimensional pose measurement module, a dynamic error modeling and compensation module, and a closed-loop feedback control module, forming a complete measurement, calculation, and execution closed loop.

[0032] The dynamic benchmark reconstruction module is the core of the system for establishing a dynamic reference benchmark. Please refer to the attached document. Figure 2 This module consists of a vibration sensing array and a reference calculation unit. The physical deployment of the vibration sensing array follows strict spatial geometric constraints. Specifically, the array includes six triaxial accelerometers and six triaxial gyroscopes, totaling twelve high-precision microelectromechanical system (MEMS) sensors. These sensors are rigidly connected and distributedly fixed to the rigid main structure of the nozzle mounting base using specialized clamps. The installation locations were selected using finite element analysis to ensure coverage of the main modal vibration regions of the mounting base. Before installation, the sensitive axis direction of each sensor was precisely calibrated using an optical measuring instrument, and its direction cosine matrix relative to the preset mounting base coordinate system was recorded. Simultaneously, the precise three-dimensional spatial positional relationships between all sensors were measured and recorded using a laser tracker, forming a known sensor network spatial geometric database. This database serves as the basis for subsequent spatial rigid body motion fusion calculations.

[0033] The benchmark calculation unit is implemented in hardware by a high-performance multi-core digital signal processor, and its software executes a continuous real-time calculation process. This process begins with synchronous data acquisition. The benchmark calculation unit simultaneously triggers and reads the raw output data from all 12 sensors in the vibration sensor array via a high-speed synchronous serial bus at a sampling rate of no less than 10 kHz. For the triaxial accelerometer, the output is specific force data in three orthogonal axes, typically in meters per second squared. The raw data is first preprocessed, including removing fixed biases and applying calibration matrices for sensitivity and cross-axis error correction. Subsequently, the critical step is gravity component stripping. The benchmark calculation unit maintains an attitude estimate based on the integration of gyroscope data. Using this estimated attitude, the accelerometer measurements are rotated from the sensor coordinate system to the local horizontal coordinate system, and the gravitational acceleration vector is subtracted to obtain the pure mount motion acceleration component in the horizontal coordinate system.

[0034] For a three-axis gyroscope, the output is angular velocity data along three orthogonal axes, typically in radians per second. The raw data is also biased and calibrated. Due to the inherent drift error of the gyroscope, the reference calculation unit uses a complementary filtering algorithm for attitude calculation and drift correction. This algorithm combines the absolute attitude reference provided by the calibrated accelerometer data in the low-frequency band with the fast dynamic response provided by the gyroscope data in the high-frequency band, and fuses them through a digital filter with an adjustable cutoff frequency. This allows for real-time estimation of the three-dimensional attitude angles of the nozzle mount relative to the global inertial coordinate system, namely roll, pitch, and yaw angles, while simultaneously estimating and compensating for the gyroscope's zero bias online.

[0035] After obtaining the calibrated motion data from all sensors, the benchmark solving unit performs spatial rigid body kinematics fusion. Since all sensors are rigidly connected to the same mount, their observed motions must satisfy rigid body constraints. The benchmark solving unit constructs an overdetermined system of equations using a pre-calibrated sensor network geometry database. This system uses the six degrees of freedom motion of the mount—three linear accelerations and three angular accelerations—as unknowns, and the linear accelerations and angular velocities measured by each sensor as observed quantities. By solving this least-squares problem, the benchmark solving unit can calculate the six degrees of freedom motion at the nozzle mount's center of mass with significantly higher accuracy than the results obtained from a single sensor, while effectively suppressing noise from individual sensors.

[0036] Finally, the reference calculation unit integrates the calculated motion quantities over time. It uses the static mounting base reference coordinate system, precisely measured and determined when the equipment is static, as its initial state. After the equipment starts operating, the reference calculation unit continuously applies the integrated displacement and angular increments to this initial coordinate system. Specifically, linear velocity updates the origin position of the coordinate system through first-order integration, and angular velocity updates the attitude direction of the coordinate system through quaternion or rotation matrix integration. Through this continuous update process, the reference calculation unit outputs a dynamic reference coordinate system in real time. The origin and attitude of this coordinate system always track the dynamic position and attitude of the actual centroid of the nozzle mounting base, thus providing a realistic, vibration-responsive spatial reference frame for subsequent measurements. The update frequency of this dynamic reference coordinate system is consistent with the sampling rate, reaching 10 kHz, ensuring the capture of high-frequency vibration components.

[0037] The non-contact three-dimensional pose measurement module is responsible for accurately acquiring the real-time pose of the nozzle tip under dynamic reference. Please refer to the attached document again. Figure 2 This module consists of a laser displacement sensor array and a pose calculation unit. The laser displacement sensor array comprises six high-precision single-point laser triangulation sensors. These six sensors are rigidly mounted on the crossbeam of the printing equipment frame in a non-coplanar manner using a high-rigidity adjustment bracket. During installation, precise optical adjustments are made to ensure that the six laser measurement beams converge within a spherical space of approximately 30 mm in diameter, located about 10 mm below the nozzle tip; this space is the preset measurement space. The spatial installation position of each sensor, i.e., the three-dimensional coordinates of its output port center in the global coordinate system, and its laser beam direction vector, are precisely calibrated once and permanently stored during system installation using a laser tracker. A 5 mm diameter ceramic spherical target with a high-reflectivity gold coating is mounted on the nozzle tip as the measurement target.

[0038] The pose calculation unit is implemented using a separate dedicated field-programmable gate array (FPGA) for high-speed parallel computing. Its workflow is as follows: In each measurement cycle, the pose calculation unit sends a synchronization trigger signal to six laser displacement sensors. Each sensor emits a laser beam that illuminates the spherical target surface at the nozzle tip. The photodetector inside the sensor receives the light spot reflected from the target surface and, using triangulation principles, calculates and outputs the precise distance value from the sensor probe to the target surface in real time, with a resolution of up to 0.1 micrometers. The pose calculation unit synchronously acquires these six distance values ​​via a high-speed Ethernet interface.

[0039] Based on the principle of spatial multi-point intersection measurement, given the spatial positions and beam directions of six sensors, as well as their respective measured distances to the target, six spatial rays can be constructed. Theoretically, these six rays should intersect at the center of the spherical target. However, due to measurement errors, they do not strictly intersect. The pose calculation unit uses the least squares method to solve for the optimal spatial coordinates of the target's center. Specifically, the algorithm finds a spatial point that minimizes the sum of the squares of the perpendicular distances from that point to the six spatial rays. By solving this nonlinear optimization problem, the three-dimensional coordinates (X, Y, Z) of the spherical target center, i.e., the nozzle tip, in the global coordinate system can be obtained.

[0040] Besides position, attitude measurement is achieved by analyzing the imaging position of the reflected light spot on the linear array charge-coupled device (CCouD) inside the sensor. When the laser beam is incident perpendicularly on the spherical target, the reflected light spot is located at the center of the CCouD. When the target tilts, i.e., its attitude changes, the position of the reflected light spot on the CCouD shifts linearly. Each laser displacement sensor can actually provide the angle information of the projection of the target surface normal direction onto the sensor's measurement plane. By fusing the angle information provided by at least three non-coplanar sensors, the pose calculation unit can calculate the small attitude angle of the spherical target, i.e., the nozzle tip, relative to the sensor array coordinate system, i.e., the tilt angle around the X and Y axes. Combining the position obtained from the aforementioned intersection measurement, the six-degree-of-freedom pose of the nozzle tip in the global coordinate system is finally obtained.

[0041] Finally, the pose calculation unit transforms the calculated global pose data into the dynamic reference coordinate system using the transformation matrix provided in real time by the dynamic reference reconstruction module. The transformed data, namely the real-time three-dimensional coordinates and three attitude angles of the nozzle tip relative to its mounting base dynamic reference, is output to the downstream module. This transformation is crucial, as it eliminates the influence of the overall motion of the mounting base, allowing the measurement results to directly reflect the relative pose deviation of the nozzle relative to the mounting base, i.e., the dynamic error we are truly concerned with. The pose calculation unit can measure at a frequency of up to 5 kHz, meeting the requirements of high-speed dynamic measurement.

[0042] The dynamic error modeling and compensation module is the intelligent core of the system, responsible for learning patterns from the data and generating compensation instructions. Please refer to the appendix. Figure 3This module runs on a host industrial computer and includes an error extraction unit, a model building unit, and a compensation calculation unit. The error extraction unit continuously receives real-time nozzle pose data from the non-contact 3D pose measurement module. Simultaneously, it reads the original printing path planning instruction corresponding to the current timestamp from the motion controller's cache. This instruction contains the nozzle target pose calculated based on an ideal rigid body kinematics model, i.e., the ideal pose model data. The error extraction unit compares the real-time measured pose with the ideal model pose item by item, performs subtraction operations, and directly calculates the nozzle's positional error in the X, Y, and Z translational directions, and the angular error in the X, Y, and Z rotational directions, totaling six degrees of freedom—the real-time error values.

[0043] The task of the model building unit is to establish a dynamic error spatiotemporal evolution model. During the offline modeling phase, the system drives the printing device to operate within the entire workspace of the multi-axis motion system according to a pre-designed scanning trajectory. This trajectory covers various typical combinations of printing speeds and accelerations, such as uniform motion from 0.1 m / s to 2 m / s, and motion from 1 m / s... 2 Up to 10m / s 2 The acceleration and deceleration process is described. During this process, the model building unit synchronously collects and stores massive amounts of data. The data includes: time-series six-degree-of-freedom error data output by the error extraction unit; actual velocity and acceleration values ​​fed back by the servo drives of each axis of the multi-axis motion system; and vibration spectrum features extracted from the raw data of the vibration sensor array, such as energy values ​​within a specific frequency band. These operating state parameters and error data constitute one-to-one corresponding sample pairs.

[0044] The model building unit uses a system identification method to process this data. It models the dynamic error as a multi-input multi-output system. The input vector U contains multiple operating state parameters, such as X-axis velocity, Y-axis velocity, X-axis acceleration, Y-axis acceleration, and the amplitude of the vibration energy characteristic frequency bands 1 to N. The output vector Y represents the error values ​​for the six degrees of freedom. The model building unit uses an autoregressive model structure with exogenous inputs to characterize this system. For the error of the k-th degree of freedom, the model can be expressed as:

[0045] in, This represents the k-th error value at the current time. The left side represents a linear combination of its own historical values, and the right side represents a linear combination of the historical values ​​of each component of the input vector U. It is white noise. , For model order, To account for time delay. The model building unit uses the least squares method, utilizing all collected offline data to estimate all coefficients in the model in one go. and This allows for the establishment of independent dynamic equations for each degree of freedom. The set of these equations constitutes the spatiotemporal evolution model of dynamic error, which quantitatively describes how nozzle pose error is generated and evolves under specific motion states and vibration excitations. The trained model parameters are then permanently stored in non-volatile memory.

[0046] The compensation calculation unit is activated during the online compensation phase. In each control cycle, for example, 0.2 milliseconds, the compensation calculation unit receives the current motion state parameters in real time, including the commanded velocities and accelerations of each axis, as well as the real-time vibration spectrum. It uses these parameters as input to a pre-trained dynamic error spatiotemporal evolution model. Using the error value from the previous moment as initial conditions, and combining it with the current input, the model iteratively predicts the six-degree-of-freedom pose error that the nozzle will generate in the next one or several control cycles, for example, 0.5 milliseconds later. This predictive capability is crucial for achieving feedforward compensation.

[0047] After obtaining the prediction error, the compensation calculation unit needs to convert it into executable instructions for the motion axes. It calls the stored inverse kinematics model of the multi-axis motion system. This model describes the mapping relationship from the nozzle tip pose to the displacements or angles of each motion axis, typically the X, Y, and Z axis linear motors, and possibly the U and V axis rotary axes. The compensation calculation unit treats the prediction error as a small pose disturbance, adds it to the current ideal pose, and obtains the compensated desired pose. Then, using the inverse kinematics model, it calculates the theoretical positions of each axis corresponding to the ideal pose, and the new positions of each axis corresponding to the compensated desired pose. The difference between the two is the compensation displacement or compensation angle amount that needs to be applied to each motion axis driver, collectively referred to as the multi-axis linkage compensation amount. These compensation amounts are vectors, possessing magnitude and direction, and directly act on the control loops of each axis.

[0048] The closed-loop feedback control module is the execution end of the system and is deeply integrated into the device's motion controller. Please refer to the appendix. Figure 5 This module includes an instruction overlay unit and a motion control unit. The instruction overlay unit receives the raw path planning instruction stream from the upper-level print management system. This instruction stream provides a series of discrete printhead target pose points at a high frequency, such as 1 kHz. Simultaneously, it receives multi-axis linkage compensation amounts from the dynamic error modeling and compensation module at an even higher frequency, such as 5 kHz. The core operation of the instruction overlay unit is time synchronization and vector overlay. Through precise timestamp alignment, it performs real-time vector overlay of the multi-axis linkage compensation amount corresponding to the current moment with the raw target pose points to be executed. For position, this involves addition of three-dimensional coordinates; for attitude, it involves multiplication of rotation matrices or quaternions. The calculation generates the actual target pose instruction after error compensation.

[0049] The motion control unit receives the actual target pose command. First, it decomposes it into independent position commands for each motion axis using inverse kinematics. Then, the motion control unit executes a classic three-loop control algorithm. The innermost loop is the current loop, which controls the driver to output precise torque by detecting the motor phase current to quickly respond to load changes. The middle loop is the speed loop, which adjusts the current command based on the actual speed feedback from the encoder, ensuring the motor speed closely follows the commanded speed. The outermost loop is the position loop, which adjusts the speed command based on the actual position feedback from a high-precision grating ruler, ultimately ensuring that each axis moves to the compensated commanded position. The calculation cycle for all control loops is extremely short, typically on the order of 100 microseconds. Through this process, the nozzle is actually driven to the target pose after dynamic error compensation, thus physically correcting the pose deviation. The entire system forms a closed loop: the non-contact 3D pose measurement module provides actual pose feedback, and the dynamic error modeling and compensation module provides model-based feedforward prediction. Both work together in the closed-loop feedback control module to achieve precise control of the nozzle's dynamic pose.

[0050] The system operates according to a phased workflow to ensure performance and efficiency. Please refer to the appendix. Figure 4 The process is clearly divided into an offline modeling stage and an online compensation stage. The offline modeling stage is performed after the initial installation and commissioning of the equipment, after major overhauls, or periodically, such as during quarterly maintenance. During this stage, the equipment does not perform actual printing but instead executes an automated operating condition scanning program. The program controls the printhead to move unloaded across the entire printing platform using various preset speed and acceleration combinations, simulating the dynamics of real printing. During this process, the model building unit of the dynamic error modeling and compensation module simultaneously collects error and status data from dense grid points throughout the entire workspace, amounting to millions of data sets. Using this data, the model building unit completes the global training and parameter calibration of the dynamic error spatiotemporal evolution model. The training process includes model structure selection, parameter estimation, and model verification. The finally verified model parameters are permanently stored in the non-volatile memory of the motion controller for use in the online stage. This stage may last several hours, but it lays a solid foundation for subsequent long-term high-precision online compensation.

[0051] The online compensation phase is automatically activated each time the device performs a print job. At the start of a print job, the system first loads the dynamic error model parameters trained offline from memory, initializing the dynamic error modeling and compensation module. Subsequently, during printing, the dynamic reference reconstruction module, non-contact 3D pose measurement module, dynamic error modeling and compensation module, and closed-loop feedback control module all run synchronously. The system continuously executes the entire process of dynamic reference establishment, printhead pose measurement, error prediction, compensation calculation, and motion control command correction in real time. During this phase, all calculations must be completed within strict timing constraints, forming a hard real-time control loop. This ensures that throughout the entire print job, regardless of the printhead's location or speed, its actual pose remains highly consistent with the ideal pose model through automatic compensation, thereby guaranteeing print quality and preventing collisions.

[0052] To cope with the complex interference in real-world industrial environments, the system is designed with advanced fault tolerance and adaptive mechanisms. In the non-contact 3D pose measurement module, the pose calculation unit integrates target integrity verification. It continuously monitors the echo signal intensity and waveform characteristics returned by each laser displacement sensor. A normal spherical target reflection signal has a specific intensity range and a smooth waveform. When ink droplets ejected from the printhead accidentally splash onto the target surface, or when the printhead body briefly blocks part of the laser beam in an extreme posture, the signal intensity received by the affected sensor drops sharply, and the waveform becomes distorted. The pose calculation unit incorporates a threshold- and pattern-matching-based detection algorithm. Once any sensor signal is detected as abnormal, the data for that measurement cycle is immediately marked as invalid. To ensure that the control closed loop is not interrupted by a single invalid data instance, the system triggers a data interpolation mechanism. This mechanism utilizes historically valid printhead pose data from previous cycles, combined with the printhead command velocity and acceleration currently read from the motion controller, to construct a simplified motion state observer. Employing a Kalman filter algorithm, the observer predicts and estimates the most probable nozzle pose at the current moment based on the system motion model and historical observations. This estimated value temporarily replaces invalid measurements and is output to the downstream module. Once the signal returns to normal, the system automatically switches back to normal measurement data. This mechanism significantly improves the system's robustness under harsh operating conditions.

[0053] In the dynamic error modeling and compensation module, the model building unit also implements a recursive least squares method based on the forgetting factor for online fine-tuning of model parameters. Although the offline model has high basic accuracy, during long-term operation, the mechanical structure may develop tiny gaps due to wear, the lubrication status of transmission components may change, and fluctuations in ambient temperature may cause thermal deformation of the structure. These factors can all cause the dynamic characteristics of the equipment to drift slowly.

[0054] To address this issue, during the online compensation phase, the system continuously performs a covert comparison: it compares the error value predicted by the dynamic error model based on the current state with the real-time error value actually measured by the non-contact 3D pose measurement module and calculated by the error extraction unit, thus calculating the prediction residual. This residual reflects the slight mismatch between the current model and the actual device state. The model building unit uses this residual sequence, along with the current operating state parameter vector, to recursively calculate the coefficients in the key parameters of the dynamic error model—namely, the coefficients in the aforementioned equations—using the least squares method. The system performs online, small-scale, and gradual updates. The recursive least squares method introduces a forgetting factor less than 1, such as 0.9995, making the algorithm more focused on the influence of recent data, thus gradually "forgetting" information from the old model corresponding to the changed operating conditions. Through this continuous adaptive fine-tuning, the dynamic error model can slowly track changes in equipment characteristics, enabling the system to maintain long-term calibration accuracy, significantly reducing reliance on periodic manual recalibration, and demonstrating the system's intelligence and self-learning capabilities.

[0055] The system implementation also involves sophisticated timing synchronization and network communication architecture. All modules, including sensor data acquisition, pose calculation, model prediction, and control command generation and execution, are synchronized based on a unified precision clock source. This clock source is typically provided by the motion controller's master clock and distributed to each computing node via a precise time protocol network.

Claims

1. A large UV printer head three-dimensional space automatic calibration system, characterized in that, The system is integrated into the motion control architecture of a large UV printing device, and the system comprises: a dynamic reference reconstruction module, configured to collect and fuse data from multiple high-precision vibration sensors in real time during operation of the printing device, to solve and generate a dynamically changing printhead mounting seat space reference coordinate system; a non-contact three-dimensional pose measurement module, configured to measure the three-dimensional coordinates and attitude angles of the printhead tip in the dynamic reference coordinate system generated by the dynamic reference reconstruction module in real time in a non-contact manner during operation of the printing device; a dynamic error modeling and compensation module, configured to construct a time-space evolution model of the printhead dynamic pose error based on the printhead pose data measured in real time by the non-contact three-dimensional pose measurement module, in combination with a preset printhead ideal pose model, and to solve a multi-axis linkage compensation amount for correcting the motion control instructions according to the model in real time; a closed-loop feedback control module, configured to receive the multi-axis linkage compensation amount output by the dynamic error modeling and compensation module, and to superimpose it with the original printing path planning instructions in real time to generate the final motion control instructions, to drive the multi-axis motion system to execute, thereby realizing closed-loop correction of the actual operation pose of the printhead.

2. The three-dimensional space automatic calibration system for a large UV printer head according to claim 1, characterized in that, The dynamic reference reconstruction module comprises a vibration sensor array and a reference solving unit; The vibration sensor array is composed of at least 3 three-axis accelerometers and at least 3 three-axis gyroscopes, which are distributedly fixedly installed on the rigid main structure of the printhead mounting seat in a rigid connection manner; The reference solving unit is configured to perform the following processes: real-time synchronous collection of original data of all sensors in the vibration sensor array; gravity component stripping and coordinate transformation processing of the accelerometer data, and attitude solving and drift correction processing of the gyroscope data; fusion processing of data of all sensors based on the fixed geometric relationship between the sensors, solving of real-time six-degree-of-freedom pose variation of the printhead mounting seat in the global coordinate system by using the spatial rigid body kinematics principle; continuous updating of the preset static mounting seat reference coordinate system by using the real-time pose variation, thereby outputting the dynamic reference coordinate system.

3. The three-dimensional space automatic calibration system for a large UV printer head according to claim 1, wherein, The non-contact three-dimensional pose measurement module comprises a laser displacement sensor array and a pose solving unit; The laser displacement sensor array is composed of 4 or more single-point laser displacement sensors, which are rigidly installed on the printing device rack in a non-coplanar manner, and the measurement beams thereof jointly converge in a preset measurement space near the printhead tip; The pose solving unit is configured to perform the following processes: controlling each sensor in the laser displacement sensor array to synchronously emit a measurement beam to a preset spherical target surface on the printhead tip; receiving accurate distance values from the target surface returned by each sensor; based on the pre-calibrated spatial installation position and beam direction of each sensor, and the measured distance values, calculating the three-dimensional coordinates of the center point of the spherical target in the global coordinate system by using the spatial multi-point intersection measurement principle; converting the solved printhead tip pose data to the dynamic reference coordinate system provided by the dynamic reference reconstruction module, and outputting the real-time three-dimensional coordinates and attitude angles of the printhead tip relative to the mounting seat.

4. The three-dimensional space automatic calibration system for a large UV printer head according to claim 1, wherein, The dynamic error modeling and compensation module comprises an error extraction unit, a model construction unit and a compensation calculation unit. The error extraction unit is configured to receive real-time position data of the printhead output by the non-contact three-dimensional position measurement module, and call printhead ideal position model data corresponding to the printing path point at the moment from a preset storage unit; compare the real-time measurement position with the ideal model position item by item, and calculate real-time error values of the printhead in six degrees of freedom in X, Y and Z three translation directions and around X, Y and Z three rotation directions; The model construction unit is configured to continuously collect time sequence error data output by the error extraction unit within a continuous equipment running time, and simultaneously collect running state parameters of the equipment at the moment, the running state parameters comprising speeds and accelerations of each axis of the multi-axis motion system and vibration frequency spectrum features collected by a vibration sensing array; A dynamic error space-time evolution model is established by using a system identification method, with the running state parameters as inputs and the six degrees of freedom error values as outputs; The compensation calculation unit is configured to receive real-time motion state parameters in subsequent equipment running, input the parameters into the established dynamic error space-time evolution model, and predict position errors of the printhead to be generated in future control cycles; according to the predicted errors, inverse solutions of compensation displacement amounts and compensation angle amounts required to be applied to each motion axis driver are solved by combining a kinematics inverse solution model of the multi-axis motion system, i.e. the multi-axis linkage compensation amount.

5. The three-dimensional space automatic calibration system for a large UV printer head according to claim 1, wherein, The closed loop feedback control module is integrated in the motion controller of the equipment, and the module comprises an instruction superposition unit and a motion control unit; The instruction superposition unit is configured to receive original path planning instructions from an upper layer printing management system, the instructions comprising a series of discrete printhead target position points calculated based on an ideal equipment model; at the same time, the multi-axis linkage compensation amount from the dynamic error modeling and compensation module is received in real time; The multi-axis linkage compensation amount corresponding to the current moment is vector superposed with the original target position point to generate actual target position instructions after error compensation; The motion control unit is configured to calculate and output driving signals to each axis servo driver according to the actual target position instructions, drive the printhead to move to the target position after compensation by a three-loop control algorithm of the position loop, the speed loop and the current loop of the motion controller.

6. The three-dimensional space automatic calibration system for a large UV printer head according to claim 1, wherein, The system runs in a calibration workflow executed in stages, the workflow comprising an offline modeling stage and an online compensation stage; The offline modeling stage is executed after equipment installation and debugging or regular maintenance, in which the printing equipment is driven to run empty in the full workspace of the multi-axis motion system at a plurality of typical speed and acceleration combinations; in this process, the dynamic error modeling and compensation module synchronously collects a large amount of error data and running state parameters, completes global training and parameter calibration of the dynamic error space-time evolution model, and solidifies the final model parameters for storage; The online compensation stage is continuously executed in the daily printing operation of the device, and a dynamic error model trained by the offline modeling stage is loaded; the dynamic reference reconstruction module, the non-contact three-dimensional position measurement module, the dynamic error modeling and compensation module, and the closed-loop feedback control module are synchronously activated during the printing process, so that real-time measurement, error prediction, and motion compensation of the position of the nozzle are achieved.

7. The three-dimensional space automatic calibration system for a large UV printer head according to claim 3, wherein, The position calculation unit is further configured to perform target integrity verification. The target integrity verification process is as follows: by analyzing the echo signal strength and waveform characteristics received by each laser displacement sensor, it is determined whether the measurement beam is completely irradiated on the effective reflection area of the spherical target; when the signal characteristics of any one sensor indicate that the beam is blocked, the position calculation unit immediately marks the data at this moment as invalid, and triggers the data interpolation mechanism; The data interpolation mechanism estimates the most likely position of the nozzle at the current moment by using the Kalman filtering algorithm according to the historical valid position data and the current device motion state, and inputs the temporary data into the dynamic error modeling and compensation module.

8. The three-dimensional space automatic calibration system for a large UV printer head according to claim 4, wherein, The model construction unit uses the recursive least square method based on the forgetting factor to perform online fine tuning of the model parameters; The online fine tuning process is as follows: in the online compensation stage, the system continuously compares the error value predicted by the dynamic error model with the error value actually measured by the non-contact three-dimensional position measurement module, calculates the residual error, and uses the residual error and the current running state parameters to perform small and gradual online updating of the key parameters of the model by using the recursive least square method.

9. The three-dimensional space automatic calibration system for a large UV printer head according to claim 2, wherein, The installation positions of the sensors in the vibration sensor array are pre-calibrated to determine the precise geometric relationship therebetween; The reference calculation unit uses the spatial rigid body kinematics principle to perform data fusion based on the precise geometric relationship.

10. The three-dimensional space automatic calibration system for a large UV printer head according to claim 3, wherein, The position calculation unit is further configured to calculate the small attitude deflection angle of the target, i.e., the tip of the nozzle, relative to the measurement beam by analyzing the position change of the reflected light spot on the sensor receiver.

Images