Visual guidance spraying trajectory optimization control method for complex curved surface injection molding parts

Abstract

The present application belongs to the technical field of motion closed-loop control adjustment, and relates to a visual guidance spraying track optimization control method for complex curved surface injection molded parts, which comprises the following steps: combining transmission disturbance and position deviation to solve track space position tracking deviation vector; determining the absolute value of curvature time sequence variability through space coordinate difference calculation; when the absolute value is within the safe boundary, performing smooth division convergence calculation of the basic control gain constant to generate calibration control gain to modulate the deviation vector; when the absolute value exceeds the safe boundary, activating the degraded state machine control flow and calling the historical safe driving current average value to keep the output; through active smooth convergence of the gain and double-mode control switching, the present application can smooth the transient tracking oscillation caused by response lag, and realize accurate following of the whole curvature on the premise of not reducing the global rate.

Description

Technical Field

[0001] This invention belongs to the field of motion closed-loop control and adjustment technology, and relates to a visually guided spraying trajectory optimization control method for injection molded parts with complex curved surfaces. Background Technology

[0002] Currently, the use of robot motion planning and adaptive control technology for automatic spraying of irregularly shaped workpieces has become a common practice in intelligent manufacturing. However, in the process of spraying complex curved workpieces, existing technologies still have shortcomings in terms of real-time trajectory adjustment and actual motion deviation compensation.

[0003] First, existing technologies can utilize machine vision and probabilistic models for spraying trajectory planning. For example, Chinese invention patent CN119369421B discloses a machine vision-based curve trajectory spraying method and system. This method reconstructs the three-dimensional model of the workpiece using an industrial camera and a depth camera, and then uses multi-agent game theory and a recursive Gaussian process state space model to generate an optimized control command sequence. This scheme requires large-scale game optimization and Gaussian process variational inference, which is computationally complex and can easily increase the computational burden on the online control system. At the same time, the analog in-memory computing architecture it uses is easily affected by changes in ambient temperature and hardware process deviations. When the signal represented by current undergoes nonlinear fluctuations, it may cause distortion of control commands. In addition, this scheme lacks direct compensation for mechanical structure deformation, making it difficult to guarantee trajectory accuracy during continuous spraying over a long period of time.

[0004] Secondly, existing technologies can also control the distance between the nozzle and the workpiece through mechanical contouring and local adjustments. For example, Chinese invention patent CN120479639B discloses a control method for online monitoring and self-correction of the coating quality of automotive plastic parts. This method uses a movable, fitted arched adapter plate to fit the workpiece surface, allowing the nozzle to rise and fall synchronously with the curvature of the bottom of the adapter plate. Local points are adjusted based on the curvature deviation obtained from laser scanning. This scheme mainly relies on the deformation of the mechanical structure to achieve contouring. When the nozzle moves to either end of the adapter plate or enters a boundary area with a large curvature change, the mechanical spring is prone to plastic deformation and there is a risk of fatigue damage. To avoid structural damage, the system needs to stop actively pulling down at the edge of the adapter plate for adjustment. Therefore, when spraying at the edge of a workpiece with a large curvature change or a variable cross-section, it is difficult to maintain the preset distance between the nozzle and the workpiece.

[0005] Third, some existing technologies correct motion trajectories by separating multiple vibration signals and performing feedback control. For example, Chinese invention patent application CN120255318A discloses a surface treatment method for a hair dryer based on intelligent multi-axis linkage technology. This method uses multiple sensors to collect electromagnetic noise, mechanical shock, and airflow turbulence signals, determines the main vibration sources through adaptive wavelet packet decomposition, dynamically adjusts multi-axis collaborative control commands, and simultaneously uses a laser displacement sensor to perform closed-loop adjustment of proportional and integral gains. Such solutions mainly suppress the vibration of the multi-axis equipment itself, but do not adequately consider the real-time detection and correction of the overall spatial shape of the workpiece and its three-dimensional deviations. When the injection molding process causes workpiece deformation, or when there are repetitive positioning errors in the tooling fixture, the system struggles to correct the overall spraying trajectory in a timely manner based on visual inspection results. It cannot effectively compensate for path offsets caused by large deviations in the actual surface of the workpiece, thus affecting the uniformity of the coating on large-sized irregularly shaped workpieces.

[0006] Therefore, existing automatic spraying control technology still needs to improve its ability to detect the overall spatial shape of the workpiece in real time, and to correct the spraying trajectory in a timely manner according to the actual deformation of the workpiece, so that trajectory control and mechanical contouring can still be kept within a safe and stable working range under complex dynamic disturbances. This is the technical problem that this invention aims to solve. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and solve the problems of physical path deviation and trajectory oscillation caused by high computational complexity, mechanical fatigue or lack of macroscopic spatial trajectory visual dynamic mapping mechanism when processing dynamic optimization of complex curved surface trajectories. This invention provides a visually guided spraying trajectory optimization and control method for injection molded parts with complex curved surfaces.

[0008] To achieve the above-mentioned objectives, this invention provides a vision-guided spraying trajectory optimization and control method for injection molded parts with complex curved surfaces, comprising the following steps:

[0009] Step S1, obtain discrete point set and calculate position tracking deviation vector: obtain discrete sampling point set in trajectory tracking channel, calculate position deviation between theoretical trajectory point position and actual image acquisition point position, and perform cascade alignment operation with electromechanical transmission disturbance estimated by inner loop disturbance observation module to calculate trajectory spatial position tracking deviation vector of current sampling period; Step S2, constructing composite dynamic variables by calculating the coordinate difference point set: Perform spatial coordinate difference calculation on the discrete sampling point set to determine the absolute value of the curvature temporal rate of change, which characterizes the degree of transient variation of spatial geometric curvature with time, and introduce the absolute value of the curvature temporal rate of change as a multiplier term into the velocity vector within the discrete sampling period to construct composite dynamic variables; Step S3, division convergence calculation of basic gain to generate calibration gain: When the absolute value of curvature temporal variation rate is within the preset safety boundary, the preset basic control gain constant is smoothly divided and converged according to the composite dynamic variable. Before the end of the current discrete sampling period, the feedback gain is reduced to generate calibration control gain. The calibration control gain has a smooth convergence relationship with the increase of the composite dynamic variable. Step S4, Modulate the product and output the command to complete the dynamic time delay compensation: The product of the modulation trajectory spatial position tracking deviation vector and the calibration control gain is used to output the decoupled current cycle drive control current command to the multi-axis channel servo control unit. The drive torque controlled unit outputs the motor torque to complete the in-situ time delay dynamic compensation at the spatial inflection point.

[0010] The control loop described in this invention is provided with a judgment window for the transition from the normal control state to the degraded sleep state, and periodically monitors the transient jump state of the absolute value of curvature time-series change rate; when the absolute value of curvature time-series change rate exceeds the safety boundary of 0.65 twice in a row, the degraded state machine control flow is activated and the spatial coordinate difference calculation path in step S2 is blocked. The average value of the safety drive current of the past 5 sampling cycles stored in the historical instruction storage module is directly called as the nominal drive control current instruction to complete the continuous output, so as to block the transmission of high-frequency outlier noise to the gain modulation loop.

[0011] Step S1 of the present invention includes the following sub-steps: Step S11, acquiring actual point signals on the surface of the complex curved injection molded part as actual image acquisition points, and comparing them with theoretical trajectory points to calculate the position deviation; Step S12, using the inner loop disturbance observation module to estimate the electromechanical transmission disturbance of the multi-axis channel servo control unit, and performing cascade alignment operation between the position deviation and the electromechanical transmission disturbance to calculate the trajectory spatial position tracking deviation vector of the current sampling period.

[0012] Step S2 of the present invention includes the following sub-steps: Step S21, performing second-order difference calculation on the spatial three-dimensional coordinates within a continuous discrete sampling period to extract curvature feature quantities characterizing the change in geometric curvature of the trajectory; Step S22, performing transient differential calculation on the curvature feature quantities along the discrete time axis to determine the absolute value of the curvature temporal variation, and introducing the absolute value of the curvature temporal variation as a multiplier term into the magnitude of the velocity vector.

[0013] The present invention describes a control environment in which the discrete sampling point set presents a concave transition state, which shortens the sampling time interval for obtaining the discrete sampling point set in step S1, thereby improving the trajectory tracking sampling frequency in the geometric variation region.

[0014] In step S4 of the present invention, before outputting the current cycle drive control current command, the dynamic coupling inertia between each motion axis in the multi-axis channel servo control unit is eliminated in situ using a cross decoupling matrix, and the trajectory spatial position tracking deviation vector of multi-axis coupling is transformed into an independent scalar control command for each motion axis.

[0015] The safety boundary described in this invention is 0.65; when the absolute value of the curvature temporal variation rate is less than or equal to 0.65, the trajectory is within the safe measurable boundary, the control loop maintains normal control state operation, and continuously performs gain scheduling on the trajectory spatial position tracking deviation vector by calibrating the control gain.

[0016] The historical instruction storage module of the present invention adopts a circular queue structure. At the end of each discrete sampling period, the queue is updated, the earliest period's drive control current data is automatically removed, and the current period's drive control current data is written in, so as to maintain the latest 5 sampling period's safe drive current data sequence stored in the historical instruction storage module.

[0017] In the continuous spraying control operation described in this invention, optical image signals from a vision sensor are acquired in real time and specular reflection noise is filtered out. The calibrated spatial surface geometric contour parameters of the injection molded part are introduced as feedforward input into step S3 to correct the preset range of the basic control gain constant in real time.

[0018] Compared with the prior art, the present invention has at least the following beneficial effects: 1. By making the calibration control gain decrease smoothly as the composite dynamic variable increases, the control loop can reduce the feedback intensity in time at the spatial inflection point of the complex curved surface, reduce the accumulation of drive control current command and transient oscillation caused by transmission backlash and electromechanical response hysteresis, and avoid obvious phase lag in the trajectory tracking process; without reducing the global motion rate of the multi-axis channel servo control unit, the smoothness and convergence stability of the complex curved surface spraying trajectory are improved.

[0019] 2. When the absolute value of the curvature timing rate of change is abnormally abruptly due to the noise of the mirror reflection, the abnormal sampling data is stopped for gain modulation by the degraded state machine control flow, and the average value of the historical safe drive current is used to maintain the continuous output of control commands. This can prevent high-frequency outlier noise from continuing to enter the gain modulation circuit, reduce sudden changes in drive commands and instability of control actions, and enable the spraying control system to maintain continuous and stable trajectory tracking even when the optical acquisition conditions are disturbed. Attached Figure Description

[0020] Figure 1 This is a flowchart of the trajectory tracking deviation calculation and feedback gain adaptive control of the present invention; Figure 2This is a schematic diagram of the closed-loop topology of the intelligent sensing and multi-axis linkage servo actuator of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings.

[0022] Example 1: This embodiment discloses a vision-guided spraying trajectory optimization and control method for injection molded parts with complex curved surfaces, including the following steps: Step S1, obtain discrete point set and calculate position tracking deviation vector: obtain discrete sampling point set in trajectory tracking channel, calculate position deviation between theoretical trajectory point position and actual image acquisition point position, and perform cascade alignment operation with electromechanical transmission disturbance estimated by inner loop disturbance observation module to calculate trajectory spatial position tracking deviation vector of current sampling period; Step S2, constructing composite dynamic variables by calculating the coordinate difference point set: Perform spatial coordinate difference calculation on the discrete sampling point set to determine the absolute value of the curvature temporal rate of change, which characterizes the degree of transient variation of spatial geometric curvature with time, and introduce the absolute value of the curvature temporal rate of change as a multiplier term into the velocity vector within the discrete sampling period to construct composite dynamic variables; Step S3, division convergence calculation of basic gain to generate calibration gain: When the absolute value of curvature temporal variation rate is within the preset safety boundary, the preset basic control gain constant is smoothly divided and converged according to the composite dynamic variable. Before the end of the current discrete sampling period, the feedback gain is reduced to generate calibration control gain. The calibration control gain has a smooth convergence relationship with the increase of the composite dynamic variable. Step S4, Modulate the product and output the command to complete the dynamic time delay compensation: The product of the modulation trajectory spatial position tracking deviation vector and the calibration control gain is used to output the decoupled current cycle drive control current command to the multi-axis channel servo control unit. The drive torque controlled unit outputs the motor torque to complete the in-situ time delay dynamic compensation at the spatial inflection point.

[0023] The control loop described in this embodiment is equipped with a decision window for transitioning from the normal control state to the degraded sleep state, and periodically monitors the transient jump state of the absolute value of curvature timing variation. When the absolute value of curvature timing variation exceeds the safety boundary of 0.65 twice consecutively, the degraded state machine control flow is activated and the spatial coordinate difference calculation path in step S2 is blocked. The average value of the safe drive current of the past 5 sampling cycles stored in the historical instruction storage module is directly called as the nominal drive control current instruction to complete the continuous output, so as to block the propagation of high-frequency outlier noise to the gain modulation loop.

[0024] Step S1 in this embodiment includes the following sub-steps: Step S11, acquiring actual point signals on the surface of the complex curved injection molded part as actual image acquisition points, and comparing them with theoretical trajectory points to calculate the position deviation; Step S12, using the inner loop disturbance observation module to estimate the electromechanical transmission disturbance of the multi-axis channel servo control unit, and performing cascade alignment operation between the position deviation and the electromechanical transmission disturbance to calculate the trajectory spatial position tracking deviation vector of the current sampling period.

[0025] Step S2 in this embodiment includes the following sub-steps: Step S21, performing second-order difference calculation on the spatial three-dimensional coordinates within a continuous discrete sampling period to extract curvature feature quantities that characterize the change in geometric curvature of the trajectory; Step S22, performing transient differential calculation on the curvature feature quantities along the discrete time axis to determine the absolute value of the curvature temporal rate of change, and introducing the absolute value of the curvature temporal rate of change as a multiplier term into the magnitude of the velocity vector.

[0026] In this embodiment, under the control environment where the discrete sampling point set presents a concave transition state, the sampling time interval for acquiring the discrete sampling point set in step S1 is shortened to improve the trajectory tracking sampling frequency in the geometric variation region.

[0027] In step S4 of this embodiment, before outputting the current cycle drive control current command, the dynamic coupling inertia between each motion axis in the multi-axis channel servo control unit is eliminated in situ using a cross decoupling matrix, and the trajectory spatial position tracking deviation vector of the multi-axis coupling is transformed into an independent scalar control command for each motion axis.

[0028] The safety boundary described in this embodiment is 0.65. When the absolute value of the curvature temporal variation rate is less than or equal to 0.65, the trajectory is within the safe measurable boundary, the control loop maintains normal control operation, and continuously performs gain scheduling on the trajectory spatial position tracking deviation vector by calibrating the control gain.

[0029] The historical instruction storage module described in this embodiment adopts a circular queue structure. At the end of each discrete sampling period, the queue is updated, the earliest period's drive control current data is automatically removed, and the current period's drive control current data is written in, so as to maintain the latest 5 sampling period's safe drive current data sequence stored in the historical instruction storage module.

[0030] In this embodiment, during continuous spraying control operation, optical image signals from a vision sensor are acquired in real time and specular reflection noise is filtered out. The calibrated spatial surface geometric contour parameters of the injection molded part are introduced as feedforward input into step S3 to correct the preset range of the basic control gain constant in real time.

[0031] Example 2: In this embodiment, the central control unit receives a set of discrete sampling points on the surface of the complex curved injection molded part uploaded by the front-end image sensor via a fieldbus network interface. The data transmission link between the front-end image sensor, the fieldbus network interface, and the central control unit constitutes a trajectory tracking channel. When the multidimensional nonlinear geometric features of the workpiece exhibit curvature variations, or when residual mold release agent on the surface causes specular reflection noise, the fixed sampling time interval causes phase lag in the data stream. The constant gain loop cannot adjust the feedback gain according to changes in spatial curvature, resulting in unidirectional inertial accumulation of the drive control current command. This causes the multi-axis channel servo control unit to experience torque oscillation and path derailment at the inflection point of the spatial curve. The central control unit reads the pixel coordinates of the two-dimensional image acquired by the front-end image sensor and performs pre-calibration. The homogeneous spatial geometric transformation matrix is ​​used to convert the workpiece surface into physical space coordinates in three dimensions, and the transformation result is used as the actual image acquisition point. At the same time, the theoretical trajectory point at the current sampling time in the theoretical machining trajectory model is retrieved, and the difference is calculated according to the coordinates of each motion axis to obtain the position deviation containing the coordinate difference components of each axis. The inner loop disturbance observation module estimates the electromechanical transmission disturbance of the multi-axis channel servo control unit. This disturbance is physically manifested as an equivalent load resistance torque. The cascade alignment operation uses a physical scale transformation factor that includes the nominal torque constant of the motor and the lead screw parameter of the transmission screw to convert the resistance torque scalar into the equivalent rigid displacement deformation of each motion axis along the screw direction. Then, it is superimposed with the corresponding axis component of the position deviation to obtain the trajectory spatial position tracking deviation vector of the current sampling period.

[0032] The central control unit performs second-order difference calculations on the spatial three-dimensional coordinates of the discrete sampling point set within three consecutive discrete sampling periods. The spatial second-order difference results form a curvature feature quantity characterizing the change in the geometric curvature of the trajectory. Subsequently, the change in the curvature feature quantity in adjacent sampling periods is calculated along the discrete time axis, and transient differential calculations are performed according to the corresponding sampling time intervals. The absolute value of the differential result is taken to obtain the absolute value of the temporal variation rate of curvature. , The original calculation results use a combination of the third-order reciprocals of time and distance, and participate in boundary judgment according to the same data scale determined during calibration. The central control unit will... The magnitude of the velocity vector is introduced as a multiplier term to construct a composite dynamic variable. In this embodiment, the global motion rate of the spraying control system remains a constant nominal value. The magnitude of the velocity vector is normalized to 1 through a calibration process, and the dimensions of the velocity term are subsequently normalized. The value of the composite dynamic variable is thus determined by... The safety boundary of 0.65 is determined based on the boundary between the continuity of control actions and noise immunity stability under the calibration conditions, and is consistent with... Use the same data scale; when When the value is less than or equal to 0.65, the control loop maintains normal control state, and the central control unit adjusts the preset basic control gain constant according to the composite dynamic variables. Perform smooth division convergence calculation, reduce the feedback gain before the end of the current discrete sampling period, and obtain the calibration control gain. Its update formula is as follows: ,in, To calibrate the control gain, The preset basic control gain constant, both of which are dimensionless scalars; The time constant adjustment factor, determined through prior calibration, has its unit set according to the inverse physical unit of the composite dynamic variable. In this embodiment, the velocity vector magnitude is normalized to 1. The unit is equivalent to The reverse unit, making As the composite dynamic variable increases, the denominator of the formula continuously increases, becoming a dimensionless quantity. Smooth convergence decreases.

[0033] After completing gain modulation, the central control unit will combine the trajectory spatial position tracking deviation vector with... Multiply the results and transform them into independent scalar control commands for each motion axis through a cross-decoupling matrix. Then, according to the current command ratio preset by the multi-axis channel servo control unit, convert the scalar control commands of each axis into the current cycle drive control current command and output them within the current sampling period. The motors of each axis controlled by the multi-axis channel servo control unit constitute a torque-controlled unit. The torque-controlled unit outputs the corresponding motor torque according to the drive control current command. The calibration control gain decreases as the composite dynamic variable increases, limiting the inertial accumulation of the drive control current command at the spatial inflection point within a single sampling period, and completing the in-situ time delay dynamic compensation; when the residual mold release agent on the surface of the injection molded part causes mirror reflection noise, and When the safety boundary of 0.65 is exceeded twice consecutively, the decision window for transitioning from normal control state to degraded sleep state triggers the degraded state machine control flow. The central control unit blocks the spatial coordinate differential calculation path, stops using the current abnormal sampling data for gain modulation, and calls the drive control current data of the most recent 5 safe sampling cycles from the historical instruction storage module. The safe sampling cycle refers to... When the sampling period is less than or equal to 0.65 and the control loop is in normal control state, the historical instruction storage module calculates the arithmetic mean of 5 safe drive current data to obtain the nominal drive control current instruction and continuously outputs it to block the transmission of high-frequency outlier noise to the gain modulation loop.

[0034] When the second-order difference direction of the spatial three-dimensional coordinates of the discrete sampling point set points to the inner side of the calibrated workpiece surface, the central control unit determines that the point set presents a concave transition state. The sampling time interval is jointly determined by the degree of change of curvature feature quantity and the processing overhead of the central control unit. In the concave transition state, it is shortened from the normal 30ms to 10ms to improve the trajectory tracking sampling frequency in the geometric variation region. Before outputting the current cycle drive control current command, the central control unit uses a cross decoupling matrix to eliminate the dynamic coupling inertia between each motion axis of the multi-axis channel servo control unit. The matrix elements of the cross decoupling matrix are determined by the combination of the mutual inertia coefficient and the joint acceleration dynamic term between each motion axis. Each off-diagonal matrix element represents the dynamic coupling component generated by another motion axis on the current axis. During the decoupling transformation, the central control unit multiplies the deviation components of other motion axes with the corresponding cross-decoupling matrix elements, sums the resulting dynamic coupling inertial components, and subtracts this sum from the original deviation control command of the current axis to obtain independent scalar control commands for each motion axis, thus numerically offsetting the mechanical inertial interference between the axes. The historical command storage module adopts a circular queue structure and performs queue update judgment at the end of each discrete sampling period. When the current period is a safe sampling period, the circular queue automatically removes the drive control current data of the earliest period and writes it into the drive control current data of the current period. When the current period does not meet the safety conditions, the drive control current data of that period is not written into the safe data queue. Thus, the circular queue always stores the safe drive current data sequence of the most recent 5 safe sampling periods. The 5 sampling periods are used to balance the average suppression effect of outlier impulse noise and the phase lag of the nominal drive control current command.

[0035] During continuous spraying control, the central control unit acquires optical image signals from the vision sensor in real time. It filters out specular reflection noise through two-dimensional digital image grayscale gradient analysis. The central control unit scans the optical image signal and calculates the grayscale change rate between adjacent pixels. When the grayscale value of a specific area suddenly increases and the local grayscale gradient exceeds a preset threshold for dramatic brightness changes, specular reflection noise is identified in that area. Subsequently, a 5x5 pixel spatial median filter window is invoked to perform pixel-level sliding window replacement on the bright distortion area, eliminating oversaturated optical bright spot noise. The filtered pixel coordinates are converted into three-dimensional physical space coordinates of the workpiece surface through a pre-calibrated homogeneous spatial geometric transformation matrix, and calibrated spatial surface geometric contour parameters of the injection molded part are formed according to the sampling order. The central control unit uses the calibrated spatial surface geometric contour parameters of the injection molded part as feedforward input. Before performing smooth division convergence calculation, it compares these parameters with the geometric contour parameters of corresponding points in the theoretical processing trajectory model. Based on the correspondence between the contour parameter deviation determined during the calibration stage and the value of the basic control gain constant, the current sampling period is determined within a preset range. This allows for real-time correction of the preset range of the basic control gain constant; the electromechanical transmission disturbance estimated by the position deviation and inner loop disturbance observation module is cascaded and aligned to form a trajectory spatial position tracking deviation vector, and the change of spatial geometric curvature with discrete time is observed through... The control gain is triggered to update, and the degraded state machine control flow is triggered when the mirror reflection noise continuously crosses the safety boundary. Through gain modulation, inter-axis decoupling and state switching, the time and space deviation caused by the electromechanical response hysteresis of the torque controlled unit is compensated. Finally, the trajectory tracking deviation is controlled within 0.19mm, the global motion rate of the multi-axis channel servo control unit remains unchanged, and the control loop maintains a stable closed-loop regulation state.

[0036] Example 3: This embodiment uses a multi-axis electromechanical drive simulation table and a vision optical imaging component to build a semi-physical in-loop test system to reproduce the follow-up control conditions of nonlinear complex curved injection molded parts during the spraying process. The image sensor outputs an optical image signal with a resolution of 1920×1080 pixels and a sampling frequency of 100Hz, and reproduces the specular reflection noise of the sprayed surface in the image. The laser displacement sensor has a measurement range of 0 to 50mm, a resolution of 0.001mm, and a sampling rate of 1kHz, and is used to collect the spatial trajectory deviation data of the end of the servo spraying tool. The sampling time interval is determined according to the variation rate of the geometric features of the injection molded part surface over time and the processing overhead of the central control unit. The central control unit performs spatial coordinate difference calculation on the discrete sampling point set. As the spatiotemporal variation of the curvature feature increases, the sampling time interval is gradually shortened. Under normal working conditions, a sampling time interval of 30ms is used. After the workpiece surface enters the curvature change region, it is shortened to 10ms to reduce the phase lag of the discrete sampling data and control the computational load of the central control unit.

[0037] The experiment included the present invention sample group, a partially missing control group 1, a partially missing control group 2, an out-of-range control group 1, and an out-of-range control group 2. The present invention sample group adopted a complete feedback control adjustment method. The partially missing control group 1 removed the feedback gain adaptive modulation loop based on the absolute value of curvature temporal variation. The partially missing control group 2 removed the decision window and circular queue storage path for the transition from the normal control state to the degraded sleep state. The out-of-range control group 1 lowered the safety boundary from 0.65 to 0.40. The out-of-range control group 2 raised the safety boundary from 0.65 to 0.90. Gaussian white noise with a signal-to-noise ratio of 25dB was superimposed on the optical image signal. The workpiece geometric contour was divided into three working condition intensity levels: low curvature variation, medium curvature variation, and extreme curvature jump. The trajectory tracking deviation of each group during the response process of the multi-axis channel servo control unit was recorded.

[0038] Under low curvature variation conditions, the optical image signal remains stable, and the absolute values ​​of curvature temporal variation calculated by each group converge to 0.23. The central control unit calculates the calibration control gain of 0.87 according to the smooth division convergence relationship between the composite dynamic variable and the basic control gain constant. Based on this, the sample group modulates the trajectory spatial position tracking deviation vector and outputs the current cycle drive control current command, and the measured trajectory tracking deviation is 0.05mm. After the condition is switched to medium curvature variation level and a specular reflection noise energy spike appears, the partial missing control group one does not adjust the feedback gain according to the absolute value of curvature temporal variation. The drive control current command experiences unidirectional inertial accumulation in the continuous sampling period, and the trajectory tracking deviation of the multi-axis channel servo control unit at the spatial curve inflection point increases to 0.46mm. The present invention extracts curvature features by calculating the difference between three-dimensional coordinates in space and performs transient differential calculations along the discrete time axis to obtain an absolute value of curvature temporal variation of 0.52. The central control unit reduces the calibration control gain based on this value, multiplies the trajectory spatial position tracking deviation vector with the calibration control gain, and outputs the current cycle drive control current command to the multi-axis channel servo control unit after decoupling transformation. The drive torque controlled unit compensates for the response delay at the inflection point of the spatial curve, and the measured trajectory tracking deviation is 0.11 mm. Under extreme curvature jump conditions, the absolute value of curvature temporal variation calculated by the present invention is 0.648, close to the safety boundary of 0.65. The composite dynamic variable increases accordingly, and the calibration control gain decreases to 0.32 according to the smooth division convergence relationship, which limits the change amplitude of the current cycle drive control current command, and the measured trajectory tracking deviation is 0.14 mm.

[0039] During the experiment, the multi-axis channel servo control unit operated at a constant nominal rate of 100 mm / s. Therefore, the velocity vector magnitude contained in the composite dynamic variable remained a fixed constant and was incorporated into the time constant adjustment factor during calibration. This ensured that the change in calibration control gain was primarily determined by the absolute value of the curvature temporal rate of change. Comparisons were made among the groups at the same motion rate to avoid the influence of motion rate variations on trajectory tracking deviation measurement results. As the input conditions continued to deteriorate, the absolute value of the curvature temporal rate of change exceeded 0.65, exceeding the range. The control group adopted a safety boundary of 0.40, before any drastic changes occurred in the spatial geometry. The system immediately transitions from normal control state to degraded sleep state, the normal gain modulation circuit is turned off prematurely, the dynamic response of the multi-axis channel servo control unit lags, and the trajectory tracking deviation at the inflection point of the spatial curve increases to 0.38mm; under the influence of strong specular reflection noise, the absolute value of curvature timing variation abnormally jumps to 0.78, the out-of-range control group II uses a safety boundary of 0.90, while the partially missing control group II does not set a conversion judgment at 0.65, both groups continue to perform normal gain modulation, the calibration control gain decreases to 0.12 under this condition, the transmission chain generates high-frequency oscillation, and the trajectory tracking deviation increases to 0.92mm.

[0040] After the sample group of this invention detects that the absolute value of curvature timing variation reaches 0.68 and 0.72 respectively twice, the determination window triggers the degraded state machine control flow, blocks the spatial coordinate difference calculation path, and stops using the current abnormal sampling data to update the calibration control gain. The historical instruction storage module reads the safe drive current data sequence of the past 5 safe sampling cycles from the circular queue, calculates its arithmetic mean, obtains a nominal drive control current instruction of 4.26A, and continuously outputs it. In this control state, the trajectory tracking deviation at the end of the multi-axis channel servo control unit is maintained at 0.183mm. According to the measurement results of each group under different working conditions, when the safety boundary is set to 0.65, the normal control state can cover the low curvature variation and medium curvature variation working conditions, and switches to the degraded sleep state after the absolute value of curvature timing variation continuously crosses the boundary. When the safety boundary is below 0.65, for example, set to 0.40, the normal gain modulation circuit is turned off prematurely, resulting in a lag in servo response. When the safety boundary is above 0.65, for example, set to 0.90, strong specular reflection noise cannot trigger state switching in time, the feedback gain continues to decay, and the trajectory diverges. The historical instruction storage module saves drive control current data for 5 safe sampling periods to balance the average suppression of outlier impulse noise and the phase response of the nominal drive control current instruction. When the number of safe sampling periods is less than 5, the arithmetic mean is easily affected by sudden outlier impulse noise. When the number of safe sampling periods is greater than 5, the phase lag of the nominal drive control current instruction increases, making it difficult to keep up with the current trajectory changes. With a safety boundary of 0.65 and 5 safe sampling periods, the sample group of this invention controls the final trajectory tracking deviation within 0.19mm under specular reflection noise and high gradient inflection point conditions, and the global motion rate of the multi-axis channel servo control unit remains unchanged.

[0041] Example 4: This embodiment combines Figures 1 to 2 This paper describes a vision-guided spraying trajectory optimization and control method for injection molded parts with complex curved surfaces, such as... Figure 1 As shown, step S1 is to obtain the discrete point set and calculate the position tracking deviation vector; step S2 is to calculate the coordinate difference of the point set to construct the composite dynamic variable; step S3 is to calculate the basic gain by division convergence to generate the calibration gain; and step S4 is to perform modulation multiplication and output the command to complete the time delay dynamic compensation.

[0042] like Figure 2As shown, the complex curved injection molded part has an actual spatial physical geometric surface, and the residual area of ​​the release agent introduces optical specular reflection noise. Its surface-related parameters are transmitted to the front-end image sensor for pixel de-identification and desensitization processing, specular reflection noise filtering, and two-dimensional digital image grayscale gradient analysis. At the same time, the laser displacement sensor performs actual three-dimensional point acquisition at the end of the spraying tool. Both the front-end image sensor and the laser displacement sensor are connected to the fieldbus network interface. The fieldbus network interface establishes a data transmission link and trajectory tracking channel and connects to the central control unit. The central control unit integrates a fixed memory, an inner-loop disturbance observation module, a historical instruction storage module, a timeliness reconstruction component, and a high-speed floating-point arithmetic array. The fixed memory contains a theoretical processing trajectory model, the inner-loop disturbance observation module is configured with a first-order sliding mode observation state equation, and the historical instruction storage module adopts a circular queue structure. It retains the latest 5 safety cycles. The timeliness reconstruction component is used to evaluate the error time series and reconstruct the basic control gain range. The high-speed floating-point array includes trajectory spatial position tracking deviation calculation, position deviation and transmission disturbance cascade alignment, spatial geometric curvature time series variation calculation, second-order difference and transient differential, and feedback gain active smoothing convergence scheduling normal control state and degraded sleep state transfer judgment window. The control output of the high-speed floating-point array is connected to the multi-axis channel servo control unit via a digital-to-analog conversion interface. The multi-axis channel servo control unit constructs a multi-axis linkage servo control channel and executes scalar independent axis command reception. The output of the multi-axis channel servo control unit transmits control commands to the torque controlled unit. The torque controlled unit includes motors for each axis and executes the torque output of each axis motor. The power output of the torque controlled unit is connected to the end of the servo spraying tool, which includes the spraying tool and drive mechanism.

[0043] Example 5: In this embodiment, after the spraying control system has been running continuously, the dynamic impedance of the hardware transmission mechanism fluctuates, and high-frequency external electromagnetic radiation will also form random disturbance spikes in the sampling signal. The central control unit receives the data uploaded by the front-end image sensor through the fieldbus network interface and controls the multi-axis channel servo control unit to perform processing path following. During the spatial multi-axis linkage process, the dynamic coupling inertia between each motion axis causes the transmission error generated by the fixed feedback matrix to continuously accumulate in the abrupt change area on the workpiece surface, and the trajectory tracking sensitivity and servo stability decrease accordingly. In severe cases, it will cause high-frequency self-excited oscillation and spraying trajectory deviation. In each discrete sampling period, the central control unit reads the two-dimensional image pixel coordinates collected by the front-end image sensor, performs de-identification and desensitization processing on the image data, and then uses the pre-calibrated homogeneous space geometric transformation matrix to convert the pixel coordinates into three-dimensional physical space coordinates of the workpiece surface, forming the discrete sampling point set of the current sampling period, and uses the converted three-dimensional coordinates as the actual image acquisition point position. The central control unit simultaneously retrieves the theoretical processing trajectory model from the fixed memory to obtain the theoretical trajectory point at the current sampling time. It calculates the coordinate difference between the theoretical trajectory point and the actual image acquisition point in each motion axis direction to obtain the position deviation. The laser displacement sensor synchronously acquires the three-dimensional actual point at the end of the spraying tool to check the spatial trajectory deviation of the current sampling period and calculates the Euclidean distance between the theoretical trajectory point and the three-dimensional actual point as the deviation measurement value.

[0044] To eliminate unmodeled disturbances caused by bearing friction variations within the transmission chain, the inner-loop disturbance observation module reads the armature voltage sequence and speed feedback sequence of the motor at the end of the multi-axis channel servo control unit. These two sequences are then input into a first-order sliding mode observation state equation. This state equation uses the motor speed feedback time sequence as the dynamic state variable and the armature voltage time sequence as the feedforward control excitation term. A first-order continuous state differential relationship is established based on the speed estimation deviation. Simultaneously, a switching gain term with a nonlinear sign function is introduced. A high-speed floating-point arithmetic array performs forward difference iteration on the state differential equation in each discrete sampling period to obtain the estimated residual. The high-frequency equivalent control quantity corresponding to this estimated residual is then filtered by a first-order low-pass filter and used as the external equivalent torque scalar value characterizing the electromechanical transmission disturbance. The central control unit uses a weighted cascade operator to perform cascade alignment operations on the position deviation and the electromechanical transmission disturbance. The external equivalent torque scalar value is multiplied by a scaling factor to convert it into equivalent displacement components in each motion axis direction. These components are then superimposed onto the corresponding axis components of the position deviation to obtain the trajectory spatial position tracking deviation vector for the current sampling period.

[0045] After obtaining the trajectory spatial position tracking deviation vector, the high-speed floating-point arithmetic array performs forward second-order difference calculation on the spatial three-dimensional coordinates within three consecutive discrete sampling periods to extract the curvature feature quantity characterizing the rate of change of the tangent vector direction of the surface. Then, transient differential calculation is performed on the curvature feature quantity along the discrete time axis to obtain the absolute value of the curvature temporal rate of change. The central control unit will The magnitude of the velocity vector is introduced as a multiplier term to construct a composite dynamic variable; when When the value is below or equal to the safety boundary of 0.65, the control loop maintains normal control state, and the central control unit reads the basic control gain constant from the register. The fundamental control gain constant was pre-calibrated in the dimensionless range of 1.25 to 1.50 through offline step response tests; subsequently, according to composite dynamic variables, and time constant adjustment factor The smooth division convergence relationship between them is used to calculate the calibration control gain used to adjust the servo sensitivity. The central control unit invokes the multiplication operator to combine the trajectory spatial position tracking deviation vector with... The components are multiplied and a cross-decoupling matrix is ​​loaded before the output. The cross-decoupling matrix eliminates the dynamic coupling inertia between each motion axis in situ, transforms the trajectory spatial position tracking deviation vector of the multi-axis coupling into independent scalar control commands for each motion axis, and the central control unit converts the scalar control commands of each axis into the current cycle drive control current command, and sends it to the multi-axis channel servo control unit through the digital-to-analog conversion interface. The drive torque controlled unit outputs the motor torque.

[0046] When the specular reflection noise on the workpiece surface or external electromagnetic disturbance causes When the safety boundary of 0.65 is exceeded twice consecutively, the decision window triggers a degraded state machine control flow. The central control unit blocks the spatial coordinate differential calculation path, stops updating the calibration control gain based on the current abnormal sampling data, and reads the safety drive current data sequence of the past 5 safe sampling cycles from the circular queue of the historical instruction storage module. The historical instruction storage module sums the 5 drive control current data using a hardware adder, and then the shift operator performs an arithmetic average calculation to obtain the nominal drive control current command and maintains the output. During the degraded sleep state operation, the decision window continues to monitor. ,when When the value drops below 0.65 for three consecutive discrete sampling cycles, the central control unit resets the degraded state machine control flow, releases the lockout on the spatial coordinate differential calculation path, and restores the gain modulation under normal control. After adopting the above control process, the trajectory tracking deviation at the end of the multi-axis channel servo control unit remains within 0.19mm, the global motion rate remains normal, and the spraying follow-up control loop maintains a stable closed-loop adjustment state.

[0047] Example 6: In this embodiment, the central control unit calls the pre-calibration program before the complex curved surface injection molded part with a brand-new structure is put into spraying to determine the initial parameters of the vision-guided follow-up closed-loop control. During the calibration process, the central control unit controls the multi-axis channel servo control unit to drive the spraying transmission tool to translate along the surface of the standard spline at a speed of 100 mm / s. The front-end image sensor acquires images at a fixed sampling rate of 100 Hz under the condition of no refractive light source, and obtains the actual image acquisition points after calibration. The central control unit retrieves the theoretical trajectory points in the theoretical processing trajectory model, calculates the initial position deviation according to the coordinate difference of each motion axis, and performs cascade alignment calculation with the electromechanical transmission disturbance estimated by the inner loop disturbance observation module, which is manifested as static transmission torque. The resulting trajectory spatial position tracking deviation vector is used as the reference tracking residual.

[0048] The vector magnitude of the baseline tracking residual is denoted as... The unit is mm. The pre-calibration procedure is based on... Determine the time constant adjustment factor The calculation formula is as follows: ,in, The calibration constant is determined by the pre-calibration field program based on the gain convergence state during stable operation of the control loop, and its unit is set according to the requirement of dimensional consistency on both sides of the equation; the value obtained from the above equation... Used for smoothing division convergence calculations in subsequent sampling periods; during continuous spraying control operation, the central control unit acquires optical image signals from the vision sensor in real time and filters out specular reflection noise to form calibrated spatial surface geometric contour parameters of the injection molded part; before performing smoothing division convergence calculations, these parameters serve as feedforward inputs, and are compared with the geometric contour parameters formed by the pre-calibration field procedure and the basic control gain constant. The correspondence between the values ​​determines the current discrete sampling period within a preset range. This allows for real-time correction of the preset range of the basic control gain constant.

[0049] After continuous operation of the spraying control system, the multi-axis actuator will experience mechanical transmission wear and impedance drift. The time-sensitivity reconfiguration component periodically evaluates the stability of the preset range of the basic control gain constant to reduce the impact of aging operating condition characteristic parameter data on gain modulation. After the central control unit has completed trajectory tracking of 500 coated workpieces, it reads the corresponding error time series. When the average curve trajectory tracking residual exceeds the preset loss threshold of 0.15mm, a sliding time window is invoked to remove the oldest operating condition characteristic parameters from 48 hours ago, and the latest 100 sets of optical characteristic parameters collected from the current production batch are used to... The upper and lower limits of the preset range are recalibrated in situ. The recalibrated range is used for subsequent sampling cycles and continues to be fed forward by the geometric contour parameters of the current injection molded part's spatial surface to suppress control divergence caused by mechanical transmission wear. Under this condition, the multi-axis channel servo control unit controls the final trajectory tracking deviation within 0.19mm, and the spraying control adjustment loop maintains a stable closed-loop adjustment state.

Claims

1. A vision-guided spraying trajectory optimization and control method for injection molded parts with complex curved surfaces, characterized in that, Includes the following steps: Step S1: Obtain the discrete sampling point set in the trajectory tracking channel, calculate the positional deviation between the theoretical trajectory point and the actual image acquisition point, and perform a cascade alignment operation with the electromechanical transmission disturbance estimated by the inner loop disturbance observation module to solve the trajectory spatial position tracking deviation vector of the current sampling period. Step S2: Perform spatial coordinate difference calculation on the discrete sampling point set to determine the absolute value of curvature temporal variation rate, which characterizes the degree of transient variation of spatial geometric curvature with time, and introduce the absolute value of curvature temporal variation rate as a multiplier term into the velocity vector within the discrete sampling period to construct a composite dynamic variable. Step S3: When the absolute value of the curvature time-series variation rate is within the preset safety boundary, perform smooth division convergence calculation on the preset basic control gain constant according to the composite dynamic variable, reduce the feedback gain before the end of the current discrete sampling period, and generate the calibration control gain. The calibration control gain and the increase of the composite dynamic variable have a smooth convergence correspondence. Step S4: Multiply the spatial position tracking deviation vector of the modulation trajectory with the calibration control gain, output the decoupled current cycle drive control current command to the multi-axis channel servo control unit, and the drive torque controlled unit outputs the motor torque to complete the in-situ time delay dynamic compensation at the spatial inflection point.

2. The method for optimizing and controlling the spraying trajectory of a vision-guided injection molded part with a complex curved surface according to claim 1, characterized in that, The control loop has a decision window for transitioning from the normal control state to the degraded sleep state, and periodically monitors the transient jump state of the absolute value of curvature timing variation. When the absolute value of curvature timing variation exceeds the safety boundary of 0.65 twice in a row, the degraded state machine control flow is activated and the spatial coordinate difference calculation path in step S2 is blocked. The average value of the safety drive current of the past 5 sampling cycles stored in the historical instruction storage module is directly called as the nominal drive control current instruction to complete the continuous output, so as to block the transmission of high-frequency outlier noise to the gain modulation loop.

3. The method for optimizing and controlling the spraying trajectory of a vision-guided injection molded part with a complex curved surface according to claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11, acquire the actual point signals on the surface of the complex curved injection molded part as the actual image acquisition points, and compare them with the theoretical trajectory points to calculate the position deviation; Step S12, use the inner loop disturbance observation module to estimate the electromechanical transmission disturbance of the multi-axis channel servo control unit, and perform cascade alignment operation on the position deviation and the electromechanical transmission disturbance to calculate the trajectory spatial position tracking deviation vector of the current sampling period.

4. The method for optimizing and controlling the spraying trajectory of a complex curved injection molded part according to claim 1, characterized in that, Step S2 includes the following sub-steps: Step S21, perform second-order difference calculation on the spatial three-dimensional coordinates within the continuous discrete sampling period to extract the curvature feature quantity that characterizes the change of the trajectory geometric curvature; Step S22: Perform transient differential calculation on the curvature feature along the discrete time axis to determine the absolute value of the curvature temporal variation rate, and introduce the absolute value of the curvature temporal variation rate as a multiplier term into the magnitude of the velocity vector.

5. The method for optimizing and controlling the spraying trajectory of a complex curved injection molded part according to claim 1, characterized in that, In a control environment where the discrete sampling point set presents a concave transition state, the sampling time interval for acquiring the discrete sampling point set in step S1 is shortened to improve the trajectory tracking sampling frequency in the geometric variation region.

6. The method for optimizing and controlling the spraying trajectory of a vision-guided injection molded part with a complex curved surface according to claim 1, characterized in that, In step S4, before outputting the current cycle drive control current command, the dynamic coupling inertia between each motion axis in the multi-axis channel servo control unit is eliminated in situ using a cross decoupling matrix, and the trajectory spatial position tracking deviation vector of multi-axis coupling is transformed into an independent scalar control command for each motion axis.

7. The method for optimizing and controlling the spraying trajectory of a vision-guided injection molded part with a complex curved surface according to claim 2, characterized in that, The safety boundary is 0.

65. When the absolute value of the curvature temporal variation rate is less than or equal to 0.65, the trajectory is within the safe measurable boundary, the control loop maintains normal control state operation, and continuously performs gain scheduling on the trajectory spatial position tracking deviation vector by calibrating the control gain.

8. The method for optimizing and controlling the spraying trajectory of a complex curved injection molded part according to claim 2, characterized in that, The historical instruction storage module adopts a circular queue structure. At the end of each discrete sampling period, the queue is updated, the earliest period's drive control current data is automatically removed, and the current period's drive control current data is written to maintain the latest 5 sampling periods' safe drive current data sequence stored in the historical instruction storage module.

9. The method for optimizing and controlling the spraying trajectory of injection molded parts with complex curved surfaces according to claim 1, characterized in that, During continuous spraying control operation, optical image signals from vision sensors are acquired in real time and mirror reflection noise is filtered out. The calibrated spatial surface geometric contour parameters of the injection molded part are introduced into step S3 as feedforward input to correct the preset range of the basic control gain constant in real time.

Images