Tapered wave-absorbing material intelligent coating process control system based on digital twinning

Abstract

The invention discloses a conical wave-absorbing material intelligent coating process control system based on digital twinning, and relates to the technical field of digital twinning and intelligent manufacturing, and the system comprises an intelligent optimization decision-making module which is used for comparing a predicted coating thickness distribution data field with preset target thickness distribution, and calculating a target thickness distribution value based on a comparison deviation; and dynamically adjusting the process parameter set through an optimization algorithm, and generating an intelligently optimized process instruction. According to the invention, a highly integrated flexible coating production system is realized by constructing a complete closed loop composed of a data perception acquisition module, a digital twinning construction module, a coating process simulation module, an intelligent optimization decision module, a real-time control execution module and a quality evaluation and feedback module; real-time operation of a physical production line and simulation optimization of a virtual space are deeply fused, and the intelligent capability and the intelligent level of a production system are improved.

Description

Technical Field

[0001] This invention relates to the fields of digital twin and intelligent manufacturing technology, and in particular to an intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins. Background Technology

[0002] With the rapid development of aerospace, high-end electronics, and other fields towards high performance, lightweight, and customization, stringent requirements have been placed on the manufacturing quality and consistency of complex curved surface components such as conical microwave absorbing parts. The performance of these components is highly dependent on the thickness and uniformity of the surface-coated absorbing material. Traditional coating production methods relying on manual experience or fixed procedures are no longer sufficient to meet the ever-increasing precision requirements and the flexible production needs of multiple varieties and small batches. Therefore, developing a computer-integrated coating production system that can intelligently adapt to individual workpiece differences and process fluctuations, and achieve precise online control, has become a key issue that urgently needs to be addressed in the field of advanced manufacturing.

[0003] In the current development of computer-controlled production systems, especially Flexible Manufacturing Systems (FMS) technology, improving process controllability through digital means has become the mainstream direction. In existing technologies, using 3D scanning to acquire workpiece morphology and combining it with offline programming and simulation software for path planning and parameter pre-tuning has become a common method for handling complex surface coating. Some systems further integrate fieldbus and sensor networks to monitor production status such as robot pose and environmental parameters, providing a data foundation for process control. In recent years, digital twin technology, as an important means of connecting the virtual and physical worlds, has begun to be introduced into the manufacturing field. By establishing virtual mappings of physical entities, it provides new approaches for process analysis and predictive maintenance, constituting the forefront of technological evolution in this field.

[0004] However, existing computer-controlled production systems still face significant technological bottlenecks in achieving high-precision intelligent coating of conical workpieces. First, the "digitalization" of most systems remains at the stage of static modeling and offline simulation. The constructed virtual model lacks real-time, high-fidelity synchronization and interaction with the dynamically changing physical production process, leading to discrepancies between simulation predictions and actual outputs, and failing to truly guide online decision-making. Second, the various modules of the system (sensing, modeling, simulation, and control) are often loosely coupled or operate independently, failing to form a tightly integrated closed-loop optimization control loop. They lack the "intelligent" ability to dynamically and autonomously adjust process parameters based on real-time feedback, appearing rigid when faced with differences in workpiece shape and environmental disturbances. Furthermore, even with the introduction of basic control, existing systems generally lack intelligent strategies for identifying and specifically compensating for localized process anomalies caused by specific geometric features such as conical surfaces and edges, making it difficult to fundamentally solve the problem of uniformity in coating complex curved surfaces. These problems constrain the flexibility, intelligence, and final process efficiency of existing production systems in pursuing high-quality, high-consistency coating manufacturing.

[0005] Therefore, it is essential to invent a digital twin-based intelligent coating process control system for cone-shaped microwave absorbing materials to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a digital twin-based intelligent coating process control system for cone-shaped microwave absorbing materials to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a digital twin-based intelligent coating process control system for cone-shaped microwave absorbing materials, specifically comprising the following modules:

[0008] The data sensing and acquisition module is used to acquire in real time the three-dimensional topography data of the conical workpiece substrate, the pose data of the coating robot, the coating environment status data, and the inherent parameter data of the coating robot.

[0009] The digital twin construction module is used to construct and dynamically correct a virtual coating scene that is synchronously mapped with the physical coating production line based on the acquired three-dimensional topography data and the coating environment state data, and to generate a digital twin of the conical workpiece and a virtual proxy model of the coating robot in the virtual scene.

[0010] The coating process simulation module is used to drive the virtual agent model to simulate coating the digital twin in the virtual coating scenario according to the preset initial process parameter set, and calculate and generate the predicted coating thickness distribution data field.

[0011] The intelligent optimization decision module is used to compare the predicted coating thickness distribution data field with the preset target thickness distribution, and dynamically adjust the process parameter set based on the comparison deviation through an optimization algorithm to generate intelligently optimized process instructions. The intelligent optimization decision module is configured to execute at least one of the following optimization strategies: a first optimization strategy for coordinating the optimization of overall thickness deviation and coating distribution uniformity, and / or a second optimization strategy for generating local compensation parameters for specific geometric features of the conical workpiece.

[0012] The real-time control execution module is used to send the intelligently optimized process instructions to the coating robot in the physical coating production line, and to perform closed-loop motion control based on the pose data of the coating robot to execute intelligent coating operations.

[0013] The quality assessment and feedback module is used to collect the actual coating thickness data after physical coating is completed, and to use this data to calibrate and update the material growth model parameters of the digital twin in order to optimize the accuracy of subsequent simulations.

[0014] The technical effects and advantages of this invention are as follows:

[0015] 1. This invention realizes a highly integrated flexible coating production system by constructing a complete closed loop consisting of a data perception and acquisition module, a digital twin construction module, a coating process simulation module, an intelligent optimization decision-making module, a real-time control execution module, and a quality assessment and feedback module. It deeply integrates the real-time operation of the physical production line with the simulation optimization in the virtual space, thereby improving the intelligent capability and intelligence level of the production system.

[0016] 2. Through the digital twin construction module, this invention creates a digital twin of a conical workpiece that integrates geometric topology and dynamic material growth model, as well as a virtual agent model of a coating robot based on a precise kinematic model. This provides a high-fidelity virtual experimental environment for the coating process, can accurately predict the coating thickness distribution, realize the visualization verification and prior optimization of the process scheme, and significantly reduce the cost of physical trial and error.

[0017] 3. The present invention compares the simulated coating thickness distribution with the target distribution in real time through the intelligent optimization decision module, and drives the optimization algorithm to dynamically adjust the process parameter set, so that the system has autonomous optimization and decision-making capabilities, and can generate the optimal coating instructions for different workpieces and real-time working conditions, effectively ensuring the accuracy and consistency of the overall coating thickness.

[0018] 4. By configuring the first optimization strategy and the second optimization strategy, this invention can not only synergistically optimize the overall thickness deviation and distribution uniformity, but also generate and apply local compensation parameters specifically for the problem of paint accumulation or insufficiency caused by the geometric features such as edges and vertices of conical workpieces. This effectively improves the ability to solve the inherent edge effects and uniformity problems in the coating of complex curved surfaces.

[0019] 5. The present invention uses the real-time control execution module to accurately and in real-time send the intelligent optimization process instructions (including local compensation parameters) generated by the optimization decision module to the physical coating robot, and performs closed-loop motion control based on the pose data of the coating robot, ensuring that the optimization strategy in the virtual space is reliably and stably reproduced on the physical production line, and realizing the connection from optimal decision to high-quality execution.

[0020] 6. Through the quality assessment and feedback module, this invention uses measurement data after the actual coating is completed to reverse-calibrate and update the key parameters of the dynamic material growth model in the digital twin, enabling the system to have continuous learning and self-evolution capabilities. The accuracy of the simulation model continuously improves with production accumulation, forming a continuously improving intelligent closed loop, which effectively enhances the system's process adaptability and reliability in the long term. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall architecture of the present invention.

[0022] Figure 2 This is a flowchart illustrating the construction process of the digital twin according to the present invention.

[0023] Figure 3 This is a flowchart illustrating the simulation step size execution of the coating process according to the present invention.

[0024] Figure 4 This is an iterative flowchart of the intelligent optimization decision-making module of the present invention.

[0025] Figure 5 This is a flowchart of the calibration process for the quality assessment and feedback module of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] This invention provides, for example Figure 1 The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins, as shown, specifically includes the following modules:

[0028] The data sensing and acquisition module is used to acquire in real time the three-dimensional topography data of the conical workpiece substrate, the pose data of the coating robot, the coating environment status data, and the inherent parameter data of the coating robot.

[0029] Furthermore, in the above technical solution, the data sensing and acquisition module specifically includes:

[0030] The laser scanning unit is used to perform high-speed line scanning on the conical workpiece substrate to acquire point cloud data with millimeter-level precision, including the conical surface curvature and edge features, and to reconstruct the three-dimensional topography data.

[0031] The pose sensing unit is integrated into the end effector of the coating robot and is used to collect the joint angles, end position and posture of the robot in real time to form the pose data of the coating robot.

[0032] An environmental monitoring unit is used to collect temperature, humidity, and airflow velocity data within the coating working chamber, which serve as the coating environmental status data.

[0033] The robot parameter configuration unit is used to acquire and store the inherent parameter data of the coating robot, which includes at least the robot's kinematic parameters and the geometric dimensions of key components.

[0034] It's important to know that before the coating operation begins, the system activates the data sensing and acquisition module. This module first loads the coating robot's inherent parameter data from the robot parameter configuration unit and establishes communication connections with the laser scanning unit, pose sensing unit, and environmental monitoring unit. When it receives the "start acquisition" command or detects that the conical workpiece is in place, the module starts each unit according to a preset logical sequence.

[0035] The laser scanning unit preferably employs a linear array scanner based on the principle of laser triangulation. The scanner is fixed at a specific position within the coating working chamber, with its optical plane forming a preset angle (e.g., 30° to 60°) with the axis of rotation of the conical workpiece. During scanning, the coating robot drives the workpiece to rotate at a uniform speed, or the scanner itself moves along the workpiece's axis, thereby achieving a high-speed linear scan covering the entire outer surface of the conical workpiece. The scanner acquires dense point cloud data at a sampling frequency of not less than 1 kHz, and each data point contains at least its three-dimensional coordinates (X, Y, F, Z) in the scanner's coordinate system. s Y s Z s The reflection intensity information I is then processed using a preset coordinate transformation matrix T. s2w Transform point clouds from the scanner coordinate system to a unified physical world coordinate system (X). w Y w Z wThe converted point cloud is denoised (e.g., using a statistical outlier removal algorithm) and smoothed. Then, using Poisson surface reconstruction or Delaunay triangulation algorithm, a three-dimensional surface model of the conical workpiece substrate with millimeter-level accuracy is reconstructed, which is the three-dimensional topography data.

[0036] The pose sensing unit specifically includes high-precision absolute encoders installed on each joint of the robot, as well as a six-dimensional force / torque sensor and visual positioning markers integrated on the end effector. The encoders read the angle values ​​θ of each joint in real time. i (i=1, 2, ..., n). Based on the loaded robot kinematic parameters (such as DH parameters), the system calculates the theoretical pose of the end effector in the robot's base coordinate system B in real time using a forward kinematic model. Simultaneously, a calibrated industrial camera (with known intrinsic parameter matrix K and distortion coefficient D) located above the working cavity captures specific visual markers (such as ArUco codes or high-contrast concentric circle markers) on the end effector. The pose of the end effector in the camera coordinate system is calculated using a PnP (Perspective-n-Point) based visual positioning algorithm. Then, the pose is determined using a known fixed transformation matrix T between the camera and the world coordinate system. cam2w The observed pose of the end effector in the world coordinate system W is obtained by transformation. Hand-eye calibration (i.e., determining T) cam2w The system installation is performed using the Tsai-Lenz or planar calibration method. Kalman filtering or complementary filtering algorithms are then used to fuse the data. and We obtained high-confidence, real-time comprehensive pose data of the coating robot end effector in the world coordinate system. This data includes position (x, y, z) and orientation (rx, ry, rz) (e.g., represented in Euler angles or quaternions). With the angles θ of each joint i Together, they constitute the pose data of the coating robot.

[0037] The environmental monitoring unit includes distributed temperature sensors, humidity sensors, and ultrasonic anemometers. These sensors synchronously collect data from multiple key locations within the coating chamber (such as near the workpiece, near the nozzle outlet, and near the return air vent) at a fixed sampling period (e.g., 100ms). The collected raw data is converted from analog to digital and then transmitted to the core processor of the data sensing and acquisition module. The processor performs mean filtering on multiple readings from the same type of sensors within the same period to obtain a representative temperature value T at the current moment. env Humidity value H env and airflow velocity value V env These three values ​​are encapsulated into an environment state vector E=[Tenv H env V env ]^T, as the coating environment state data.

[0038] The data sensing and acquisition module has a built-in real-time data buffer and a synchronization clock. It adds a unified timestamp t to the real-time acquired 3D topography data (as static or quasi-static data), coating robot pose data (high-frequency updates), coating environment state data (periodic updates), and intrinsic parameter data read from the configuration unit (static). k In each control cycle (e.g., 10ms), the module packages the timestamp-aligned multi-source data into a standardized data frame and outputs it to the digital twin building module via high-speed industrial Ethernet (e.g., EtherCAT or Profinet).

[0039] The digital twin construction module is used to construct and dynamically correct a virtual coating scene that is synchronously mapped with the physical coating production line based on the acquired three-dimensional topography data and the coating environment state data, and to generate a digital twin of the conical workpiece and a virtual proxy model of the coating robot in the virtual scene.

[0040] Furthermore, in the above technical solution, refer to Figure 2 The process of constructing a digital twin using the digital twin construction module includes:

[0041] The three-dimensional topography data is meshed to generate a conical workpiece base mesh model with a geometric topological structure;

[0042] Based on the coating environment status data and the process parameter set, a dynamic influence factor for correcting the material growth rate is calculated.

[0043] A dynamic material growth model is associated with each grid cell of the base grid model; the model takes coating time, paint flow rate, nozzle movement speed and the dynamic influence factor as inputs and the cumulative coating thickness on the cell as output, thereby forming the digital twin.

[0044] The digital twin construction module constructs and maintains the digital twin according to the following specific steps and methods:

[0045] The three-dimensional topography data, namely the reconstructed three-dimensional surface model of the conical workpiece base, has a data structure of a set of triangular facets. The digital twin construction module first performs adaptive mesh re-division on this surface model. Specifically, it uses the leading-edge propulsion method or the Delaunay triangulation algorithm, with a preset mesh size (e.g., side length 2-5 mm) as a constraint, to discretize the continuous curved surface into a mesh model composed of numerous triangular or quadrilateral mesh units. Each mesh unit is a Cell.i It is assigned the following attributes:

[0046] (1) The coordinates of its geometric center in the world coordinate system (x i y i , z i );

[0047] (2) Its unit normal vector ;

[0048] (3) Its area A i ;

[0049] (4) The topological connection relationship pointing to its adjacent grid cells.

[0050] This process generates the tapered workpiece base mesh model M with accurate geometry and topology. base .

[0051] The dynamic influence factor α(t) is a scalar coefficient that evolves over time or process time, used to quantify the combined impact of environmental conditions and process parameters on the coating curing / deposition rate. Its specific calculation model is as follows:

[0052] α(t) = f env (T) env (t), H env (t), V env (t))*f proc (P(t))

[0053] Among them, T env (t), H env (t), V env (t) represents the environmental conditions (temperature, humidity, airflow velocity) of the coating environment at time t.

[0054] P(t) represents the process parameters related to time t, such as the viscosity of the coating (which can be indirectly reflected by the flow rate parameter).

[0055] f env This is an environmental impact function. In one embodiment, this function can be expressed as:

[0056] f env =[1+k T *(T) env -T ref )]*[1-k H *(H env -H ref )]*exp(-k V *V env ),

[0057] Among them, T ref Href For reference environmental conditions, k T k H k V To obtain the material specificity coefficients obtained through experimental calibration, representing the weights of the effects of temperature, humidity, and airflow on the growth rate, the calibration method is as follows: In a dedicated calibration chamber, with other process parameters (flow rate Q, velocity v) fixed, the temperature, humidity, and airflow velocity are varied individually, and the changes in coating thickness on the flat plate are measured. Taking temperature as an example: at humidity H... ref airflow velocity V ref Below, set temperature gradients T1, T2, T... n A spraying experiment was conducted and the thickness h at each point was measured. i The straight line h is fitted using the least squares method. i =a*T i +b, then kT=a / (a*T) ref +b). Similarly, k can be calibrated. H k V The calibration experiment must use the same paint batch as the production batch.

[0058] f proc The process influence function can be simplified to a linear relationship that is positively or inversely proportional to the flow rate, depending on the coating characteristics, or obtained by querying a preset process-influence relationship table.

[0059] The value α(t) is recalculated at each simulation step or environmental data update cycle. A value greater than 1 indicates accelerated growth, while a value less than 1 indicates slowed growth.

[0060] For the base mesh model M base Each grid cell in i Instantiate an independent dynamic material growth model. The core of this model is a differential or difference equation describing the accumulation of coating thickness. In one specific embodiment, the following discrete-time model is used:

[0061] When the unit is within the paint spraying influence zone during the time interval Δt, its coating thickness increment Δh i (t) is calculated as follows:

[0062] Δh i (t)=η*[Q(t) / (v(t)*L)]*cos(θ i (t))*α(t)*Δt,

[0063] Wherein, η is the coating deposition efficiency coefficient, and its specific value is obtained through the following experimental calibration procedure: under standard environmental and process conditions (temperature T) ref Humidity H ref(Airflow is still), using the same paint and nozzles as the production line, for a known area A (unit: m). 2 The flat plate specimen was subjected to a fixed time t. spray Spraying for 10 seconds (e.g.), while recording the constant paint flow rate Q (unit: m³) during the spraying process. 3 / s). After spraying, wait for the coating to fully cure, and then use a precision balance to measure the weight gain Δm (unit: kg) of the specimen, combined with the coating density ρ (unit: kg / m³). 3 Calculate the actual sediment volume V deposit =Δm / ρ. Calculate the total volume V of paint ejected from the nozzle. spray =Q*t spray Then η = V deposit / V spray To ensure calibration accuracy, the experiment was repeated 5 times and the average value was taken. For the aforementioned absorbing material coating process, the typical value of η ranges from 0.3 to 0.7.

[0064] Q(t) represents the volumetric flow rate of the coating at time t, expressed in cubic meters per second (m³ / s). 3 / s), this parameter is one of the core control variables of the coating process, derived from the set of process parameters or real-time control commands. In simulation, it is determined by the process commands driving the virtual agent model; in the physical system, it is precisely executed by the fluid control unit (such as a metering pump) of the coating robot according to the received process commands.

[0065] v(t) represents the moving speed of the nozzle relative to the workpiece surface at time t, measured in meters per second (m / s). This parameter is a result of coating path planning and robot motion control, and its value is calculated in real time by the motion controller of the virtual agent model or the physical robot based on process instructions (such as path speed curves). When calculating the growth of the mesh cells, the average speed within the simulation step size Δt is taken.

[0066] L is the width of the spray pattern, a physical quantity describing the width of the spray pattern, measured in meters (m). It is an inherent property of a specific nozzle model under rated air pressure and flow rate, and can be obtained directly from the technical manual provided by the nozzle manufacturer, or determined by measuring the width of the wet or dry film formed on a flat plate at a standard distance through a simple spray test. In simulations, L is treated as a known process constant that does not change over time.

[0067] θ i (t) represents the distance between the nozzle's central axis and the normal vector of the mesh element at time t. The angle between θ and y is expressed in radians (rad), and the cosine value of this angle is cos(θ). i(t) is used to correct the "cosine law" effect in spraying, which is the ratio of the amount of paint received per unit area during tilted spraying to that received during normal spraying. Within each simulation step, based on the instantaneous pose (position and orientation) of the nozzle geometry model and the mesh cell in the virtual agent model... i The center coordinates and normal vector are calculated using the dot product formula for spatial vectors.

[0068] α(t) is the dynamic influence factor calculated at time t.

[0069] Δt is the simulation step size, which is the discrete time interval on the virtual time axis during the coating process simulation by the digital twin, measured in seconds (s). Its value requires a trade-off between simulation accuracy and computational efficiency: a step size that is too large will result in a coarse simulation, while a step size that is too small will increase unnecessary computational load. Typically, Δt should be set much smaller than the characteristic time of the coating process (e.g., the time required for the nozzle to sweep across a single grid cell). In a specific embodiment, Δt can be dynamically determined based on the coating robot's maximum speed and grid size, ensuring that within one step, the nozzle's movement distance does not exceed the size of one grid cell. Its typical value range is from 1 millisecond to 100 milliseconds. The value of Δt can be dynamically determined using the following formula: Δt = min(0.1 * L grid / v max ,0.1), where L grid v represents the average side length of the grid cell (in meters). max This represents the maximum nozzle movement speed (in meters per second). This formula ensures that the nozzle movement distance within one step does not exceed 10% of the grid size, balancing accuracy and efficiency.

[0070] The total coating thickness of this unit at time t is: h i (t) = ΣΔh i (t) is used to accumulate all increments from the start of coating to time t.

[0071] The digital twin building block is the entire mesh model M base Create a thickness attribute matrix H, where each element H[i] corresponds to a cell. i Current thickness h i . The mesh model M base The thickness attribute matrix H and the aforementioned growth calculation rules associated with each grid cell are collectively encapsulated into a digital twin of the conical workpiece. This twin is a computable, dynamic object whose state can evolve over time.

[0072] Before the coating operation begins, the digital twin construction module completes the aforementioned meshing, model association, and initialization of the thickness matrix H (all hi are set to zero) based on the received initial 3D topography data. During the coating simulation or actual coating process, the module continuously receives real-time environmental status data from the data sensing and acquisition module and updates α(t) accordingly. Simultaneously, based on the process parameters (such as real-time flow rate and speed) fed back by the coating process simulation module or the actual control execution module, it drives the growth model of each mesh unit to update the thickness prediction or synchronously map the actual growth, thereby achieving dynamic correction of the digital twin state and maintaining a high degree of synchronization with the coating growth process of the physical workpiece.

[0073] Furthermore, in the above technical solution, the virtual agent model is constructed through the following process:

[0074] The inherent parameter data of the coating robot are obtained from the robot parameter configuration unit;

[0075] Based on the geometric dimension data in the inherent parameter data, a three-dimensional geometric appearance model of the virtual agent model is generated in the virtual coating scene; and based on the kinematic parameters in the inherent parameter data, a forward kinematics model, an inverse kinematics solver, joint limits, and velocity / acceleration constraints consistent with the physical robot are established and associated with the three-dimensional geometric appearance model.

[0076] Based on the real-time acquired pose data of the coating robot, the mapping relationship between the robot's base coordinate system and the physical world coordinate system in the virtual coating scene is determined and calibrated.

[0077] A control command interface for the virtual agent model is established to receive process commands and parse them into joint motion trajectories.

[0078] It should be noted that, furthermore, in the above technical solution, the construction of the virtual proxy model is achieved through the following specific steps and methods:

[0079] The inherent parameter data stored in the robot parameter configuration unit exists in the form of a structured configuration file (such as JSON, XML, or YAML format). When the virtual agent model building program starts, it first reads and parses this file. The kinematic parameters include at least the standard DH (Denavit-Hartenberg) parameter table used to describe the robot's link structure, which defines four parameters for each joint i: link torsion angle α. i Length a of the connecting rod i Linkage offset d i and joint angle θ iThe geometric dimensions of the key components include at least the key external dimensions (such as the dimensions of the bounding box and flange) of the robot's base, links, wrist, and end effector, as well as parameters of the simplified geometric elements (such as cylinders and cuboids) used to define the collider.

[0080] Based on the obtained geometric dimension data, generate the 3D geometric appearance model of the virtual proxy model using any of the following methods:

[0081] Method 1 (CAD Model Import): Call the computer-aided design (CAD) 3D model file (such as STEP or IGES format) that is completely consistent with the physical robot, and accurately load and position it into the virtual coating scene according to the scaling ratio and installation offset defined in the geometric dimension data.

[0082] Method 2 (Procedural Generation): Based on the simplified geometric element parameters defined in the geometric dimension data, the corresponding three-dimensional geometric shapes (such as cylinders and cuboids) are generated in real time in a virtual scene rendering engine (such as Unity3D, Unreal Engine, or a dedicated robot simulation software kernel), and assembled according to the robot topology to form a simplified but accurate three-dimensional appearance model of the robot.

[0083] Forward kinematics model: Based on the DH parameter table, a forward kinematics transformation function is established from joint space to maneuver space. For a robot with n joints, the pose of its end effector in base coordinate system B is... By continuously multiplying by the transformation matrix of adjacent links The model is obtained by using the given joint angles θ. i Calculate the end-effector pose.

[0084] Inverse kinematics solver: Establish an inverse kinematics solver consistent with the physical robot controller algorithm. For a given end-effector pose... The solver calculates one or more feasible joint angle solutions [θ1, θ2, ..., θ] using numerical iteration methods (such as the Newton-Raphson method) or analytical methods (for robots with specific configurations). n The solver incorporates selection logic consistent with that of the physical robot, such as selecting the solution closest to the current pose or avoiding singularities.

[0085] Joint limit and velocity / acceleration constraints: Read the range of motion of each joint [θ] from the intrinsic parameter data. i，min θ i，max Maximum speed ω i，max and maximum acceleration a i，maxIn the virtual proxy model, these constraints are bound to each joint object. During motion simulation or trajectory planning, limit checks are performed on all calculated joint angles, and constraints are applied to joint angular velocities and angular accelerations to ensure that the motion of the virtual proxy model perfectly matches the physical capabilities of the physical robot.

[0086] After initializing the virtual coating scene, based on at least one set of coating robot pose data (physical) acquired in real time and its corresponding pose (virtual) calculated in the virtual scene using a forward kinematics model, the rigid body transformation relationship between the two coordinate systems is calculated. Specifically, the following method is used:

[0087] Let W be the coordinate system of the physical world, V be the coordinate system of the virtual scene world, and B be the robot's base coordinate system in the physical world. w In the virtual scene, it is B. v .

[0088] Collect N (N≥3, non-collinear) precise poses of the physical robot end effector under W. W，i Simultaneously, the corresponding pose calculated by the virtual agent model end in V using forward kinematics is recorded. V，i .

[0089] Using the "three-point method" or an absolute orientation algorithm based on singular value decomposition (SVD), the optimal rigid body transformation matrix T from V to W (or vice versa) can be solved. V→W This transformation matrix establishes a spatial mapping relationship between the virtual scene and the physical world, ensuring spatial consistency between the digital twin and the physical workpiece, and between the motion of the virtual agent model and the motion of the physical robot.

[0090] The control command interface is implemented as a software function or service. It receives process commands from the intelligent optimization decision module. In one embodiment, the process command is defined as a structure containing: a sequence of coating path points (each point includes position, orientation, and velocity), a coating flow rate curve, and possible local compensation parameters. The interface's workflow is as follows:

[0091] Path interpolation: For the received discrete path point sequence, linear or spline interpolation is performed according to a preset interpolation period (e.g., consistent with the simulation step size Δt) to generate a continuous end-point pose trajectory Pose. V，des (t).

[0092] Inverse kinematics solution: For each interpolated end-target pose, the inverse kinematics solver is invoked to calculate the corresponding joint angle sequence Φ(t) = [θ1(t), θ2(t), ..., θ n (t)].

[0093] Trajectory Generation and Output: The joint angle sequence is smoothed (considering velocity / acceleration constraints) to generate joint space trajectory commands that can directly drive the motion of each joint of the virtual agent model. These commands are encapsulated into specific data packets and sent to the simulation engine of the virtual coating scene via inter-process communication or network sockets to drive the motion of the virtual agent model. Simultaneously, this interface also has the ability to convert the same process commands into command formats (such as UDP packets and ROS topic messages) recognizable by the physical robot controller, providing a foundation for actual control.

[0094] The coating process simulation module is used to drive the virtual agent model to simulate coating the digital twin in the virtual coating scenario according to the preset initial process parameter set, and calculate and generate the predicted coating thickness distribution data field.

[0095] Furthermore, in the above technical solution, refer to Figure 3 The coating process simulation module simulates the coating process, which includes:

[0096] On the virtual timeline, the virtual agent model is driven through the control command interface of the virtual agent model according to the coating flow rate curve, nozzle movement path and speed curve specified by the initial process parameter set.

[0097] Within each simulation step, based on the instantaneous pose of the three-dimensional geometric appearance model of the virtual agent model, its spatial relationship with each mesh unit on the digital twin is calculated to determine the paint spraying influence area.

[0098] The dynamic material growth model corresponding to the affected mesh cell is invoked. Based on the coating flow rate parameter of the current step and the dynamic influence factor corresponding to the current step, the coating growth within the current step is calculated and added to the coating thickness attribute of the cell.

[0099] After traversing the entire virtual coating process, the final coating thickness attributes of all grid cells are summarized to generate the predicted coating thickness distribution data field.

[0100] It should be noted that the specific process, algorithm, and data flow of the coating process simulation module are as follows:

[0101] After the coating process simulation module is started, it first loads the preset initial process parameter set from the system. This parameter set is stored in the form of a data structure and includes at least: an ordered list of path points (each point contains position [x, y, z], orientation [rx, ry, rz], and a preset velocity v through that point), and a coating flow rate curve Q(s) associated with time or path length. The module simultaneously obtains the current state of the digital twin (i.e., the current thickness attribute of all mesh cells, which is usually zero initially) and the virtual proxy model.

[0102] The module establishes an independent virtual timeline, whose time variable t vir Starting from 0, the simulation progresses step by step according to discrete time steps Δt (as defined above). Each simulation step constitutes a main loop, with the specific steps as follows:

[0103] Step S1: Drive the virtual agent model:

[0104] Based on the current virtual time t vir Based on the path and velocity curves in the process parameter set, the desired nozzle pose at the current moment is calculated using linear or spline interpolation. des (t) vir The desired pose is then compared with the current flow Q(t). vir The control commands are encapsulated and sent to the control command interface of the virtual proxy model. This interface calculates the joint trajectories and updates the angles of each joint in the virtual proxy model, thereby updating the instantaneous pose of its 3D geometric appearance model (especially the nozzle tip) in the virtual scene through its forward kinematics model. noz (t) vir ).

[0105] Step S2: Determine the area affected by paint spraying:

[0106] The goal of this step is to efficiently identify all mesh cells in the digital twin mesh model that may be sprayed at the current step size. i In practice, a space acceleration structure is used to improve efficiency.

[0107] Constructing the query range: using Pose noz (t) vir Using the center of the nozzle orifice as the origin, a "spray cone" or "spray cube" is constructed along the nozzle axis as the query range. The size of this range is determined by the spray fan width L and a preset effective spray distance D. eff Decide.

[0108] Fast Retrieval: Utilizing pre-built spatial indexes (such as octrees, KD-trees, or bounding box hierarchies) in the digital twin mesh model, quickly retrieve all mesh cells intersecting the query range, forming an initial set of potential influential cells C. pot .

[0109] Precise determination: For each potentially affected cell i ∈C pot To make an accurate judgment, the following conditions must be met simultaneously for a cell to be included in the set of affected grid cells C at the current step size. aff :

[0110] (1) Distance condition: the distance from the center of the unit to the nozzle orifice. i Less than the effective spraying distance D eff ;

[0111] (2) Angle condition: element normal vector Vector of nozzle axis The included angle θ i Less than the preset maximum effective spraying angle θ max (Typically 60° to 90°).

[0112] Among them, the effective spraying distance D eff and maximum effective spray angle θ max The effective spraying radius r was determined through nozzle spraying characteristic tests: When spraying vertically onto a flat plate at a standard distance d0, the coating thickness distribution curve along the radial direction was measured. The radial distance where the thickness drops to 80% of the center thickness was defined as the effective spraying radius r. eff Then D eff =d0+r eff / tan(θ) max / 2). θ max The following method was determined through inclined spraying experiments: With a fixed distance, the angle β between the nozzle and the normal to the plate was varied, and the coating thickness was measured as a function of β. The angle at which the thickness decreased to 80% of the thickness achieved by vertical spraying was defined as θ. max For typical air spraying, D eff The usual value is 100~300mm, θ max The angle is usually 60° to 90°.

[0113] Step S3: Calculate and sum the coating growth:

[0114] Traverse the set of affected mesh cells C at the current step size aff Each cell in i :

[0115] Obtaining Input Parameters: Obtain all inputs required for the dynamic material growth model from the current simulation state: current coating flow rate Q(t) vir), Nozzle moving speed v (t) vir (This can be calculated through the difference between adjacent poses), included angle θ i Current step size dynamic influence factor α (t) vir ) and simulation step size Δt.

[0116] Model calculation: Substitute the above parameters into the cell that is already associated. i Dynamic material growth models (e.g., formula Δh) i =η*[Q / (v*L)]*cos(θ i ) * α * Δt), calculate the coating thickness increment Δh of the element within the current step size Δt. i (t) vir ).

[0117] Cumulative thickness attribute: This increment Δh i (t) vir ) and Cell i Stored current thickness attribute h i Add, i.e., h i =h i +Δh i (t) vir ), and complete the update of the unit thickness.

[0118] Step S4: Update the time and loop.

[0119] Advance the virtual time by one step: t vir =t vir +Δt. Return to step S1 until virtual time t. vir The total coating time specified in the process parameter set is reached or exceeded, or all path points are traversed.

[0120] After the entire virtual coating process simulation is completed, the coating process simulation module traverses all mesh cells of the digital twin and reads its final thickness attribute h. i The module associates these thickness data with the geometric coordinates of the corresponding grid cells to generate a structured predicted coating thickness distribution data field.

[0121] One specific implementation of this data field is a data array that corresponds one-to-one with the vertices or faces of the mesh model, or a two-dimensional matrix (e.g., a thickness mapping after unfolding a cone into a parametric plane). The data field is output in a standardized format (such as CSV, VTK, or a custom binary format) for the intelligent optimization decision module to read and compare. Essentially, this data field is a complete, quantitative digital prediction of the coating results under the stated initial set of process parameters.

[0122] The intelligent optimization decision module is used to compare the predicted coating thickness distribution data field with the preset target thickness distribution, and dynamically adjust the process parameter set based on the comparison deviation through an optimization algorithm to generate intelligently optimized process instructions. The intelligent optimization decision module is configured to execute at least one of the following optimization strategies: a first optimization strategy for coordinating the optimization of overall thickness deviation and coating distribution uniformity, and / or a second optimization strategy for generating local compensation parameters for specific geometric features of the conical workpiece.

[0123] Furthermore, in the above technical solution, refer to Figure 4 The working process of the intelligent optimization decision-making module includes:

[0124] The predicted coating thickness distribution data field is converted into a two-dimensional thickness matrix, and element-wise difference is performed between the matrixed target thickness distribution and the matrix to obtain the thickness deviation matrix.

[0125] With the goal of reducing overall thickness deviation, a parameter optimization model was constructed using paint flow rate, nozzle moving speed, and path overlap rate as variables to be optimized.

[0126] The parameter optimization model is solved using a particle swarm optimization algorithm or a gradient descent algorithm. The process parameter set is iteratively adjusted until the thickness deviation obtained from the simulation is less than a preset threshold or the upper limit of the number of iterations is reached. The corresponding process parameter set at this time is then output as the intelligently optimized process instruction.

[0127] It should be noted that the intelligent optimization decision-making module operates according to the following specific process:

[0128] The intelligent optimization decision module receives a predicted coating thickness distribution data field from the coating process simulation module. This data field is essentially a set of thickness values ​​associated with digital twin mesh cells. The module first maps this data field to a regular two-dimensional parameterized planar matrix M. pred Above, each element M of the matrix pred (i, j) represents the predicted thickness at a specific parameter coordinate on the surface of the conical workpiece. Simultaneously, the module reads a preset target thickness distribution from the process database, which is also represented by a two-dimensional matrix M with the same dimensions and parameterization rules. target The data is stored in a formal format. Subsequently, the module calculates the thickness deviation matrix E by subtracting the two matrices element-wise, i.e., E(i,j) = M. pred (i,j)-M target (i, j).

[0129] The module establishes a parameter optimization model with the optimization objective of "reducing overall thickness deviation". This model is defined as follows:

[0130] Decision variables (variables to be optimized): Key process parameters that significantly affect coating thickness are selected as decision variables, forming a decision vector X. In one embodiment, X may include: a global coating flow rate baseline value Q. base Nozzle movement speed reference value v base and the overlap rate r between adjacent rows of the coating path overlap .

[0131] Objective Function: Construct a quantifiable objective function F(X) to evaluate the thickness deviation obtained after simulation of the process parameter set determined by the decision variable X. A typical F(X) is the average of the absolute values ​​of all elements in the thickness deviation matrix E (i.e., mean absolute deviation), or the mean of the sum of squares of all elements (mean square error). The smaller the value of F(X), the closer the simulation result is to the target thickness distribution.

[0132] Constraints: Set physically feasible ranges of values ​​for decision variable X (upper and lower bound constraints), such as flow rate not exceeding pump capacity, speed not exceeding robot maximum speed, overlap rate between 0 and 1, etc. These constraints must be met during the optimization process.

[0133] The module uses an iterative optimization algorithm to automatically solve the above model in order to find the optimal decision variable X that minimizes the objective function F(X). opt The process is as follows:

[0134] If the particle swarm optimization algorithm is used, a group of "particles" (i.e., a set of potential solutions X) is randomly generated within the range of values ​​of the decision variables, and each particle has its own position and velocity. If the gradient descent algorithm is used, the initial set of process parameters is used as the search starting point X0, and a learning rate parameter is set.

[0135] In each iteration:

[0136] (1) Simulation evaluation: For each candidate solution X in the current iteration k (The position of a particle in the particle swarm, or the current point in gradient descent), the intelligent optimization decision module reconstructs it into a complete set of process parameters and sends it to the coating process simulation module.

[0137] (2) Obtaining simulation results: Based on the set of process parameters, the coating process simulation module drives the virtual agent model to perform a complete simulation coating of the digital twin and returns a new predicted thickness distribution data field.

[0138] (3) Calculate fitness: The intelligent optimization decision module calculates the candidate solution X based on the new predicted thickness distribution, i.e., the thickness deviation matrix E. k The corresponding objective function value F(X) kIn particle swarm optimization, this value is called the particle's "fitness"; in gradient descent, it is the function value of the current point.

[0139] Solution position update:

[0140] Particle swarm optimization (PSO): Based on the historical best position of each particle and the historical best position of the entire particle swarm, the velocity and position of all particles are updated according to a preset velocity and position update formula, generating a new generation of candidate solution groups.

[0141] Gradient descent algorithm: Estimating the objective function F(X) at the current point X. k Find the nearest gradient (the direction of steepest change), then update the solution position along the opposite direction of the gradient (i.e., the direction in which the function value decreases) to obtain the next candidate point X. k+1 .

[0142] After each iteration, the module checks whether any of the following termination conditions are met:

[0143] (1) Performance meets the target: The objective function value F(X) corresponding to the currently obtained optimal solution. best The deviation is less than the preset deviation threshold (e.g., average thickness deviation is less than 10 micrometers).

[0144] (2) Iteration limit: The number of iterations performed has reached the preset upper limit of the number of iterations (e.g., 1000 times).

[0145] If any termination condition is met, the iteration stops; otherwise, return to the "Iteration Evaluation" step to continue the next iteration.

[0146] When the optimization iteration terminates, the module will return the currently found optimal decision variable X. opt This is converted into a complete, executable set of process instructions, which includes at least optimized coating flow parameters, nozzle movement speed curves, and path planning parameters. This set of instructions, namely the intelligently optimized process instructions, will be sent to the real-time control execution module for physical coating.

[0147] Furthermore, in the above technical solution, the first optimization strategy is implemented in the following way:

[0148] After each simulation iteration, the coating thickness variance or the consistency coefficient of the thickness distribution on the entire conical workpiece surface is calculated based on the predicted coating thickness distribution data field.

[0149] The calculated uniformity evaluation index is compared with the preset expected range;

[0150] If the uniformity evaluation index fails to meet the standard, when the optimization algorithm adjusts the set of process parameters, an additional optimization term aimed at improving coating uniformity is introduced to guide the optimization process to reduce overall thickness deviation while synergistically improving the uniformity of thickness distribution.

[0151] It's important to understand that the module first obtains the current predicted coating thickness distribution data field from the coating process simulation module. This data field contains the predicted thickness values ​​of all mesh cells on the digital twin of the conical workpiece. Subsequently, the module calculates a uniformity evaluation index for the coating distribution based on this set of thickness values, such as calculating the variance σ of all thickness values. 2 To measure the degree of dispersion, or to calculate the difference R between the maximum and minimum thickness to reflect the thickness range.

[0152] The module will calculate the uniformity evaluation index, such as variance σ. 2 And the thickness range R, with respect to a preset expected range (e.g., upper limit of variance threshold). and the upper limit threshold of the range R th The comparison is performed. If the indicator fails to meet the standard, i.e., σ... 2 > Or R>R th If so, it is determined that the uniformity of the current coating needs to be improved.

[0153] At this point, the module will dynamically adjust the optimization process. Specifically, when the optimization algorithm (such as particle swarm optimization or gradient descent) adjusts the process parameter set, the module will dynamically adjust the original main optimization function F, which aims to reduce the overall thickness deviation. bias Based on (X) (e.g., mean absolute deviation), an additional sub-optimization function F is introduced to improve coating uniformity. uniform (X) (e.g., F) uniform (X) = σ 2 By assigning weight coefficients w to these two functions. bias and w uniform (satisfies w) bias +w uniform =1), the module constructs a new comprehensive objective function F. total (X) = w bias *F bias (X) + w uniform *F uniform (X) is used to guide subsequent optimization iterations. The weighting coefficients can be dynamically fine-tuned according to the severity of the non-compliance with uniformity standards, guiding the optimization process to improve the uniformity of the thickness distribution while simultaneously reducing overall thickness deviation. Once the optimization iterations bring the uniformity index into the desired range, this collaborative optimization mechanism can maintain or adjust the focus, thereby ensuring that the final intelligently optimized process instructions achieve a uniform coating distribution while meeting the overall thickness requirements.

[0154] Furthermore, in the above technical solution, the second optimization strategy is implemented in the following way:

[0155] Based on the geometric topology of the digital twin, all edge mesh units and vertex mesh units on the conical workpiece are automatically identified;

[0156] During the simulated coating process in the coating process simulation module, the flow rate of the paint flowing to these edge and vertex units is virtually marked and tracked;

[0157] Analyzing the simulation results, we identified the edges and vertices where paint tends to accumulate or become insufficient due to the geometric shape.

[0158] When generating the intelligently optimized process instructions, a local compensation parameter set is independently generated for the identified specific edges and vertex regions. The local compensation parameter set includes at least the following: when coating the region, the temporary dwell time adjustment of the robot nozzle, the instantaneous flow rate adjustment coefficient, and the small offset of the path.

[0159] When the robot moves to the corresponding edge or vertex area during the coating process, the real-time control execution module dynamically loads and applies the corresponding local compensation parameter set to correct the coating unevenness caused by geometric effects.

[0160] First, feature cells are automatically identified based on the geometric topology of the digital twin. The module traverses all mesh cells of the conical workpiece base mesh model and determines their relationship with adjacent cells: if a mesh cell has exactly two adjacent cells, it is marked as an edge mesh cell; if a mesh cell has exactly one adjacent cell, it is marked as a vertex mesh cell. All marked cells are recorded in a feature cell set.

[0161] Next, during the simulated coating process in the coating simulation module, the paint flowing to these feature units is tracked. In each simulation step, when calculating the coating growth of the affected mesh unit, the system checks whether the current unit belongs to the set of feature units. If it does, after the calculation of that step, the virtual paint increment received by that unit within that step is accumulated and recorded, along with the process state during spraying (such as the nozzle's position, orientation, and velocity relative to the unit). By traversing the entire simulation process, the total amount of paint received by each feature unit during the entire coating process (i.e., the final predicted thickness) and its corresponding historical process data can be obtained.

[0162] Then, the simulation results are analyzed to identify problem areas. After the simulation, the intelligent optimization decision module reads the final predicted thickness of all feature units. The average predicted thickness of the entire conical workpiece surface is calculated, and then the thickness of each edge or vertex unit is compared with this average. A relative deviation threshold (e.g., ±15%) is set. If the thickness deviation of a feature unit exceeds this threshold, the area where that unit is located is determined to be a problem area where the coating is prone to buildup (positive deviation) or insufficient (negative deviation) due to the geometry.

[0163] Subsequently, for each identified problem area, a local compensation parameter set is generated independently. The compensation logic is determined based on the direction of the average thickness deviation in that area.

[0164] For areas with excessive paint buildup (larger thickness), the generated compensation parameters are designed to reduce paint deposition in these areas, for example, by setting the instantaneous flow rate fine-tuning factor to 0.8 to 0.95 (i.e., reducing the flow rate).

[0165] For areas with insufficient coating (too thin), the generated compensation parameters are designed to increase coating deposition in those areas, for example, by setting the instantaneous flow rate fine-tuning factor to 1.05 to 1.2 (i.e., increasing the flow rate).

[0166] These compensation parameters are associated with specific robot program position points or workpiece geometric coordinates and are sent to the real-time control execution module as part of the intelligently optimized process instructions.

[0167] Finally, compensation is applied during physical execution. The real-time control module monitors the robot's real-time pose or program position while driving the coating robot to move according to the optimized main process instructions. When the robot is detected entering the boundary of a registered problem area, the module dynamically loads the corresponding local compensation parameter set to fine-tune the main instructions in real time. For example, when reaching specific edge coordinates, the flow rate is temporarily multiplied by a pre-stored fine-tuning coefficient, or a small path offset is added during linear interpolation. When the robot leaves the compensated area, the control parameters automatically revert to the main instruction parameters, thereby achieving intelligent and precise correction for geometric features.

[0168] To verify the synergistic optimization effect of the first and second optimization strategies, in a typical embodiment of this system, the following three strategy combinations were used for simulation optimization on the same conical workpiece, and the key performance indicators were recorded as shown in the table below:

[0169] Optimize strategy combination Simulation iterations Overall thickness deviation (μm) Thickness variance (μm²) Maximum deviation (μm) in edge / vertex region Only the first optimization strategy (overall optimization) 85 12.5 6.8 35.2 Only the second optimization strategy (local compensation) 78 18.3 12.4 8.7 First + Second Strategy Collaborative Optimization 92 9.1 4.2 6.3

[0170] Table Explanation:

[0171] Overall thickness deviation: the average absolute deviation between the predicted coating thickness distribution and the target thickness distribution;

[0172] Thickness variance: an index for evaluating the uniformity of coating distribution; the smaller the value, the more uniform the distribution.

[0173] Maximum deviation in edge / vertex region: The maximum thickness deviation in the edge and vertex regions of a tapered workpiece, used to evaluate the effect of local compensation;

[0174] Simulation iteration count: The number of simulation iterations required to reach the optimization termination condition.

[0175] Conclusion Analysis:

[0176] Using only the first optimization strategy: Although it can control the overall thickness deviation well, there is still a large local deviation in the edge / vertex area due to geometric effects.

[0177] Using only the second optimization strategy significantly improves coating uniformity in edge / vertex regions, but slightly reduces overall thickness deviation and uniformity.

[0178] The combined use of the first and second optimization strategies yielded the best overall results in terms of overall thickness deviation, distribution uniformity, and coating consistency in local geometric feature areas, demonstrating the advantages of the intelligent optimization decision module in multi-objective collaborative optimization.

[0179] The real-time control execution module is used to send the intelligently optimized process instructions to the coating robot in the physical coating production line, and to perform closed-loop motion control based on the pose data of the coating robot to execute intelligent coating operations.

[0180] It is important to understand that the real-time control execution module first receives the intelligently optimized process instructions from the intelligent optimization decision module via a high-speed communication link. This instruction is a structured data packet, containing at least a spatial path point sequence, a paint flow curve, and a set of local compensation parameters generated by the second optimization strategy. The real-time control execution module then parses this instruction, separating it into two parts: a robot motion trajectory instruction and a paint spraying process instruction.

[0181] In terms of motion control, the real-time control execution module performs real-time motion planning based on the parsed path point sequence, generates a continuous end-effector trajectory that considers the constraints of physical robot dynamics, and issues low-level motion commands at fixed control cycles through the robot controller's dedicated drive interface. In terms of process control, the process controller embedded in the real-time control execution module is strictly synchronized with the motion trajectory, queries the flow curve in real time based on the trajectory progress, and controls the paint supply unit to achieve precise flow tracking.

[0182] One of the core functions of the real-time control execution module is to dynamically perform local compensation. Based on the real-time acquired pose data of the coating robot, this module determines whether the robot's end effector has entered a preset local compensation area. Once entered, the real-time control execution module immediately loads the corresponding local compensation parameter set from memory and instantly overwrites the main command parameters, such as adjusting the instantaneous flow rate, fine-tuning the path, or inserting a brief pause. When the robot leaves the area, the control parameters automatically recover.

[0183] Meanwhile, the real-time control execution module implements high-frequency closed-loop motion control. Within each control cycle, it compares the real-time pose data with the commanded target pose, calculates the tracking error, and generates a correction value using a PID or model predictive control algorithm. This correction value is then added to the issued motion command to compensate for the trajectory tracking error in real time.

[0184] In addition, the real-time control execution module is also responsible for the system's safety monitoring, continuously monitoring the robot's status, collision avoidance distance, coating pressure, and communication status. Once an abnormal signal is detected, the real-time control execution module immediately triggers a safety response, stops issuing commands, and reports the system status.

[0185] The quality assessment and feedback module is used to collect the actual coating thickness data after physical coating is completed, and to use this data to calibrate and update the material growth model parameters of the digital twin in order to optimize the accuracy of subsequent simulations.

[0186] Furthermore, in the above technical solution, refer to Figure 5 The working process of the quality assessment and feedback module includes:

[0187] After physical coating is completed, an offline thickness measurement device is used to sample and measure the conical workpiece to obtain the actual coating thickness data at multiple spatial locations.

[0188] The actual coating thickness data is compared with the final simulated predicted thickness of the digital twin at the corresponding location point, and the model prediction error is calculated.

[0189] Based on the prediction error of the model, the key parameters in the dynamic material growth model are back-calibrated using the backpropagation algorithm or the least squares method to reduce the deviation between the predicted coating thickness distribution data field and the actual situation in subsequent simulations.

[0190] It is important to understand that the quality assessment and feedback module first controls or guides the operator to use a high-precision offline thickness measurement device (such as an ultrasonic thickness gauge or a laser confocal microscope) to perform sampling measurements on the cooled conical workpiece. Measurement points are selected on the workpiece surface according to a preset sampling strategy, typically covering key feature areas such as the conical surface, edges, and vertices, and the precise three-dimensional position coordinates of each measurement point in the world coordinate system are recorded. The multiple sets of "position-thickness" data pairs obtained from the measurements constitute the actual coating thickness data.

[0191] Subsequently, the quality assessment and feedback module accesses the digital twin construction module to obtain the state of the digital twin synchronized with the end of the physical coating process, namely the final simulated predicted thickness of each grid cell on its surface. Based on the three-dimensional coordinates of the actual measurement points, the module uses a spatial coordinate mapping algorithm (such as nearest neighbor search or barycentric coordinate interpolation) to find the corresponding predicted position on the grid model of the digital twin and reads the simulated predicted thickness value at that location. For each measurement point, the difference between its actual thickness and the simulated predicted thickness is calculated to obtain the model prediction error at that point. By summarizing the errors of all sampling points, an overall evaluation index, such as mean absolute error (MAE) or root mean square error (RMSE), can be calculated to quantify the overall prediction accuracy of the current simulation model.

[0192] Next, the module initiates the parameter calibration process. Its core task is to use the collected input-output data pairs (i.e., the process and environmental inputs during the coating process, and the corresponding actual thickness output) to perform reverse optimization of key parameters in the dynamic material growth model. Specifically, the module extracts the input parameters used in the simulation at each measurement point location during the coating process from the system log, including but not limited to: the coating flow rate Q, nozzle movement speed v, and nozzle angle θ experienced at that location. i The input parameters, including the dynamic influence factor α(t) and the action time, together with the actual measured thickness, constitute a calibration sample.

[0193] The module is calibrated using a constrained nonlinear least squares method, with the objective function being:

[0194] minΣ i (h) pred，i (η, k) T k H k V )-h meas，i ) 2 ,

[0195] Constraints: 0.2 ≤ η ≤ 0.8, 0 ≤ k T ≤0.1, 0≤k H ≤0.05, 0≤k V ≤0.2,

[0196] Where h pred，i Let be the predicted thickness of the model at the i-th measurement point, and let be the parameters η and k. T k H k V The function; h meas，i This is the actual measured thickness at the i-th measurement point. The optimization uses the Levenberg-Marquardt algorithm, with the initial value set as follows:

[0197] η=0.5, k T =0.01, k H =0.01, k V =0.05,

[0198] This process can be achieved using nonlinear least squares optimization algorithms such as the Gauss-Newton method or the Levenberg-Marquardt method.

[0199] After the calibration calculations are completed, the quality assessment and feedback module will obtain a set of updated, more accurate model parameter values. The module will then use a secure write operation to update these new calibrated parameter values ​​to the corresponding dynamic material growth model parameter storage area in the digital twin construction module, replacing the original parameters. At this point, the core physical model upon which the digital twin relies has been corrected based on the actual production data.

[0200] After this calibration update, when the system performs coating process simulation and optimization for the same type of workpiece again, the digital twin building module will use the calibrated new parameters to drive the model calculation. This allows the next predicted coating thickness distribution data field to more accurately reflect the real behavior of the physical world, thereby achieving iterative improvement in simulation accuracy and forming a continuous improvement closed loop of "physical production - data acquisition - model calibration - simulation optimization".

[0201] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital twin-based intelligent coating process control system for cone-shaped microwave absorbing materials, characterized in that, Specifically, it includes the following modules: The data sensing and acquisition module is used to acquire in real time the three-dimensional topography data of the conical workpiece substrate, the pose data of the coating robot, the coating environment status data, and the inherent parameter data of the coating robot. The digital twin construction module is used to construct and dynamically correct a virtual coating scene that is synchronously mapped with the physical coating production line based on the acquired three-dimensional topography data and the coating environment state data, and to generate a digital twin of the conical workpiece and a virtual proxy model of the coating robot in the virtual scene. The coating process simulation module is used to drive the virtual proxy model to simulate coating the digital twin in the virtual coating scenario according to the preset initial process parameter set, and calculate and generate the predicted coating thickness distribution data field. The intelligent optimization decision module is used to compare the predicted coating thickness distribution data field with the preset target thickness distribution, and dynamically adjust the process parameter set based on the comparison deviation through an optimization algorithm to generate intelligently optimized process instructions. The intelligent optimization decision module is configured to execute at least one of the following optimization strategies: a first optimization strategy for coordinating the optimization of overall thickness deviation and coating distribution uniformity, and / or a second optimization strategy for generating local compensation parameters for specific geometric features of the conical workpiece. The real-time control execution module is used to send the intelligently optimized process instructions to the coating robot in the physical coating production line, and to perform closed-loop motion control based on the pose data of the coating robot to execute intelligent coating operations. The quality assessment and feedback module is used to collect the actual coating thickness data after physical coating is completed, and to use this data to calibrate and update the material growth model parameters of the digital twin in order to optimize the accuracy of subsequent simulations.

2. The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins according to claim 1, characterized in that, The data sensing and acquisition module specifically includes: The laser scanning unit is used to perform high-speed line scanning on the conical workpiece substrate to acquire point cloud data with millimeter-level precision, including the conical surface curvature and edge features, and to reconstruct the three-dimensional topography data. The pose sensing unit is integrated into the end effector of the coating robot and is used to collect the joint angles, end position and posture of the robot in real time to form the pose data of the coating robot. An environmental monitoring unit is used to collect temperature, humidity, and airflow velocity data within the coating working chamber, which serve as the coating environmental status data. The robot parameter configuration unit is used to acquire and store the inherent parameter data of the coating robot, which includes at least the robot's kinematic parameters and the geometric dimensions of key components.

3. The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins according to claim 1, characterized in that, The process of constructing a digital twin using the digital twin construction module includes: The three-dimensional topography data is meshed to generate a conical workpiece base mesh model with a geometric topological structure; Based on the coating environment status data and the process parameter set, a dynamic influence factor for correcting the material growth rate is calculated. A dynamic material growth model is associated with each grid cell of the base grid model; the model takes coating time, paint flow rate, nozzle movement speed and the dynamic influence factor as inputs and the cumulative coating thickness on the cell as output, thereby forming the digital twin.

4. The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins according to claim 1, characterized in that, The virtual agent model is constructed through the following process: The inherent parameter data of the coating robot are obtained from the robot parameter configuration unit; Based on the geometric dimensions in the inherent parameter data, a three-dimensional geometric appearance model of the virtual agent model is generated in the virtual coating scene; and based on the kinematic parameters in the inherent parameter data, a forward kinematics model, an inverse kinematics solver, joint limits, and velocity / acceleration constraints consistent with the physical robot are established and associated with the three-dimensional geometric appearance model. Based on the real-time acquired pose data of the coating robot, the mapping relationship between the robot's base coordinate system and the physical world coordinate system in the virtual coating scene is determined and calibrated. A control command interface for the virtual agent model is established to receive process commands and parse them into joint motion trajectories.

5. The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins according to claim 1, characterized in that, The coating process simulation module simulates the coating process, including: On the virtual timeline, the virtual agent model is driven through the control command interface of the virtual agent model according to the coating flow rate curve, nozzle movement path and speed curve specified by the initial process parameter set. Within each simulation step, based on the instantaneous pose of the three-dimensional geometric appearance model of the virtual agent model, its spatial relationship with each mesh unit on the digital twin is calculated to determine the paint spraying influence area. The dynamic material growth model corresponding to the affected mesh cell is invoked. Based on the coating flow rate parameter of the current step and the dynamic influence factor corresponding to the current step, the coating growth within the current step is calculated and added to the coating thickness attribute of the cell. After traversing the entire virtual coating process, the final coating thickness attributes of all grid cells are summarized to generate the predicted coating thickness distribution data field.

6. The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins according to claim 1, characterized in that, The working process of the intelligent optimization decision-making module includes: The predicted coating thickness distribution data field is converted into a two-dimensional thickness matrix, and element-wise difference is performed between the matrixed target thickness distribution and the matrix to obtain the thickness deviation matrix. With the goal of reducing overall thickness deviation, a parameter optimization model was constructed using paint flow rate, nozzle moving speed, and path overlap rate as variables to be optimized. The parameter optimization model is solved using a particle swarm optimization algorithm or a gradient descent algorithm. The process parameter set is iteratively adjusted until the thickness deviation obtained from the simulation is less than a preset threshold or the upper limit of the number of iterations is reached. The corresponding process parameter set at this time is then output as the intelligently optimized process instruction.

7. The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins according to claim 1, characterized in that, The first optimization strategy is implemented in the following way: After each simulation iteration, the coating thickness variance or the consistency coefficient of the thickness distribution on the entire conical workpiece surface is calculated based on the predicted coating thickness distribution data field. The calculated uniformity evaluation index is compared with the preset expected range; If the uniformity evaluation index fails to meet the standard, when the optimization algorithm adjusts the set of process parameters, an additional optimization term aimed at improving coating uniformity is introduced to guide the optimization process to reduce overall thickness deviation while synergistically improving the uniformity of thickness distribution.

8. The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins according to claim 1, characterized in that, The second optimization strategy is implemented in the following way: Based on the geometric topology of the digital twin, all edge mesh units and vertex mesh units on the conical workpiece are automatically identified; During the simulated coating process in the coating process simulation module, the flow rate of the paint flowing to these edge and vertex units is virtually marked and tracked; Analyzing the simulation results, we identified the edges and vertices where paint tends to accumulate or become insufficient due to the geometric shape. When generating the intelligently optimized process instructions, a local compensation parameter set is independently generated for the identified specific edges and vertex regions. The local compensation parameter set includes at least the following: when coating the region, the temporary dwell time adjustment of the robot nozzle, the instantaneous flow rate adjustment coefficient, and the small offset of the path. When the robot moves to the corresponding edge or vertex area during the coating process, the real-time control execution module dynamically loads and applies the corresponding local compensation parameter set to correct the coating unevenness caused by geometric effects.

9. The intelligent coating process control system for cone-shaped microwave absorbing materials based on digital twins according to claim 1, characterized in that, The working process of the quality assessment and feedback module includes: After physical coating is completed, an offline thickness measurement device is used to sample and measure the conical workpiece to obtain the actual coating thickness data at multiple spatial locations. The actual coating thickness data is compared with the final simulated predicted thickness of the digital twin at the corresponding location point, and the model prediction error is calculated. Based on the prediction error of the model, the key parameters in the dynamic material growth model are back-calibrated using the backpropagation algorithm or the least squares method to reduce the deviation between the predicted coating thickness distribution data field and the actual situation in subsequent simulations.

Images