Machine vision-based steel member surface coating thickness control system

Abstract

This invention belongs to the technical field of industrial coating engineering and automation control, and relates to a machine vision-based coating thickness control system for steel components. It uses a dielectric parameter detection module to perceive the electromagnetic properties of the liquid coating in real time; a micro-dynamics measurement module, combined with impact force sensing and structured light imaging, analyzes the rheological and dynamic behavior of the coating; and integrates the direct measurement results of the coating's three-dimensional morphology from the machine vision module. With the help of a multi-source data fusion module, the above information and spray gun pose data are synchronously encapsulated into a physical state vector. A collaborative decision control module matches the optimal strategy and generates control commands. Finally, a controlled fluid resonance micro-perturbation field is formed in the jet, thereby achieving in-situ, adaptive control of the coating deposition process. This invention solves the problem that traditional spraying processes lack real-time, quantitative, and adaptive control capabilities for the coating deposition process, ensuring coating thickness uniformity and process stability.

Description

Technical Field

[0001] This invention belongs to the technical field of industrial coating engineering and automation control, and relates to a machine vision-based coating thickness control system for steel components. Background Technology

[0002] The service life of steel components in harsh environments such as marine and heavy industrial settings largely depends on the protective performance of their surface coatings. However, traditional spraying processes rely heavily on operator experience and lack real-time, quantitative monitoring methods for the coating deposition process. Especially for complex geometric areas such as the inner corners of H-beams and the recesses of I-beams, traditional methods struggle to achieve uniform and sufficient coating coverage, easily leading to localized areas of excessive or insufficient coating thickness, resulting in paint waste, curing defects, or reduced protective capabilities.

[0003] To improve coating quality, the industry typically employs manual multiple coats, increases the number of spray guns, improves spray gun design, or uses thickness gauges for sampling inspection and rework after completion. For example, for hard-to-cover areas such as inner corners, operators will adjust the angle and movement trajectory of the spray gun to enhance the spraying. For quality control, the industry relies on sampling inspection of the coating using thickness gauges after completion. Although existing technologies have attempted to introduce machine vision technology, such as using laser triangulation or structured light 3D reconstruction for offline inspection of cured coatings, or limited online monitoring during spraying, these methods usually have the following limitations: First, there is a delay in measurement feedback, making it impossible to capture and control the instantaneous dynamics of coating deposition; second, only thickness or morphology information is obtained, lacking real-time perception of the coating's own physicochemical state, such as changes in viscosity and solid content, resulting in weak adaptability of control strategies.

[0004] However, while these methods can partially improve coating uniformity, they are all reactive or qualitative controls, making it difficult to achieve real-time, dynamic process adjustments. Furthermore, traditional processes have poor adaptability to changes in the coating's own state, such as viscosity and water content, making it difficult to adaptively adjust parameters based on the characteristics of different batches of coating, resulting in large fluctuations in coating quality and insufficient stability. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a machine vision-based coating thickness control system for steel components.

[0006] A machine vision-based coating thickness control system for steel components includes:

[0007] The machine vision inspection module adopts the principle of laser triangulation. It projects a laser line onto the surface of the newly deposited liquid coating using a line laser, and uses an industrial camera to acquire an image containing the outline of the laser line from a specific angle. The image processing algorithm calculates the three-dimensional surface morphology and thickness distribution data of the coating in real time.

[0008] The dielectric parameter detection module is used to acquire the real-time dielectric characteristic parameters of the liquid coating.

[0009] The micro-dynamics measurement module calculates coating dynamics information based on real-time dielectric characteristic parameters;

[0010] The multi-source data fusion module timestamps and encapsulates the coating thickness distribution data, coating dynamics information and spray gun position sensor data to generate the coating intrinsic physical state vector.

[0011] The collaborative decision control module inputs the coating intrinsic physical state vector into the collaborative decision-maker, and generates a collaborative control strategy package based on the pre-set engineering strategy library and the coating defect prediction results.

[0012] The piezoelectric vector excitation module drives two sets of miniature piezoelectric transducer arrays set inside the nozzle according to the cooperative control strategy package to generate a vector resonance excitation signal;

[0013] The fluid perturbation control module generates a controlled fluid resonance perturbation field in the ejected liquid coating fluid through a vector resonance excitation signal, thereby achieving in-situ coordinated control of the coating thickness.

[0014] A further aspect of the present invention involves a machine vision inspection module acquiring coating thickness distribution data, comprising the following steps:

[0015] Synchronous trigger line laser and industrial camera to acquire distorted images of laser lines on the surface of liquid coating;

[0016] The image is preprocessed to extract the center pixel coordinates of the laser line;

[0017] Based on the pre-calibrated internal and external parameters of the camera and laser, the pixel coordinates are converted into a three-dimensional point cloud in the workpiece coordinate system;

[0018] By comparing the three-dimensional point cloud with the pre-stored three-dimensional model of the steel component substrate, the normal distance of each point is calculated, thereby obtaining the real-time thickness distribution data of the coating.

[0019] A further aspect of the present invention involves obtaining real-time dielectric characteristic parameters, including the following steps:

[0020] Monitor and collect the real-time shift of the resonant frequency caused by the liquid coating coverage;

[0021] The real-time dielectric characteristic parameters are obtained by converting the real-time offset of the resonant frequency according to the preset physical relationship.

[0022] Among them, the real-time dielectric characteristic parameters include the equivalent dielectric constant and the loss tangent.

[0023] A further aspect of the present invention, obtaining coating dynamics information, includes the following steps:

[0024] Acquire the micro-rheological stress signal generated by the impact of liquid coating on micron-sized metal target points;

[0025] The dynamic information of microscale schlieren deformation after structured light passes through a micro-region of coating around a micro-scale metal target is collected to obtain the velocity field gradient;

[0026] Based on the pre-set rheological constitutive relation, combined with real-time dielectric characteristic parameters, micro-rheological stress signals and velocity field gradients, the coating dynamics information is obtained.

[0027] The coating kinetics information includes the trends in apparent viscosity and thickness variation.

[0028] A further aspect of the present invention generates the intrinsic physical state vector of the coating, comprising the following steps:

[0029] The thickness distribution data of the machine vision inspection module is integrated, and the apparent viscosity and thickness change trend are extracted from the coating dynamics information as core physical state indicators.

[0030] The three-dimensional spatial attitude and movement speed of the spray gun are obtained as a spatial position vector;

[0031] The core physical state indicators and spatial location vectors are time-stamped and structured to generate the coating's intrinsic physical state vector.

[0032] A further aspect of the present invention generates a collaborative control strategy package, comprising the following steps:

[0033] Based on the physical and spatial states in the coating's intrinsic physical state vector, a matching composite control mode is retrieved from the engineering strategy library.

[0034] The composite control mode is encapsulated into a set of instructions that includes multi-dimensional adjustment parameters, generating a cooperative control strategy package.

[0035] A further aspect of the present invention generates a vector resonance excitation signal, comprising the following steps:

[0036] The first set of micro piezoelectric transducer arrays is driven to generate in-plane circular vibration at the nozzle tip, which is used to adjust the tangential momentum when the droplet hits the workpiece.

[0037] The second set of micro piezoelectric transducer arrays is driven to generate axial extension and contraction vibrations in the micro-perturbation structure inside the nozzle, which is used to apply directional shear stress.

[0038] Under the regulation of the collaborative control strategy package, the in-plane circular arc vibration and axial extension vibration are coupled to generate a vector resonance excitation signal.

[0039] A further aspect of the present invention involves forming a controlled fluid resonant perturbation field, comprising the following steps:

[0040] Shear micro-vibration fields are induced in the liquid coating fluid by vector resonance excitation signal;

[0041] By adjusting the internal shear rate gradient of the droplet and the momentum transfer path when it impacts the workpiece through a shear micro-vibration field, the leveling, wetting and adhesion behavior of the coating is interfered with, thus forming a controlled fluid resonant micro-perturbation field.

[0042] A further embodiment of the present invention includes a coating defect prediction module, which, after generating the coating intrinsic physical state vector, predicts the coating defect probability based on the time series analysis of the coating intrinsic physical state vector and a preset defect model, comprising the following steps:

[0043] Match the trajectory of the coating's intrinsic physical state vector in the multidimensional state space with a preset fault trajectory;

[0044] Based on the matching results, identify the fault type and output the probability of coating defects.

[0045] A further embodiment of the present invention includes a strategy library self-optimization module, which, after forming a controlled fluid resonant perturbation field, further includes the following steps:

[0046] The surface morphology data of the coating after application is obtained using a machine vision inspection module;

[0047] Based on the correlation analysis between post-application surface morphology data and historical coating intrinsic physical state vector logs, the engineering strategy library is updated using a reinforcement learning algorithm.

[0048] In summary, the present invention has the following beneficial technical effects:

[0049] 1. By integrating machine vision and electromagnetic sensors at the tip of the spray gun, the newly deposited liquid coating can be monitored in real time and non-contactly. This allows for the acquisition of key physical parameters during coating deposition, and dynamic adjustment decisions for spraying parameters can be made based on this data. Through a closed-loop feedback mechanism, the uniformity and repeatability of coating thickness can be improved, reducing the probability of defects such as uneven thickness and missed coating that are common in traditional processes.

[0050] 2. For areas in steel components, such as the inner corners of H-beams and the recesses of I-beams, where traditional processes are difficult to apply coating evenly, these areas are identified by integrating spatial location information and coating dynamics information. A specially optimized control strategy is then applied to these areas. By regulating the internal flow characteristics of the sprayed droplets and their interaction dynamics with the workpiece surface, the leveling, wetting, and adhesion behavior of the coating in these complex areas is optimized, which helps to improve the reliability of the coating protection.

[0051] 3. By measuring the thickness, dielectric characteristics, and rheological properties of the liquid coating in real time, changes in the coating state can be sensed, including real-time changes in parameters such as viscosity and surface tension caused by factors such as temperature fluctuations, solvent evaporation, and changes in moisture content. After obtaining this information, the spraying parameters can be automatically adjusted, which helps to improve the stability of coating thickness and quality. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 A schematic diagram of the framework in the embodiments of this application is disclosed.

[0054] Figure 2 A flowchart illustrating an embodiment of this application is disclosed. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The following is in conjunction with the appendix Figures 1-2 A preferred description of the present invention is provided below.

[0057] See attached document Figures 1-2 This invention proposes a machine vision-based coating thickness control system for steel components, comprising the following modules:

[0058] The machine vision inspection module, fixed to the moving head of the spray gun, follows immediately after the spraying fan. This module consists of a line laser, a CMOS industrial camera, and an embedded computing unit. During operation, the line laser projects a laser line onto the newly deposited liquid coating surface. Due to the height variations on the coating surface, the laser line captured by the CMOS camera, mounted at a specific angle, will exhibit corresponding distortion. The embedded computing unit uses image processing algorithms to extract the center coordinates of the distorted laser line in real time and, based on pre-calibrated system parameters, calculates the three-dimensional point cloud data of the coating surface using the principle of laser triangulation. This point cloud data is registered and compared with a pre-stored three-dimensional model of the steel component substrate to calculate the real-time thickness distribution information of the coating.

[0059] The dielectric parameter detection module is used to acquire the real-time dielectric characteristic parameters of the liquid coating.

[0060] The micro-dynamics measurement module calculates coating dynamics information based on real-time dielectric characteristic parameters;

[0061] The multi-source data fusion module timestamps and encapsulates the thickness distribution data, coating dynamics information and spray gun position sensor data obtained by the machine vision inspection module to generate the coating intrinsic physical state vector.

[0062] The collaborative decision control module inputs the coating intrinsic physical state vector into the collaborative decision-maker, and generates a collaborative control strategy package based on the output results of the preset engineering strategy library and the optional coating defect prediction module.

[0063] As an optional implementation, the system may also include a coating defect prediction module, which receives the time series of the coating's intrinsic physical state vectors and inputs them into a pre-defined defect model based on a Long Short-Term Memory (LSTM) network. This model analyzes the evolution trajectory of the vector sequence in the multi-dimensional state space and compares it with the evolution patterns of typical defects (such as accumulation and sagging) stored within the model and trained using a large amount of historical data. It then outputs in real time the probability of a specific defect occurring in the current state. This probability value is sent to the collaborative decision control module to assist it in making forward-looking decisions.

[0064] The piezoelectric vector excitation module drives two sets of miniature piezoelectric transducer arrays set inside the nozzle according to the cooperative control strategy package to generate a vector resonance excitation signal;

[0065] The fluid perturbation control module generates a controlled fluid resonance perturbation field in the ejected liquid coating fluid through a vector resonance excitation signal, thereby achieving in-situ coordinated control of the coating thickness.

[0066] In one embodiment of the present invention, obtaining real-time dielectric characteristic parameters includes the following steps:

[0067] Monitor and collect the real-time shift of the resonant frequency caused by the liquid coating coverage;

[0068] The real-time dielectric characteristic parameters are obtained by converting the real-time offset of the resonant frequency according to the preset physical relationship.

[0069] Among them, the real-time dielectric characteristic parameters include the equivalent dielectric constant and the loss tangent.

[0070] Specifically, the open microwave resonant cavity structure is first fixed to the end of the moving head of the spray gun of the automatic spraying equipment. The open microwave resonant cavity structure is configured such that its detection area is located exactly on the path of the newly sprayed liquid coating. When the spray gun moves and operates, the liquid coating that has just been sprayed onto the surface of the steel component will flow over and temporarily cover the detection area of ​​the open microwave resonant cavity structure.

[0071] Liquid coating refers to the uncured paint fluid that has just been sprayed from the spray gun and deposited on the surface of the steel component. Since liquid coating is usually a dielectric material, its entry will change the original electromagnetic field boundary conditions inside the cavity, thereby disturbing the stable distribution of the electromagnetic field. This change will directly cause the inherent resonant frequency of the open microwave resonant cavity structure to change, thus serving as the basis for measuring coating data.

[0072] In one embodiment of the present invention, a monitoring system coupled to the cavity continuously measures the resonant frequency of the cavity and compares it in real time with a reference resonant frequency when the cavity is not covered by the coating, thereby calculating the real-time offset of the resonant frequency. Obtain the real-time offset. Then, by calling the preset physical relationship model and performing reverse solving, the continuous real-time offsets are obtained. The data stream is converted in real time into real-time dielectric characteristic parameters that characterize the current internal polar molecular state and free water content of the liquid coating; the pre-defined physical relationship model reveals the real-time offset. Equivalent dielectric constant of coating material and loss tangent A deterministic mapping relationship between them.

[0073] Finally, the obtained real-time dielectric characteristic parameters include at least the equivalent dielectric constant. and loss tangent These two key indicators together describe the electromagnetic properties of the liquid coating at the microscopic level, and thus reflect the state of the liquid coating.

[0074] Among them, the preset physical relationship is a mathematical model or lookup table that is pre-established and stored in the control system, which will determine the real-time offset of the input resonant frequency. Equivalent dielectric constant of the coating output and loss tangent In connection, this relationship can be established based on electromagnetic parameter calibration experiments on n groups of industrial anti-corrosion coating samples with different ratios, and can be expressed through polynomial fitting or neural network models to cover a wider working range and nonlinear effects; this physical relationship model reveals the real-time offset. Equivalent dielectric constant of coating material and loss tangent A deterministic mapping relationship between them.

[0075] For the resonant cavity perturbation theory, the conversion relationship can be simplified to the following linear relationship:

[0076]

[0077] in, The real-time offset of the resonant frequency, measured in Hz, is obtained by the monitoring system and characterizes the frequency change caused by the liquid coating.

[0078] This represents the reference resonant frequency of the open microwave resonant cavity structure when it is not covered by a liquid coating, expressed in Hz. This value is measured and stored during device initialization as a stable reference; it is also considered for electromagnetic compatibility in industrial spraying environments. The typical value is set at 2.45 GHz, and the engineering design range is typically 2.4 GHz to 2.5 GHz.

[0079] It is a dimensionless geometric fill factor, the value of which depends on the volume percentage of the liquid coating within the detection region of the open microwave resonant cavity structure and the cavity's own geometry. During the equipment calibration phase, the dielectric constant is calculated by measuring a standard liquid with a known dielectric constant. The value is fixed in the system, with a typical setting of 0.05. This value is based on the average value of calibration test results for more than 10 epoxy resin coatings with different viscosities. The calibration results range from 0.03 to 0.08, and the specific value depends on the specific configuration of the cavity and the installation position.

[0080] The equivalent dielectric constant of the coating is a dimensionless parameter that directly reflects the arrangement and responsiveness of polar molecules within the coating. This formula aims to calculate this core real-time dielectric characteristic parameter from measurable frequency shifts.

[0081] It should be noted that the open microwave resonant cavity structure is a special electromagnetic sensor used to generate stable, localized microwave electromagnetic fields, and is highly sensitive to changes in the dielectric properties of materials entering this field. Its data structure includes a reference resonant frequency. and geometric fill factor The static parameter set. Electromagnetic field coupling detection is a non-contact measurement process that obtains physical property information by analyzing the influence of the liquid coating on the electromagnetic field inside the open microwave resonant cavity structure. Real-time offset of the resonant frequency. It is a scalar data stream that varies continuously over time, used to quantify the extent of the liquid coating's influence on the resonant cavity. Real-time dielectric characteristic parameters include at least the equivalent dielectric constant. and loss tangent Two key indicators, together describing the electromagnetic properties of liquid coatings at the microscopic level, thus reflecting their physicochemical state.

[0082] For example, suppose that during a spraying operation, the reference resonant frequency of the open microwave resonant cavity structure integrated into the moving head of the spray gun when it is not covered. The resonant frequency, initially 2.45 GHz, changes to 2.4497 GHz when a 100-micron-thick epoxy zinc-rich primer, acting as a liquid coating, flows through the detection area. At this point, the real-time shift of the resonant frequency is calculated. The frequency is -0.0003 GHz. The system then invokes the preset physical relationship to... It is -0.0003GHz. For 2.45GHz, and the preset Substituting 0.05, the equivalent dielectric constant of the coating is calculated. Approximately 5.92. The value is then output as a component of the real-time dielectric characteristic parameter to characterize the internal polarity distribution state of the current liquid coating.

[0083] In one embodiment of the present invention, an installation method for an open microwave resonant cavity structure is provided. The structure is fixed to the moving head of the spray gun of the spraying equipment by a rigid mounting bracket. The plane of the detection window is perpendicular to the spraying direction and can protrude slightly from the exit plane of the spray gun nozzle by about 5 to 8 mm, so that the sprayed paint fan first passes through the detection window. The detection window is designed as a rectangular slit, and its elongated shape is intended to allow the flowing liquid coating to form a continuous thin film coverage, while maximizing its interaction area with the electromagnetic field inside the cavity.

[0084] In one embodiment of the present invention, obtaining coating dynamics information includes the following steps:

[0085] Acquire the micro-rheological stress signal generated by the impact of liquid coating on micron-sized metal target points;

[0086] The dynamic information of microscale schlieren deformation after structured light passes through a micro-region of coating around a micro-scale metal target is collected to obtain the velocity field gradient;

[0087] Based on the pre-set rheological constitutive relation, combined with real-time dielectric characteristic parameters, micro-rheological stress signals and velocity field gradients, the coating dynamics information is obtained.

[0088] The coating kinetics information includes the trends in apparent viscosity and thickness variation.

[0089] Specifically, the acquired real-time dielectric characteristic parameters are used as a synchronous trigger signal to activate the sensing unit located downstream of the open microwave resonant cavity structure. The core of the sensing unit is a suspended and fixed micron-sized metal target and a structured light projection and imaging system. To ensure that measurable microscopic disturbances are generated in the liquid coating while avoiding excessive size that would interfere with the normal spraying flow field, the micron-sized metal target is an inert metal ball with a diameter of 10μm to 100μm, preferably 50μm, used to generate small, controllable disturbances in the coating fluid and to serve as a mechanical sensor probe.

[0090] When the liquid coating flows through and impacts this micron-sized metal target, a highly sensitive strain gauge mechanically connected to it captures the nanometer-sized physical displacement caused by the flow impact force in real time. This deformation is converted into a continuous voltage signal, which, after amplification and calibration, is interpreted as a micro-rheological stress signal characterizing the viscoelastic properties of the coating. The micro-rheological stress signal is time-series data, which reflects the ability of the liquid coating to resist deformation at the microscale, i.e., viscoelasticity.

[0091] It should be noted that a high-sensitivity strain gauge is a sensor that can convert minute mechanical deformations into changes in resistance. It is used to measure the displacement of micron-sized metal targets, thereby quantifying the impact force of the coating on them.

[0092] At the same time, the structured light system projects a periodically adjustable sinusoidal or rectangular wave grating pattern, for example, with a spatial frequency of 10–50 line pairs / mm, onto the microscopic region of the coating surrounding a micrometer-scale metal target. Due to the uneven flow velocity within the coating, especially the velocity gradient near the target obstacle, minute changes in refractive index occur on and within the coating surface, causing the projected grating pattern to distort. The same optical imaging unit continuously captures this dynamic information of microscale schlieren deformation, which is a series of digital images containing the structured light distortion pattern. These images are used to visualize and record the velocity field distribution within the coating. Then, the degree of light deflection is calculated using schlieren image processing algorithms, such as those based on Fourier transform or phase stepping, thereby retrieving the velocity field gradient characterizing the local rate of change of flow velocity. It is a scalar value used to quantify the relative motion rate between adjacent fluid layers within the coating.

[0093] Finally, the real-time dielectric characteristic parameter ε and the real-time measured microrheological stress signal are used to... and velocity field gradient The three are jointly processed based on a pre-defined rheological model relationship: firstly, using... Apparent viscosity was calculated using an empirical transformation function. Then With velocity field gradient Micro-rheological stress signals Substituting data into the data for consistency verification and fusion, such as a data fusion algorithm based on weighted least squares, for data from... and The theoretical stress calculated based on the constitutive relation, and the measured stress Consistency verification and iterative fitting are performed, and finally the solution is calculated and encapsulated into a solution containing apparent viscosity. With thickness variation trend Coating kinetics information.

[0094] Among them, the pre-set rheological constitutive modeling relationship is a mathematical algorithm stored in the system, which defines the intrinsic relationship between stress, strain rate, and material properties, and is used to fuse multiple sensor signals into physically meaningful rheological parameters; the coating dynamics information is a composite data structure, including apparent viscosity. and thickness variation trend Two time series are used to describe the flow and evolution characteristics of the liquid coating at the current moment.

[0095] Rheological construct modeling relationship:

[0096]

[0097]

[0098] in, It represents micro-rheological stress, with the unit Pa, and is obtained by measuring the impact force on a micrometer-scale metal target and dividing it by the effective area.

[0099] Representing apparent viscosity, measured in Pa·s, it is not a directly measured quantity, but rather derived from real-time dielectric characteristic parameters via an empirical transformation function fitted to experimental data. The transformed function has the following relationship: ,in and This is an empirical constant, obtained by fitting a correlation analysis of numerous rheological and dielectric spectral tests on specific types of coatings, such as polyurethane coatings. It has the dimension of viscosity. It is a dimensionless constant, with typical values ​​as follows: Set to 3.5 Pa·s, The constants are set to -0.5; these two constants are derived from the correlation analysis of rheological and dielectric spectral tests of more than 30 different polyurethane coatings with different ratios at different temperatures.

[0100] Represents the velocity field gradient, in seconds. -1, is a physical quantity that describes the change in velocity between fluid layers, and is calculated by analyzing image information of structured light schlieren deformation.

[0101] It represents the thickness variation trend, with units of m / s, and is used to predict the leveling or deposition rate of the coating under the current rheological state.

[0102] This represents the density of the liquid coating, expressed in kg / m³. This is an inherent property of the coating and is input as a known parameter. The typical value is set to 1500 kg / m³.

[0103] This represents the spray gun's movement speed, measured in m / s, and is provided by the spray gun's own position sensor to ensure the dynamic accuracy of the calculation.

[0104] For example, the real-time dielectric characteristic parameter 5.92 obtained in the above example is used as input; for ease of calculation, the following examples use... 3.5 -0.5 For 1500 and The calculation is performed with a value of 0.5. It is assumed that the impact force on the micron-sized metal target point is measured by a high-sensitivity strain gauge, and the micro-rheological stress signal is obtained after conversion. The average velocity field gradient near the target point was calculated after processing images acquired by the same optical path imaging unit at a pressure of 15.2 Pa. 150s -1 By invoking a pre-defined rheological model relationship, the empirical transformation function is first applied. The apparent viscosity was calculated. The thickness is approximately 0.18 Pa·s. Subsequently, these data were integrated, and combined with the known coating density of 1500 kg / m³ and the spray gun moving speed of 0.5 m / s, to calculate the thickness variation trend: 15.2 - 0.18 × 150 / (1500 × 0.5) ≈ -0.0157 mm / s. Finally, the apparent viscosity was included... The two values, 0.18 Pa·s and a thickness variation trend of -0.0157 mm / s, are encapsulated into coating kinetic information for transmission.

[0105] In another embodiment of the invention, the microdynamics measurement module can also be implemented using a completely non-contact measurement principle. In this alternative, the module is installed on the moving head of the spray gun, downstream of the dielectric parameter detection module, and consists of a high-frequency ultrasonic Doppler probe, a laser scattering velocimeter, and a matching signal processing unit.

[0106] A high-frequency ultrasonic Doppler probe emits ultrasonic pulses at a frequency of 1-10 MHz at a predetermined tilt angle onto the surface of a newly deposited liquid coating and receives the reflected echoes. By measuring the propagation speed and signal attenuation coefficient of the ultrasonic waves, and combining this with real-time dielectric characteristic parameters obtained from the dielectric parameter detection module, including the equivalent dielectric constant and loss tangent, the apparent viscosity of the coating is calculated in real time using a preset acoustic-electric correlation model, such as a sound velocity-viscosity-dielectric constant relationship fitted based on empirical data.

[0107] Simultaneously, a laser scattering velocimeter projects a focused laser beam onto the coating surface in the same measurement area and collects the scattered light. By analyzing the Doppler frequency shift of the scattered light, the microscopic velocity distribution and its spatial gradient (i.e., velocity field gradient) on the coating surface are calculated non-contactly. Finally, the signal processing unit fuses the apparent viscosity and surface velocity gradient obtained through this non-contact method with the coating thickness variation trend data provided by the machine vision inspection module, and substitutes them into a preset rheological constitutive relation for calculation, outputting coating dynamics information containing real-time apparent viscosity and thickness variation trends.

[0108] In one embodiment of the present invention, generating the intrinsic physical state vector of the coating includes the following steps:

[0109] After timestamp alignment, the real-time thickness value of the current spraying position is obtained from the machine vision inspection module;

[0110] The apparent viscosity and thickness variation trends are extracted from coating dynamics information as core physical state indicators.

[0111] The three-dimensional spatial attitude and movement speed of the spray gun are obtained as a spatial position vector;

[0112] The measured thickness value, core physical state index and spatial position vector are timestamped and structured to generate the coating intrinsic physical state vector.

[0113] Specifically, the generated coating dynamics information is placed into a data fusion processing queue, and data is retrieved in real time from a multi-axis position sensor array integrated with the spray gun body.

[0114] Apparent viscosity extracted from coating kinetics information and thickness variation trend These two values ​​are designated as core physical state indicators to describe the inherent rheological properties and behavioral trends of coating materials.

[0115] The data stream from the position sensor is analyzed to extract the three-dimensional spatial coordinates of the spray gun in the steel component coordinate system, the attitude angle describing the nozzle orientation, and the current moving speed. These data are then integrated into a spatial position vector to describe the spatial environment in which the spraying behavior occurs.

[0116] Each core physical state index and each spatial location vector is timestamped. Timestamp alignment is achieved through a synchronization mechanism based on the Network Time Protocol (NTP) or hardware trigger signals. Core physical state indices with the same timestamp are paired with spatial location vectors, and the paired data is encapsulated into a predefined data structure, such as JSON format or a custom binary structure. The encapsulation process is structured; data is filled into specified fields according to a fixed format, ultimately generating a new, composite data unit: the coating intrinsic physical state vector. This vector, as an atomized information packet, records the coating's intrinsic physical state at a specific moment and spatial location. The coating intrinsic physical state vector is a composite data structure used to provide a comprehensive and real-time snapshot of the coating state at a specific spray point. This snapshot simultaneously includes the coating's measured thickness, physical properties, and spatial properties at its location.

[0117] Among them, the core physical state index is derived from apparent viscosity. With thickness variation trend The two-dimensional data pairs are designed to reflect the most critical intrinsic physical properties of the liquid coating. Timestamp alignment is a data synchronization technique that marks the precise acquisition time of data streams from different sources, ensuring that the data associated with them in subsequent processing is a snapshot of the same moment. Structured encapsulation is the process of organizing aligned data according to a preset format, such as a JSON object or a binary structure, to facilitate subsequent parsing and retrieval.

[0118] For example, continuing with the coating kinetics information from the previous example, the core physical state index is extracted: apparent viscosity. The value is 0.18 Pa·s and the thickness variation trend. The velocity is -0.0157 mm / s. At the same timestamp of 15.253 s, the machine vision module measures the coating thickness at the corresponding point to be 11 μm. Assuming that the spatial position vector obtained and parsed from the spray gun position sensor is: three-dimensional spatial coordinates (X=1.25m, Y=0.83m, Z=0.12m), attitude (pitch angle=15°, yaw angle=90°), and moving speed of 0.5 m / s, where the attitude is measured by the inertial measurement unit (IMU) mounted on the spray gun. Timestamp alignment is performed to confirm that the two sets of data belong to the same instant. Then, structured encapsulation is performed to generate the coating intrinsic physical state vector, whose data content can be represented as: {timestamp: 15.253, thickness 115, viscosity: 0.18, thickness trend: -0.0157, coordinates: (1.25, 0.83, 0.12), attitude: (15, 90), speed: 0.5}.

[0119] In one embodiment of the present invention, generating a collaborative control strategy package includes the following steps:

[0120] Based on the physical and spatial states in the coating's intrinsic physical state vector, a matching composite control mode is retrieved from the engineering strategy library.

[0121] The composite control mode is encapsulated into a set of instructions that includes multi-dimensional adjustment parameters, generating a cooperative control strategy package.

[0122] Specifically, the coating's intrinsic physical state vector is transmitted as a data packet to the core computing unit, or collaborative decision-maker, in real time and continuously via the internal bus. The collaborative decision-maker is a dedicated software module or hardware logic unit whose function is to receive and parse the coating's intrinsic physical state vector and automatically make optimal control decisions based on preset rules. Upon receiving the coating's intrinsic physical state vector, the collaborative decision-maker parses it, separating the physical and spatial attribute information contained within. It reads values ​​such as apparent viscosity and thickness change trends and compares them with preset thresholds, such as viscosity thresholds or thickness trend thresholds, to determine the current physical state of the coating, for example, whether it is in a high viscosity or rapid thickness increase trend. Simultaneously, it parses the spatial position vector to determine whether the spraying point is located in complex geometric regions, such as the inner corner of an H-beam.

[0123] Next, the collaborative decision-maker uses the current state as a search keyword to quickly query and match within its internally stored pre-configured engineering strategy library. This library stores numerous optimized control schemes for different operating conditions. Once it finds an entry that matches or is closest to the current state, the collaborative decision-maker locks onto the corresponding composite control mode. A composite control mode is not a single instruction, but rather a data structure containing a series of specific execution parameters, such as pre-configured parameter templates. These templates define the specific values ​​of multi-dimensional adjustment parameters, specifying how to combine and adjust multiple control variables to address specific painting challenges. The multi-dimensional adjustment parameters are the specific control variables that constitute the composite control mode, including but not limited to the waveform, frequency, phase, and amplitude of the drive signal. These parameters collectively define the specific shape of the micro-perturbation field ultimately applied to the coating. For example, the waveform of the high-frequency load should be a sine wave, the phase difference between the two sets of vibrations should be 90°, and the respective vibration intensity weights should be 0.7 and 0.3.

[0124] Finally, the collaborative decision-maker encapsulates these specific parameter values ​​into a standardized set of instructions, forming a collaborative control strategy package that can be directly parsed and executed by downstream execution units. This package contains a formatted data packet containing the specific values ​​of all necessary multidimensional adjustment parameters, used to issue explicit and executable control instructions to the downstream drive system.

[0125] It should be noted that the pre-built engineering strategy library is a structured database or configuration file that stores a large number of state-strategy mapping pairs. Each mapping pair includes a specific coating state description and a corresponding complete composite control mode. The engineering strategy library is usually built based on experimental data and expert experience from n typical steel component spraying scenarios, for example, more than 1,000.

[0126] For example, the collaborative decision-maker receives the aforementioned intrinsic physical state vector of the coating, namely: {timestamp: 15.253, thickness: 115, viscosity: 0.18, thickness trend: -0.0157, coordinates: (1.25, 0.83, 0.12), attitude: (15, 90), velocity: 0.5}; if the physical state is determined after parsing to be: relatively high thickness, medium viscosity, with a thinning trend, and the spatial state is located in the inner corner region of the H-beam. At the same time, the coating defect prediction model reports that the probability of accumulation defects in the current state is 25%. The collaborative decision-maker uses this combination of state and predicted risk as an index to search in the preset engineering strategy library. Assuming that a certain composite control mode with a certain number is matched, it specifies multi-dimensional adjustment parameters, such as a triangular wave high-frequency load waveform, a vibration phase difference of 45°, and an intensity weight combination of 0.6 for group A and 0.4 for group B. The collaborative decision-maker then generates a collaborative control strategy package based on this model, where the intensity weight combination refers to the ratio of the driving voltage amplitude allocated to the micro piezoelectric transducer arrays of group A and group B.

[0127] In one embodiment of the present invention, generating a vector resonance excitation signal includes the following steps:

[0128] The first set of micro piezoelectric transducer arrays is driven to generate in-plane circular vibration at the nozzle tip, which is used to adjust the tangential momentum when the droplet hits the workpiece.

[0129] The second set of micro piezoelectric transducer arrays is driven to generate axial extension and contraction vibrations in the micro-perturbation structure inside the nozzle, which is used to apply directional shear stress.

[0130] Under the regulation of the collaborative control strategy package, the in-plane circular arc vibration and axial extension vibration are coupled to generate a vector resonance excitation signal.

[0131] Specifically, the drive system at the nozzle tip receives the cooperative control strategy package and parses it to obtain a specific set of instructions. In this embodiment, the drive system controls two independent sets of micro piezoelectric transducer arrays. These two arrays are embedded in different positions of the nozzle. The micro piezoelectric transducer array is an array composed of multiple small piezoelectric ceramic elements, used to efficiently convert the input electrical signal into high-frequency mechanical vibration. In a feasible embodiment, each array contains four piezoelectric ceramic sheets with a diameter of 5 mm, arranged in a ring.

[0132] A subset of parameters from the parsed instruction set, such as the specified drive waveform, frequency, and amplitude, are routed to the first set of piezoelectric transducers. Excited by these electrical signals, this set of transducers undergoes precise, periodic expansion and contraction. This deformation is amplified by the mechanical structure and transmitted to the nozzle tip, causing it to produce a small-amplitude, high-frequency in-plane circular vibration. This is a small, periodic circular trajectory movement of the nozzle tip in a plane parallel to the workpiece surface. The amplitude is typically set to 10–50 μm, and the frequency is usually 20–40 kHz to avoid the natural frequency of the spray gun's mechanical structure and achieve effective fluid micro-disturbance. This vibration trajectory is parallel to the surface of the steel component being sprayed, used to adjust the tangential momentum distribution of the droplets ejected from the nozzle at the moment of impact with the workpiece.

[0133] Meanwhile, another set of parameters from the instruction set, such as phase difference and intensity weights, is sent to a second set of piezoelectric transducers. This set of transducers is designed to drive a micro-perturbation structure inside the nozzle, causing it to generate axial extensional vibrations perpendicular to the sprayed surface. The direction of this vibration is consistent with the direction of paint spraying, and its main function is to apply directional shear stress within the paint fluid.

[0134] Ultimately, these two sets of vibrations, which occur synchronously but each has its own characteristics under the precise command and control of the collaborative control strategy package, superimpose and couple with each other in space and time to form a unified composite excitation signal with specific vector characteristics. This signal is the vector resonance excitation signal, which directly acts on the liquid coating that is about to be sprayed. The vector resonance excitation signal is a composite mechanical vibration field, and its vibration direction, amplitude and phase change with time and space.

[0135] Among them, tangential momentum refers to the momentum corresponding to the velocity component parallel to the workpiece surface when the droplet impacts the workpiece; the perturbation structure is a movable part inside the nozzle, such as a thin sheet or piston, which is mechanically coupled with the second set of piezoelectric transducers and its function is to transfer the vibration energy of the transducer to the coating flowing through it.

[0136] Axial extensional vibration is the reciprocating motion of the micro-perturbation structure along the central axis of the nozzle, i.e. the spraying direction. In order to generate sufficient shear stress in the coating fluid without clogging the nozzle, its amplitude is usually set in the range of 5 to 20 μm, and the frequency is consistent with the in-plane vibration, but there is an adjustable phase difference. Directional shear stress is an interlayer fault force generated in the coating fluid along the spraying direction under the action of axial extensional vibration.

[0137] For example, the nozzle drive system parses the collaborative control strategy package and sends a triangular wave waveform and an amplitude command calculated according to a weight of 0.6 for group A to the first piezoelectric transducer array, driving the nozzle tip to generate an in-plane circular arc vibration with an amplitude of 30 μm. Simultaneously, a phase difference of 45° and an amplitude command calculated according to a weight of 0.4 for group B are sent to the second piezoelectric transducer array, driving the micro-perturbation structure inside the nozzle to generate an axial extensional vibration with an amplitude of 8 μm, maintaining a 45° phase difference with the in-plane circular arc vibration. These two controlled vibrations couple at the nozzle exit, jointly acting on the coating fluid to form a specific vector resonance excitation signal corresponding to a matched composite control mode with a specific number.

[0138] In one embodiment of the present invention, forming a controlled fluid resonant perturbation field includes the following steps:

[0139] Shear micro-vibration fields are induced in the liquid coating fluid by vector resonance excitation signal;

[0140] By adjusting the internal shear rate gradient of the droplet and the momentum transfer path when it impacts the workpiece through a shear micro-vibration field, the leveling, wetting and adhesion behavior of the coating is interfered with, thus forming a controlled fluid resonant micro-perturbation field.

[0141] Specifically, the mechanical vibration energy of the generated vector resonance excitation signal is directly and efficiently transmitted through the nozzle structure to the liquid coating fluid that is being ejected from the nozzle at high speed. The energy injection process is not uniform, but rather, based on the complex spatiotemporal characteristics of the vector resonance excitation signal itself, a shear micro-vibration field with a specific spatial energy distribution pattern is induced and formed inside the coating fluid.

[0142] Under the influence of a shear micro-vibration field, the particles within the coating fluid no longer undergo simple laminar or turbulent motion, but rather superimposed with controlled, high-frequency micro-vibrations. By finely controlling the multidimensional adjustment parameters in the vector resonance excitation signal, such as changing the phase difference or intensity weight of two sets of vibrations, the spatial morphology of the shear micro-vibration field can be directly altered, thereby precisely controlling the internal shear rate gradient of the droplets after ejection, before reaching the workpiece, and during their spread on the workpiece surface. Simultaneously, this internal micro-vibration also changes the overall dynamic characteristics of the droplets, enabling the active adjustment of the momentum transport path when the droplets impact the steel component surface. Through this active intervention, the macroscopic leveling process of the coating, the wetting performance of the coating on the workpiece surface, and the adhesion behavior in complex structural regions can be influenced.

[0143] Ultimately, a controlled fluid resonant perturbation field, with its physical state actively optimized and adjusted, is superimposed on the original spray jet. This perturbation field plays a continuous role throughout the coating formation process, thereby achieving in-situ, real-time coordinated control of the final coating thickness and uniformity at the spraying source. The controlled fluid resonant perturbation field is the result of coupling the basic spray flow field with the micro-vibration field induced by the vector resonant excitation signal. Through active, real-time intervention, it optimizes the dynamic behavior of the coating before deposition and curing to achieve the preset thickness and uniformity targets.

[0144] Among them, the shear micro-vibration field is the region formed inside the liquid coating fluid under the action of the vector resonance excitation signal. The fluid particles in this region are subjected to periodic and controlled shear force. The internal shear rate gradient refers to the rate of change of the velocity difference between adjacent micro-layers inside the fluid, which will directly affect the viscosity behavior and fluidity of the fluid.

[0145] The momentum transport path describes how the momentum of a droplet is distributed and transferred in time and space when it impacts a solid surface, which determines the droplet's spreading diameter and rebound behavior.

[0146] It should be noted that leveling, wetting, and creeping are physical processes that describe the ability of a liquid coating to spread, contact, and cover a solid surface. Leveling refers to the coating’s ability to spontaneously form a smooth surface; wetting refers to the tendency of a liquid to spread on a solid surface; and creeping refers to the phenomenon of a liquid flowing upward against gravity along a vertical or inclined surface.

[0147] The controlled fluid resonant micro-perturbation field is the physical effect field ultimately formed on the surface of the steel component by this method. It is the product of the coupling between the basic spraying flow field and the micro-vibration field induced by the vector resonant excitation signal. Its function is to optimize the dynamic behavior of the coating before deposition and curing through active, real-time intervention, so as to achieve the preset thickness and uniformity targets.

[0148] For example, assuming the aforementioned specific vector resonance excitation signal acts on the sprayed epoxy zinc-rich primer coating fluid, due to the 45° phase difference between the axial vibration and the in-plane vibration in this signal, a shear micro-vibration field with a spiral-progressing energy distribution characteristic can usually be induced inside the coating. When droplets carrying this internal disturbance impact the inner corner of the H-shaped steel, their momentum transport path is optimized, reducing direct normal impact and enhancing tangential flow, thus suppressing coating accumulation at the corner to some extent. This active intervention improves the coating's adhesion behavior, allowing the coating to more uniformly cover the entire inner corner area, ultimately forming a controlled fluid resonance micro-perturbation field on the steel component surface. This achieves synergistic control of the coating thickness uniformity in this complex area, reducing common defects such as thick edges or exposed substrate.

[0149] In one embodiment of the present invention, the defect prediction function after generating the coating intrinsic physical state vector is specifically described as: predicting the coating defect probability based on the time series analysis of the coating intrinsic physical state vector and a preset defect model;

[0150] Predicting the probability of coating defects includes the following steps:

[0151] Match the trajectory of the coating's intrinsic physical state vector in the multidimensional state space with a preset fault trajectory;

[0152] Based on the matching results, identify the fault type and output the probability of coating defects.

[0153] Specifically, after generating the coating's intrinsic physical state vectors, the system activates the online defect prediction module. This module operates based on a pre-set defect model, which is a trained classifier capable of identifying pre-failure patterns in the coating, such as a time-series classification model based on a Long Short-Term Memory (LSTM) network or a Support Vector Machine (SVM). Its implementation steps include:

[0154] Before system deployment, a defect model is trained using historical spraying experimental data. The training data includes the coating intrinsic physical state vector in the form of a time series generated during the spraying process, as well as the actual defect labels determined after the coating at the corresponding point is cured, such as normal, accumulation, sagging, etc.

[0155] During real-time spraying, the system maintains a sliding window of length N, storing the latest N intrinsic physical state vectors of the coating to form a short-time sequence, which is then input into the loaded defect model. The defect model analyzes this sequence and outputs the confidence scores of various types of defects that will occur at the current spraying point, i.e., the coating defect probability.

[0156] The specific implementation of predicting coating defect probability is based on matching real-time data trajectories with preset fault modes, and the steps include:

[0157] Key parameters of the coating's intrinsic physical state vector, such as viscosity, thickness trend, and coordinates, are used to construct a multi-dimensional state space. Based on historical fault data, multiple typical preset fault trajectories are plotted within this space, each corresponding to a defect type. Vector sequences within a sliding window are mapped to this state space to form real-time trajectories. A trajectory similarity algorithm, such as Dynamic Time Warping (DTW), is used to calculate the matching degree between the real-time trajectory and each preset fault trajectory. The fault type with the highest matching degree is identified as the most likely fault type, and its matching degree value is used as the coating defect probability output.

[0158] For example: The system inputs 10 consecutive coating intrinsic physical state vectors near the timestamp T=15.253s, including the vector {timestamp: 15.253, thickness: 115, viscosity: 0.18, thickness trend: -0.0157, coordinates: (1.25, 0.83, 0.12), attitude: (15, 90), velocity: 0.5} from the previous example into the defect model. The model analysis shows that this sequence exhibits a pattern of excessive thickness, stable viscosity, and concentrated spatial coordinates, which has a certain degree of consistency with the early precursor of internal corner accumulation defects. Therefore, the output coating defect probability is: internal corner accumulation probability 25%. At the same time, trajectory matching is performed in the state subspace, and the similarity with the internal corner accumulation fault trajectory is calculated to be 0.25, thereby identifying the matched fault type and outputting this probability value.

[0159] In one embodiment of the present invention, after forming the controlled fluid resonant perturbation field, the method further includes the following steps:

[0160] Obtain surface morphology data for post-application of the coating;

[0161] Based on the correlation analysis between post-application surface morphology data and historical coating intrinsic physical state vector logs, the engineering strategy library is updated.

[0162] Specifically, after the spraying operation is completed, the system optimizes the pre-set engineering strategy library through a closed-loop learning mechanism. The implementation steps include:

[0163] After the coating has cured, a high-precision 3D topography scanner, such as a laser profilometer, is used to scan the sprayed area to obtain the post-application surface topography data of the coating, including the 3D coordinates and thickness of each point. Based on spatial and temporal information, each measurement point in the post-application surface topography data is precisely correlated with the corresponding record in the historical coating intrinsic physical state vector log. The actual coating quality of each correlated point pair is analyzed, along with the correlation between it and the implemented collaborative control strategy package, such as thickness deviation and uniformity.

[0164] Based on this analysis, the engineering strategy library will be updated in the following ways:

[0165] For the existing composite control modes in the strategy library, based on their statistical effects in practical applications, optimize their multidimensional adjustment parameters, such as adjusting the weighted average value of vibration intensity weights.

[0166] If a new and valid state-policy mapping is found, it will be added as a new entry to the engineering policy library.

[0167] Using associated data as training samples, reinforcement learning algorithms, such as Q-learning, are employed to fine-tune the policy selection logic of the collaborative decision-maker, making it more inclined to select policies that produce better coating quality.

[0168] For example, after a spraying task, a topography scan showed that the coating thickness in the inner corner region at timestamp T=15.253s was 118μm, slightly higher than the target value. Analysis revealed that the control effect of the mode used at that time—triangular wave, phase difference 45°, weight 0.6:0.4—needed improvement. The system then updated the engineering strategy library: through reinforcement learning, it slightly reduced the Q-value of the default strategy associated with the physical state of the inner corner region and increased the Q-value of another set of historically better-performing alternative strategies. Simultaneously, this state-effect data pair was stored in the sample library for subsequent reinforcement learning training to continuously optimize the decision-making logic.

[0169] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0170] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A machine vision-based coating thickness control system for steel components, characterized in that, include: The machine vision inspection module adopts the principle of laser triangulation. It projects a laser line onto the surface of the newly deposited liquid coating using a line laser, and uses an industrial camera to acquire an image containing the outline of the laser line from a specific angle. The image processing algorithm calculates the three-dimensional surface morphology and thickness distribution data of the coating in real time. The dielectric parameter detection module is used to acquire the real-time dielectric characteristic parameters of the liquid coating. The micro-dynamics measurement module calculates coating dynamics information based on real-time dielectric characteristic parameters; The multi-source data fusion module timestamps and encapsulates the coating thickness distribution data, coating dynamics information and spray gun position sensor data to generate the coating intrinsic physical state vector. The collaborative decision control module inputs the coating intrinsic physical state vector into the collaborative decision-maker, and generates a collaborative control strategy package based on the pre-set engineering strategy library and the coating defect prediction results. The piezoelectric vector excitation module drives two sets of micro piezoelectric transducer arrays installed inside the nozzle according to a collaborative control strategy package. The piezoelectric vector excitation module is used to: drive the first set of micro piezoelectric transducer arrays to generate in-plane circular arc vibration at the nozzle tip, used to adjust the tangential momentum when the droplet impacts the workpiece; drive the second set of micro piezoelectric transducer arrays to generate axial extensional vibration in the micro-perturbation structure inside the nozzle, used to apply directional shear stress; and under the control of the collaborative control strategy package, couple the in-plane circular arc vibration with the axial extensional vibration to generate a vector resonance excitation signal. The fluid perturbation control module generates a controlled fluid resonance perturbation field in the ejected liquid coating fluid through a vector resonance excitation signal, thereby achieving in-situ coordinated control of the coating thickness. Specifically, the fluid perturbation control module is used to: induce the formation of a shear micro-vibration field in the liquid coating fluid through the vector resonance excitation signal; and adjust the internal shear rate gradient of the droplets and the momentum transfer path when impacting the workpiece through the shear micro-vibration field, thereby interfering with the leveling, wetting, and adhesion behavior of the coating and forming a controlled fluid resonance perturbation field.

2. The steel component surface coating thickness control system based on machine vision according to claim 1, wherein the machine vision detection module acquires coating thickness distribution data, includes the following steps: Synchronous trigger line laser and industrial camera to acquire distorted images of laser lines on the surface of liquid coating; The image is preprocessed to extract the center pixel coordinates of the laser line; Based on the pre-calibrated internal and external parameters of the camera and laser, the pixel coordinates are converted into a three-dimensional point cloud in the workpiece coordinate system; By comparing the three-dimensional point cloud with the pre-stored three-dimensional model of the steel component substrate, the normal distance of each point is calculated, thereby obtaining the real-time thickness distribution data of the coating.

3. The machine vision-based steel component surface coating thickness control system according to claim 1, characterized in that, Obtaining real-time dielectric characteristic parameters includes the following steps: Monitor and collect the real-time shift of the resonant frequency caused by the liquid coating coverage; The real-time dielectric characteristic parameters are obtained by converting the real-time offset of the resonant frequency according to the preset physical relationship. Among them, the real-time dielectric characteristic parameters include the equivalent dielectric constant and the loss tangent.

4. The machine vision-based steel component surface coating thickness control system according to claim 1, characterized in that, Obtaining coating kinetics information includes the following steps: Acquire the micro-rheological stress signal generated by the impact of liquid coating on micron-sized metal target points; The dynamic information of microscale schlieren deformation after structured light passes through a micro-region of coating around a micro-scale metal target is collected to obtain the velocity field gradient; Based on the pre-set rheological constitutive relation, combined with real-time dielectric characteristic parameters, micro-rheological stress signals and velocity field gradients, the coating dynamics information is obtained. The coating kinetics information includes the trends in apparent viscosity and thickness variation.

5. The machine vision-based steel component surface coating thickness control system according to claim 1, characterized in that, Generating the intrinsic physical state vector of the coating includes the following steps: The coating thickness distribution data of the machine vision inspection module is integrated, and the apparent viscosity and thickness change trend are extracted from the coating dynamics information as core physical state indicators. The three-dimensional spatial attitude and movement speed of the spray gun are obtained as a spatial position vector; The coating thickness distribution data, core physical state indicators, and spatial location vectors are time-stamped and structured to generate the coating intrinsic physical state vector.

6. The machine vision-based steel component surface coating thickness control system according to claim 1, characterized in that, Generating a collaborative control strategy package includes the following steps: Based on the physical and spatial states in the coating's intrinsic physical state vector, a matching composite control mode is retrieved from the engineering strategy library. The composite control mode is encapsulated into a set of instructions that includes multi-dimensional adjustment parameters, generating a cooperative control strategy package.

7. The machine vision-based steel component surface coating thickness control system according to claim 1, characterized in that, It also includes a coating defect prediction module, which, after generating the coating intrinsic physical state vector, predicts the coating defect probability based on the time series analysis of the coating intrinsic physical state vector and a preset defect model. This includes the following steps: Match the trajectory of the coating's intrinsic physical state vector in the multidimensional state space with a preset fault trajectory; Based on the matching results, identify the fault type and output the probability of coating defects.

8. The machine vision-based steel component surface coating thickness control system according to claim 1, characterized in that, It also includes a strategy library self-optimization module, which, after forming a controlled fluid resonant perturbation field, includes the following steps: The surface morphology data of the coating after application is obtained using a machine vision inspection module; Based on the correlation analysis between post-application surface morphology data and historical coating intrinsic physical state vector logs, the engineering strategy library is updated using a reinforcement learning algorithm.

Images