A plastic spraying link quality detection feedback method and system

Abstract

The application discloses a plastic spraying link quality detection feedback method and system, and relates to the field of plastic spraying quality control systems.The scheme points comprise a nozzle management module, the nozzle management module allocates a unique electronic code to each nozzle, monitors the parameters of each nozzle in real time, and calibrates the position of the nozzle regularly; a scheme library module, the scheme library module stores a plurality of plastic spraying master schemes, each master scheme corresponds to a specific type of product, the master scheme contains a plurality of sub-schemes of the product under different working conditions, and the scheme library module establishes an association rule between the master scheme and the sub-scheme; the application fuses the working environment changes through a multi-source sensor, and combines a machine learning model to predict the best process parameter combination; the electronic code is used to realize independent control of a single nozzle, and a position calibration algorithm is used to ensure uniform coverage of complex curved surface workpieces; a three-dimensional evaluation network covering the overall appearance, local microstructure and performance indicators is used to realize quality control from the macroscopic to the microscopic.

Description

Technical Field

[0001] This invention relates to the field of powder coating quality control system technology, and more specifically, to a method and system for quality inspection and feedback in the powder coating process. Background Technology

[0002] In the industrial manufacturing sector, powder coating (electrostatic powder spraying), as a mainstream surface treatment technology, has long faced the following pain points:

[0003] The production process suffers from several problems: unstable quality, reliance on manual parameter adjustments based on experience, and significant susceptibility to environmental temperature and humidity, as well as operator techniques, leading to uneven coating thickness, color differences, and insufficient adhesion. Inefficiency is also a concern, as single-nozzle operation struggles to adapt to complex workpiece shapes, and the lack of intelligent scheduling in multi-nozzle collaboration results in resource waste. Furthermore, feedback is delayed, with traditional testing relying on offline sampling, failing to monitor the production process in real time, resulting in high defect rates and costly rework. Data silos exist, with equipment operation data scattered across different control systems, lacking a unified analysis platform to support process optimization. Finally, poor adaptability is hampered by fixed process parameters that struggle to cope with seasonal climate fluctuations (such as increased coating fluidity due to summer heat) and batch-to-batch variations in raw materials. These issues severely restrict both production efficiency improvement and cost control.

[0004] Although some automated powder coating equipment has appeared on the market, it generally suffers from the following defects:

[0005] The level of intelligence is low, with most systems only implementing basic motion control functions and failing to integrate environmental perception and adaptive decision-making algorithms; the detection dimensions are limited, with visual inspection focusing mainly on appearance defect identification and lacking quantitative evaluation of key indicators such as micro-morphology and mechanical properties; closed-loop optimization is lacking, with detection data not being effectively fed back to the production end to form closed-loop control, and process improvement relying on manual trial and error. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for quality inspection and feedback in the powder coating process.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A quality inspection and feedback system for the powder coating process includes:

[0009] The nozzle management module assigns a unique electronic code to each nozzle, monitors the parameters of each nozzle in real time, and periodically calibrates the position of the nozzles.

[0010] The solution library module stores multiple powder coating master solutions, each corresponding to a specific product. The master solution includes multiple sub-solutions for that product under different working conditions. The solution library module establishes association rules between master solutions and sub-solutions, enabling the system to automatically match the most suitable sub-solution based on the current environmental conditions.

[0011] The intelligent decision-making module is based on machine learning algorithms to build a prediction model and trains the optimal combination of powder coating parameters under different environments based on historical data.

[0012] The quality inspection module performs a comprehensive assessment and local evaluation of the powder coating quality of the entire workpiece surface.

[0013] The data analysis module collects and organizes production data;

[0014] The feedback adjustment module is equipped with an instant feedback mechanism. When the quality inspection module discovers a quality problem or the data analysis module identifies a potential risk point, it immediately issues a warning signal to the operator and provides specific improvement suggestions.

[0015] Preferably, the nozzle management module includes:

[0016] The coding and identification unit uses a hexadecimal coding method to assign a unique identifier to each nozzle;

[0017] The status monitoring unit uses sensors to collect the working data of the nozzle and issues an alarm when the data exceeds the limit.

[0018] The position calibration unit uses a laser tracker in conjunction with a six-degree-of-freedom robotic arm to complete the initial installation and positioning of the nozzle.

[0019] Preferably, the solution library module includes:

[0020] The parent solution construction unit is classified into multiple levels according to the product;

[0021] The sub-solution generation unit derives multiple sub-solutions based on different environmental factors. Each sub-solution contains targeted operation instructions and parameter settings. A genetic algorithm is used to locally optimize the basic parent scheme. Through selection, crossover, and mutation operations, individuals in the population are continuously evolved to find the optimal solution that maximizes fitness.

[0022] The solution retrieval and matching unit supports keyword search and fuzzy matching based on the user interface; users can quickly find solutions by inputting product name, material type, or other relevant attributes; the backend uses the Levenshtein distance algorithm to calculate the similarity score between the user input and entries in the database, and selects the top K results with the highest scores to display to the user.

[0023] Preferably, the intelligent decision-making module includes:

[0024] The environmental sensing unit integrates data streams from multiple devices such as temperature and humidity sensors, light intensity meters, and air quality monitors; it uses a Kalman filter algorithm to perform weighted averaging on data from different sources to eliminate noise interference and obtain more accurate environmental state estimates.

[0025] The algorithm model building unit uses a deep convolutional neural network as the core algorithm framework. First, a set of candidate learning rates is fixed, and then experiments are conducted at different network depths, recording the results of each experiment. This process is repeated until the configuration that achieves the highest accuracy on the validation set is found.

[0026] The dynamic optimization unit comprehensively considers three factors: product quality, production efficiency, and cost-effectiveness, and constructs a comprehensive evaluation index.

[0027] Preferably, the quality inspection module includes:

[0028] The overall effect evaluation unit is based on a visual inspection station composed of a high-resolution industrial camera and a telecentric lens. After the product images are captured, they are transmitted to the image processing unit for analysis; semantic segmentation technology in deep learning is used to parse the images.

[0029] The local effect evaluation unit divides the entire product surface into several small areas according to the coverage of the nozzle; each area corresponds to a specific nozzle responsible for spraying; the boundary contour lines of each area are extracted using an image segmentation algorithm; a scanning electron microscope is used to observe the microstructure of the selected area; after acquiring secondary electron images, image processing software is used to measure the particle size distribution histogram and porosity parameters.

[0030] The scoring mechanism design unit determines the importance and weight of each evaluation indicator; and develops detailed scoring rules for each evaluation indicator.

[0031] Preferably, the data analysis module includes:

[0032] The data statistics and analysis unit, based on the establishment of unified data collection standards, ensures that all relevant production equipment can upload data in accordance with the same protocol;

[0033] The correlation analysis unit performs correlation analysis between variables.

[0034] The trend prediction model unit decomposes historical data into trend components, seasonal components, cyclical components, and irregular components; it uses the moving average method to remove the influence of random fluctuations and extract the characteristics of long-term trends and short-term fluctuations.

[0035] Preferably, the feedback adjustment module includes:

[0036] The real-time feedback mechanism unit, based on the human-machine interface (HMI), displays the working status, environmental parameters, and product quality test results of each nozzle in real time.

[0037] The adaptive learning unit establishes a knowledge base for storing historical experiences and lessons learned; each problem-solving process is recorded and stored in the database.

[0038] Preferably, the system further includes a secondary evaluation module, which performs a special aging test on the powder-coated product after it has been placed for a set time, simulating various harsh conditions that the product may encounter in the actual use environment; and performs performance tests on the product before and after the test, recording the changes in various indicators.

[0039] Preferably, the secondary evaluation module includes:

[0040] The aging test procedure unit determines the specific test time and number of cycles based on the product's expected service life and the characteristics of the usage environment.

[0041] The durability assessment index unit conducts wear resistance and corrosion resistance tests on materials.

[0042] The parameter re-optimization process unit describes the relationship between input variables and output response by establishing a quadratic regression model, thereby finding the global optimal solution.

[0043] A quality inspection and feedback method for the powder coating process includes the following steps:

[0044] S1 Initialization: After the system starts, the preset scheme library content is first loaded into memory, and all nozzles connected to the system are initialized.

[0045] S2 Scheme Selection: Based on the data transmitted from the environmental perception unit, the algorithm model in the intelligent decision-making module is invoked to determine the most suitable powder coating sub-scheme for the current conditions; after the scheme is selected, it is sent to the corresponding execution agency for implementation.

[0046] S3 Spraying Execution: Begin the actual spraying operation according to the selected plan;

[0047] S4 Data Acquisition: During the spraying process, various process data are collected and stored;

[0048] S5 Data Analysis: After the spraying is completed, the data analysis module processes and analyzes all the collected data and generates a report;

[0049] S6 Feedback Adjustment: Based on the results of data analysis, the feedback adjustment module automatically generates a series of adjustment instructions and sends them to the relevant equipment or devices for execution;

[0050] S7 Secondary Evaluation: For product samples that have passed the initial inspection, a secondary evaluation will be conducted after a set time.

[0051] S8 Solution Update: When the adjusted powder coating solution deviates significantly from the original solution, the adjusted powder coating solution will be added to the solution library as a new sub-solution.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] This invention uses multi-source sensors, such as temperature, humidity, and light intensity, to sense changes in the working environment and combines them with machine learning models to predict the optimal combination of process parameters (such as spraying speed and electrostatic voltage) to solve quality problems caused by seasonal climate fluctuations. It also achieves independent control of a single spray head based on electronic coding and, together with a position calibration algorithm, ensures uniform coverage of complex curved workpieces.

[0054] This invention encompasses a three-dimensional evaluation network that includes overall appearance (high-definition imaging + deep learning defect segmentation), local microstructure (SEM electron microscopy analysis), and performance indicators (adhesion testing machine quantitative scoring), enabling quality control from macro to micro levels.

[0055] When the adjusted process parameters deviate from the original scheme by more than a threshold, the present invention automatically generates a new sub-scheme and stores it in the knowledge base, forming an innovation cycle of "practice-feedback-iteration". Attached Figure Description

[0056] Figure 1 The present invention provides a flowchart of a quality inspection and feedback method for the powder coating process;

[0057] Figure 2 This invention provides a flowchart for evaluating the overall effect in a quality inspection and feedback system for the powder coating process;

[0058] Figure 3 This invention provides a flowchart for evaluating local effects in a quality inspection feedback system for the powder coating process. Detailed Implementation

[0059] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0060] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0061] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0062] Reference Figures 1-3 As shown.

[0063] The embodiments further illustrate the quality inspection and feedback method and system for the powder coating process proposed in this invention.

[0064] A quality inspection and feedback system for the powder coating process, comprising:

[0065] The system includes a nozzle management module, a solution library module, an intelligent decision-making module, a quality inspection module, a data analysis module, a feedback adjustment module, and a secondary evaluation module.

[0066] The nozzle management module assigns a unique electronic code to each nozzle so that the system can accurately identify and manage the operating status and parameters of each nozzle. This code uses a binary format; for example, for an n-bit code, its value ranges from 0 to 2. n-1 This method ensures that each nozzle has a unique identifier, facilitating subsequent data recording and analysis.

[0067] The nozzle management module monitors key parameters such as working pressure, flow rate, and temperature of each nozzle in real time to ensure normal operation. When abnormalities are detected, such as excessively high or low pressure, or unstable flow rate, the system promptly issues an alarm and takes corresponding measures. For example, if the working pressure of a nozzle falls below the set threshold Pmin, the system automatically adjusts the air supply to that nozzle to restore its pressure to the normal range.

[0068] The nozzle management module regularly calibrates the position of the nozzle to ensure the accuracy and uniformity of spraying. By using a high-precision positioning sensor in conjunction with a robotic arm, precise adjustment of the nozzle position is achieved. Assuming the initial position coordinates of the nozzle are (x0, y0, z0) and the target position coordinates after calibration are (x1, y1, z1), the position deviations are Δx = x1 - x0, Δy = y1 - y0, and Δz = z1 - z0. Based on these deviation values, the system can calculate the displacement required and drive the robotic arm through a motor to complete the movement of the nozzle.

[0069] The solution library module stores multiple powder coating master solutions, and each master solution corresponds to a specific type of product. The master solution contains multiple sub-solutions for the product under different working conditions, such as sub-solutions designed for high temperatures in summer, low temperatures in winter, humid weather, and dry weather. The sub-solutions are saved in the form of a data structure, including spraying parameters (such as paint viscosity η, spraying speed v, spraying distance d, etc.), process route planning, and expected effect indicators.

[0070] The solution library module establishes the association rules between the master solution and the sub-solutions, enabling the system to automatically match the most suitable sub-solution according to the current environmental conditions. For example, define the environmental temperature T as one of the judgment bases. When T > Tsummer, select the sub-solution suitable for high temperatures in summer; when T < Twinter, select the sub-solution suitable for low temperatures in winter. Here, Tsummer and Twinter are the preset temperature thresholds respectively. Similarly, the sub-solutions for humid or dry environments can also be selected based on the humidity H.

[0071] When the deviation between the adjusted powder coating solution and the original solution in the system is relatively large, the solution library module adds the adjusted powder coating solution as a new sub-solution to the solution library. For example, if the differences in various parameters between the new solution and the original solution exceed a certain ratio k (e.g., k = 0.2), it is considered that the new solution has sufficient innovation and practicality and should be saved for future reference. In this way, the content of the solution library is continuously enriched and improved, enhancing the adaptability and flexibility of the system.

[0072] The intelligent decision-making module is equipped with multiple sensors to collect information about the working environment, including temperature sensors, humidity sensors, and light intensity sensors. The data obtained in real time by each sensor is transmitted to the central processing unit for analysis and processing. For example, the environmental temperature measured by the temperature sensor can be calculated using the formula Tenv = R * V / I, where R is the value of the thermistor, V is the voltage applied across it, and I is the current passing through it.

[0073] The intelligent decision-making module constructs a predictive model based on machine learning algorithms, training the optimal combination of powder coating parameters for different environments using historical data. Based on a neural network, the input layer receives environmental parameters (such as temperature and humidity) and other relevant information; the hidden layer extracts features through a series of weighted summations and activation function operations; and the output layer provides recommended powder coating parameter values. Let the input vector X = [x1, x2, ..., x...]. n ] represents environmental factors and other variables, and the weight matrix W = [w ij Connecting neurons in each layer, with the bias term b used to adjust the activation threshold, the output Ym of the m-th layer can be expressed as: Ym = f(∑ i w ij *X i +b), where f is the activation function (such as ReLU, Sigmoid, etc.).

[0074] The intelligent decision-making module combines real-time monitored data with a pre-set objective function J(θ) and uses gradient descent or other optimization algorithms to dynamically adjust the powder coating parameters. The objective function J(θ) typically measures the difference between the actual coating effect and the desired effect, such as a comprehensive score of color deviation ΔE and thickness uniformity U. By continuously iterating and updating the parameter θ, J(θ) is gradually reduced to its minimum value, thereby achieving the optimal coating effect.

[0075] The quality inspection module, when performing an overall evaluation, utilizes image recognition technology and computer vision algorithms to comprehensively assess the powder coating quality of the entire workpiece surface. First, high-resolution photos or video streams are acquired as input data sources. Then, edge detection operators (such as the Canny operator) are used to extract edge information from the images, thereby calculating geometric characteristic indicators such as surface roughness Ra and flatness FL. Simultaneously, the color values ​​are compared with those of a standard color chart sample to calculate the color difference ΔE = [(L-L0)]. 2 +(a-a0) 2 +(b-b0) 2 ] 0.5 Where L, a, b are the measured color coordinate values, and L0, a0, b0 are the corresponding values ​​of the standard colors.

[0076] When performing local evaluation, the quality inspection module analyzes the powder coating situation within the area of ​​responsibility of each nozzle based on its working range. A region segmentation algorithm divides the entire image into several small regions, each corresponding to the work output of one nozzle. For each small region, the overall effect evaluation steps described above are repeated to obtain scores for various quality indicators within that region. Furthermore, the analysis can be further refined to the pixel level, statistically analyzing the grayscale distribution of each pixel to determine the presence of defects such as spots or runs.

[0077] The quality inspection module comprehensively considers both overall and local evaluation results. Based on a preset scoring system, the overall effect accounts for a weight α (α∈[0,1]) of the total score, while the local effect accounts for β (β=1-α). For each evaluation dimension (such as roughness, color accuracy, etc.), a corresponding weight coefficient ω is assigned. i (Σω) i =1), and the final comprehensive score is S=αSglobal+βΣω i *S i local, where Sglobal is the overall performance score, S i `local` is the score of the i-th local region.

[0078] The data analysis module collects and organizes a large amount of production data, including the total amount of paint M used in each spraying and the amount m used by each spray head. j The data includes spraying time t, pass rate p, etc. Statistical methods are used to process these data, calculating statistical quantities such as the mean μ, standard deviation σ, and coefficient of variation CV. For example, the average paint usage μavg for all batches within a certain period is calculated as follows: μavg = (ΣM i ) / N, where N is the total number of batches; calculate the standard deviation σ of the amount used per nozzle. j =sqrt[Σ(m ji -μ j ) 2 / (N-1)] reflects the stability of the nozzle usage.

[0079] The data analysis module explores the relationships between different variables to identify key factors affecting powder coating quality. The Pearson correlation coefficient (r) is used to measure the strength and direction of the linear relationship between two variables. Given two sets of data X={x1,x2,...,x...} n} and Y={y1,y2,...,y n The Pearson correlation coefficient r is calculated as follows: r = [nΣxy - (Σx)(Σy)] / sqrt([nΣx) 2 -(Σx) 2 ][nΣy 2 -(Σy) 2 By calculating the correlation coefficients between each factor and the quality indicators, it can be determined which factors have a significant impact on quality.

[0080] The data analysis module establishes a predictive model based on time series analysis methods to predict future production trends and potential problems. Commonly used models include the ARIMA model and LSTM networks. Taking the ARIMA model as an example, it consists of an autoregressive component AR(p), a differencing component I(d), and a moving average component MA(q), in the form φ(B)(1-B).dXt =θ(B)ε t Where B is the lag operator, φ and θ are the coefficient vectors of the AR and MA polynomials, respectively, and ε t It is a white noise sequence. By fitting and validating historical data, and selecting appropriate model parameters p, d, and q, accurate predictions of future data can be achieved.

[0081] The feedback adjustment module is equipped with an instant feedback mechanism. When the quality inspection module detects quality problems or the data analysis module identifies potential risk points, it immediately issues a warning signal to the operator and provides specific improvement suggestions. For example, if the thickness of a certain area is insufficient, the system prompts to increase the number of sprays in that area or adjust the angle of the spray gun; if the deviation of a certain color is too large, it suggests replacing the paint batch of the corresponding color.

[0082] The feedback adjustment module periodically reviews and optimizes the entire production process based on accumulated historical data and long-term statistical analysis results. For example, at regular intervals (such as monthly), the performance of each nozzle is reassessed, and their overall score ranking determines whether maintenance or replacement of parts is required. At the same time, the powder coating solutions in the solution library are updated according to the latest market demands and technological development trends to maintain the system's advanced nature and competitiveness.

[0083] The feedback adjustment module records all parameters and result data during the process after each successful production task is completed, and adds them to the training set for the next model training.

[0084] The secondary evaluation module conducts specialized aging tests on powder-coated products after a set period of time, simulating various harsh conditions the product might encounter in real-world use. Performance tests are performed on the product before and after the initial test, recording changes in various indicators. For example, during ultraviolet irradiation testing, color changes and adhesion loss rates of the coating are measured at specified time intervals.

[0085] Based on the results of the secondary evaluation, the secondary evaluation module re-enters the feedback adjustment stage to fine-tune the original powder coating parameters to achieve better long-term stability and reliability. For example, if cracking is found in some areas of the coating after aging tests, the coating thickness in these areas is increased to improve their flexibility.

[0086] Specifically, the nozzle management module includes:

[0087] The coding and identification unit assigns a unique identifier to each printhead using a hexadecimal coding system, in the format "SP-XXXX" (where XXXX represents a four-digit hexadecimal number). For example, the first printhead could be coded as "SP-0001". During production, the code is physically attached to the printhead using an RFID tag or QR code, ensuring the system can quickly and accurately read and identify it. When a printhead is repaired and put back into use, the coding refresh procedure is automatically triggered. The new code is generated according to an incremental principle; for example, if the original code is "SP-0005", the next usable code after repair should be "SP-0006". Simultaneously, the mapping relationship between old and new codes is recorded for easy tracing of historical data. The correctness of the code is verified based on the CRC algorithm. Each time the printhead code is read, a CRC calculation is performed and compared with a pre-stored standard value. If they do not match, it is determined to be an incorrect code, and an alarm is triggered.

[0088] The status monitoring unit utilizes high-precision pressure, flow, and temperature sensors to collect data per second on the operating pressure P (Pa), paint flow rate Q (mL / min), and nozzle temperature T (°C) for each nozzle. The data acquisition card converts these analog signals into digital signals and transmits them to the central controller. Based on the technical specifications of different nozzle models, reasonable upper and lower limits for the operating range are set. For example, the normal operating pressure range for a certain nozzle is [Pmin, Pmax] = [300kPa, 500kPa]. When the detected pressure is lower than Pmin or higher than Pmax, an audible and visual alarm is immediately activated, and a detailed alarm information box pops up on the operation interface, displaying the abnormal nozzle's code, current value, and suggested handling measures. A sliding window technique is used to analyze the continuously collected data, with the window size set to N = 60 seconds. The average value μ, standard deviation σ, and rate of change δ = (latest value − earliest value) / time interval Δt are calculated for each window. If δ exceeds the preset growth rate threshold γ, a potential malfunction risk is predicted for the nozzle, and a maintenance plan is arranged in advance.

[0089] The position calibration unit uses a laser tracker in conjunction with a six-degree-of-freedom robotic arm to complete the initial installation and positioning of the nozzle. First, a three-dimensional coordinate system OXYZ is established within the working area. Then, the position error of the robotic arm's end effector is determined to be no more than ±0.05mm using the principle of laser interferometry. Afterward, the nozzle is fixed to the robotic arm, and its attitude is adjusted to ensure that the nozzle axis coincides with the predetermined trajectory at a rate of over 99%. A comprehensive position check is performed every M=7 days. A vision-guided positioning method is used, capturing panoramic images of the working area. Feature point coordinates are extracted using image processing algorithms and compared with the theoretical model. For deviations greater than ε=0.1mm, an automatic compensation program is initiated to adjust the robotic arm's motion parameters until the accuracy requirements are met. Based on the least squares dynamic compensation algorithm, assuming the point set on the ideal trajectory is {Pi*}, i=1,...,n, and the actual measured point set is {Pi}, i=1,...,n, the optimal translation vector Δx=(Δu,Δv,Δw) can be obtained by solving the following system of equations: ∑[(ui-Δu)] 2 +(vi-Δv) 2 +(wi-Δw) 2 →min. Where (ui,vi,wi) represents the coordinates of the i-th ideal point, and (ui',vi',wi') represents the coordinates of the corresponding actual measured point; the above system of equations is solved through an iterative optimization process to obtain the accurate compensation amount.

[0090] Specifically, the solution library module includes:

[0091] The parent solution construction unit is categorized into multiple levels based on factors such as product material (metal, plastic, etc.), shape complexity (simple geometry, complex surfaces), and size range (small parts, large structural components). For example, products like automotive wheels are categorized as "metal materials -> complex surfaces -> medium size." A standardized powder coating process template is developed for each product category; a data table with multiple fields is created to store the specific parameters of the parent solution. Key fields include coating type (epoxy polyester powder, polyurethane powder, etc.), curing curve (heating rate, holding time, cooling rate), target coating thickness dtarget (unit: μm), and electrostatic voltage Uesd (unit: kV). Each record corresponds to a specific process configuration. Domain expert rules of thumb are introduced as constraints. For example, "When the product surface roughness Ra > 1.6 μm, it is recommended to add a pre-degreasing process"; "For thin-walled parts, the spraying air pressure should be appropriately reduced to avoid deformation." These rules are embedded in the system in the form of IF-THEN logical expressions to guide the selection and optimization of solutions.

[0092] The sub-scheme generation unit generates multiple sub-schemes based on different climatic conditions (temperature, humidity), seasonal changes (spring, summer, autumn, winter), and geographical location (altitude, latitude, and longitude). For example, a special dehumidification pretreatment step is set for high-temperature and high-humidity environments; while for cold regions, preheating the workpiece is required to improve adhesion. Each sub-scheme includes a series of targeted operation instructions and parameter settings. A genetic algorithm is used to locally optimize the basic parent scheme. The fitness function F=f(c1,c2,...,c k ), where c i This represents the i-th adjustable parameter. The population is continuously evolved through selection, crossover, and mutation operations to find the optimal solution that maximizes fitness. For example, the optimization objective is to minimize the coating defect density ρfault while ensuring that the production efficiency η does not fall below a certain lower limit. A virtual spraying environment is constructed using computer-aided engineering software to simulate the spraying effects under different conditions. The parameters of the sub-scheme to be tested are input, the simulation model is run, and the predicted results, such as coverage and film thickness distribution uniformity, are output. Only sub-schemes that pass simulation verification can enter the practical application stage.

[0093] The solution retrieval and matching unit supports keyword search and fuzzy matching based on the user interface. Users can input product name, material type, or other relevant attribute information for quick searching. The backend uses the Levenshtein distance algorithm to calculate the similarity score between the user input and entries in the database, selecting the top K results with the highest scores to display to the user. A classification model is trained using machine learning techniques, based on the feature vector X=[x1,x2,...,x...] of historical successful cases. n The system predicts the most suitable sub-solution category Y for the current task. For example, given a new order of automotive parts, the system automatically recommends the best powder coating solution based on its material, shape, and other characteristics. Manual adjustments to the automatically recommended solution are supported based on actual conditions. The system records all modifications and their rationale, forming part of a knowledge base for future algorithm accuracy improvements.

[0094] Specifically, the intelligent decision-making module includes:

[0095] The environmental sensing unit integrates data streams from various devices, including temperature and humidity sensors, light intensity meters, and air quality monitors. It employs a Kalman filter algorithm to perform weighted averaging of data from different sources, eliminating noise interference and obtaining a more accurate environmental state estimate: Eest=[Test,Hest,Lest,AQIest]. Here, Test represents the estimated temperature, Hest represents the relative humidity percentage, Lest represents the light intensity in lux, and AQIest represents the air quality index. Safety ranges are set for each indicator. For example, the suitable temperature range for spraying operations is typically 18℃~25℃, and the relative humidity is controlled between 40%~70%. If any indicator exceeds the normal range, the system will trigger an early warning signal and provide corresponding response suggestions. Historical meteorological data is collected as a training sample set D={(ti,Ei)}, i=1,...,N. A suitable time series forecasting model (such as the ARIMA model) is selected, and its goodness of fit is tested and parameters are fine-tuned. The final model can be used for short-term (next few hours) environmental change prediction, helping to prepare for production in advance.

[0096] The algorithm model building unit uses a deep convolutional neural network as the core algorithm framework. The input layer receives a normalized environment parameter vector Enorm=[Tnorm,Hnorm,Lnorm,AQInorm]; the hidden layer hierarchy can be flexibly configured according to the complexity of the problem, generally including several convolutional layers, pooling layers, and fully connected layers; the output layer outputs the recommended powder coating parameter combination Π=[p1,p2,...,p...]. k The loss function chosen is the mean squared error. ,in It is the actual value. Here, m represents the predicted value, and m is the number of samples. A grid search combined with cross-validation is used to find the optimal combination of network structure and learning rate. Specifically, a set of candidate learning rates lr∈{0.01,0.001,0.0001} is fixed, and experiments are conducted at different network depths D∈{2,4,6,8}, recording the results of each experiment. This process is repeated until the configuration that yields the highest accuracy on the validation set is found. An incremental learning mechanism is introduced, allowing the model to continuously update its weights as new data arrives without retraining from scratch. Whenever a new production batch is completed, its corresponding input-output pair is added to the training set, the gradient descent direction is recalculated, and the weight matrix is ​​updated. Where α is the learning rate, It is the gradient vector under the current weights.

[0097] The dynamic optimization unit comprehensively considers three factors: product quality (QP), production efficiency (PE), and cost-effectiveness (CB), constructing a comprehensive evaluation index J = ω1·QP + ω2·PE + ω3·CB. Here, ω1 + ω2 + ω3 = 1, and the weight coefficients reflect the importance of each factor. For example, for high-end customized products, a higher quality weight ω1 is assigned; while for mass production tasks, more emphasis is placed on efficiency and cost control. In addition to the objective function, constraints such as the maximum permissible spraying speed Vmax not exceeding the equipment's design limit and the minimum coating thickness tmin meeting corrosion resistance requirements are transformed into mathematical inequalities and incorporated into the optimization problem. For the above constrained optimization problem, the Lagrange multiplier method or interior-point method is used for solution. Taking the Lagrange multiplier method as an example, the Lagrange function L(x,λ) = f(x) + λg(x) is constructed, where f(x) is the objective function to be minimized, and g(x) <= 0 represents all constraints. By taking the partial derivatives of L with respect to x and λ and setting them equal to zero, a set of nonlinear equations can be obtained, and then the optimal solution x* can be solved.

[0098] Specifically, the quality inspection module includes:

[0099] The overall effect evaluation unit, based on a high-resolution industrial camera and a telecentric lens forming a visual inspection station, uses a ring-shaped LED light source to provide uniform illumination. After shooting, the acquired product image (img) is transmitted to the image processing unit for analysis. Semantic segmentation technology from deep learning is used to parse the image. A large number of pre-labeled sample images with various defect types are used as a training set to train the UNet++ network model to automatically identify common defects such as scratches, pinholes, and bubbles. The probability map (Pmap) output by the model shows the probability of each pixel belonging to a certain defect category. A threshold τ=0.5 is set; when the Pmap value of a certain area is greater than τ, a defect of the corresponding type is considered to exist. A standard color chart is selected as a reference, and the color coordinates (Lab values) of the sample are measured using a spectrophotometer. The color difference ΔE between the sample and the standard color chart is calculated as ΔE=[(L2-L1)]. 2 +(a2-a1) 2 +(b2-b1) 2 ] 0.5 Where L1, a1, b1 are the Lab values ​​of the standard color, and L2, a2, b2 are the Lab values ​​of the sample. The color is judged to be acceptable according to the CIEDE2000 standard.

[0100] The local effect evaluation unit divides the entire product surface into several small areas R={R1,R2,...,R...} based on the coverage of the nozzle. nEach area is assigned to a specific spray nozzle for coating. Image segmentation algorithms are used to extract the boundary contours of each area. A scanning electron microscope (SEM) is used to observe the microstructure of the selected areas. After acquiring secondary electron images, image processing software is used to measure parameters such as particle size distribution histogram and porosity. Microscopic features reflect the internal quality and bonding strength of the coating. Mechanical performance evaluations, such as adhesion and hardness tests, are performed at designated test stations. For example, the cross-cut adhesion test is used to determine the adhesion level between the coating and the substrate; a micro Vickers hardness tester is used to measure the hardness value (HV) of the coating surface. Test results are recorded and compared with standard requirements.

[0101] The scoring mechanism design unit determines the importance weights (ωᵢ) of each evaluation indicator based on customer needs and corporate standards. For example, appearance quality accounts for 40%, functional performance for 30%, and durability for 30%. It ensures that the sum of the weights equals 1 and is reasonably distributed; detailed scoring rules are developed for each evaluation indicator. For example, appearance quality is divided into five levels: Excellent (90-100 points), Good (80-89 points), Average (70-79 points), Poor (60-69 points), and Unacceptable (<60 points). Each level corresponds to a specific evaluation standard description. The final comprehensive score, Score=Σ(ωᵢ), is calculated using a weighted average method. i ×Score i ), where Score i This is the actual score of the i-th indicator. Products are categorized and managed according to their total scores to facilitate subsequent quality traceability and improvement efforts.

[0102] Specifically, the data analysis module includes:

[0103] The data statistical analysis unit, based on standardized data collection protocols, ensures that all relevant production equipment can upload data according to the same protocol. For example, it specifies recording current production status information, including equipment operating parameters and output statistics, every minute. SPSS software is used to perform basic statistical analysis on the collected data. Statistics such as mean, median, mode, standard deviation, and variance are calculated, and histograms and boxplots are created to visually display the data distribution characteristics. The CP / CPK index is used to assess the stability and capability level of the production process. The calculation formulas are as follows: CP = (USL - LSL) / 6σ; CPK = min{(USL - μ) / 3σ, (μ - LSL) / 3σ}, where USL and LSL are the upper and lower limits of specifications, respectively, μ is the population mean, and σ is the standard deviation. Generally, CPK ≥ 1.33 indicates that the process is well controlled.

[0104] The correlation analysis unit selects a few key variables from numerous factors that may affect product quality for analysis. Variables closely related to quality issues and easily controlled are typically chosen as the research subjects. Examples include paint viscosity, spraying distance, and atomization pressure. A scatter plot matrix is ​​generated between each pair of selected variables to initially observe whether linear or other patterned relationships exist. If a clear linear trend is found, further regression analysis is performed; if a non-linear relationship is found, multinomial regression or other non-linear models are considered to fit the data points. A multiple linear regression model is established: Y = β0 + β1X1 + β2X2 + ... + β n X n +ε, where Y is the dependent variable (e.g., product quality score), X... i These are explanatory variables (i.e., the key influencing factors selected earlier), β i Here, ε is the regression coefficient, and ε is the random error term. The values ​​of the regression coefficients are estimated using the least squares method, and a significance test (t-test) is performed to determine the degree of influence of each independent variable on the dependent variable.

[0105] The trend forecasting model unit decomposes historical data into four components: trend, seasonality, cyclicality, and irregularity. Moving averages can be used to remove the influence of random fluctuations and extract characteristics of long-term trends and short-term fluctuations. The Holt-Winters three-parameter exponential smoothing method is employed to forecast future demand. This method considers the influence of three factors: level, trend, and seasonality. The recursive formula is as follows: Q t =αy t +(1-α)(Q{t-1}+b{t-1}); b t =β(Q t -Q{t-1})+(1-β)b{t-1};s t =γ(y t -Q t )+(1-γ)s{tm}, where α, β, and γ are smoothing coefficients, and m is the length of the seasonal cycle. Cross-validation is used to evaluate the performance of the prediction model. The dataset is divided into training and test sets. The model is trained using the training set and tested for its generalization ability using the test set.

[0106] Specifically, the feedback adjustment module includes:

[0107] The real-time feedback mechanism unit, based on a human-machine interface (HMI), displays real-time information such as the operating status of each nozzle, environmental parameters, and product quality inspection results. When an abnormality occurs, the interface visually alerts the operator. A built-in Fault Tree Analysis (FTA) tool quickly locates the root cause of problems. Starting from the top-level event, it traces down the hierarchy of possible causes until the root cause is found. When a serious fault (such as overpressure or overload) is detected, the power supply is immediately cut off and all equipment operation is stopped to prevent the accident from escalating and causing greater losses.

[0108] An adaptive learning unit establishes a knowledge base for storing historical experiences and lessons learned. Each problem-solving process is recorded and stored in the database for future reference. The database employs a relational database management system to achieve efficient data retrieval and management functions. Instance-based reasoning techniques are used to solve similar problems. When encountering a new problem, the knowledge base is first checked for similar cases; if found, previous solutions can be directly referenced; otherwise, the problem needs to be analyzed and solved from scratch, and the new solution added to the knowledge base.

[0109] Specifically, the secondary evaluation module includes:

[0110] The aging test program unit is based on a multi-functional environmental test chamber used to simulate usage conditions under various extreme environments. This test chamber should possess features such as a wide temperature adjustment range (-70℃ to +180℃), high humidity control accuracy (±5%RH), and adjustable UV irradiation intensity to meet different testing needs. The specific test time and number of cycles are determined based on the product's expected service life and the characteristics of the usage environment. For example, outdoor plastic products may require hundreds of hours of UV aging testing to reach the expected service life requirements. A mathematical model of material performance over time is established through regression analysis of a large amount of experimental data: Performance(t) = A·exp(-B·t) + C, where A, B, and C are constant parameters obtained through fitting, representing the initial performance level, degradation rate, and residual performance limit, respectively.

[0111] The durability assessment unit uses the Taber abrasion testing machine to test the abrasion resistance of materials. This equipment is equipped with grinding wheels of different hardness grades, allowing for the selection of appropriate abrasive types and load weights according to different testing standards. Electrochemical impedance spectroscopy (EIS) is used to measure the corrosion rate of metallic materials in different media: Corrosion Rate (mm / year) = K / ρ·(Rsol / Rpol), where K is a constant depending on the type of electrolyte solution used; ρ is the metal density; Rsol is the solution resistance; and Rpol is the polarization resistance. Regularly measuring the trends of these parameters allows for the assessment of the material's corrosion resistance. Based on fracture mechanics principles, the relationship between the fatigue crack propagation rate da / dN and the stress intensity factor range ΔK is established using the Paris Law: da / dN = C·(ΔK). m , where C and m are material constants determined experimentally.

[0112] The parameter re-optimization process unit employs a central composite design (CCD) method to arrange experimental schemes and explore the optimal combination of process parameters. OptimalParametersSet={p1,p2,...,p k The goal of maximizing or minimizing the response variable (such as product quality score) is achieved by establishing a quadratic regression model (ResponseSurfaceModel) to approximate the relationship between the input variable and the output response, thereby finding the global optimum. Orthogonal experimental design and analysis of variance (ANOVA) techniques are then combined to further refine the search scope and narrow the spatial dimension of the feasible region, thereby improving optimization efficiency and reducing experimental costs.

[0113] The implementation methods of the system include:

[0114] S1 Initialization: After the system starts, the preset scheme library content is first loaded into memory, and all nozzles connected to the system are initialized.

[0115] S2 Scheme Selection: Based on the data transmitted from the environmental perception unit, the algorithm model in the intelligent decision-making module is invoked to determine the most suitable powder coating sub-scheme for the current conditions; after the scheme is selected, it is sent to the corresponding execution agency for implementation.

[0116] S3 Spraying Execution: Begin the actual spraying operation according to the selected plan;

[0117] S4 Data Acquisition: During the spraying process, various process data are collected and stored;

[0118] S5 Data Analysis: After the spraying is completed, the data analysis module processes and analyzes all the collected data and generates a report;

[0119] S6 Feedback Adjustment: Based on the results of data analysis, the feedback adjustment module automatically generates a series of adjustment instructions and sends them to the relevant equipment or devices for execution;

[0120] S7 Secondary Evaluation: For product samples that have passed the initial inspection, a secondary evaluation will be conducted after a set time.

[0121] S8 Solution Update: When the adjusted powder coating solution deviates significantly from the original solution, the adjusted powder coating solution will be added to the solution library as a new sub-solution.

[0122] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. 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.

Claims

1. A quality inspection and feedback system for the powder coating process, characterized in that, include: The nozzle management module assigns a unique electronic code to each nozzle, monitors the parameters of each nozzle in real time, and periodically calibrates the position of the nozzles. The solution library module stores multiple powder coating master solutions, each corresponding to a specific product. The master solution includes multiple sub-solutions for that product under different working conditions. The solution library module establishes association rules between master solutions and sub-solutions, enabling the system to automatically match the most suitable sub-solution based on the current environmental conditions. The intelligent decision-making module is based on machine learning algorithms to build a prediction model and trains the optimal combination of powder coating parameters under different environments based on historical data. The quality inspection module performs a comprehensive assessment and local evaluation of the powder coating quality of the entire workpiece surface. The data analysis module collects and organizes production data; The feedback adjustment module is equipped with an instant feedback mechanism. When the quality inspection module discovers a quality problem or the data analysis module identifies a potential risk point, it immediately issues a warning signal to the operator and provides specific improvement suggestions. The data analysis module includes: The data statistics and analysis unit, based on the establishment of unified data collection standards, ensures that all relevant production equipment can upload data in accordance with the same protocol; The correlation analysis unit performs correlation analysis between variables. The trend prediction model unit decomposes historical data into trend components, seasonal components, periodic components, and irregular components; it uses the moving average method to remove the influence of random fluctuations and extract the characteristics of long-term trends and short-term fluctuations. The feedback adjustment module includes: The real-time feedback mechanism unit, based on the human-machine interface (HMI), displays the working status, environmental parameters, and product quality test results of each nozzle in real time. The adaptive learning unit establishes a knowledge base for storing historical experiences and lessons learned; each problem-solving process is recorded and stored in the database.

2. The powder coating process quality inspection and feedback system according to claim 1, characterized in that, The nozzle management module includes: The coding and identification unit uses a hexadecimal coding method to assign a unique identifier to each nozzle; The status monitoring unit uses sensors to collect the working data of the nozzle and issues an alarm when the data exceeds the limit. The position calibration unit uses a laser tracker in conjunction with a six-degree-of-freedom robotic arm to complete the initial installation and positioning of the nozzle.

3. The powder coating process quality inspection and feedback system according to claim 1, characterized in that, The solution library module includes: The parent solution construction unit is classified into multiple levels according to the product; The sub-solution generation unit derives multiple sub-solutions based on different environmental factors. Each sub-solution contains targeted operation instructions and parameter settings. A genetic algorithm is used to locally optimize the basic parent scheme. Through selection, crossover, and mutation operations, individuals in the population are continuously evolved to find the optimal solution that maximizes fitness. The solution retrieval and matching unit supports keyword search and fuzzy matching based on the user interface; users can quickly find solutions by inputting product name, material type, or other relevant attributes; the backend uses the Levenshtein distance algorithm to calculate the similarity score between the user input and entries in the database, and selects the top K results with the highest scores to display to the user.

4. The powder coating process quality inspection and feedback system according to claim 1, characterized in that, The intelligent decision-making module includes: The environmental sensing unit integrates data streams from temperature and humidity sensors, light intensity meters, and air quality monitors; it uses a Kalman filter algorithm to perform weighted averaging on data from different sources to eliminate noise interference and obtain more accurate environmental state estimates. The algorithm model building unit uses a deep convolutional neural network as the core algorithm framework. First, a set of candidate learning rates is fixed, and then experiments are conducted at different network depths, recording the results of each experiment. This process is repeated until the configuration that achieves the highest accuracy on the validation set is found. The dynamic optimization unit comprehensively considers three factors: product quality, production efficiency, and cost-effectiveness, and constructs a comprehensive evaluation index.

5. The powder coating process quality inspection and feedback system according to claim 1, characterized in that, The quality inspection module includes: The overall effect evaluation unit is based on a visual inspection station composed of a high-resolution industrial camera and a telecentric lens. After the product images are captured, they are transmitted to the image processing unit for analysis; semantic segmentation technology in deep learning is used to parse the images. The local effect evaluation unit divides the entire product surface into several small areas according to the coverage of the nozzle; each area corresponds to a specific nozzle responsible for spraying; the boundary contour lines of each area are extracted using an image segmentation algorithm; after acquiring secondary electronic images, image processing software is used to measure the particle size distribution histogram and porosity parameters. The scoring mechanism design unit determines the importance and weight of each evaluation indicator; and develops detailed scoring rules for each evaluation indicator.

6. The powder coating process quality inspection feedback system according to claim 1, characterized in that, The system also includes a secondary evaluation module, which conducts a special aging test on the powder-coated product after it has been placed for a set time, simulating various harsh conditions that the product may encounter in the actual use environment; the system also conducts performance tests on the product before and after the test, and records the changes in various indicators.

7. A quality inspection and feedback system for the powder coating process according to claim 6, characterized in that, The secondary evaluation module includes: The aging test procedure unit determines the specific test time and number of cycles based on the product's expected service life and the characteristics of the usage environment. The durability assessment index unit conducts wear resistance and corrosion resistance tests on materials. The parameter re-optimization process unit describes the relationship between input variables and output response by establishing a quadratic regression model, thereby finding the global optimal solution.

8. A quality inspection and feedback method for the powder coating process, characterized in that, The powder coating process quality inspection and feedback system according to any one of claims 1-7 includes the following steps: S1 Initialization: After the system starts, the preset scheme library content is first loaded into memory, and all nozzles connected to the system are initialized. S2 Scheme Selection: Based on the data transmitted from the environmental perception unit, the algorithm model in the intelligent decision-making module is invoked to determine the most suitable powder coating sub-scheme for the current conditions; after the scheme is selected, it is sent to the corresponding execution agency for implementation. S3 Spraying Execution: Begin the actual spraying operation according to the selected plan; S4 Data Acquisition: During the spraying process, various process data are collected and stored; S5 Data Analysis: After the spraying is completed, the data analysis module processes and analyzes all the collected data and generates a report; S6 Feedback Adjustment: Based on the results of data analysis, the feedback adjustment module automatically generates a series of adjustment instructions and sends them to the relevant equipment or devices for execution; S7 Secondary Evaluation: For product samples that have passed the initial inspection, a secondary evaluation will be conducted after a set time. S8 Solution Update: When the adjusted powder coating solution deviates significantly from the original solution, the adjusted powder coating solution will be added to the solution library as a new sub-solution.

Images