Method for optimizing coating process of polytetrafluoroethylene inside steel lining

Abstract

The invention provides an innovative method for optimizing a coating process of polytetrafluoroethylene (PTFE) inside a steel lining, and belongs to the technical field of industrial coating. According to the method, an advanced optimization adversarial neural network (GAN) and gradient enhanced decision tree (GBDT) technology is fused, a multi-layer spraying control strategy is introduced, the GAN can generate single-layer or multi-layer spraying parameters according to requirements, and the spraying rhythm is dynamically adjusted, so that the bonding strength and corrosion resistance of a coating are greatly improved. The historical data is deeply analyzed by the GBDT, so that a more accurate input parameter weight is provided for the GAN, and the prediction capability is further enhanced. The optimization method is widely applied to equipment such as steel lining PTFE corrosion-resistant pipelines, fluorine-lined storage tanks, valves, compensators and the like, the coating consistency and the overall performance of products are remarkably improved, and higher value is brought to industrial application. In a word, through intelligent and refined coating process optimization, double improvement of the performance and economic benefits of the PTFE coating is achieved.

Description

Technical Field

[0001] The present invention relates to a method for optimizing a polytetrafluoroethylene (PTFE) coating process inside a steel liner based on a generative adversarial neural network (GAN) and a gradient boosted decision tree (GBDT), and belongs to the technical field of anti-corrosion coatings. Background Art

[0002] Steel-lined polytetrafluoroethylene (PTFE) corrosion-resistant pipes are widely used in the chemical, pharmaceutical, and food industries. Their excellent corrosion resistance and mechanical strength make them an important medium conveying equipment. However, existing PTFE coating processes still have many problems. First, the traditional spraying process cannot accurately control the thickness of PTFE, resulting in material waste or local weaknesses, which affect corrosion resistance. Second, the bond strength between the steel lining and the PTFE coating is significantly affected by process parameters, resulting in localized delamination or insufficient durability. Moreover, due to the high cost of PTFE, reducing material usage while ensuring corrosion resistance becomes a key issue. Furthermore, traditional spraying parameter settings rely on manual experience and lack a systematic optimization method, resulting in poor process consistency. Therefore, a coating process optimization method is urgently needed that can utilize advanced data modeling technology and intelligent optimization algorithms to optimize spraying parameters through a data-driven approach, improve coating quality, and reduce material waste. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the purpose of the present invention is to provide a steel-lined PTFE coating process optimization method based on GAN and GBDT. The spraying parameters are optimized by GAN, and the input data weight of GAN is adjusted in combination with GBDT to improve the adaptability of the spraying process, realize intelligent spray path optimization, quality detection and adaptive compensation, and improve the thickness uniformity, bonding strength and material utilization of the coating.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions: The technical solution of the present invention is: a method for optimizing the polytetrafluoroethylene coating process inside a steel liner based on optimizing GAN and GBDT, comprising the following steps: Step 1: Install sensors to collect data, including ambient temperature, humidity, spraying pressure, rotation speed, heating temperature, coating thickness, etc. Step 2: Use GAN combined with GBDT to optimize coating process parameters and improve the prediction accuracy of coating parameters generated by GAN; Step 3: Use the optimal parameters generated by GAN to execute the spraying process and combine it with the adaptive compensation mechanism to optimize the spraying path; Step 4: Perform quality inspections through sensors and feed the data back to the GAN model for adaptive optimization; Further, the specific process of step one is: During the optimization process of polytetrafluoroethylene (PTFE) coating inside steel liners, to ensure data accuracy and stability, it is necessary to install high-precision sensors at key process links to collect production environment and process parameters in real time, and provide reliable input data for the Generative Adversarial Network (GAN) optimization model. The first step is sensor installation, including installing environmental parameter sensors and installing process parameter sensors; Environmental parameters have a significant impact on PTFE coating quality. For example, high humidity may lead to reduced adhesion. Therefore, the following sensors need to be placed around the coating equipment: Ambient temperature sensor T e Parameter range: T e ∈[10,50]°C Function: Monitor the production environment temperature to ensure that the coating process is carried out within the optimal temperature range; Installation location: the four corners of the spray room to reduce the impact of local temperature gradients; Environmental humidity sensor H e Parameter range: H e ∈[20,80]% Function: Monitor the effect of humidity on PTFE melting and adhesion, and ensure that process parameters are adapted to different humidity conditions; Installation location: coating workshop walls and around coating equipment; In order to monitor the key variables of the coating process in real time, the following process parameter sensors are required: Spraying pressure sensor P s Parameter range: P s ∈[0.5,2]MP Function: Monitor the spraying pressure to ensure that the PTFE particles adhere to the steel lining surface under the optimal kinetic energy, and avoid excessive pressure causing particle rebound or too little pressure causing uneven coating; Installation location: near the spray gun, the nozzle outlet pressure is detected by a high-precision sensor; Rotational speed sensor R s Parameter range: R s ∈[500,2000]rpm Function: Monitor the rotation speed, optimize the coating thickness uniformity, and avoid uneven coating caused by unstable rotation speed; Installation location: Coating the axis of the rotating device, using a non-contact speed measuring device to improve accuracy; Infrared temperature sensor T c Parameter range: Tc ∈[300,500]°C Function: Measure the heating temperature to ensure that the PTFE coating melts and spreads evenly on the steel lining surface; Installation location: on both sides of the spray track in the heating area, using infrared sensors for non-contact temperature measurement to ensure accurate temperature measurement; Laser Thickness Gauge D t Parameter range: D t ∈[10,200]μm Function: Real-time monitoring of coating thickness to ensure uniform thickness and avoid excessive thickness or thinness that affects anti-corrosion performance; Installation location: Above the coating target surface in the spraying area, using the laser measurement principle to achieve non-destructive measurement; Then perform data collection and preprocessing: The process parameters collected by the sensor need to be cleaned and preprocessed to improve data quality and ensure that the GAN optimization model can accurately identify the optimal coating parameters; All installed sensors are connected to the central control system through the Industrial Internet of Things (IIoT) system, collecting data in real time and storing it in the database at a fixed frequency; Acquisition frequency f c : Environmental parameter sensors: (sampled once per second) Process parameter sensors: (Sampled every 0.1 seconds) The collected data set is defined as follows: in: :Ambient temperature (°C), :Ambient humidity (%), :Spraying pressure (MPa), :Rotation speed (rpm), :Heating temperature (°C), :Coating thickness (μm), : time(s); The raw data collected by the sensor may contain noise or outliers and require preprocessing: Outlier detection: Use the 3σ criterion to eliminate abnormal data: in, is the data mean, is the standard deviation, and the abnormal data beyond the range are eliminated; Data filtering: Kalman filter is used to remove noise and improve measurement stability. The filtering equation is as follows: in: is the current state, is the control input, is the noise term; Dynamically adjust the prediction value through Kalman gain to improve data accuracy; Data normalization: Due to different dimensions of different parameters, maximum and minimum normalization is used: in: is the original data, , is the minimum and maximum value in the data set; Normalized data Between 0 and 1, improving the stability of GAN training; After outlier detection, filtering and normalization, these data will be input into the GAN optimization model to optimize the spray process parameters and improve the coating quality and consistency.

[0005] Furthermore, the specific process of step 2 is as follows: In this paper, in order to optimize the coating process parameters of polytetrafluoroethylene (PTFE) inside the steel liner, a genetic adversarial network (GAN) is used to generate the coating parameters, and a gradient boosted decision tree (GBDT) is introduced to optimize the GAN input parameters, thereby improving the coating uniformity and bonding strength, while optimizing material usage and production efficiency; First, a GAN model is constructed. The GAN generator (G) is used to generate spraying process parameters. Its input variables include: temperature (Heating temperature, unit: °C) pressure (Spraying pressure, unit: MPa) Rotation speed (Speed ​​of coating rotating device, unit: rpm) time (Spraying time, unit: seconds) Mathematical description of GAN generator: Generator Receive a random noise vector (usually obeys a normal distribution) and maps it to the spraying process parameters through a deep neural network in: is the spray parameter set generated by GAN, (0,1) represents random noise from a standard normal distribution; The loss function of the generator is: in: Represents the discriminator's evaluation of the generated parameters. The goal is to make the discriminator unable to distinguish between generated data and real data; The discriminator (D) is used to distinguish whether the spraying parameters generated by GAN are reasonable. Its input includes: GAN-generated spraying parameters Spraying parameters based on real production data Coating quality evaluation index: coating uniformity U t (Unit: μm, target error range: ±5 μm), bonding strength S t (Unit: MPa, target value> 10MPa), material utilization rate M u (target value>85%); Mathematical description of the discriminator: The goal of the discriminator is to judge the authenticity of the input data by learning the distribution of real spray data: in: is the input data (GAN generated or real spray parameters), , is the discriminator weight parameter, represents the Sigmoid activation function; The loss function of the discriminator is: Among them: the first term represents the classification loss of real data, and the second term represents the classification loss of GAN-generated data; Then, adversarial training is performed through the constructed GAN model. GAN continuously optimizes the spraying process parameters through adversarial training: The generator (G) attempts to generate parameter combinations that are closer to real spray data to improve coating uniformity and bonding strength. The discriminator (D) distinguishes the GAN-generated parameters from the real parameters and feeds them back to the generator for optimization. The generator generates a set of spraying parameters G(z), the discriminator calculates D(G(z)) and gives feedback, and optimizes it through gradient descent. and , until GAN ​​training converges; After that, the gradient boosted decision tree (GBDT) is combined to optimize the GAN input parameters; Gradient boosted decision tree (GBDT) is used to analyze spraying history data, adjust the input variable weights of GAN, and improve the prediction accuracy of the model; The goal of GBDT is to analyze the effects of temperature, pressure, speed, and time on coating uniformity, bonding strength, and material utilization using historical coating data, assigning different weights to GAN input parameters to improve the accuracy of GAN-generated parameters. GBDT calculation formula: GBDT uses multiple decision trees Perform optimization, each round of iteration: in: is the model prediction value of the kth iteration, is the decision tree model for this round, is the learning rate (usually 0.01-0.1); The loss function uses the mean square error (MSE): in: It is the coating quality indicator of the real spraying parameters. are the input parameters (temperature, pressure, speed, time); GBDT optimization effect: Calculate the influence weight of each parameter on coating quality: Temperature impact weight: Pressure impact weight: Speed ​​influence weight: Time impact weight: Adjust the GAN training set based on the weights to more accurately generate the optimal spraying parameters; GAN and GBDT collaborative optimization: GAN generates spraying process parameters GBDT calculates the weights of GAN input parameters Adjusting GAN training data Give more optimization space to input parameters with higher weights, and reduce the GAN learning effort for parameters with lower weights; Update GAN input After multiple rounds of training, the optimal spraying process parameters are finally converged; Furthermore, the specific process of step three is: First, the spraying parameters are adaptively adjusted: The GAN model generates optimal spraying process parameters based on historical data and real-time process feedback: Spraying temperature T s (Unit: °C): Target range: T s ∈[300,500] Too low will result in insufficient PTFE melting, while too high may cause material decomposition or reduced adhesion; GAN optimization formula: in: is the initial spraying temperature, is the learning rate, is the gradient of the effect of temperature on coating quality; Spraying pressure P s (Unit: MPa): Target range: P s ∈[0.5,2.0] Too high will result in uneven coating thickness, too low will affect adhesion; GAN optimization formula: in: is the initial spraying pressure, is the learning rate, is the gradient of the effect of pressure on coating quality; Rotation speed R s (Unit: rpm): Target range: R s ∈[500,2000] The rotation speed affects the coating uniformity and adhesion strength; GAN optimization formula: in: is the initial rotation speed, is the learning rate, is the gradient of the effect of rotation speed on coating quality; In order to ensure the spraying quality, sensors are used for real-time monitoring and automatic adjustment of spraying time or spraying pressure: Set coating thickness target value Target thickness: Sensor measures real-time thickness : like Automatically reduce spray pressure or shorten spray time: in is the adaptive adjustment coefficient; like : Automatically increase spray pressure or extend spray time: Secondly, optimize the multi-layer spraying; GAN can automatically generate multi-layer spraying plans based on process requirements to optimize coating adhesion and corrosion resistance: First layer (adhesion layer): Objective: To improve the bonding strength between PTFE and steel lining and enhance the stability of coating; parameter: Spraying temperature: Spraying pressure: Coating thickness: Adhesion optimization formula: in: is the adhesion force, , , is the empirical coefficient; Second layer (corrosion resistant layer): Goal: Enhance acid and alkali resistance and improve durability; parameter: Spraying temperature: Spraying pressure: Coating thickness: Corrosion resistance optimization formula: in: For corrosion resistance, , , is the empirical coefficient; However, in the actual coating process, there may be local uneven coating thickness, which requires an adaptive compensation mechanism for optimization: GAN adjusts the spray path: The coating thickness sensor monitors the data to determine whether overspray or underspray occurs: Overspray area: Reduce spraying time Under-sprayed area: Increase spraying time Spray path adjustment formula: in: is the current spraying position, Adjustment coefficient Optimization of material utilization: Goal: Improve material utilization to >85% Calculation formula: in: is the material utilization rate, is the actual thickness; like , GAN automatically adjusts the spraying path and process parameters; Furthermore, the specific process of step four is as follows: First, non-contact testing is performed, including coating thickness testing and coating bonding strength testing; coating thickness It is a key factor affecting corrosion resistance and mechanical strength. Usually the uniformity is required to be within the range of ±5μm. For this purpose, a laser thickness gauge is used for non-contact detection. Measurement principle: Using laser triangulation measurement method, the laser thickness gauge emits a laser beam to the coating surface, calculates the displacement difference between the reflected light and the incident light, and thus obtains the thickness ; Calculation formula: in: is the thickness of the reference surface, is the distance from the sensor to the spraying surface, is the distance from the sensor to the reference surface; Quality judgment: like , determine the coating thickness is qualified; like , then you need to readjust the spraying parameters; For unqualified samples, the data is recorded and fed back to GAN for optimization; Coating bonding strength It directly determines whether the PTFE layer is firmly adhered to the steel lining surface. Usually, its adhesion is required to be greater than 10 MPa. The ultrasonic testing method is used to evaluate the bonding strength. Measurement principle: The bonding strength between the coating and the substrate is measured using the ultrasonic pulse echo method. When ultrasonic waves propagate between different media, partial reflection and transmission will occur. The bonding strength can be calculated by measuring the echo signal. Calculation formula: in: is the material characteristic coefficient, is the ultrasonic reflection signal intensity, is the ultrasonic transmission signal intensity Quality determination; like , the bonding strength is judged to be qualified; like , then it is necessary to adjust the spraying process parameters (such as temperature, pressure, etc.); GAN then performs feedback optimization. The optimization goal of the GAN model is to continuously correct the spraying process parameters based on the detection data to improve the quality of the next batch of spraying. GAN optimization mainly includes: Generator (G): takes spraying temperature, pressure, rotation speed, time and other parameters as input to generate a data set that fits the high-quality spraying process; Discriminator (D): trained with real production data, used to evaluate the quality of generated parameters and provide feedback to the generator; GAN optimization objective function: in: is the actual detection data distribution, is the random input parameter distribution, The spraying process parameters output by the generator, The quality assessment result output by the discriminator; GAN optimization strategy: Data input: The collected spraying process parameters and inspection data are input into the GAN model as a training set; Use Gradient Boosting Decision Tree (GBDT) to analyze historical data and improve GAN prediction accuracy; Parameter correction: like If the deviation is too large, GAN adjusts the spraying time or spraying pressure; like Below 10 MPa, GAN adjusts the spraying temperature or rotation speed; Optimized output: After multiple rounds of training, GAN generates new spraying process parameters to guide the next batch of production; Finally, data storage and quality traceability are carried out. The quality traceability system can ensure the stability of the spraying process and support future optimization; Data Storage: All spraying process parameters and test data are recorded in the spraying process database, including: in: is the ambient temperature and humidity, is the spraying process parameter, is the coating thickness, is the bonding strength, is the timestamp; Quality traceability: Batch management: Each batch of spray products is assigned a unique identifier BiB_iBi, and the corresponding process parameters and test results are stored; Abnormal backtracking: If the detection data deviates from the standard range, the spraying parameters can be backtracked through the database and the GAN optimization strategy can be adjusted; Long-term optimization: Through GAN and GBDT joint optimization, a coating quality prediction model is established to improve coating stability; Beneficial effects: The present invention provides a method for optimizing the polytetrafluoroethylene (PTFE) coating process inside a steel liner based on a generative adversarial neural network (GAN) and a gradient boosted decision tree (GBDT). Compared with the existing technology, the present invention has the following significant beneficial effects: Improve coating uniformity and optimize thickness control. The present invention uses a GAN model combined with laser thickness gauge data to intelligently optimize parameters such as spraying temperature, pressure, and rotation speed to control the coating thickness error within ±5μm. At the same time, adaptive spray path adjustment is adopted to automatically correct the spray trajectory based on feedback from the laser thickness gauge to reduce material waste. Traditional coating processes mainly rely on fixed parameters or manual experience for spraying, and it is impossible to adjust the spray parameters in real time, resulting in uneven coating thickness distribution. Localized over-thickness or over-thinness may occur, affecting the corrosion resistance of the coating.

[0006] Improve coating bond strength and enhance durability. Ultrasonic testing combined with GAN optimization ensures optimal matching of spray temperature, pressure, and rotation speed, optimizing coating bond strength and increasing adhesion to over 10 MPa. A multi-layer spray optimization strategy optimizes adhesion in the first layer and corrosion resistance in the second, achieving layer-by-layer optimization. Traditional coating processes, however, cannot adjust coating adhesion parameters in real time during the spraying process, resulting in significant fluctuations in the bond strength between the coating and the steel lining, making it prone to peeling and blistering.

[0007] Reduce material waste and improve production efficiency. Use GAN to optimize the coating path and intelligently adjust the spray pressure, time, and speed to increase material utilization to over 85%, reducing raw material waste.

[0008] By combining historical data with a gradient boosted decision tree (GBDT), spray parameters can be adaptively adjusted, reducing unnecessary trials and improving production efficiency. Traditional spraying processes, however, have fixed parameters and lack optimization based on spraying feedback, leading to severe overspray and underspray, low coating material utilization, and high costs.

[0009] Quality inspection and adaptive optimization improve production consistency. Traditional quality inspection methods typically use manual sampling, which is subject to detection lag and the inability to adjust spray parameters in real time, resulting in unstable batch quality. This invention uses a laser thickness gauge + ultrasonic detection for non-contact quality inspection, real-time monitoring of coating thickness and bonding strength, and optimizes spray parameters through GAN feedback to ensure consistent product quality for each batch. In addition, it also integrates a spray process database to record all spray process parameters and inspection data, enabling long-term traceability and optimization.

[0010] In summary, this invention utilizes a GAN+GBDT optimization method, combined with intelligent data collection and feedback optimization, to achieve multiple optimizations, including improved coating uniformity, enhanced bonding strength, increased material utilization, and enhanced production consistency. This significantly enhances the intelligence, precision, and efficiency of the steel-lined PTFE coating process. Compared to existing technologies, this invention reduces material waste while improving product quality and production efficiency, possessing high industrial application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a flow chart of an intelligent optimization algorithm for the coating process of polytetrafluoroethylene inside steel lining.

[0012] Figure 2 : Schematic diagram of spray path optimization DETAILED DESCRIPTION

[0013] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0014] The specific contents of the present invention are further explained below with reference to the embodiments.

[0015] Example 1 1. Sensor installation and data collection During the polytetrafluoroethylene (PTFE) coating process inside the reactor's steel lining, sensors must first be installed to collect real-time environmental and process parameters. According to the document, the sensor installation locations and parameter ranges are as follows: Ambient temperature sensor (Te): installed at the four corners of the spray room, the parameter range is [10, 50] ° C; Ambient humidity sensor (He): installed on the walls of the coating workshop and around the coating equipment, with a parameter range of [20,80]%; Spraying pressure sensor (Ps): installed near the spray gun, the parameter range is [0.5, 2] MPa; Rotation speed sensor (Rs): installed at the axis of the coating rotating device, the parameter range is [500, 2000]rpm; Infrared temperature sensor (Tc): installed on both sides of the spray track in the heating area, with a parameter range of [300, 50]°C; Laser thickness gauge (Dt): installed above the coated target surface in the spraying area, with a parameter range of [10, 200] μm.

[0016] Assume that during a batch coating process, the following data is collected: 2. Data Preprocessing The collected data needs to be processed for outlier detection, filtering and normalization to ensure data quality.

[0017] Outlier detection: Use the 3σ criterion to eliminate abnormal data. Assuming the mean of the data is μ and the standard deviation is σ, eliminate data outside the range of μ ± 3σ.

[0018] For ambient temperature Te: Mean μ = 25.33°C Standard deviation σ = 0.0577°C Normal range: 25.33 ± 0.1731°C, or [25.1569, 25.5031]°C Since the collected ambient temperature data are all within the normal range, there is no need to eliminate them.

[0019] Data filtering: Kalman filtering is used to remove noise. Assume the state equation is: in: is the current state, is the control input, is the noise term; the predicted value is dynamically adjusted through the Kalman gain to improve data accuracy.

[0020] Data normalization: Use maximum and minimum normalization to scale the data to [0, 1]. For example, for the ambient temperature Te: in: is the original data, , , therefore, 25.3°C is normalized to: 3. GAN and GBDT Optimization According to historical data, temperature has the greatest impact on coating uniformity (weight 0.35), spray pressure has a significant impact on bond strength (weight 0.25), and rotation speed affects material utilization (weight 0.20). During GAN training, the temperature range was adjusted to [350, 450]°C, the spray pressure range was [1.2, 1.8] MPa, and the rotation speed range was [800, 1600] rpm.

[0021] GAN generator (G): Generates spraying process parameters. Input variables include heating temperature Tc, spraying pressure Ps, rotation speed Rs and spraying time t s .

[0022] The mathematical description of the generator is: in: represents random noise from a standard normal distribution; The loss function of the generator is: in: Represents the discriminator's evaluation of the generated parameters. The goal is to make the discriminator unable to distinguish between generated data and real data; The discriminator (D) is used to distinguish whether the spraying parameters generated by GAN are reasonable. Its input includes: GAN-generated spraying parameters Spraying parameters based on real production data Coating quality evaluation index: coating uniformity U t (Unit: μm, target error range: ±5 μm), bonding strength S t (Unit: MPa, target value> 10MPa), material utilization rate M u (target value>85%); Mathematical description of the discriminator: The goal of the discriminator is to judge the authenticity of the input data by learning the distribution of real spray data: in: is the input data (GAN generated or real spray parameters), , is the discriminator weight parameter, represents the Sigmoid activation function; The loss function of the discriminator is: Among them: the first term represents the classification loss of real data, and the second term represents the classification loss of GAN-generated data; through adversarial training of the constructed GAN model, the generator (G) attempts to generate parameter combinations that are closer to the real spray data, and the discriminator (D) distinguishes the GAN-generated parameters from the real parameters and feeds them back to the generator for optimization.

[0023] GBDT optimization: Using historical coating data, we analyze the impact of temperature, pressure, speed, and time on coating uniformity, bonding strength, and material utilization, assigning different weights to the GAN input parameters. Based on these weights, we adjust the GAN training set to more accurately generate optimal spray parameters. For example, we assign a weight of 0.35 to the temperature effect, 0.25 to the spray pressure effect, and 0.20 to the rotation speed effect.

[0024] 4. Spraying process execution After GAN and GBDT optimization, the optimal output parameters are: Heating temperature Tc = 420°C Spraying pressure Ps = 1.5 MPa Rotation speed Rs = 1400 rpm Spraying time t s = 30 s This parameter combination was used in the spraying process to improve coating uniformity, increase the bonding strength to 12 MPa, and achieve a material utilization rate of 89%.

[0025] 5. Quality inspection and feedback optimization During the spraying process, non-contact quality inspection is carried out using laser thickness gauges and ultrasonic detectors.

[0026] Coating thickness detection: The target thickness is 50±5μm. The collected thickness data is as follows: For unqualified samples, the data is recorded and fed back to GAN for optimization.

[0027] Bond strength test: The target bond strength is 10 MPa. The collected bond strength data are as follows: For unqualified samples, the spraying process parameters are further optimized in combination with the GAN model.

[0028] 6. Final output optimization results Following these steps, the GAN model, after multiple rounds of training, generated new spraying process parameters to guide the next batch of production. The final optimization results: coating thickness error was controlled within ±5μm, material utilization increased to 89%, and bond strength increased to 12 MPa, meeting the target requirements. After optimizing the spraying process parameters, coating uniformity and bond strength were significantly improved, material waste was reduced, and production efficiency was increased.

[0029] Example 2 1. Sensor installation and data collection During the polytetrafluoroethylene (PTFE) coating process inside the reactor's steel lining, sensors must first be installed to collect real-time environmental and process parameters. According to the document, the sensor installation locations and parameter ranges are as follows: Ambient temperature sensor (Te): installed at the four corners of the spray booth, with a parameter range of [10, 50]°C; Environmental humidity sensor (He): installed on the walls of the coating workshop and around the coating equipment, with a parameter range of [20,80]; Spray pressure sensor (Ps): installed near the spray gun, with a parameter range of [0.5, 2] MPa; Rotation speed sensor (Rs): installed at the axis of the coating rotating device, the parameter range is [500, 2000] rpm; Infrared temperature sensor (Tc): installed on both sides of the spray track in the heating area, with a parameter range of [300, 500]°C; Laser thickness gauge (Dt): installed above the coated target surface in the spraying area, with a parameter range of [10, 200] μm.

[0030] Assume that during a batch coating process, the following data is collected: 2. Data Preprocessing The collected data needs to be processed for outlier detection, filtering and normalization to ensure data quality.

[0031] Outlier detection: Use the 3σ criterion to eliminate abnormal data. Assuming the mean of the data is μ and the standard deviation is σ, eliminate data outside the range of μ ± 3σ.

[0032] For ambient temperature Te: Mean μ = 26.53°C Standard deviation σ = 0.0577°C Normal range: 26.53±0.1731°C, i.e. [26.3569, 26.7031]°C. Since the collected ambient temperature data are all within the normal range, there is no need to exclude them.

[0033] Data filtering: Kalman filtering is used to remove noise. Assume the state equation is: in: is the current state, is the control input, is the noise term; the predicted value is dynamically adjusted through the Kalman gain to improve data accuracy.

[0034] Data normalization: Use maximum and minimum normalization to scale the data to [0, 1]. For example, for the ambient temperature Te: in: is the original data, , , therefore, 26.5°C is normalized to: 3. GAN and GBDT Optimization According to historical data, temperature has the greatest impact on coating uniformity (weight 0.35), spray pressure has a significant impact on bond strength (weight 0.25), and rotation speed affects material utilization (weight 0.20). During GAN training, the temperature range was adjusted to [350, 450]°C, the spray pressure range was [1.2, 1.8] MPa, and the rotation speed range was [800, 1600] rpm.

[0035] GAN generator (G): Generates spraying process parameters. Input variables include heating temperature Tc, spraying pressure Ps, rotation speed Rs and spraying time t s .

[0036] The mathematical description of the generator is: in: represents random noise from a standard normal distribution; The loss function of the generator is: in: Represents the discriminator's evaluation of the generated parameters. The goal is to make the discriminator unable to distinguish between generated data and real data; The discriminator (D) is used to distinguish whether the spraying parameters generated by GAN are reasonable. Its input includes: GAN-generated spraying parameters Spraying parameters based on real production data Coating quality evaluation index: coating uniformity U t (Unit: μm, target error range: ±5 μm), bonding strength S t (Unit: MPa, target value> 10MPa), material utilization rate M u (target value>85%); Mathematical description of the discriminator: The goal of the discriminator is to judge the authenticity of the input data by learning the distribution of real spray data: in: is the input data (GAN generated or real spray parameters), , is the discriminator weight parameter, represents the Sigmoid activation function; The loss function of the discriminator is: Among them: the first term represents the classification loss of real data, and the second term represents the classification loss of GAN-generated data; through adversarial training of the constructed GAN model, the generator (G) attempts to generate parameter combinations that are closer to the real spray data, and the discriminator (D) distinguishes the GAN-generated parameters from the real parameters and feeds them back to the generator for optimization.

[0037] Through adversarial training on the constructed GAN model, the generator (G) attempts to generate parameter combinations that are closer to the real spray data. The discriminator (D) distinguishes the GAN-generated parameters from the real parameters and feeds them back to the generator for optimization.

[0038] GBDT optimization: Using historical coating data, we analyze the impact of temperature, pressure, speed, and time on coating uniformity, bonding strength, and material utilization, assigning different weights to the GAN input parameters. Based on these weights, we adjust the GAN training set to more accurately generate optimal spray parameters. For example, we assign a weight of 0.35 to the temperature effect, 0.25 to the spray pressure effect, and 0.20 to the rotation speed effect.

[0039] 4. Spraying process execution After GAN and GBDT optimization, the optimal output parameters are: Heating temperature Tc = 430°C Spraying pressure Ps = 1.6MPa Rotation speed Rs = 1450 rpm Spraying time t s = 35s This parameter combination is used in the spraying process to improve the coating uniformity, increase the bonding strength to 13 MPa, and achieve a material utilization rate of 90%.

[0040] 5. Quality inspection and feedback optimization During the spraying process, non-contact quality inspection is carried out using laser thickness gauges and ultrasonic detectors.

[0041] Coating thickness detection: The target thickness is 60±5μm. The collected thickness data is as follows: For unqualified samples, the data is recorded and fed back to GAN for optimization.

[0042] Binding strength test: The target bonding strength is 12 MPa, and the collected bonding strength data are as follows: For unqualified samples, the spraying process parameters are further optimized in combination with the GAN model.

[0043] 6. Final output optimization results Through the above steps, after multiple rounds of training, the GAN model generates new spraying process parameters to guide the next batch of production. The final optimization result is: Through GAN model optimization, the coating thickness error was controlled within ±5μm, material utilization increased to 91%, and bond strength increased to 13.5 MPa. After optimizing the spraying process parameters, coating uniformity and bond strength were significantly improved, material waste was reduced, and production efficiency was increased.

[0044] This invention achieves intelligent, precise, and efficient PTFE coating of steel linings through the collaborative optimization of GAN and GBDT, combined with intelligent data collection and feedback optimization. Compared to traditional coating processes, this invention significantly improves coating quality and production efficiency while reducing material waste, possessing high industrial application value.

Claims

1. A method for optimizing the polytetrafluoroethylene (PTFE) coating process inside a steel liner based on a generative adversarial neural network (GAN) and a gradient boosted decision tree (GBDT), characterized in that: The method comprises the following steps: Step 1: Install sensors to collect data, including ambient temperature, humidity, spraying pressure, rotation speed, heating temperature, coating thickness, etc. Step 2: Use GAN combined with GBDT to optimize coating process parameters and improve the prediction accuracy of coating parameters generated by GAN; Step 3: Use the optimal parameters generated by GAN to execute the spraying process and combine it with the adaptive compensation mechanism to optimize the spraying path; Step 4: Perform quality inspection through sensors and feed the data back to the GAN model for adaptive optimization.

2. The method according to claim 1, characterized in that The sensor is connected to the central control system via a wireless transmission module to improve the real-time performance and flexibility of data collection. The specific process of step 1 is as follows: During the optimization process of polytetrafluoroethylene (PTFE) coating inside steel liners, to ensure data accuracy and stability, it is necessary to install high-precision sensors in key process links to collect production environment and process parameters in real time and provide reliable input data for the Generative Adversarial Network (GAN) optimization model. Specifically, this step can be broken down into the following parts: First, the sensor installation: Installing environmental parameter sensors: Environmental parameters have a significant impact on PTFE coating quality. For example, high humidity may lead to reduced adhesion. Therefore, the following sensors need to be placed around the coating equipment: Ambient temperature sensor T e Parameter range: T e ∈[10,50]°C Function: Monitor the production environment temperature to ensure that the coating process is carried out within the optimal temperature range; Installation location: the four corners of the spray room to reduce the impact of local temperature gradients; Environmental humidity sensor H e Parameter range: H e ∈[20,80]% Function: Monitor the effect of humidity on PTFE melting and adhesion, and ensure that process parameters are adapted to different humidity conditions; Installation location: coating workshop walls and around coating equipment; Installing process parameter sensors: In order to monitor the key variables of the coating process in real time, the following process parameter sensors are required: Spraying pressure sensor P s Parameter range: P s ∈[0.5,2]MP Function: Monitor the spraying pressure to ensure that the PTFE particles adhere to the steel lining surface under the optimal kinetic energy, and avoid excessive pressure causing particle rebound or too little pressure causing uneven coating; Installation location: near the spray gun, the nozzle outlet pressure is detected by a high-precision sensor; Rotational speed sensor R s Parameter range: R s ∈[500,2000]rpm Function: Monitor the rotation speed, optimize the coating thickness uniformity, and avoid uneven coating caused by unstable rotation speed; Installation location: Coating the axis of the rotating device, using a non-contact speed measuring device to improve accuracy; Infrared temperature sensor T c Parameter range: T c ∈[300,500]°C Function: Measure the heating temperature to ensure that the PTFE coating melts and spreads evenly on the steel lining surface; Installation location: on both sides of the spray track in the heating area, using infrared sensors for non-contact temperature measurement to ensure accurate temperature measurement; Laser Thickness Gauge D t Parameter range: D t ∈[10,200]μm Function: Real-time monitoring of coating thickness to ensure uniform thickness and avoid excessive thickness or thinness that affects anti-corrosion performance; Installation location: Above the coating target surface in the spraying area, using the laser measurement principle to achieve non-destructive measurement; Then perform data collection and preprocessing: The process parameters collected by the sensor need to be cleaned and preprocessed to improve data quality and ensure that the GAN optimization model can accurately identify the optimal coating parameters; All installed sensors are connected to the central control system through the Industrial Internet of Things (IIoT) system, collecting data in real time and storing it in the database at a fixed frequency; Acquisition frequency f c : Environmental parameter sensors: (sampled once per second) Process parameter sensors: (sampled every 0.1 seconds) The collected data set is defined as follows: in: :Ambient temperature (°C), :Ambient humidity (%), :Spraying pressure (MPa), :Rotation speed (rpm), :Heating temperature (°C), :Coating thickness (μm), : time(s); The raw data collected by the sensor may contain noise or outliers and require preprocessing: Outlier detection: Use the 3σ criterion to eliminate abnormal data: in, is the data mean, is the standard deviation, and the abnormal data beyond the range are eliminated; Data filtering: Kalman filter is used to remove noise and improve measurement stability. The filtering equation is as follows: in: is the current state, is the control input, is the noise term; Dynamically adjust the prediction value through Kalman gain to improve data accuracy; Data normalization: Since different parameters have different dimensions, maximum and minimum normalization is used: in: is the original data, , are the minimum and maximum values ​​in the data set; Normalized data Between 0 and 1, improving the stability of GAN training.

3. The method according to claim 1, characterized in that Gradient boosted decision tree (GBDT) is used to analyze historical coating data and adjust the weights of GAN input parameters to improve prediction accuracy. The specific process of step 2 is as follows: In this paper, in order to optimize the coating process parameters of polytetrafluoroethylene (PTFE) inside the steel liner, a genetic adversarial network (GAN) is used to generate the coating parameters, and a gradient boosted decision tree (GBDT) is introduced to optimize the GAN input parameters, thereby improving the coating uniformity and bonding strength, while optimizing material usage and production efficiency; First, a GAN model is constructed. The GAN generator (G) is used to generate spraying process parameters. Its input variables include: temperature (Heating temperature, unit: °C) pressure (Spraying pressure, unit: MPa) Rotation speed (Speed ​​of coating rotating device, unit: rpm) time (Spraying time, unit: seconds) Mathematical description of GAN generator: Generator Receive a random noise vector (usually obeys a normal distribution) and maps it to the spraying process parameters through a deep neural network in: is the spray parameter set generated by GAN, (0,1) represents random noise from a standard normal distribution; The loss function of the generator is: in: Represents the discriminator's evaluation of the generated parameters. The goal is to make the discriminator unable to distinguish between generated data and real data; The discriminator (D) is used to distinguish whether the spraying parameters generated by GAN are reasonable. Its input includes: GAN-generated spraying parameters Spraying parameters based on real production data Coating quality evaluation index: coating uniformity U t (Unit: μm, target error range: ±5 μm), bonding strength S t (Unit: MPa, target value> 10MPa), material utilization rate M u (target value>85%); Mathematical description of the discriminator: The goal of the discriminator is to judge the authenticity of the input data by learning the distribution of real spray data: in: is the input data (GAN generated or real spray parameters), , is the discriminator weight parameter, represents the Sigmoid activation function; The loss function of the discriminator is: Among them: the first term represents the classification loss of real data, and the second term represents the classification loss of GAN-generated data; Then, adversarial training is performed through the constructed GAN model. GAN continuously optimizes the spraying process parameters through adversarial training: The generator (G) attempts to generate parameter combinations that are closer to real spray data to improve coating uniformity and bonding strength. The discriminator (D) distinguishes the GAN-generated parameters from the real parameters and feeds them back to the generator for optimization. The generator generates a set of spraying parameters G(z), the discriminator calculates D(G(z)) and gives feedback, and optimizes it through gradient descent. and , until GAN ​​training converges; After that, the gradient boosted decision tree (GBDT) is combined to optimize the GAN input parameters; Gradient boosted decision tree (GBDT) is used to analyze spraying history data, adjust the input variable weights of GAN, and improve the prediction accuracy of the model; The goal of GBDT is to analyze the effects of temperature, pressure, speed, and time on coating uniformity, bonding strength, and material utilization using historical coating data, assigning different weights to GAN input parameters to improve the accuracy of GAN-generated parameters. GBDT calculation formula: GBDT uses multiple decision trees Perform optimization, each round of iteration: in: is the model prediction value of the kth iteration, is the decision tree model for this round, is the learning rate (usually 0.01-0.1); The loss function uses the mean square error (MSE): in: It is the coating quality indicator of the real spraying parameters. are the input parameters (temperature, pressure, speed, time); GBDT optimization effect: Calculate the influence weight of each parameter on coating quality: Temperature impact weight: Pressure impact weight: Speed ​​influence weight: Time impact weight: Adjust the GAN training set based on the weights to more accurately generate the optimal spraying parameters; GAN and GBDT collaborative optimization: GAN generates spraying process parameters GBDT calculates the weights of GAN input parameters Adjusting GAN training data Give more optimization space to input parameters with higher weights, and reduce the GAN learning effort for parameters with lower weights; Update GAN input After multiple rounds of training, the optimal spraying process parameters are finally converged.

4. The method according to claim 1, wherein The spraying system includes a multi-layer spraying control strategy, in which GAN can generate single-layer or multi-layer spraying parameters and dynamically adjust the spraying rhythm to optimize bonding strength and corrosion resistance. The specific process of step three is as follows: First, the spraying parameters are adaptively adjusted: The GAN model generates optimal spraying process parameters based on historical data and real-time process feedback: Spraying temperature T s (Unit: °C): Target range: T s ∈[300,500] Too low will result in insufficient PTFE melting, while too high may cause material decomposition or reduced adhesion; GAN optimization formula: in: is the initial spraying temperature, is the learning rate, is the gradient of the effect of temperature on coating quality; Spraying pressure P s (Unit: MPa): Target range: P s ∈[0.5,2.0] Too high will result in uneven coating thickness, too low will affect adhesion; GAN optimization formula: in: is the initial spraying pressure, is the learning rate, is the gradient of the effect of pressure on coating quality; Rotation speed R s (Unit: rpm): Target range: R s ∈[500,2000] The rotation speed affects the coating uniformity and adhesion strength; GAN optimization formula: in: is the initial rotation speed, is the learning rate, is the gradient of the effect of rotation speed on coating quality; In order to ensure the spraying quality, sensors are used for real-time monitoring and automatic adjustment of spraying time or spraying pressure: Set coating thickness target value Target thickness: Sensor measures real-time thickness : like Automatically reduce spray pressure or shorten spray time: in is the adaptive adjustment coefficient; like : Automatically increase spray pressure or extend spray time: Secondly, multi-layer spraying is optimized; GAN can automatically generate multi-layer spraying plans based on process requirements to optimize coating adhesion and corrosion resistance: First layer (adhesion layer): Objective: To improve the bonding strength between PTFE and steel lining and enhance coating stability; parameter: Spraying temperature: Spraying pressure: Coating thickness: Adhesion optimization formula: in: is the adhesion force, , , is the empirical coefficient; Second layer (corrosion resistant layer): Goal: Enhance acid and alkali resistance and improve durability; parameter: Spraying temperature: Spraying pressure: Coating thickness: Corrosion resistance optimization formula: in: For corrosion resistance, , , is the empirical coefficient; However, in the actual coating process, there may be local uneven coating thickness, which requires an adaptive compensation mechanism for optimization: GAN adjusts the spray path: The coating thickness sensor monitors the data to determine whether overspray or underspray occurs: Overspray area: Reduce spraying time Under-sprayed area: Increase spraying time Spray path adjustment formula: in: is the current spraying position, Adjustment coefficient Optimization of material utilization: Goal: Improve material utilization to >85% Calculation formula: in: is the material utilization rate, is the actual thickness; like , GAN automatically adjusts the spraying path and process parameters.

5. The method according to claim 1, wherein An adaptive loss function is used in the GAN optimization process to balance material usage and coating performance requirements, thereby improving industrial application value. The specific process of step 4 is as follows: First, non-contact testing is performed, including coating thickness testing and coating bonding strength testing; coating thickness It is a key factor affecting corrosion resistance and mechanical strength. Usually the uniformity is required to be within the range of ±5μm. For this purpose, a laser thickness gauge is used for non-contact detection. Measurement principle: Using laser triangulation measurement method, the laser thickness gauge emits a beam of laser to the coating surface, calculates the displacement difference between the reflected light and the incident light, and thus obtains the thickness ; Calculation formula: in: is the thickness of the reference surface, is the distance from the sensor to the spraying surface, is the distance from the sensor to the reference surface; Quality judgment: like , determine the coating thickness is qualified; like , then you need to readjust the spraying parameters; For unqualified samples, the data is recorded and fed back to GAN for optimization; Coating bonding strength It directly determines whether the PTFE layer is firmly adhered to the steel lining surface. Usually, its adhesion is required to be greater than 10MPa. The ultrasonic testing method is used to evaluate the bonding strength. Measurement principle: The bonding strength between the coating and the substrate is measured using the ultrasonic pulse echo method. When ultrasonic waves propagate between different media, partial reflection and transmission will occur. The bonding strength can be calculated by measuring the echo signal. Calculation formula: in: is the material characteristic coefficient, is the ultrasonic reflection signal intensity, is the ultrasonic transmission signal intensity Quality determination; like , the bonding strength is judged to be qualified; like , then it is necessary to adjust the spraying process parameters (such as temperature, pressure, etc.); GAN then performs feedback optimization. The optimization goal of the GAN model is to continuously correct the spraying process parameters based on the detection data to improve the quality of the next batch of spraying. GAN optimization mainly includes: Generator (G): takes spraying temperature, pressure, rotation speed, time and other parameters as input to generate a data set that fits the high-quality spraying process; Discriminator (D): trained with real production data, used to evaluate the quality of generated parameters and provide feedback to the generator; GAN optimization objective function: in: is the actual detection data distribution, is the random input parameter distribution, The spraying process parameters output by the generator, The quality assessment result output by the discriminator; GAN optimization strategy: Data input: The collected spraying process parameters and inspection data are input into the GAN model as a training set; Use Gradient Boosting Decision Tree (GBDT) to analyze historical data and improve GAN prediction accuracy; Parameter correction: like If the deviation is too large, GAN adjusts the spraying time or spraying pressure; like Below 10 MPa, GAN adjusts the spraying temperature or rotation speed; Optimized output: After multiple rounds of training, GAN generates new spraying process parameters to guide the next batch of production; Finally, data storage and quality traceability are carried out. The quality traceability system can ensure the stability of the spraying process and support future optimization; Data Storage: All spraying process parameters and test data are recorded in the spraying process database, including: in: is the ambient temperature and humidity, is the spraying process parameter, is the coating thickness, is the bonding strength, is the timestamp; Quality traceability: Batch management: Each batch of spray products is assigned a unique identifier BiB_iBi, and the corresponding process parameters and test results are stored; Abnormal backtracking: If the detection data deviates from the standard range, the spraying parameters can be backtracked through the database and the GAN optimization strategy can be adjusted; Long-term optimization: Through GAN and GBDT joint optimization, a coating quality prediction model is established to improve coating stability.

6. The method according to claims 1 to 5, characterized in that This method can be applied to the intelligent spray optimization of steel-lined PTFE anti-corrosion pipes, fluorine-lined storage tanks, valves, compensators and other equipment to improve product consistency and industrial application value.

Images