Thermal spraying coating preparation process parameter optimization method based on machine learning

Abstract

The invention relates to a thermal spraying coating preparation process parameter optimization method based on machine learning so as to obtain a high-quality thermal spraying coating. According to the method, a small-sample-size process parameter data set is constructed through high-throughput preparation of the thermal spraying coating, the porosity and long-term soaking corrosion performance of the coating serve as output parameters, and the relevance between the coating preparation process parameters and the porosity and corrosion resistance of the coating is established through a machine learning method; furthermore, more excellent thermal spraying coating preparation process parameters can be predicted in a wider process parameter range, and a high-quality coating can be obtained. The method can be widely applied to rapid optimization of a thermal spraying coating preparation process and stable control of coating quality.

Description

Technical Field

[0001] The present invention relates to the field of thermal spraying technology, and in particular to a method for optimizing process parameters of thermal spraying coating preparation based on machine learning. Background Art

[0002] Thermal spraying is a commonly used surface coating technology. Through the spraying equipment, the molten or semi-molten coating material is sprayed onto the surface of the component or part to be protected to form a protective, functional or decorative coating. Thermal spraying technology is usually used to enhance the surface properties of materials, such as improving their wear resistance and corrosion resistance, so it has extremely high requirements for the coating structure and quality. Although people have a deep understanding of the microscopic process of thermal spray coating formation, that is, the heated and accelerated powder particles hit the substrate material at a certain temperature and speed, and then the spreading and solidification of the particles will determine the degree of particle flattening and the final morphology, thereby affecting the microstructure of the coating. However, the actual spraying process is the deposition behavior of a large number of powder particles of different particle sizes. The different diameters of the powder particles will lead to different spreading and solidification dynamics. Therefore, the actual formation process of thermal spray coatings is very complicated, and it is difficult to theoretically predict its deposition process and coating quality. In addition, there are many process parameters in the thermal spraying process, and the various spraying process parameters cooperate with each other and have complex relationships, which will affect the speed and temperature before particle deposition, and thus affect the coating quality. Therefore, in order to reduce time and economic costs, a fast, efficient and accurate method for optimizing process parameters for thermal spray coating preparation is urgently needed.

[0003] Among the process parameter optimization methods for thermal spray coatings, the traditional trial and error method is the most direct method. This method analyzes the changes in the structure and performance of the coating by continuously trying different combinations of process parameters, and gradually finds the optimal process parameters. The advantage of this method is that it is simple to operate, but the disadvantage is that it consumes a lot of time and economic costs, is inefficient, and may not be able to find the optimal process parameters. In addition, people will also set one or several groups of process parameters based on personal experience and professional knowledge, and then conduct experimental verification. This method does not fully consider the impact of all process parameters, which makes it difficult to screen out the optimal process parameters, and thus it is impossible to obtain high-quality coatings. Based on this, the present invention proposes a method for optimizing the process parameters of thermal spray coating preparation based on machine learning, which can efficiently obtain excellent coating preparation process parameters, further reduce the porosity of the coating, and improve the quality and performance of the coating. Summary of the invention

[0004] In view of the problems existing in the background technology, the purpose of the present invention is to provide a method for optimizing the process parameters of thermal spray coating preparation based on machine learning, which can obtain the relationship between the process parameters of thermal spray coating preparation and the coating quality, and can optimize the optimal spraying process parameters, so as to quickly, efficiently and accurately prepare high-quality and high-performance thermal spray coatings, and comprehensively cover the coating quality control problems under many thermal spray process parameters.

[0005] In order to achieve the above object, the present invention adopts the following technical solution: a method for optimizing process parameters of thermal spray coating preparation based on machine learning, comprising the following steps:

[0006] S1. Perform variable parameter high-throughput preparation of coating samples;

[0007] S2. Test and analyze the porosity of the coating specimens, and establish the corresponding relationship between the coating preparation process parameters and the coating porosity; and construct a data set of process parameters and coating porosity;

[0008] S3. Test and analyze the corrosion performance of the coating specimens to obtain the corrosion performance under the coating preparation process parameter conditions; and construct a process parameter and corrosion performance data set;

[0009] S4. Using at least one machine learning model, based on the process parameter and coating porosity data set, the machine learning model is trained and tested on the correlation between the spraying process parameters and the coating porosity, and a process parameter and coating porosity relationship model is obtained; based on the process parameter and corrosion performance data set, the machine learning model is trained and tested on the correlation between the spraying process parameters and the corrosion performance, and a process parameter and corrosion performance relationship model is obtained;

[0010] S5. Input the spraying process parameters after epitaxy and expansion into the trained process parameter and coating porosity relationship model to predict several coating preparation process parameters with the lowest coating porosity; input the spraying process parameters after epitaxy and expansion into the trained process parameter and corrosion performance relationship model to predict several coating preparation process parameters with the best corrosion performance, and take the intersection of the coating preparation process parameters with the lowest porosity and the coating preparation process parameters with the best corrosion performance to obtain the optimal coating preparation process parameters.

[0011] In step S1, the variable parameter high-throughput preparation of the coating sample comprises the following steps:

[0012] A plurality of movable spraying templates are arranged so that the plurality of spraying templates are located at different spraying distances relative to the same spray gun; by changing the spraying distance, a plurality of coating specimens with different spraying distances but the same other coating preparation process parameters can be obtained by a single spraying.

[0013] The coating preparation process parameters include spraying distance, fuel pressure, spray gun traverse rate, powder feeding rate, and spray gun stepping.

[0014] In step S2, the porosity of the coating specimen is tested and analyzed, specifically: an optical microscope or a scanning electron microscope is used to take a cross-sectional photograph of the coating sample, and the coating porosity is obtained by image processing.

[0015] In step S3, the corrosion performance of the coating specimen is tested and analyzed, specifically:

[0016] The corrosion resistance of the coating was evaluated by immersion corrosion test or long-term electrochemical impedance spectroscopy test with a duration of more than 1000 hours.

[0017] In step S4, the training of the process parameter and coating porosity model and the process parameter and corrosion performance model is as follows:

[0018] For the process parameter and coating porosity model: take the process parameter as the input parameter and the porosity value as the output parameter, use at least one of the Autogluon, TPOT and FLAML algorithms to train and test the model, and select the model with the best result as the process parameter and coating porosity relationship model to establish the relationship between the coating preparation process parameters and the coating porosity;

[0019] For the process parameter and corrosion performance model: take the process parameters as input parameters and the corrosion resistance as output parameters, use at least one of the Autogluon, TPOT and FLAML algorithms for model training and testing, and select the model with the best result as the process parameter and corrosion performance relationship model to establish the relationship between the coating preparation process parameters and the coating corrosion performance.

[0020] In step S5, the optimal coating preparation process parameters are predicted, and the set range of one or more spraying process parameters is extended and expanded to obtain more spraying process parameters. Based on the trained process parameter and coating porosity relationship model, one or several coating preparation processes with the lowest coating porosity are predicted; based on the trained process parameter and corrosion performance relationship model, one or several coating preparation processes with the best corrosion performance are predicted.

[0021] In step S5, the top 1‰ coating preparation process parameters with the lowest coating porosity are predicted by using the trained process parameter and coating porosity relationship model; the top 1‰ coating preparation process parameters with the best corrosion performance are predicted by using the trained process parameter and corrosion performance relationship model.

[0022] After the prediction is completed in step S5, the porosity of the coating prepared by the predicted spraying process parameters is verified, specifically:

[0023] The coating porosity is determined by taking cross-sectional photographs of the coating sample using an optical microscope or a scanning electron microscope, and obtaining the area percentage of the coating pores by image processing to obtain the coating porosity; or by using an X-ray three-dimensional imaging system to perform three-dimensional reconstruction and analysis on the coating sample, and finally obtaining the volume fraction of the coating pores.

[0024] After the prediction is completed in step S5, the corrosion performance of the coating prepared by the predicted spraying process parameters is verified, specifically: the corrosion performance verification test adopts an immersion corrosion experiment or an electrochemical impedance spectroscopy long-term test with a duration greater than 1000 hours.

[0025] In one embodiment, the specific steps for variable parameter high-throughput preparation of coating samples are as follows: using self-developed thermal spraying tooling, it is possible to obtain multiple coating samples with different spraying distances (other spraying process parameters are the same) in a single spraying, which can greatly reduce the spraying time and achieve high-throughput preparation of coating samples.

[0026] In one embodiment, the specific steps for analyzing the porosity of the coating specimens are as follows: the coating specimens obtained under the above-mentioned spraying process conditions are cut into small pieces and inlaid, ground and polished, and then the surface or cross-sectional photographs of the coating are obtained by an optical microscope, a scanning electron microscope, a laser confocal microscope, etc. The two-dimensional area percentage of the pores in the coating cross section is obtained by image analysis, or the volume fraction of the coating pores is obtained by X-ray three-dimensional imaging analysis, so as to obtain the relationship between the coating porosity and the spraying process parameters.

[0027] In one embodiment, the specific steps of the corrosion performance test of the coating specimen are as follows: all the coating samples prepared by high throughput are cut into small samples and numbered, and the edges are sealed with sealant to expose the middle part of the coating surface to be tested; for the immersion corrosion test, each sealed coating sample is placed in a separate plastic container containing a corrosive medium, and photographs are taken at regular intervals to record the corrosion morphology of the coating surface; the time when the rust area on the surface of each coating sample reaches a certain area fraction is counted, and a chart is drawn to obtain the relationship between the immersion corrosion performance of the coating and the spraying process parameters.

[0028] In one embodiment, the specific steps of machine learning the correlation between spraying process parameters and coating porosity and immersion corrosion performance are as follows: based on the above-mentioned coating porosity and immersion corrosion days results, the model is trained and tested using AutoGluon, FLAML and TPOT algorithms to establish a spraying process parameter-porosity model and a spraying process parameter-corrosion days model; in order to reduce the overfitting caused by sample division, different training set and test set ratios are defined, substituted into AutoGluon, FLAML and TPOT algorithms, and then the best training set and test set ratios are selected to improve the accuracy of the prediction model; based on the above-mentioned relationship model, spraying process parameters with lower coating porosity and better corrosion resistance are predicted in a wider range of spraying process parameters.

[0029] In one embodiment, the specific steps for verifying the predicted spraying process are as follows: verify the optimal process (lowest porosity), medium process (medium porosity) and worst process (highest porosity) predicted within a wider range of spraying process parameters. Use these three spraying process parameters to prepare an amorphous alloy coating, and compare the porosity and corrosion resistance of the coating prepared with the optimal coating process in the data set. The comparison of porosity uses image analysis or three-dimensional imaging analysis, and the corrosion performance test uses immersion corrosion test or long-term electrochemical impedance analysis. Determine the optimal coating preparation process based on the accuracy of the verification results.

[0030] The thermal spray coating preparation process optimization method based on high-throughput coating preparation and machine learning of the present invention can quickly obtain the optimal process parameters for coating preparation for a certain spraying material. Its beneficial effects are as follows: (1) On the basis of the traditional trial-and-error method for screening thermal spraying process methods, the screening efficiency is improved, saving time and labor costs; (2) It can explore more excellent spraying process parameters within a larger spraying process range and can be flexibly applied according to actual conditions; (3) The coating prepared according to the screened process has good target performance; (4) It is suitable for optimization of a variety of different thermal spray coating preparation processes and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of high-throughput preparation of thermally sprayed Fe-based amorphous alloy coating in the present invention and a physical picture of a self-made high-throughput thermal spraying tooling;

[0032] Figure 2The morphology and structure characterization of the Fe-based amorphous alloy powder and the typical amorphous alloy coating prepared by high throughput in the present invention: (a) powder morphology photo, (b) powder particle size distribution, (c) scanning electron microscope (SEM) photo of the coating cross section, (d) transmission electron microscope (TEM) photo of the coating, (e) X-ray diffraction pattern of the powder and the coating, and (f) differential scanning calorimetry analysis pattern of the powder and the coating.

[0033] Figure 3 Actual photos of the Fe-based amorphous alloy coating prepared by high throughput in the present invention and the analysis results of the coating porosity: (a) a photo of the coating sample after cutting and edge sealing, (b) a SEM photo of the corresponding coating sample, (c) and (d) are photos of the SEM photo before and after binarization processing using ImageJ software when calculating the coating porosity by image analysis method, and (e) the porosity result of the corresponding coating.

[0034] Figure 4 The experimental photos of the Fe-based amorphous alloy coating prepared by the high-throughput method of the present invention being immersed in a 3.5wt.% NaCl solution for a long time are shown as follows: (a) a photo of the coating sample immersed for 1 day, (b) a photo of the coating sample immersed for 50 days, (c) a photo of the coating sample immersed for 100 days, and (d) the corresponding results of the number of days when rust appears on the surface of the coating sample

[0035] Figure 5 The thermal spraying process parameter-coating porosity prediction model performance of the three algorithms, AutoGluon, FLAML and TPOT, under different test set ratios in the present invention: (a) the influence of the test set ratio on the determination coefficients of the three algorithms, (b) the comparison results between the measured porosity values ​​and the predicted values ​​of the AutoGluon model when the test set ratio is 20%.

[0036] Figure 6 The thermal spraying process parameters-corrosion days prediction model performance in the present invention: (a) the distribution characteristics of the original unprocessed corrosion days data, (b) the distribution characteristics of the corrosion days data processed by random oversampling, (c) the influence of the test set ratio on the determination coefficients of the three algorithms, (d) the comparison results between the measured values ​​of the corrosion days and the predicted values ​​of the TPOT model when the test set ratio is 15%.

[0037] Figure 7 The spraying process parameter data set in the present invention includes: (a) a spraying process parameter data set of sprayed samples, which includes a training set and a test set, and each grid point represents a spraying process parameter. A total of 336 process parameters are selected in the present invention; (b) a predicted spraying process parameter data set, which includes an original process parameter data set, a total of 11,375 process parameters, and the predicted optimal, medium, and poor spraying process parameters, as well as the optimal process parameters in the original data set are marked.

[0038] Figure 8 Verification of the porosity of the coatings prepared by the predicted spraying process parameters and the optimal process in the original data set in the present invention: (a) and (b) coatings prepared by the predicted optimal spraying process parameters, (c) and (d) coatings prepared by the predicted medium spraying process parameters, (e) and (f) coatings prepared by the predicted poor spraying process parameters, (g) and (h) coatings prepared by the optimal spraying process parameters in the original data set.

[0039] Fig. 9 This is the verification of the corrosion performance of the coating prepared by the predicted spraying process parameters and the optimal process in the original data set in the present invention: the red five-pointed star represents the coating prepared by the predicted optimal spraying process parameters, the green square dots represent the coating prepared by the predicted medium spraying process parameters, the black dots represent the coating prepared by the predicted poor spraying process parameters, and the blue triangle dots represent the coating prepared by the optimal spraying process parameters in the original data set. DETAILED DESCRIPTION

[0040] The accompanying drawings illustrate embodiments of the present invention, and it will be understood that the disclosed embodiments are merely examples of the present invention, and the present invention can be implemented in various forms. Therefore, the specific details disclosed herein should not be interpreted as limiting, but only as the basis for the claims and as a representative basis for teaching those skilled in the art to implement the present invention in various ways.

[0041] The present invention provides a method for optimizing process parameters for thermal spray coating preparation based on machine learning, which comprises: S1: variable parameter high-throughput preparation of coating samples and establishment of a preparation process parameter data set; S2: testing and analyzing the porosity of coating samples, and establishing a corresponding relationship between coating preparation process parameters and coating porosity; S3: testing and analyzing the corrosion performance of coating samples, and obtaining the corrosion performance under the conditions of coating preparation process parameters; S4: using one or more algorithms to perform machine learning model training and testing on the correlation between spraying process parameters and coating porosity and corrosion performance, and predicting spraying process parameters with lower coating porosity and better corrosion resistance in a wider range of spraying process parameters; S5: verifying the above-mentioned predicted spraying process parameters, and finally obtaining the optimal coating preparation process parameters.

[0042] The specific steps for high-throughput preparation of coating samples and establishment of preparation process parameter data set are as follows: for specific thermal spraying process parameters, multiple levels are set for each parameter (e.g. Figure 1In the HVAF spraying process, the fuel pressure is set to 3 levels, the spray gun traverse speed is set to 3 levels, the powder feeding speed is set to 3 levels, and the spray gun stepping is set to 2 levels); for the change of spray distance parameters, a spray tool with adjustable lateral distance is used, which can achieve a single spraying and obtain multiple coating samples with different spray distances and the same other process parameters (for example Figure 1 Seven coating samples with different spraying distances can be prepared at one time), thereby realizing high-throughput preparation of the coating; all process parameters are numbered according to a certain rule to establish a process parameter data set.

[0043] The tool has multiple (8) parallel spindles, each with positioning pits of equal distance (e.g., 10mm spacing), each spindle with a matching sleeve, a spray sample fixture fixed on the sleeve, and a positioning screw on the side of the sleeve for fixing the spray sample after adjusting the spray distance; after the spray sample position is adjusted, the positioning screw is screwed into the positioning pit on the spindle corresponding to the position to achieve positioning. The tool can simultaneously fix multiple sample substrates with different spray distances, and realize the one-time preparation of coating samples with the same other process parameters and different spray distances.

[0044] Compared with the traditional method of preparing one coating sample for each set of process parameters, the high-throughput thermal spray coating process parameter optimization method of the present invention can prepare multiple coating samples (several to dozens or even more) at one time, thereby greatly improving the preparation efficiency of coating samples.

[0045] After high-throughput preparation of the coating samples, basic structural and performance characterization of the powder and coating is required, e.g. Figure 2 The structure and performance of the FeCrMoNiPBCSi amorphous alloy powder and the typical coating prepared in the present invention are characterized. It can be seen that the powder has good sphericity ( Figure 2 In (a), the powder particle size is basically normally distributed ( Figure 2 (b)), the coating thickness is uniform and well integrated with the cross section of the substrate ( Figure 2 (c)), showing a typical amorphous structure ( Figure 2 Middle (d) and Figure 2 In (e), the powder and coating have similar crystallization peaks and melting peaks ( Figure 2 Middle (f).

[0046] The specific steps for testing and analyzing the porosity of the coating specimens are as follows: cutting, mounting, grinding and polishing the high-throughput prepared coating, sealing the edges and bottom of the coating specimens with silicone rubber and arranging them regularly ( Figure 3 (a)), SEM images of the coating cross section were taken using SEM ( Figure 3(b)), combined with the image analysis technology of ImageJ software ( Figure 3 (c) and Figure 3 (d) Calculate the percentage of the pore area of ​​each coating to obtain the porosity of each coating and perform statistics ( Figure 3 In (e), the corresponding relationship between the coating preparation process parameters and the coating porosity is established.

[0047] The specific steps of testing and analyzing the corrosion performance of the coating specimens are as follows: after cutting the high-throughput prepared coating, use sealant to seal the edges, only expose the middle part of the coating surface of the sample, and seal the bottom surface opposite to the coating surface, all side surfaces and the edge of the coating surface with sealant; put each sealed coating sample into a separate plastic container, and then arrange them regularly according to the number; add a corrosive medium (3.5wt.% NaCl solution) into each plastic container containing the coating sample for immersion, take photos every 24 hours to record the corrosion morphology of the coating surface, and the statistical results of the coating samples immersed for 1 day, 50 days, 100 days and the corrosion failure days are as follows Figure 4 Middle (a), Figure 4 (b) and Figure 4 As shown in (c); the time when the surface corrosion area of ​​each coating sample reaches 10% is counted and used as the number of days of corrosion failure to draw a chart ( Figure 4 In (d), the corresponding relationship between the coating preparation process parameters and the coating corrosion performance is established.

[0048] The specific steps of machine learning model training and testing for the correlation between spraying process parameters and coating porosity are as follows: using the three algorithms of AutoGluon, FLAML and TPOT, the fuel pressure, spray gun traverse rate, powder feeding rate, spray gun stepping and spraying distance in spraying (HVAF) are used as input parameters, and the coating porosity is used as the output parameter to train the data and establish an HVAF process parameter-coating porosity relationship model. In order to reduce the overfitting caused by sample division, different training set and test set ratios are defined, and the AutoGluon, FLAML and TPOT algorithms are substituted, and then the optimal training set and test set ratios are selected to improve the performance of the model. In the present invention, the HVAF process parameter-porosity model performance of the AutoGluon, FLAML and TPOT algorithms under different test set ratios is shown as follows: Figure 5 As shown in (a), compared with FLAML and TPOT, the AutoGluon model shows high stability and extremely high performance in different proportions of training and test sets; when the test set ratio is 20%, the comparison between the measured value of porosity and the predicted value of the AutoGluon model is Figure 5As shown in (b), these data points are close to the diagonal line, reflecting the extremely high accuracy of the HVAF process parameter-porosity model and confirming the high reliability of the AutoGluon algorithm in predicting the coating porosity.

[0049] The specific steps of machine learning model training and testing for the correlation between spraying process parameters and coating corrosion performance are as follows: Figure 6 In (a), the corrosion days distribution is randomly oversampled and its data distribution characteristics are improved ( Figure 6 In (b), the skewness and kurtosis values ​​are lower, and the data set is more evenly distributed and continuous; further using the three algorithms of AutoGluon, FLAML and TPOT, the fuel pressure, spray gun traverse rate, powder feeding rate, spray gun step and spray distance in spraying (HVAF) are used as input parameters, and the number of days of coating corrosion is used as the output parameter to train the data and establish the HVAF process parameter-coating corrosion day relationship model. The determination coefficient (accuracy) of the AutoGluon, FLAML and TPOT algorithms under different test set ratios is analyzed as follows Figure 6 As shown in (c), it can be seen that after applying random oversampling, the AutoGluon, FLAML and TPOT algorithm models under different test set ratios have significantly improved the determination coefficient; when the test set ratio is 15%, the comparison between the measured value of corrosion days and the predicted value of the TPOT model is Figure 6 As shown in (d), these data points are close to the diagonal line, reflecting the high accuracy of the HVAF process parameter-corrosion days relationship model and confirming the high reliability of the TPOT algorithm in predicting the corrosion performance of coatings.

[0050] The specific steps for predicting the spraying process parameters of coatings with lower porosity and better corrosion resistance in a wider range of spraying process parameters are as follows: normalize multiple spraying parameters, divide different sector areas in the circle according to different spraying distances, fuel pressure / powder feeding rate can represent the energy shared by unit mass of powder particles, and is represented by radii of different angles in the sector, and the spray gun traverse rate × spray gun stepping can reflect the overlap / overlap degree in the deposition process of spraying particles, and is represented by radii of different lengths in the sector, so that each grid point in the circular area can represent a spraying process parameter (such as Figure 7 As shown in (a) in the figure, the entire circular area can represent all 336 spraying processes that have been sprayed.

[0051] The specific steps of predicting the spraying process parameters of the coating with lower porosity and better corrosion resistance in a wider range of spraying process parameters also include: further subdividing or extending each process parameter such as spraying distance, fuel pressure, powder feeding rate, spray gun traverse rate, and spray gun stepping, so as to establish a wider set of process parameters (such as Figure 7 As shown in (b), a total of 11375 spraying process parameters are included. Based on the training and optimization of the HVAF process parameter-coating porosity relationship model and the HVAF process parameter-coating corrosion days relationship model, the optimal spraying process parameters with the lowest porosity and better corrosion performance, the medium spraying process parameters, and the poor spraying process parameters (such as Figure 7 In addition, the optimal process in the original spraying process data set is also marked for subsequent comparative verification of coating porosity and coating corrosion resistance.

[0052] The specific steps for verifying the porosity of the coating prepared by the predicted spraying process parameters are as follows: select the three predicted optimal, medium and poor spraying process parameters, and prepare the corresponding coatings, and use the optimal coating in the original sprayed data set as a comparison; cut, grind, and polish the prepared coating samples and take photos using SEM, and use image analysis to determine the two-dimensional porosity of the coating; or use a three-dimensional X-ray imaging system to test the volume fraction of the porosity of the coating, such as Figure 8 As shown, the optimal ( Figure 8 (a) and Figure 8 Medium (b)), medium ( Figure 8 (c) and Figure 8 Medium (d)) and poor ( Figure 8 (e) and Figure 8 The porosity of the coatings prepared by the spraying process parameters in (f) are 0.0312 vol.%, 0.0764 vol.% and 1.2799 vol.%, respectively. The best coating in the original data set ( Figure 8 Medium(g) and Figure 8 The porosity of (h) is 0.0462 vol.%, which shows that the quality of the coating prepared by the newly predicted optimal spraying process is significantly better than that of the original optimal coating.

[0053] The specific steps for verifying the corrosion resistance of the coating prepared by the predicted spraying process parameters are as follows: select the three predicted optimal, medium and poor spraying process parameters, and prepare the corresponding coatings, and use the optimal coating in the original sprayed data set as a comparison; after cutting, mounting, grinding and polishing the prepared coating samples, the samples are subjected to long-term electrochemical impedance spectroscopy testing, and the impedance spectrum of the samples is tested every 24 hours and the impedance modulus value at 10mHz is calculated; the long-term impedance modulus value changes of the four coatings are shown in Figure 2. Fig. 9As shown in the figure, it can be seen that the impedance modulus of the coating prepared by the predicted poor spraying process is always low, about 1.6×10 3 Ω·cm 2 The predicted impedance modulus of the coating prepared by the medium spraying process gradually decreases with the immersion time and remains basically stable after 768 hours, about 10 4 Ω·cm 2 The impedance modulus of the coating prepared by the predicted optimal spraying process is the highest and stable at 2.4×10 4 Ω·cm 2 , which is significantly better than the optimal coating in the original dataset.

Claims

1. A method for optimizing process parameters for thermal spray coating preparation based on machine learning, characterized in that: The following steps are involved: S1. Perform variable parameter high-throughput preparation of coating samples; S2. Testing and analyzing the porosity of the coating specimens, and establishing the corresponding relationship between the coating preparation process parameters and the coating porosity; and constructing a data set of process parameters and coating porosity; S3. Testing and analyzing the corrosion performance of the coating specimens to obtain the corrosion performance under the coating preparation process parameter conditions; and constructing a data set of process parameters and corrosion performance; S4. Using at least one machine learning model, based on the process parameter and coating porosity data set, train and test the machine learning model for the correlation between the spraying process parameters and the coating porosity, and obtain a relationship model between the process parameters and the coating porosity; Based on the process parameter and corrosion performance data set, the machine learning model is trained and tested for the correlation between the spray process parameters and corrosion performance, and the relationship model between the process parameters and corrosion performance is obtained; S5. Input the spraying process parameters after epitaxy and expansion into the trained process parameter and coating porosity relationship model to predict several coating preparation process parameters with the lowest coating porosity; input the spraying process parameters after epitaxy and expansion into the trained process parameter and corrosion performance relationship model to predict several coating preparation process parameters with the best corrosion performance, and take the intersection of the coating preparation process parameters with the lowest porosity and the coating preparation process parameters with the best corrosion performance to obtain the optimal coating preparation process parameters.

2. The method for optimizing process parameters of thermal spray coating preparation based on machine learning according to claim 1, characterized in that: In step S1, the variable parameter high-throughput preparation of the coating sample comprises the following steps: A plurality of movable spraying templates are arranged so that the plurality of spraying templates are located at different spraying distances relative to the same spray gun; by changing the spraying distance, a plurality of coating specimens with different spraying distances but the same other coating preparation process parameters can be obtained by a single spraying.

3. A method for optimizing process parameters for thermal spray coating preparation based on machine learning according to claim 1 or 2, characterized in that: The coating preparation process parameters include spraying distance, fuel pressure, spray gun traverse rate, powder feeding rate, and spray gun stepping.

4. The method for optimizing process parameters of thermal spray coating preparation based on machine learning according to claim 1, characterized in that: In step S2, the porosity of the coating specimen is tested and analyzed, specifically: an optical microscope or a scanning electron microscope is used to take a cross-sectional photograph of the coating sample, and the coating porosity is obtained by image processing.

5. The method for optimizing process parameters of thermal spray coating preparation based on machine learning according to claim 1, characterized in that: In step S3, the corrosion performance of the coating specimen is tested and analyzed, specifically: The corrosion resistance of the coating was evaluated by immersion corrosion test or long-term electrochemical impedance spectroscopy test with a duration of more than 1000 hours.

6. The method for optimizing process parameters for thermal spray coating preparation based on machine learning according to claim 1, characterized in that: In step S4, the training of the process parameter and coating porosity model and the process parameter and corrosion performance model is as follows: For the process parameter and coating porosity model: take the process parameter as the input parameter and the porosity value as the output parameter, use at least one of the Autogluon, TPOT and FLAML algorithms to train and test the model, and select the model with the best result as the process parameter and coating porosity relationship model to establish the relationship between the coating preparation process parameters and the coating porosity; For the process parameter and corrosion performance model: take the process parameters as input parameters and the corrosion resistance as output parameters, use at least one of the Autogluon, TPOT and FLAML algorithms for model training and testing, and select the model with the best result as the process parameter and corrosion performance relationship model to establish the relationship between the coating preparation process parameters and the coating corrosion performance.

7. The method for optimizing process parameters for thermal spray coating preparation based on machine learning according to claim 1, characterized in that: In step S5, the optimal coating preparation process parameters are predicted, and the set range of one or more spraying process parameters is extended and expanded to obtain more spraying process parameters. Based on the trained process parameter and coating porosity relationship model, one or several coating preparation processes with the lowest coating porosity are predicted; based on the trained process parameter and corrosion performance relationship model, one or several coating preparation processes with the best corrosion performance are predicted.

8. The method for optimizing process parameters for thermal spray coating preparation based on machine learning according to claim 1 or 7, characterized in that: In step S5, the top 1‰ coating preparation process parameters with the lowest coating porosity are predicted by using the trained process parameter and coating porosity relationship model; the top 1‰ coating preparation process parameters with the best corrosion performance are predicted by using the trained process parameter and corrosion performance relationship model.

9. The method for optimizing process parameters of thermal spray coating preparation based on machine learning according to claim 1, characterized in that: After the prediction is completed in step S5, the porosity of the coating prepared by the predicted spraying process parameters is verified, specifically: The coating porosity was determined by taking cross-sectional photos of the coating sample using an optical microscope or a scanning electron microscope, and the area percentage of the coating pores was obtained by image processing to obtain the coating porosity. Alternatively, an X-ray three-dimensional imaging system is used to perform three-dimensional reconstruction and analysis on the coating sample to ultimately obtain the volume fraction of the coating pores.

10. The method for optimizing process parameters for thermal spray coating preparation based on machine learning according to claim 1, characterized in that: After the prediction is completed in step S5, the corrosion performance of the coating prepared by the predicted spraying process parameters is verified, specifically: the corrosion performance verification test adopts an immersion corrosion experiment or an electrochemical impedance spectroscopy long-term test with a duration greater than 1000 hours.

Images