Large ship surface self-adaptive rust removal method based on laser rust removal

Abstract

The invention relates to the technical field of ship rust removal, in particular to a large ship surface self-adaptive rust removal method based on laser rust removal, and provides the following scheme that a ship structure design drawing is obtained, three-dimensional position marking is conducted on a rusted part, and depth data is measured; setting laser initial parameters according to the environmental data; constructing a parameter updating network, inputting the depth data and the initial parameters for training, and generating laser adjustment parameters; rust removal is conducted on the rusted part according to the adjusting parameters and the path planning; and finally, acquiring the surface roughness of the cleaned part and optimizing network parameters. According to the method, the laser power, frequency and scanning speed are adjusted in a self-adaptive mode, and the accuracy and efficiency of laser rust removal are improved.

Description

Technical Field

[0001] The invention relates to the technical field of ship rust removal, in particular to a large ship surface self-adaptive rust removal method based on laser rust removal. Background Art

[0002] With the development of ship rust removal technology, cleaning the rust on the surface of the ship does not only require simple cleaning, but also needs to consider whether the environment will be polluted in the process of removing the surface rust, whether the rust on the ship surface can be removed faster and more non-destructively. Therefore, laser rust removal technology has been used to clean the rust on the ship surface. However, when using laser rust removal technology, the laser energy density needs to be precisely controlled. However, in the face of different degrees of rust on the surface of large ships, the existing technology on how to control the relevant parameters of laser rust removal and how to clean different degrees of rust without damaging the performance of the ship's metal materials has not taken into account the fact that the rust on the surface of the ship has different thicknesses during the laser rust removal process, and that when the ship is too large, there will be missed areas.

[0003] For example, a Chinese patent with publication number CN116604190A discloses a device for removing the surface material of a workpiece by a marine laser. The device for removing the surface material of a workpiece by a marine laser comprises: a host module, including a housing, a continuous laser generator and a pulse laser generator, both of which are arranged in the housing; a laser emission module, including a continuous laser emission muzzle connected to the continuous laser generator, a pulse laser emission muzzle connected to the pulse laser generator, and an air blow pipe, the air blow pipe is configured to blow air to the continuous laser emission muzzle and the pulse laser emission muzzle, and the continuous laser emission muzzle and the pulse laser emission muzzle can irradiate the surface of the workpiece simultaneously or sequentially within a preset time. The device for removing the surface material of a workpiece by a marine laser of the invention can be used for grinding and rust removal of the workpiece, with a good construction environment, and the continuous laser generator and the pulse laser generator can act simultaneously during rust removal, so that the rust removal is thorough.

[0004] For example, the Chinese patent application with publication number CN116038130A discloses a laser rust removal device for ships and marine engineering structures, including a mobile vehicle, a dust collector, a laser and a controller, a photoelectric rust removal device connected to the laser through an optical fiber, and a photoelectric rust removal device connected to the dust collector through an air duct; the photoelectric rust removal device includes a recovery device, a drive device, and a focal length stabilizer rotatably connected to the drive device through a torque output rod, and the photoelectric rust removal device is also provided with a rangefinder and a focal length indicator. When in use, the mobile vehicle can realize mobile following operation, which is convenient and fast. At the same time, the photoelectric rust removal device and the dust collector work at the same time, which is efficient, environmentally friendly and safe; the focal length stabilizer can realize the adjustment of the focal length and the stability of the focal length during the rust removal process, ensuring the effect of rust removal, and at the same time, the casters can move 360° on the surface to be rusted, which has the function of auxiliary rust removal, can reduce the labor intensity of the operators, and improve the quality of rust removal.

[0005] The above patents all have the problems raised by the background technology: they do not take into account the fact that the rust on the surface of the ship during the laser rust removal process also has different thicknesses, and that when the ship is too large, there will be missed areas; to solve these problems, the present invention proposes an adaptive rust removal method for the surface of large ships based on laser rust removal. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a large-scale ship surface adaptive rust removal method based on laser rust removal in view of the deficiencies of the prior art, obtain the ship structure design drawing, mark the three-dimensional position of the rusted parts, and measure the depth data; set the laser initial parameters according to the environmental data; build a parameter update network, input the depth data and initial parameters into the training, and generate the laser adjustment parameters; remove the rust from the rusted parts according to the adjustment parameters and path planning; finally, obtain the surface roughness of the cleaned parts and optimize the network parameters. This method improves the accuracy and efficiency of laser rust removal by adaptively adjusting the laser power, frequency and scanning speed.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A large ship surface adaptive rust removal method based on laser rust removal, the method comprising:

[0009] Obtain the structural design drawings of large ships, mark the three-dimensional positions of the corroded parts on the surface of the ship, and measure and record the depth data of the corroded parts;

[0010] Obtain environmental data and set initial parameters of the laser rust removal equipment according to the environmental data;

[0011] Constructing a parameter update network, taking the depth data and the initial parameters as input parameters of the parameter update network, training the input parameters through the parameter update network, and outputting laser adjustment parameters;

[0012] The laser rust removal equipment is adjusted according to the laser adjustment parameters, and the rust is removed from the rusted part according to the marked three-dimensional position;

[0013] The surface roughness of the cleaned rusted part is obtained, and the network parameters of the network are updated according to the surface roughness optimization parameters.

[0014] The measuring and recording of the depth data of the rusted part includes:

[0015] Performing a preliminary scan of the corroded part of the ship surface using a laser rangefinder to generate surface contour information, and determining the corrosion boundary based on the surface contour information;

[0016] According to the corrosion boundary, the thickness of the corrosion part is measured by an ultrasonic thickness gauge to obtain preliminary depth data;

[0017] The corroded part is imaged by an image sensor, and the texture features of the corroded part are obtained by image processing technology;

[0018] The preliminary depth data is corrected according to the texture feature to calculate the depth data.

[0019] The method of obtaining the texture features of the rusted part by using image processing technology includes:

[0020] The image of the rusted part is converted from the RGB model to the HSV color model, and the three HSV channels are subjected to median filtering to obtain a median filtered image. The reflection intensity of each pixel is calculated according to the color balance and contrast of the image of the rusted part and the median filtered image.

[0021] Calculating the gray level co-occurrence matrix of the image of the corroded part, and calculating the texture information of the corroded part according to the reflection intensity and the gray level co-occurrence matrix, wherein the texture information includes texture directionality, roughness and unevenness;

[0022] The texture information is characterized by using a BRIEF descriptor, a feature descriptor of the texture information is calculated, the texture information is converted into high-dimensional data by using an MLP network, and the data is added to the feature descriptor to generate texture features.

[0023] The correcting the preliminary depth data according to the texture feature comprises:

[0024] By performing Fourier transform analysis on the texture features, the periodic features of the concave-convex area on the surface of the rusted part are extracted, and the curvature distribution information of the concave-convex area on the surface of the rusted part is calculated according to the periodic features;

[0025] constructing a surface concave-convex feature model according to the curvature distribution information, matching the surface concave-convex feature model with the preliminary depth data, and identifying nonlinear errors in the preliminary depth data;

[0026] The nonlinear error is nonlinearly corrected by a curvature adjustment function to calculate depth data. The curvature adjustment function is used to adaptively adjust the correction amplitude of the preliminary depth data according to the curvature change in the corroded part.

[0027] The environmental data include temperature, humidity and wind speed, and the initial parameters include laser power, laser frequency and laser scanning speed.

[0028] The parameter updating network comprises:

[0029] A parameter generator module, which is used to implement multi-level feature mapping through deep convolutional layers, convert input parameters into preliminary prediction vectors containing laser adjustment parameters, and adjust the preliminary prediction vectors through adaptive factors according to changes in depth data at different positions to generate initial laser adjustment parameters;

[0030] The parameter discriminator module is used to receive the initial laser adjustment parameters, compare them with the optimized parameters in the historical best parameter library, analyze whether the generated parameters are consistent with the actual rust removal requirements, calculate the accuracy score, and if not, feed the accuracy score back to the parameter generator module, adjust the adaptive factor according to the accuracy score, and output the laser adjustment parameters if they are consistent.

[0031] The parameter generator module comprises:

[0032] The local concave-convex features and overall depth information in the depth data are extracted through convolution operation, and the local concave-convex features and overall depth information are fused through the normalized embedded Gaussian function and converted into a feature vector matrix;

[0033] Mapping the feature vector matrix and the initial parameters to a high-dimensional space, performing feature matching in the high-dimensional space, calculating a parameter matching vector according to a maximum pooling function, flattening the output of the maximum pooling function into a one-dimensional vector by a flattening function, and obtaining a preliminary prediction vector;

[0034] Calculate the local depth gradient of each point in the eigenvector matrix, and obtain the depth change rate of the point according to the local depth gradient;

[0035] Extracting surface roughness information according to local concave-convex features, drawing a roughness distribution map, and obtaining the roughness value of each point through the roughness distribution map;

[0036] The position factor of the point is calculated according to the depth change rate and the roughness value, and each point is traversed. According to the position factor, the adaptive factor is calculated by weighted average. The calculation formula of the position factor is:

[0037]

[0038] Among them, α x,y Represents the position factor of the point (x, y), d x,y Indicates the depth change rate of point (x, y), D x,y represents the maximum depth data of the point (x, y), μ represents the average value of the depth data, and σ d represents the variance of the depth data, β 1 represents the adaptive adjustment coefficient, which is used to control the contribution weight of the depth gradient to the position factor, γ represents the smoothing factor, and β 2 Represents the adaptive adjustment coefficient, which is used to control the contribution weight of roughness to the position factor, C x,y Represents the roughness value of the point (x,y).

[0039] The three-dimensional position marking of the rusted parts on the surface of the ship includes: constructing a three-dimensional model of the ship according to the structural design drawing, and marking all the rusted parts on the surface of the ship in the three-dimensional model, constructing a three-dimensional rectangular coordinate system of the ship with any point in the three-dimensional model as the origin, obtaining the three-dimensional coordinates of the center points of all the marked rusted parts on the surface of the ship as the position coordinates of all the corresponding rusted parts, and storing the position coordinates of all the corresponding rusted parts and the thickness of all the corresponding rusted parts in a rusted part coordinate set.

[0040] The method further includes a laser rust removal path planning strategy, which is used to plan a rust removal work path according to a set of rusted part coordinates, including:

[0041] Get the number of corroded parts on the ship, set it as Q; take the center point of any corroded part on the ship as the starting point of the path, set it as the jth corroded part, and let j = 1;

[0042] Obtain the center point position coordinates of the jth rusted part from the rusted part coordinate set, calculate the distance between the center point position coordinates of the jth rusted part and the center point position coordinates of the remaining Qj rusted parts by using the Euclidean distance formula, and take the corresponding rusted part with the smallest distance as the j+1th path node, and let j=j+1;

[0043] Repeat the steps until j=Q, and take the Qth path node as the path end point to obtain the laser rust removal path.

[0044] The surface roughness is used to update the adaptive factor and the parameter discriminator module.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The present invention dynamically adjusts the laser power, frequency and scanning speed according to the corrosion depth, roughness and environmental data. The introduction of adaptive factors and parameter update networks optimizes the laser parameters in real time to adapt to different corrosion conditions, ensuring the stability and consistency of the rust removal effect;

[0047] 2. The present invention optimizes the laser parameter setting through the surface roughness feedback mechanism to avoid excessively high or low laser power, effectively reducing excessive wear on the substrate and extending the life of the hull. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0049] Figure 1 This is a schematic flow chart of a method for adaptively removing rust from the surface of a large ship based on laser rust removal according to Example 1 of the present invention;

[0050] Figure 2 This is a schematic diagram of the depth data measurement process of Example 1 of the present invention;

[0051] Figure 3 This is a schematic diagram of a texture information encoder according to Embodiment 1 of the present invention;

[0052] Figure 4 This is a parameter update network structure diagram of Example 1 of the present invention. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0054] Example 1

[0055] See also Figure 1 The present invention provides an embodiment: a large ship surface adaptive rust removal method based on laser rust removal, the specific steps of the method are as follows:

[0056] S1: Obtain the structural design drawings of large ships and mark the three-dimensional positions of the rusted parts on the surface of the ship;

[0057] In this step, the 3D coordinate system of the ship is used to mark each rusted part. This marking provides accurate spatial data for the positioning and path planning of the subsequent laser rust removal equipment;

[0058] S2: Measure and record the depth data of the rusted parts;

[0059] In this step, a variety of measurement technologies (such as laser ranging, ultrasonic thickness measurement and image sensing) are used to measure the thickness and surface characteristics of the rusted parts in multiple dimensions, and the data is recorded. Combined with the distribution of depth data, accurate depth information is generated for each rusted point, so that subsequent parameter settings are more accurate;

[0060] S3: Acquire environmental data and set initial parameters of the laser rust removal equipment according to the environmental data;

[0061] S4: constructing a parameter updating network, taking the depth data and the initial parameters as input parameters of the parameter updating network, training the input parameters through the parameter updating network, and outputting laser adjustment parameters;

[0062] In this step, by establishing a parameter update network, taking depth data, initial parameters, etc. as input, multiple rounds of training and optimization are performed in the network to gradually generate laser adjustment parameters suitable for different rust characteristics. The parameter update network adaptively analyzes and adjusts the input data to generate the optimal laser settings in order to dynamically respond to different rust removal needs;

[0063] S5: adjusting the laser rust removal equipment according to the laser adjustment parameters, and removing rust from the rusted parts according to the marked three-dimensional positions and the laser rust removal path planning strategy;

[0064] In this step, the laser adjustment parameters provided by the parameter update network are used to set the power, frequency, and scanning speed of the laser equipment. At the same time, combined with the three-dimensional annotation position and path planning strategy, the laser equipment gradually covers each rusted part along the optimal path to ensure the comprehensiveness and efficiency of rust removal;

[0065] S6: Obtaining the surface roughness of the cleaned rusted part, and updating the network parameters of the network according to the surface roughness optimization parameters;

[0066] In this step, the roughness of the cleaned rusted parts is measured to check the cleaning effect, and the roughness data is fed back to the parameter update network to further optimize the network parameters. Through this feedback mechanism, the network parameters can be gradually adjusted to adapt to the characteristics of different rusted surfaces, improving the accuracy and consistency of subsequent rust removal operations;

[0067] Marking the positions of rusted parts on the surface of a ship in a structural design drawing includes: constructing a three-dimensional model of the ship according to the structural design drawing, marking all rusted parts on the surface of the ship in the three-dimensional model, constructing a three-dimensional rectangular coordinate system of the ship with any point in the three-dimensional model as the origin, obtaining the three-dimensional coordinates of the center points of all the marked rusted parts on the surface of the ship as the position coordinates of all the corresponding rusted parts, and storing the position coordinates of all the corresponding rusted parts and the thickness of all the corresponding rusted parts in a rusted part coordinate set.

[0068] See also Figure 2 , a schematic diagram of the depth data measurement process according to an embodiment of the present invention, wherein the specific steps of S2 are as follows:

[0069] S2.1: Preliminarily scan the corroded part of the ship surface using a laser rangefinder to generate surface contour information, and determine the corrosion boundary based on the surface contour information;

[0070] Use a laser rangefinder to conduct a full coverage scan of the rusted parts on the ship's surface. The laser rangefinder emits a light beam and receives the reflected echo, calculates the distance between each measurement point and the rangefinder, and generates preliminary contour data of the rusted parts on the ship's surface;

[0071] The scanned contour data is stored in the form of three-dimensional coordinates, and the algorithm clusters and analyzes these coordinate points. The boundary area with drastic surface changes, that is, the edge position of the rust, is detected by curvature analysis. The convolution curvature analysis technology is further used to enhance the recognition accuracy of the boundary area and construct the contour information of the rusted area into a three-dimensional model;

[0072] S2.2: According to the corrosion boundary, the thickness of the corrosion part is measured by an ultrasonic thickness gauge to obtain preliminary depth data;

[0073] After determining the corrosion boundary, use an ultrasonic thickness gauge to measure the thickness of the corrosion site point by point within the boundary. Ultrasonic thickness measurement is based on the principle of material interface reflection. Ultrasonic waves are emitted to penetrate the corrosion layer and reach the metal substrate interface. The returned reflected waves are used to calculate the corrosion thickness. The thickness information obtained by point-by-point measurement is used to generate a preliminary depth data matrix, and the depth information of each measurement point is presented in the form of a two-dimensional matrix.

[0074] S2.3: imaging the corroded part through an image sensor, and obtaining the texture features of the corroded part through image processing technology;

[0075] Based on the depth data of the rusted area, the image sensor is used to perform high-resolution imaging to obtain surface texture information. Image processing is based on HSV color model conversion. The information of the three channels of hue, saturation and brightness obtained by conversion is used to generate more recognizable texture features. The extraction of texture features can accurately identify the structural differences of different areas on the rusted surface, which helps to optimize the correction of subsequent depth data and provide details of the rust thickness distribution;

[0076] S2.4: Correcting the preliminary depth data according to the texture feature to calculate depth data;

[0077] Specifically, by superimposing a texture-based local depth correction factor on the preliminary depth data of each rust point, the correction process combines the surface roughness and depth gradient information to achieve multi-dimensional optimization of depth measurement, ensuring that stable depth data can be obtained under different types of rust layer thickness changes.

[0078] The specific steps of texture feature extraction are as follows:

[0079] S2.3.1: Convert the image of the rusted part from the RGB model to the HSV color model, perform median filtering on the three HSV channels to obtain a median filtered image, and calculate the reflection intensity of each pixel point according to the color balance and contrast of the rusted part image and the median filtered image;

[0080] The HSV color model decomposes color information into three channels that are easier for the human eye to perceive, making it easier to separate color and brightness information in an image. This conversion provides a basis for subsequent reflection intensity calculations, because the V (brightness) channel of the HSV model is more suitable for light intensity analysis;

[0081] Perform median filtering on each channel in the HSV model to reduce the impact of noise on image texture feature extraction. Median filtering is an effective nonlinear smoothing technique that can retain the edge details of the image while removing stray noise caused by illumination or sensor noise. The filtered image contains clearer rust texture information, which is conducive to subsequent texture information extraction;

[0082] For example, for a rust spot on the surface of a ship, the original image may contain color impurities under the RGB model, but in the HSV model, the brightness channel can reflect the influence of lighting alone. After filtering to eliminate interference, only the true brightness changes of the rusted part are retained, which is convenient for subsequent accurate analysis;

[0083] S2.3.2: Calculate the gray level co-occurrence matrix of the image of the corroded part, the gray level co-occurrence matrix obtains the local texture information of the image by analyzing the gray value distribution between adjacent pixels in the gray image, and calculates the texture information of the corroded part according to the reflection intensity and the gray level co-occurrence matrix, wherein the texture information includes texture directionality, roughness and unevenness;

[0084] The texture directionality is calculated based on the angular distribution of the grayscale co-occurrence matrix (e.g., the grayscale value difference in different directions), indicating the main texture orientation of the rusted part, which helps to analyze the directional consistency of the rust characteristics;

[0085] The roughness is calculated based on the contrast of the gray-level co-occurrence matrix, reflecting the degree of change in the details of the rusted surface. A high contrast means a rougher rusted surface, which is suitable for determining the power setting for rust removal;

[0086] The inhomogeneity is calculated by the entropy of the gray-level co-occurrence matrix. A higher entropy value indicates that the texture of the rusted surface changes greatly, that is, the inhomogeneity is high, which is suitable for judging the adaptability of the laser rust removal power and frequency;

[0087] S2.3.3: Using the BRIEF descriptor to describe the texture information, calculating the feature descriptor of the texture information, using the MLP network to convert the texture information into high-dimensional data, and adding it to the feature descriptor to generate texture features;

[0088] Generate binary feature descriptors for texture information based on the BRIEF algorithm. The BRIEF algorithm generates concise feature vectors by analyzing the grayscale value differences of the neighborhood of each pixel in the image, reducing the computational complexity. The feature descriptors generated by BRIEF are fused with the texture information calculated by GLCM to generate preliminary texture features. At this time, the feature descriptor provides a structured expression for the rust image by identifying the uniqueness of the texture features. The texture features are nonlinearly mapped through the MLP network and expanded to a high-dimensional space, enabling it to more accurately characterize the complex texture characteristics of the rust surface, and are added to the feature descriptors to generate the final high-dimensional texture features.

[0089] See also Figure 3 , a schematic diagram of a texture information encoder according to an embodiment of the present invention, which is used to increase the dimension of texture information, so that the texture information and the feature descriptor are added to form a feature vector, each texture information is combined with its coordinates and confidence, and a multi-layer perceptron is used to embed the feature points into the high-dimensional vector. First, a five-layer fully connected perceptron is used to embed the feature points G i Converted to 256-dimensional data, the number of channels of the five-layer perceptron are 3, 32, 64, 128 and 256 respectively;

[0090] The texture information of the rusted surface (such as roughness and unevenness) can reflect the microscopic concave-convex state of the surface and help correct potential errors in depth measurement. In addition, the depth data combined with the texture features can form a more comprehensive rust model, so that the laser parameters can be more accurately and adaptively adjusted to meet the rust removal needs of different surfaces;

[0091] The purpose of correcting the preliminary depth data is to ensure the accuracy and efficiency of laser rust removal. Since the preliminary depth data may deviate due to equipment, measurement errors and complex features of the rusted surface, it is necessary to optimize the data through specific correction steps to make the generated laser parameters more accurate, ensure that the laser energy distribution matches the actual rust depth, and avoid excessive or insufficient treatment of the hull surface. The specific steps for correcting the preliminary depth data are as follows:

[0092] S2.4.1: By performing Fourier transform analysis on the texture features, the surface features in the spatial domain can be transformed into the frequency domain, thereby capturing the periodic components in the surface microstructure, extracting the periodic features of the concave and convex areas on the surface of the rusted part, and calculating the curvature distribution information of the concave and convex areas on the surface of the rusted part based on the periodic features;

[0093] By analyzing the Fourier transform spectrum, frequency components with large periodic changes can be identified, which correspond to the concave and convex features of the rusted surface. For example, frequency components with large amplitudes in the spectrum indicate that there are significant periodic concavities and convexities on the surface, and this information will be used to construct the curvature distribution of the surface;

[0094] The curvature distribution can be used to quantify the convex and concave shape of the rust surface, providing accurate geometric information for subsequent depth data correction;

[0095] 2.4.2: Construct a surface concave-convex feature model based on the curvature distribution information. The model uses curvature data to represent the characteristics of different regions of the rusted surface, including relatively smooth regions, concave regions, and convex regions. The surface concave-convex feature model is matched with the preliminary depth data to identify nonlinear errors in the preliminary depth data.

[0096] The feature model takes curvature distribution as the core, maps the depth value of each area into the model, and provides a benchmark for subsequent matching with preliminary depth data. For example, the curvature value in the concave area is larger, while the curvature value in the smooth area is smaller;

[0097] 2.4.3: Match the preliminary depth data with the constructed surface concave-convex feature model point by point to identify nonlinear errors. The preliminary depth data may be affected by noise, equipment deviation and surface irregularities, which may easily lead to local errors;

[0098] By comparing the preliminary depth data with the curvature data in the concave-convex feature model, we can identify which depth data points deviate from the expected value of the model, determine the distribution and magnitude of the nonlinear error, and provide a clear direction for further correction;

[0099] The identified nonlinear errors are corrected using a curvature adjustment function. This curvature adjustment function uses an adaptive adjustment mechanism based on curvature changes, so that the depth data is corrected more significantly in areas with dramatic changes in concavity and convexity, while maintaining a smaller adjustment in smooth areas.

[0100] Apply the curvature adjustment function to each depth data point, superimpose the correction amount on the preliminary depth data, and generate the corrected depth data. Through this adaptive correction, the corrected depth data is closer to the actual morphology of the rusted surface, especially in the micro-concave and convex areas, the corrected data can effectively restore the real surface features;

[0101] Specifically, there are a large number of small groove areas on the rusted surface. These grooves show specific frequency characteristics in the frequency domain due to their periodic distribution. These periodic features can be identified through Fourier transform, and the curvature distribution of the groove area can be further calculated. During the correction process, the curvature adjustment function will adaptively correct the depth data of these groove areas, so that the energy distribution during laser rust removal is more uniform, avoiding incomplete or excessive rust removal due to excessive errors. The main function of depth data correction is to improve the accuracy of laser rust removal and make the distribution of laser energy in the rusted area more reasonable. The complex structure of the rusted surface, especially the uneven areas, can easily interfere with the absorption of laser energy. By correcting the preliminary depth data, the accuracy of laser parameter setting can be significantly improved, energy waste can be reduced, and the hull surface can be protected;

[0102] See also Figure 4 , a parameter updating network structure diagram of an embodiment of the present invention, wherein the parameter updating network comprises:

[0103] A parameter generator module, which is used to implement multi-level feature mapping through deep convolutional layers, convert input parameters into preliminary prediction vectors containing laser adjustment parameters, and adjust the preliminary prediction vectors through adaptive factors according to changes in depth data at different positions to generate initial laser adjustment parameters;

[0104] The parameter generator module first extracts features from the input data through a multi-layer convolutional neural network (CNN) and converts it into a feature vector in a high-dimensional space. In the convolution operation, the model is able to identify features of different depths and roughness by extracting multi-level information such as depth change rate, roughness, and rust surface distribution from the input data;

[0105] After the convolutional layer, the generator maps the feature vector layer by layer to gradually refine the laser parameters of each rusted part. This mapping process uses deep convolutional layers to create multi-layer feature representations in a high-dimensional space, so that the rust depth and surface characteristics of different locations are fully expressed;

[0106] The parameter generator module can provide initial adjustment parameters for the laser rust removal equipment according to different depth and roughness characteristics, and enable the parameters to dynamically adapt to different rust characteristics, thereby improving rust removal efficiency. The introduction of adaptive factors enables precise control of specific areas, ensuring that the power and frequency of the laser can be adjusted in real time during the rust removal process as the depth characteristics and roughness change;

[0107] The parameter discriminator module is used to receive the initial laser adjustment parameters, compare them with the optimized parameters in the historical optimal parameter library, analyze whether the generated parameters are consistent with the actual rust removal requirements, and calculate the accuracy score. This score is based on the rust removal effect of the laser parameters on different rusted parts and the stability calculation of the equipment to determine whether the generated parameters can meet the actual needs. If the accuracy score is lower than the preset threshold, the discriminator will feed back the score to the generator module and optimize the calculation of the adaptive factor. Feedback correction adjusts the weight of the adaptive factor through error back propagation, thereby correcting the deviation between the generated parameters and the optimal parameter library. If the accuracy score of the generated parameters meets expectations, the discriminator will output the generated parameters as the final laser adjustment parameters. This ensures that the output parameters are highly matched with the actual rust removal requirements and achieves accurate rust removal operations;

[0108] The introduction of the parameter discriminator module enables the initial laser adjustment parameters output by the parameter generator module to be verified for accuracy in real time, ensuring the application effect of the generated parameters in the actual rust removal environment. Through multiple rounds of corrections by the feedback mechanism, the adaptive factors are continuously optimized, so that the laser adjustment parameters gradually approach the optimal state, thereby maintaining high rust removal efficiency under different rust conditions;

[0109] Assume that a certain rusted area on the surface of a ship is a surface with a relatively thick depth and high roughness. Through the feature extraction of the parameter generator module, the depth gradient change rate is relatively high, and the adaptive factor is therefore also relatively large. The initial generation parameters may be high power and low frequency settings. At this time, the parameter discriminator module will compare the generated parameters with the historical data and find that the optimal parameters for the rusted area should be a slightly lower laser power to protect the substrate. The parameter discriminator module feeds back the accuracy score to the parameter generator module, which reduces the power by adjusting the adaptive factor to obtain the corrected laser parameters to ensure accurate rust removal and substrate safety;

[0110] The core role of the parameter generator module in the entire laser rust removal process is to generate and optimize the laser adjustment parameters according to the depth change, roughness information and other input parameters of the rusted part, so as to ensure the adaptive change of the power, frequency and scanning speed of the laser rust removal equipment, so as to accurately remove the rust layer of different depths and roughness. The design of this module includes multi-step feature extraction, matrix calculation, adaptive factor generation and adjustment process;

[0111] The parameter generator module comprises:

[0112] First, a feature matrix is ​​generated through surface depth data and local concave-convex features. Each element of the feature matrix contains the local depth gradient, overall depth features and surface roughness information of the point. Here, the parameter generator module smoothes the local features at different positions through the normalized embedded Gaussian function to reduce noise interference, and uses convolution operations to extract effective feature information. The construction of this feature matrix ensures the multi-dimensional fusion of data, so that the generated laser parameters can accurately reflect the corrosion status of each point;

[0113] The information in the feature matrix is ​​mapped to a high-dimensional space to facilitate subsequent feature matching. Through the operation of the maximum pooling function, representative parameters in the feature matrix are selected and feature matching is performed in the high-dimensional space to ensure that the initial prediction of the laser adjustment parameters in this space can correspond to the actual corrosion condition.

[0114] The application of the maximum pooling function here can effectively extract the local extreme values ​​in the feature matrix, retain important depth change information, and ignore the unimportant features interfered by noise. After being processed by the flattening function, the pooling output is flattened into a one-dimensional vector to obtain a preliminary prediction vector;

[0115] For each point in the feature vector matrix, the parameter generator module calculates its local depth gradient and depth change rate to quantify the corrosion change rate at that point;

[0116] Furthermore, the roughness value of each point is extracted from the surface roughness data to form a roughness distribution map. The roughness distribution map provides the parameter generator module with the surface roughness information of each rust point, providing support for the subsequent calculation of the adaptive factor;

[0117] The position factor of the point is calculated according to the depth change rate and the roughness value, and each point is traversed. According to the position factor, the adaptive factor is calculated by weighted average. The calculation formula of the position factor is:

[0118]

[0119] Among them, α x,y Represents the position factor of the point (x, y), dx,y Indicates the depth change rate of point (x, y), D x,y represents the maximum depth data of the point (x, y), μ represents the average value of the depth data, and σ d represents the variance of the depth data, β 1 represents the adaptive adjustment coefficient, which is used to control the contribution weight of the depth gradient to the position factor, γ represents the smoothing factor, and β 2 Represents the adaptive adjustment coefficient, which is used to control the contribution weight of roughness to the position factor, C x,y Indicates the roughness value of the point (x, y);

[0120] The calculation process of the adaptive factor ensures that in areas with significant depth changes and roughness, the laser adjustment parameters will tend to increase the power or frequency to achieve a sufficient cleaning effect; in areas with smoother surfaces, the adaptive factor is correspondingly smaller to prevent excessive rust removal from damaging the hull substrate. Through the introduction of the adaptive factor, the laser adjustment parameters can be fine-tuned according to the depth and roughness information of each position to ensure that the best rust removal effect is achieved in rusted areas with different characteristics. In deeper or rougher rusted areas, the adaptive factor increases, and the laser power and frequency are increased accordingly; while in shallower or smoother areas, the adaptive factor decreases to avoid excessive damage to the substrate, thereby achieving a balance between cleaning efficiency and surface protection. Through the feedback of the parameter discriminator module, the parameter generator module can continuously optimize the calculation method of the adaptive factor, so that the laser adjustment parameters gradually become more accurate and achieve higher rust removal quality.

[0121] The method further includes a laser rust removal path planning strategy, which is used to plan a rust removal work path according to a set of rusted part coordinates, including:

[0122] Get the number of corroded parts on the ship, set it as Q; take the center point of any corroded part on the ship as the starting point of the path, set it as the jth corroded part, and let j = 1;

[0123] Obtain the center point position coordinates of the jth rusted part from the rusted part coordinate set, calculate the distance between the center point position coordinates of the jth rusted part and the center point position coordinates of the remaining Qj rusted parts by using the Euclidean distance formula, and take the corresponding rusted part with the smallest distance as the j+1th path node, and let j=j+1;

[0124] Repeat the steps until j=Q, and take the Qth path node as the path end point to obtain the laser rust removal path.

[0125] The surface roughness is used to update the adaptive factor and the parameter discriminator module, and the laser adjustment parameters are further optimized through the feedback mechanism, so that the laser rust removal process can adapt to the changing characteristics of different rusted surfaces, thereby reducing damage to the substrate while ensuring the cleaning effect;

[0126] In the process of laser rust removal, the roughness of the hull surface is a key factor affecting laser absorption and rust removal efficiency. The greater the roughness, the higher the surface unevenness of the rusted part, and the laser will produce unstable energy distribution during the cleaning process. Therefore, in this embodiment, the surface roughness is not only used for the initial setting of the laser parameters, but also as a feedback parameter for dynamically updating the adaptive factor and parameter discriminator module to ensure the adaptability of the laser adjustment parameters in the real-time rust removal process;

[0127] After each round of laser rust removal, the surface roughness of the cleaned rusted part is measured by the surface roughness sensor. The roughness data matrix is ​​obtained by scanning, in which each element represents the roughness value of the cleaned rusted part. This data matrix can reflect the surface flatness and microscopic concave-convex state after laser cleaning, and provide basic data for the update of the adaptive factor and parameter discriminator module;

[0128] According to the measured roughness data, it is compared with the initially set roughness value, and the difference is used to update the adaptive factor. The updating process of the adaptive factor introduces a feedback gain factor to control the weight of the roughness feedback, and the feedback gain factor is calculated based on the difference;

[0129] For example, if the actual roughness is too high, it means that the laser power or frequency may be too low to achieve the best rust removal effect. At this time, the adaptive factor will increase to increase the laser power or frequency. Conversely, if the roughness is lower than the target value, the adaptive factor will decrease to prevent excessive concentration of laser energy from damaging the substrate.

[0130] The parameter discriminator module is used to detect whether the laser adjustment parameters output by the parameter generator module meet the actual rust removal requirements. By introducing roughness feedback, the parameter discriminator module can adjust the parameter generation process more accurately, so that the generated laser parameters are closer to the optimal state in subsequent rust removal. Specifically, the parameter discriminator module calculates the consistency score of the cleaned surface based on the roughness data, and uses it as a feedback indicator to optimize the output parameters of the parameter generator module.

[0131] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A large-scale ship surface adaptive rust removal method based on laser rust removal, characterized in that: The method comprises: Obtain the structural design drawings of large ships, mark the three-dimensional positions of the corroded parts on the surface of the ship, and measure and record the depth data of the corroded parts; Obtain environmental data and set initial parameters of the laser rust removal equipment according to the environmental data; Constructing a parameter update network, taking the depth data and the initial parameters as input parameters of the parameter update network, training the input parameters through the parameter update network, and outputting laser adjustment parameters; The laser rust removal equipment is adjusted according to the laser adjustment parameters, and the rust is removed from the rusted part according to the marked three-dimensional position; The surface roughness of the cleaned rusted part is obtained, and the network parameters of the network are updated according to the surface roughness optimization parameters.

2. According to claim 1, the method for adaptive rust removal on the surface of large ships based on laser rust removal is characterized in that: The measuring and recording of the depth data of the rusted part includes: Performing a preliminary scan of the corroded part of the ship surface using a laser rangefinder to generate surface contour information, and determining the corrosion boundary based on the surface contour information; According to the corrosion boundary, the thickness of the corrosion part is measured by an ultrasonic thickness gauge to obtain preliminary depth data; The corroded part is imaged by an image sensor, and the texture features of the corroded part are obtained by image processing technology; The preliminary depth data is corrected according to the texture feature to calculate the depth data.

3. According to claim 2, the method for adaptive rust removal on the surface of large ships based on laser rust removal is characterized in that: The method of obtaining the texture features of the rusted part by using image processing technology includes: The image of the rusted part is converted from the RGB model to the HSV color model, and the three HSV channels are subjected to median filtering to obtain a median filtered image. The reflection intensity of each pixel is calculated according to the color balance and contrast of the image of the rusted part and the median filtered image. Calculating the gray level co-occurrence matrix of the image of the corroded part, and calculating the texture information of the corroded part according to the reflection intensity and the gray level co-occurrence matrix, wherein the texture information includes texture directionality, roughness and unevenness; The texture information is characterized by using a BRIEF descriptor, a feature descriptor of the texture information is calculated, the texture information is converted into high-dimensional data by using an MLP network, and the data is added to the feature descriptor to generate texture features.

4. According to claim 3, the method for adaptive rust removal on the surface of large ships based on laser rust removal is characterized in that: Correcting the preliminary depth data according to the texture feature includes: By performing Fourier transform analysis on the texture features, the periodic features of the concave-convex area on the surface of the rusted part are extracted, and the curvature distribution information of the concave-convex area on the surface of the rusted part is calculated according to the periodic features; constructing a surface concave-convex feature model according to the curvature distribution information, matching the surface concave-convex feature model with the preliminary depth data, and identifying nonlinear errors in the preliminary depth data; The nonlinear error is nonlinearly corrected by a curvature adjustment function to calculate depth data. The curvature adjustment function is used to adaptively adjust the correction amplitude of the preliminary depth data according to the curvature change in the corroded part.

5. According to the large ship surface adaptive rust removal method based on laser rust removal according to claim 1, it is characterized in that: The environmental data include temperature, humidity and wind speed, and the initial parameters include laser power, laser frequency and laser scanning speed.

6. The method for self-adaptive rust removal on the surface of large ships based on laser rust removal according to claim 5, characterized in that: The parameter updating network comprises: A parameter generator module, which is used to implement multi-level feature mapping through deep convolutional layers, convert input parameters into preliminary prediction vectors containing laser adjustment parameters, and adjust the preliminary prediction vectors through adaptive factors according to changes in depth data at different positions to generate initial laser adjustment parameters; The parameter discriminator module is used to receive the initial laser adjustment parameters, compare them with the optimized parameters in the historical best parameter library, analyze whether the generated parameters are consistent with the actual rust removal requirements, calculate the accuracy score, and if not, feed the accuracy score back to the parameter generator module, adjust the adaptive factor according to the accuracy score, and output the laser adjustment parameters if they are consistent.

7. The method for self-adaptive rust removal on the surface of large ships based on laser rust removal according to claim 6, characterized in that: The parameter generator module comprises: The local concave-convex features and overall depth information in the depth data are extracted through convolution operation, and the local concave-convex features and overall depth information are fused through the normalized embedded Gaussian function and converted into a feature vector matrix; Mapping the feature vector matrix and the initial parameters to a high-dimensional space, performing feature matching in the high-dimensional space, calculating a parameter matching vector according to a maximum pooling function, flattening the output of the maximum pooling function into a one-dimensional vector by a flattening function, and obtaining a preliminary prediction vector; Calculate the local depth gradient of each point in the eigenvector matrix, and obtain the depth change rate of the point according to the local depth gradient; Extracting surface roughness information according to local concave-convex features, drawing a roughness distribution map, and obtaining the roughness value of each point through the roughness distribution map; The position factor of the point is calculated according to the depth change rate and the roughness value, and each point is traversed. According to the position factor, the adaptive factor is calculated by weighted average. The calculation formula of the position factor is: Among them, α x,y Represents the position factor of the point (x, y), d x,y Indicates the depth change rate of point (x, y), D x,y represents the maximum depth data of the point (x, y), μ represents the average value of the depth data, and σ d represents the variance of the depth data, β1 represents the adaptive adjustment coefficient, which is used to control the contribution weight of the depth gradient to the position factor, γ represents the smoothing factor, β2 represents the adaptive adjustment coefficient, which is used to control the contribution weight of the roughness to the position factor, and C x,y Represents the roughness value of the point (x,y).

8. The method for self-adaptive rust removal on the surface of large ships based on laser rust removal according to claim 1, characterized in that: The three-dimensional position marking of the rusted parts on the surface of the ship includes: constructing a three-dimensional model of the ship according to the structural design drawing, and marking all the rusted parts on the surface of the ship in the three-dimensional model, constructing a three-dimensional rectangular coordinate system of the ship with any point in the three-dimensional model as the origin, obtaining the three-dimensional coordinates of the center points of all the marked rusted parts on the surface of the ship as the position coordinates of all the corresponding rusted parts, and storing the position coordinates of all the corresponding rusted parts and the thickness of all the corresponding rusted parts in a rusted part coordinate set.

9. The method for self-adaptive rust removal on the surface of a large ship based on laser rust removal according to claim 8, characterized in that: The method further includes a laser rust removal path planning strategy, which is used to plan a rust removal work path according to a set of rusted part coordinates, including: Get the number of corroded parts on the ship, set it as Q; take the center point of any corroded part on the ship as the starting point of the path, set it as the jth corroded part, and let j = 1; Obtain the center point position coordinates of the jth rusted part from the rusted part coordinate set, calculate the distance between the center point position coordinates of the jth rusted part and the center point position coordinates of the remaining Qj rusted parts by using the Euclidean distance formula, and take the corresponding rusted part with the smallest distance as the j+1th path node, and let j=j+1; Repeat the steps until j=Q, and take the Qth path node as the path end point to obtain the laser rust removal path.

10. The method for self-adaptive rust removal on the surface of large ships based on laser rust removal according to claim 7, characterized in that: The surface roughness is used to update the adaptive factor and the parameter discriminator module.

Images