Laser cladding effect evaluation method, equipment and medium

Abstract

The invention discloses a laser cladding effect evaluation method and device and a medium, and relates to the field of data processing.The method comprises the steps that an attention mechanism is introduced into a full-connection neural network architecture to construct a laser cladding result prediction model; laser cladding variables are obtained, and the laser cladding variables are preprocessed; inputting the preprocessed laser cladding variables into the laser cladding result prediction model to obtain a laser cladding prediction result; and obtaining a laser cladding effect evaluation result based on the laser cladding prediction result. According to the laser cladding effect evaluation method and device, the limitation problem existing in the process of coping with inherent complex nonlinear correlation and multi-parameter mutual coupling effects in the laser cladding process can be solved, and meanwhile, the accuracy and real-time performance of laser cladding effect evaluation can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and particularly to a method, device and medium for evaluating the effect of laser cladding. Background Art

[0002] Laser cladding is a key surface modification technology, which is characterized by creating minimal dilution, controlling the heat affected zone, and having excellent metallurgical bonding with the substrate material. During the laser cladding process, various substrate materials exhibit extremely significant differences in thermophysical properties such as density, melting temperature, and coefficient of thermal expansion. Given the characteristics of rapid heating and rapid cooling of the laser cladding layer, these material property differences will induce many defects, such as irregular morphology, porosity, and cracking. Process parameters have important effects on the temperature field, flow field, and chemical reactions in the molten pool. Therefore, optimizing these parameters according to material properties is necessary to obtain high-quality coatings.

[0003] Traditional methods for optimizing laser cladding parameters include the Taguchi method, Relative Standard Error (RSE), and analysis of variance. Recent developments have introduced intelligent algorithms, including Particle Swarm Optimization (PSO), Gradual Admixture (GA) model, and gene expression programming. However, these methods have obvious limitations when dealing with the complex non-linear relationships and multi-parameter coupling effects inherent in the laser cladding process. Summary of the Invention

[0004] To solve the above problems, the present application provides a method, device and medium for evaluating the effect of laser cladding.

[0005] To achieve the above object, the present application provides the following solutions: In a first aspect, the present application provides a method for evaluating the effect of laser cladding, including: Introducing an attention mechanism into the fully connected neural network architecture to construct a laser cladding result prediction model; Obtaining laser cladding variables and preprocessing the laser cladding variables; Inputting the preprocessed laser cladding variables into the laser cladding result prediction model to obtain a laser cladding prediction result; Obtaining a laser cladding effect evaluation result based on the laser cladding prediction result.

[0006] Optionally, the attention mechanism is used to dynamically allocate weights to the input preprocessed laser cladding variables to determine the correlation weights between the query vector and the key-value pair.

[0007] Optionally, the laser cladding result prediction model includes at least three input neurons and at least three output neurons.

[0008] Optionally, when the laser cladding result prediction model includes three input neurons and three output neurons, the laser cladding variables are ceramic content, laser power, and scanning speed; the laser cladding prediction results are aspect ratio, cladding angle, and dilution rate.

[0009] Optionally, preprocessing the laser cladding variables includes: Using a standard scaling function to standardize the laser cladding variables.

[0010] Optionally, inputting the preprocessed laser cladding variables into the laser cladding result prediction model to obtain the laser cladding prediction results, including: Inputting the preprocessed laser cladding variables into the laser cladding result prediction model to obtain the original prediction results; Performing inverse standardization processing on the original prediction results to obtain the laser cladding prediction results.

[0011] Optionally, introducing an attention mechanism into the fully connected neural network architecture to construct a laser cladding result prediction model, including: Conducting multiple groups of laser cladding experiments based on the orthogonal test method, and during the experiments, collecting three types of process parameters, namely laser power, scanning speed, and ceramic content, as input variables; Using an optical microscope to perform geometric characterization and data extraction on the cladding cross-section obtained from the laser cladding experiment, and measuring the aspect ratio, average cladding angle, and dilution rate of the cladding layer cross-section as output variables; Forming a sample data set based on the input variables and output variables; Dividing the sample data set into a training set and a test set; Introducing an attention mechanism into the fully connected neural network architecture to obtain an initial model; Using the training set and the test set to train and test the initial model, and using the stochastic gradient descent algorithm to optimize the parameters of the initial model until the evaluation metrics of the initial model after parameter optimization meet the set conditions, obtaining a trained model; the evaluation metrics include: loss function value and coefficient of determination; Taking the trained model as the laser cladding result prediction model.

[0012] In a second aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the steps of the above-provided laser cladding effect evaluation method.

[0013] In a third aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-provided laser cladding effect evaluation method are implemented.

[0014] In a fourth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the above-provided laser cladding effect evaluation method are implemented.

[0015] According to the specific embodiments provided by the present application, the present application has the following technical effects: The present application provides a laser cladding effect evaluation method, device and medium. By introducing an attention mechanism into the fully connected neural network architecture, dynamic feature importance weighting can be achieved, and the interaction analysis between parameters can be enhanced. Furthermore, when dealing with the complex non-linear correlations and multi-parameter coupling effects inherent in the laser cladding process, the limitations can be solved. Moreover, the present application uses the constructed laser cladding result prediction model to predict the laser cladding parameters, and then obtains the laser cladding effect evaluation result based on the laser cladding prediction result, which can improve the accuracy and real-time performance of the laser cladding effect evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 It is a schematic flow chart of a laser cladding effect evaluation method provided by an embodiment of the present application; Figure 2 It is a schematic flow chart of the attention mechanism addressing operation provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of a laser cladding result prediction model provided by an embodiment of the present application; Figure 4 It is a schematic diagram of the cross-sectional geometry and measurement parameters provided by an embodiment of the present application; Figure 5 It is a schematic flow chart of data training and test processing provided by an embodiment of the present application; Figure 6 It is a schematic diagram of the mean absolute error of different neuron configurations provided by an embodiment of the present application; Figure 7 It is a schematic diagram of the coefficient of determination of different neuron configurations provided by an embodiment of the present application; Figure 8 Schematic diagram of the comparison result of the loss value between the predicted value and the actual value provided by an embodiment of the present application; Figure 9 Schematic diagram of the comparison result of the determination coefficient between the predicted value and the actual value provided by an embodiment of the present application; Figure 10 Schematic diagram of the structure of a computer device provided by an embodiment of the present application. Specific implementation manners

[0018] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0019] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0020] In an exemplary embodiment, the present application provides a method for evaluating the laser cladding effect. This method is executed by a computer device, and specifically can be executed alone by a computer device such as a terminal or a server, or can be jointly executed by a terminal and a server. In the embodiments of the present application, this method is described by taking its application to a server as an example. As Figure 1 shown, this method includes: Step 100: Introduce an attention mechanism into the fully connected neural network architecture to construct a laser cladding result prediction model. Among them, the attention mechanism performs dynamic weight allocation on the input variables, and the fully connected neural network (FCNN) architecture performs deep non-linear modeling on the weighted features.

[0021] Step 101: Obtain laser cladding variables and preprocess the laser cladding variables.

[0022] Step 102: Input the preprocessed laser cladding variables into the laser cladding result prediction model to obtain a laser cladding prediction result.

[0023] Step 103: Obtain a laser cladding effect evaluation result based on the laser cladding prediction result.

[0024] By implementing the above steps 100-step 103, the limitations existing when dealing with the complex non-linear associations and multi-parameter coupling effects inherent in the laser cladding process can be solved. Moreover, the accuracy and real-time performance of the laser cladding effect evaluation can also be improved.

[0025] In another exemplary embodiment of the present application, in order to focus on the parameter interaction pattern rather than sequential data, the core of the attention mechanism introduced in step 100 of the present application is scaled dot-product attention, which mainly maps a query and a set of key-value pairs to an output. Based on this, by using the attention mechanism and dynamically calculating the correlation weights between the query vector and the key-value pairs, intelligent focusing on key parameters and effective suppression of redundant information can be achieved. Among them, the mathematical formal definition of the attention mechanism is as follows: .

[0026] In the formula, Q 、 K and V represent the query matrix, the key matrix, and the value matrix respectively, d k is the dimension of the key vector, and T is the time step. The Softmax function realizes normalization. The scaling factor can prevent the magnitude of the dot product from becoming too large, thus pushing the Softmax function into the region with extremely small gradients.

[0027] Based on the above description, for each element in the fully connected neural network, a scalar weight value is assigned and used to generate a vector representation of the processed data. Among them, the scoring function can be expressed as : .

[0028] In the formula, and represent the input vector and the weight vector respectively, W represents the weight matrix of the layer, is the transpose of the matrix.

[0029] The multi-head attention mechanism is one of the core components in the Transformer model. By decomposing the attention calculation into multiple different subspaces, the model's ability to capture different feature patterns is enhanced. Based on this, in order to simultaneously capture different aspects of parameter interaction, the following attention mechanism is implemented: .

[0030] .

[0031] In the formula, is the mathematical formula of the multi-head attention mechanism. Through the operation, the outputs of each attention head are concatenated into a high-dimensional vector, is the attention function formula, is the definition of each attention head, is Q 、K and V parameter matrices, which can project the input into different representation subspaces. is the output projection matrix. In the implementation process, are 1 to h parallel attention heads, and h = 4 parallel attention heads can be used, allowing the model to jointly focus on different parameter interaction patterns.

[0032] The attention mechanism enables the fully connected neural network to assign different weights to the input parameters according to their relevance to each output metric. This method is different from traditional neural networks, in which all inputs are equally important regardless of the specific output being predicted.

[0033] For the laser cladding process, this function of the attention mechanism is particularly valuable because the effects of parameters [e.g., ceramic content (also known as carbon content, WC), laser power, scanning speed] on different quality metrics (e.g., aspect ratio, cladding angle, dilution rate) vary significantly. The attention weights provide interpretable insights into the importance of the parameters, addressing a key limitation of traditional black-box neural network methods.

[0034] As Figure 2 shown, the addressing operation process of the attention mechanism (Attention) is given, highlighting how to calculate, normalize, and apply the weight coefficients to extract relevant features from the input parameters (Query). Figure 2 In Key represents the key matrix element, and Value is 1, 2, 3, 4.

[0035] Among them, the addressing operation process of the attention mechanism mainly includes the following three stages.

[0036] (1) Weight coefficient calculation: Calculate the similarity scores between the query and each key.

[0037] (2) Normalization processing: Apply the SoftMax function to convert the similarity scores (s1, s2, s3, s4, etc.) into normalized weights. Among them, corresponding to 4 parallel attention heads, there are four normalized weights a1, a2, a3, and a4.

[0038] (3) Weighted integration: Multiply the normalized weights by the corresponding values and add them to obtain the final output of the attention mechanism.

[0039] In another exemplary embodiment of the present application, in order to achieve dynamic feature importance weighting and enhance parameter interaction analysis, the present application incorporates an attention mechanism into the FCNN framework to obtain an attention-enhanced FCNN model (i.e., the laser cladding result prediction model). Based on this, the laser cladding result prediction model adopted above in the present application includes at least three input neurons and three output neurons.

[0040] As Figure 3 shown, the obtained laser cladding result prediction model includes three input dimensions (WC, laser power, scanning speed), three output dimensions (aspect ratio, cladding angle, dilution rate), and five hidden layers. The neuron count follows a geometric progression to facilitate effective feature extraction and processing.

[0041] For example, the neuron count following a geometric progression is following a geometric series (4 → 8 → 16 → 8 → 4) and 3 output neurons (aspect ratio, cladding angle, dilution rate).

[0042] Among them, the attention mechanism dynamically weights the input features before processing through the fully connected layer.

[0043] In another exemplary embodiment of the present application, in order to improve the performance of the laser cladding result prediction model in the present application and facilitate anomaly detection, in this embodiment, the way to preprocess the laser cladding variables in step 101 above can be standardization processing.

[0044] For example, use the standard scaling function of the sklearn package to perform z-score standardization on the laser cladding variables (i.e., input data), and convert the features into a standard normal distribution ( μ = 0, σ = 1). The specific conversion process is as follows: .

[0045] Among them, represents the input data, represents the standardized data, represents the mean of the jth feature, , represents the number of data. represents the standard deviation of the jth feature, and there is: .

[0046] Based on the above description, the preliminary data has been standardized. Then, during the subsequent process of obtaining the laser cladding prediction results, inverse standardization processing can be further performed. Based on this, the implementation process of step 102 provided above in the present application can be described as: Step 1: Input the preprocessed laser cladding variables into the laser cladding result prediction model to obtain the original prediction result.

[0047] Step 2: Perform inverse normalization processing on the original prediction result to obtain the laser cladding prediction result.

[0048] In another exemplary embodiment of the present application, in order to improve the accuracy of the laser cladding result prediction model, in this embodiment, the implementation process of the above step 100 can be replaced by the following steps 1 - 7.

[0049] Step 1: Conduct multiple groups of laser cladding experiments based on the orthogonal experiment method, and during the experiment, collect three types of process parameters, namely laser power, scanning speed, and ceramic content, as input variables.

[0050] For example, design 27 groups of Fe60-based WC composite coating experiments based on the orthogonal experiment method. Among them, the orthogonal experiment method can adopt a three-factor and three-level design, and the parameter ranges of the input variables are respectively: laser power 1000W, 1100W, 1200W, scanning speed 300mm / min, 400mm / min, 500mm / min, and ceramic content 10wt.%, 15wt.%, 20wt.%.

[0051] Step 2: Use a high-resolution optical microscope to perform geometric characterization and data extraction on the cladding cross-section obtained from the laser cladding experiment, and measure the aspect ratio, average cladding angle, and dilution rate of the cladding layer cross-section as output variables.

[0052] Step 3: Form a sample data set based on the input variables and output variables.

[0053] Step 4: Divide the sample data set into a training set and a test set.

[0054] Step 5: Introduce an attention mechanism into the fully connected neural network architecture to obtain an initial model.

[0055] Step 6: Use the training set and the test set to train and test the initial model, and use the Stochastic Gradient Descent (SGD) algorithm to optimize the parameters of the initial model until the evaluation indicators of the initial model after parameter optimization meet the set conditions, and obtain the trained model; the evaluation indicators include: loss function value and coefficient of determination.

[0056] For example, verify the consistency between the model prediction data and the actual data by monitoring that the loss function value drops to 1.04 and stabilizes, and the batch coefficient of determination reaches the optimal value. Among them, the model accuracy verification uses the coefficient of determination R², and the calculation method is: 。

[0057] In the formula, is the number of single-pass cladding experiment samples, is the measured value of the single-pass cladding experiment, is the fitted value of the single-pass cladding layer performance index model, is the average value of all measured values obtained from single-pass cladding experiments.

[0058] Step 7: Use the trained model as the laser cladding result prediction model.

[0059] In another exemplary embodiment of the present application, a Q235 steel plate with dimensions of 90 mm×50 mm×2 mm is used as the substrate for laser cladding experiments to obtain experimental data.

[0060] Among them, before laser cladding, the surface of the substrate is ground with silicon carbide sandpaper and ultrasonically cleaned in alcohol.

[0061] An XL-F2000W fiber laser processing system is used as the laser system to improve the beam quality and operation stability. Based on the XL-F2000W fiber laser processing system, a high-precision optical system and a computer-controlled workpiece operation platform are equipped to form a laser cladding experimental device. The defocus distance is set to 5 mm above the substrate surface, and the laser spot diameter is 2.5 mm. The powder collection efficiency and coating quality are optimized through preliminary experiments.

[0062] Environmental control is achieved through high-purity argon (99.99%) shielding gas. A 20-minute pre-gas flow cycle is carried out before starting the cladding process to ensure complete removal of residual oxygen in the processing area.

[0063] Based on the above content, a comprehensive experimental design is implemented to study the relationship between process parameters and coating characteristics. Three key variables are determined: WC(ω), laser power (p), and scanning speed (v).

[0064] A full-factorial experimental design is adopted to systematically explore the parameter space and identify potential interaction effects.

[0065] The experimental training data matrix consists of 27 unique parameter combinations, allowing a comprehensive analysis of parameter effects and interactions. Based on theoretical considerations and preliminary experimental results, the selection range of each parameter is determined.

[0066] WC was varied at three levels (10%, 15%, 20%) to study its effect on the microstructure and properties of the coating. Based on preliminary melting tests, laser power levels (1000 W, 1100 W, 1200 W) were selected to ensure complete melting of the powder while avoiding excessive dilution of the substrate. Scanning speeds (300, 400, 500 mm / min) were chosen to achieve appropriate energy input and coating thickness while maintaining process stability.

[0067] The experimental matrix was obtained by comprehensively analyzing the parameter effects and interactions, resulting in 27 sets of experimental data. Among them, 24 sets were used as the training data set, and 3 sets were used as independent test data sets for model validation to verify the generalization ability of the model. The specific experimental scheme is shown in Table 1.

[0068] Table 1 Experimental Scheme Setting Table

[0069] According to the strict metallographic preparation procedure, cross-sectional analysis of the clad layer was carried out. The clad samples were sliced perpendicular to the cladding direction using a DK7735 wire-cut electrical discharge machining (EDM) system to obtain sliced samples.

[0070] The sliced samples were systematically ground using a series of silicon carbide papers (800 - 2000 grit size) and then finely polished to obtain a mirror finish suitable for microstructure analysis.

[0071] Based on the above experimental process, the experimental data obtained consisted of 27 unique parameter combinations, allowing for a comprehensive analysis of parameter effects and interactions. Table 2 shows the complete experimental data.

[0072] Table 2 Experimental Data Table

[0073] In Table 2, W represents the coating width, H represents the coating height, η represents the aspect ratio, and represent the left and right cladding angles respectively, θ represents the average cladding angle, Am represents the melted area of the substrate, Ac represents the clad layer area above the substrate surface, and D represents the dilution rate.

[0074] Specialized image analysis software was used to measure, analyze, and quantify the key parameters, including: Aspect ratio (η): η = W / H.

[0075] Average cladding angle (θ): .

[0076] Dilution rate (D): D = Am / (Ac + Am).

[0077] Furthermore, to ensure accurate area measurement, a pixel-based digital image analysis method is implemented using the formula A = (n × S) / m. Here, A represents the actual area, n is the pixel count of the measured area, S is the calibrated reference area, and m is the pixel count of the reference area. Figure 4 Details of the cross-sectional geometry and measurement parameters are provided. All measurements are carried out under standardized conditions to ensure reproducibility, and each sample is analyzed at multiple locations to account for potential variations in the coating geometry.

[0078] The experimental process is carried out under carefully controlled environmental conditions (temperature: 20 ± 2 °C, relative humidity: 45 ± 5%) to minimize the influence of external variables on the coating process and subsequent analysis.

[0079] Based on the above description, a final set of 27 sample datasets is obtained. From these 27 sample datasets, 24 are randomly selected as the training set to train the initial model, and the remaining 3 are used as the test set for independent testing to verify the generalization ability of the initial model. The training process uses a learning rate of 0.001 and performs 2000 iterations, and the loss value is recorded after each training epoch.

[0080] The initial model is constructed by inheriting nn of PyTorch to ensure flexibility and scalability. Based on the above description, as Figure 5 shown, the data training and test processing flow includes: the input data undergoes z-score standardization and attention-weighted feature extraction, forward propagation is performed through multiple fully connected layers with non-linear activation, and finally predictions are made for three output metrics (aspect ratio, cladding angle, and dilution rate).

[0081] During the training process of the initial model, the parameters of the FCNN architecture are optimized using the Stochastic Gradient Descent (SGD) algorithm. The optimization process can be implemented using the SGD optimizer in the optim module of PyTorch. Among them, using the smooth L1 loss as the performance metric, there is: .

[0082] In the formula, represents the loss function, represents the predicted value, represents the true value, represents the number of samples.

[0083] To prevent overfitting, L1 regularization is adopted, and there is: .

[0084] In the formula, λ represents the L1 regularization parameter sum, and represents the absolute value of the model weights, Represents the L1 regularization result, represents the loss term MSE.

[0085] This network structure has interconnected neurons between adjacent layers, and the assignment is made through matrix operations. The weight update is as follows: .

[0086] .

[0087] In the formula, represents the updated weight, represents the weight before update, represents the learning rate, represents the loss function, represents the gradient of the loss function represents the hyperbolic tangent function. As an activation function, the hyperbolic tangent function can introduce non-linear characteristics, enabling the model to learn complex patterns. represents the analog input, represents the natural logarithm.

[0088] Furthermore, to enhance the generalization of the laser cladding result prediction model, a dropout layer with a probability of 0.5 is implemented on each network node. The mean squared error (MSE) and the coefficient of determination (R-squared, R²) are used as metrics to evaluate the performance of the obtained model. Among them, R² is used as the main performance metric. The model performance when the number of neurons in the hidden layer changes is as Figure 6 and Figure 7 shown. When there are 4 neurons, both the MSE and R² perform best. It can be seen that over-simplification (2 neurons) and over-complication (>4 neurons) will both lead to a decrease in the model's performance.

[0089] Different from the traditional method using a fixed network architecture, this application performs a systematic analysis to determine the optimal neuron configuration. The analysis shows that the performance is best when there are 4 neurons in the first hidden layer, with an R² value of 0.825 and an MSE value of 0.99.

[0090] Based on the above description, the laser cladding result prediction model provided by this application shows good prediction ability for all three coating quality indicators. Figure 8 and Figure 9Shows the comparison results between the predicted values and the actual values of the aspect ratio, cladding angle, and dilution rate over the entire test set. Based on this comparison result, it can be obtained that: (a) The aspect ratio has high prediction accuracy, R² = 0.825. (b) The cladding angle has high prediction accuracy, R² = 0.817. (c) The dilution rate has the highest accuracy among the three indicators when R² = 0.831. The close alignment between the predicted (red line) values and the actual (blue line) values shows the excellent performance of the model on all three quality indicators.

[0091] The model has high accuracy, and the R² values of the aspect ratio, cladding angle, and dilution rate are 0.825, 0.817, and 0.831 respectively. The corresponding MSE values are 0.99, 1.12, and 0.95 respectively. These results indicate that the model effectively captures the complex relationship between the processing parameters and the coating characteristics.

[0092] To ensure robust validation when the dataset size is limited, five-fold cross-validation can be performed. This validation method allows for a more reliable performance evaluation while maximizing the use of available data. The cross-validation results show that the performance of all folds is consistent, and the standard deviations of the R² values of the three output indicators are 0.031, 0.028, and 0.035 respectively, confirming the stability and generalization ability of the model.

[0093] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal, and its internal structure diagram can be as Figure 10 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store laser cladding effect evaluation data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a method for evaluating the laser cladding effect.

[0094] Those skilled in the art can understand that Figure 10The structure shown is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0095] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the steps in the above method embodiments are implemented.

[0096] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0097] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0098] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0099] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memory (RRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0100] The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0101] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0102] In this text, specific examples are used to illustrate the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application; at the same time, for those of ordinary skill in the art, according to the idea of the present application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present application.

Claims

1. A method for evaluating laser cladding effect, characterized in that: include: Introducing the attention mechanism into the fully connected neural network architecture to build a laser cladding result prediction model; Acquire laser cladding variables and pre-process the laser cladding variables; Inputting the preprocessed laser cladding variables into the laser cladding result prediction model to obtain a laser cladding prediction result; A laser cladding effect evaluation result is obtained based on the laser cladding prediction result.

2. The laser cladding effect evaluation method according to claim 1, characterized in that: The attention mechanism is used to dynamically assign weights to the input preprocessed laser cladding variables to determine the correlation weights between the query vector and the key-value pairs.

3. The laser cladding effect evaluation method according to claim 1, characterized in that: The laser cladding result prediction model includes at least three input neurons and at least three output neurons.

4. The laser cladding effect evaluation method according to claim 3, characterized in that: When the laser cladding result prediction model includes three input neurons and three output neurons, the laser cladding variables are ceramic content, laser power and scanning speed; and the laser cladding prediction results are aspect ratio, cladding angle and dilution rate.

5. The laser cladding effect evaluation method according to claim 1, characterized in that: Pre-processing of laser cladding variables, including: The laser cladding variables are standardized using the standard calibration function.

6. The laser cladding effect evaluation method according to claim 5, characterized in that: The preprocessed laser cladding variables are input into the laser cladding result prediction model to obtain the laser cladding prediction results, including: Inputting the preprocessed laser cladding variables into the laser cladding result prediction model to obtain an original prediction result; The original prediction result is subjected to de-normalization processing to obtain the laser cladding prediction result.

7. The laser cladding effect evaluation method according to claim 1, characterized in that: The attention mechanism is introduced into the fully connected neural network architecture to build a laser cladding result prediction model, including: Based on the orthogonal test method, multiple groups of laser cladding experiments were carried out. During the experiments, three types of process parameters, namely laser power, scanning speed and ceramic content, were collected as input variables. An optical microscope was used to perform geometric characterization and data extraction on the cladding cross section obtained from the laser cladding experiment, and the aspect ratio of the cladding layer cross section, the average cladding angle, and the dilution rate were measured as output variables. Form a sample data set based on the input variables and the output variables; Dividing the sample data set into a training set and a test set; Introduce the attention mechanism into the fully connected neural network architecture to obtain the initial model; The initial model is trained and tested using the training set and the test set, and the parameters of the initial model are optimized using a stochastic gradient descent algorithm until the evaluation index of the initial model after parameter optimization meets the set conditions, thereby obtaining a trained model; the evaluation index includes: a loss function value and a determination coefficient; The trained model is used as the laser cladding result prediction model.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the laser cladding effect evaluation method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the laser cladding effect evaluation method described in any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the laser cladding effect evaluation method described in any one of claims 1 to 7 is implemented.

Images