Laser cladding powder proportioning hardness prediction method, storage medium and electronic equipment

Abstract

The invention discloses a method for predicting the proportion hardness of laser cladding powder, a storage medium and electronic equipment, and the method specifically comprises the steps: inputting the proportion of to-be-predicted laser cladding powder into a cladding material hardness full-connection layer neural network, and predicting the hardness of the to-be-predicted laser cladding powder. In practical application, the ratio of the laser cladding powder can be obtained according to the prediction result, the laser cladding powder is prepared according to the ratio of the laser cladding powder, and laser cladding is carried out on a base material, so that additive manufacturing can be realized. According to the method, prediction of the proportion hardness of the laser cladding powder is achieved, time and cost are saved, the problem that in the prior art, the powder proportion is difficult to predict is solved, and the R2 evaluation result is as high as 0.98.

Description

Technical Field

[0001] The invention belongs to the technical field of laser cladding, and in particular relates to a method for predicting the hardness of laser cladding powder ratio, a storage medium, and an electronic device. Background Art

[0002] At present, to manufacture parts similar to dumbbell shapes, traditional machining equipment is usually used to reduce the middle part of the rod to obtain the finished metal parts, which is not only costly but also seriously wastes materials. In recent years, the processing method of additive manufacturing has been promoted to a certain extent. On the basis of the original parts, the size can be increased and processed into the required shape to save costs and carry out mass production. However, in the actual application process, it is found that what kind of powder to use in the additive process and the powder ratio are a major problem. In existing production, the determination of the powder ratio usually depends on the experience and experiments of engineers. Engineers conduct a large number of experiments based on the properties of each raw material to find the best ratio. It is not only time-consuming and laborious, but also may not be able to obtain the optimal ratio, resulting in the hardness of the added part not matching the hardness of the original product. The gap is too large, which also wastes time and cost.

[0003] Therefore, in response to the above problems and technical requirements, it is necessary to develop a new method that can predict the corresponding hardness based on the powder ratio to solve the problem of mismatch between the hardness of the added part and the original product in additive manufacturing. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a method for predicting the hardness of laser cladding powder ratio, a storage medium, and an electronic device, which can realize the prediction of the hardness of laser cladding powder ratio without spending a lot of time on experiments, saving time and cost. It can be applied to laser cladding projects under complex working conditions, and solves the problem of difficult prediction of powder ratio in the prior art.

[0005] A method for predicting the hardness of laser cladding powder ratio is proposed. The laser cladding powder ratio to be predicted is input into the cladding material hardness fully connected layer neural network to predict its hardness. The laser cladding powder ratio to be predicted is input into the cladding material hardness fully connected layer neural network to directly obtain its hardness without spending a lot of experiments to find the target hardness and the optimal ratio. It can be applied to complex working conditions.

[0006] A method for additive manufacturing using a laser cladding powder ratio hardness prediction method, wherein the laser cladding powder ratio to be predicted is input into a cladding material hardness fully connected layer neural network to predict its hardness, the laser cladding powder ratio is obtained according to the prediction result, laser cladding powder is configured according to the laser cladding powder ratio, and then laser cladding is performed on a substrate to achieve additive manufacturing.

[0007] In the present invention, the construction of the fully connected layer neural network of the cladding material hardness includes the following steps: (1) Obtain a data set and divide the data into a training set and a test set; (2) Using nn.Module as the parent class, build a fully connected neural network; (3) Import the training set into the fully connected layer neural network and train the fully connected layer neural network; (4) The test set is imported into the trained fully connected layer neural network, and the fully connected layer neural network is evaluated and tuned to obtain the fully connected layer neural network of the cladding material hardness.

[0008] In this invention, the data sets are derived from publicly available literature and experiments.

[0009] In the present invention, the fully connected layer neural network includes an input layer, 7 hidden layers and an output layer connected in sequence. Preferably, a ReLU activation function is added after the input layer and each hidden layer; a Sigmoid activation function is added after the output layer. The input layer is used to receive external input data and pass it to the hidden layer, and the hidden layer is used to perform nonlinear transformation on the input data and extract useful features; adding a ReLU activation function after the input layer allows the input data to be nonlinearly transformed before being passed to the first hidden layer to ensure that the input data has good distribution and nonlinear characteristics.

[0010] In the present invention, the test set is imported into the trained fully connected layer neural network, the predicted hardness is calculated, the predicted hardness and the actual hardness are imported into the loss function, the loss function value is calculated, and the fully connected layer neural network is updated based on the loss function value. Further preferably, based on the loss function value, the weights of the fully connected layer neural network are updated using the gradient descent method, and the weights corresponding to the minimum value of the loss function are saved. The gradient descent method is a conventional technique used to update the weights of the fully connected layer neural network so that the network's prediction is closer to the actual target value; the loss function value is used to measure the difference between the prediction result of the fully connected layer neural network and the actual target value, and the weights corresponding to the minimum value of the loss function are saved to ensure that the optimal fully connected layer neural network is used, improve the generalization ability of the fully connected layer neural network, and avoid overfitting.

[0011] The present invention discloses a storage medium storing a program executable by a processor, wherein the program is used to execute the above-mentioned laser cladding powder ratio hardness prediction method. Furthermore, the present invention discloses an electronic device comprising a processor and a memory storing a program for executing the above-mentioned laser cladding powder ratio hardness prediction method.

[0012] The invention discloses the application of the laser cladding powder proportion hardness prediction method in predicting the laser cladding powder proportion hardness.

[0013] The present invention discloses the application of the above-mentioned method for predicting the hardness of the laser cladding powder ratio in additive manufacturing. Preferably, the above-mentioned method for predicting the hardness of the laser cladding powder ratio is applied in additive manufacturing using laser cladding powder.

[0014] Due to the application of the above technical solution, the beneficial effects of the present invention are: (1) Based on the deep learning model, the present invention creatively proposes a method for predicting the hardness of laser cladding powder ratio. By inputting the laser cladding powder ratio to be predicted into the fully connected layer neural network of the cladding material hardness, the corresponding hardness can be directly obtained. There is no need to spend a lot of experiments to find the target hardness and the optimal ratio, and it can be applied to complex working conditions. (2) The fully connected neural network of the present invention can automatically learn the features in the data without manually designing the features, and can better discover the potential laws and associations in the data, and realize the prediction of the hardness of the laser cladding powder ratio, R 2 The evaluation result is as high as 0.98; (3) The fully connected neural network of the present invention is composed of multiple levels, which can extract complex features in the data through multi-level nonlinear transformation, has better data fitting ability, and can more accurately fit the relationship between the laser cladding powder ratio and hardness; (4) The present invention updates the fully connected layer neural network through the gradient descent method during the training process of the fully connected layer neural network, thereby improving the generalization ability of the fully connected layer neural network, avoiding the overfitting problem, being able to better adapt to different data sets, and improving the stability of prediction; (5) The fully connected neural network of the present invention achieves high prediction accuracy through training and learning of a large number of data sets. At the same time, due to its automated characteristics, it can improve the prediction efficiency and save labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the structure of a fully connected layer neural network.

[0016] Figure 2 This is the ratio hardness prediction interface of the first embodiment.

[0017] Figure 3 It is the loss function image of the fully connected layer neural network in Example 1.

[0018] Figure 4 is the R obtained by running the fully connected neural network of the cladding material hardness in Example 1 2 value.

[0019] Figure 5 This is the sample after laser cladding in verification experiment 1.

[0020] Figure 6It is the hardness test sample of verification experiment 1, where 1 is the cladding layer and 2 is the substrate.

[0021] Figure 7 This is the metallographic diagram of the hardness test sample after the hardness is measured in the verification experiment one.

[0022] Figure 8 This is the mix hardness prediction interface for verification experiment one.

[0023] Fig. 9 It is the interface for predicting the ratio hardness of the second verification experiment.

[0024] Fig.10 These are the failure load images of the unclad rod and clad rod in the application experiment.

[0025] Fig.11 This is the state diagram of the unclad rod and clad rod after tensile test in the application experiment. DETAILED DESCRIPTION

[0026] In the present invention, the laser cladding powder ratio to be predicted is input into the fully connected layer neural network of the cladding material hardness, and its corresponding hardness can be directly obtained without a large number of experiments to obtain the target hardness and the optimal ratio, which solves the problem of difficult powder ratio prediction in the prior art and saves time and cost.

[0027] In the present invention, the construction of the fully connected layer neural network of the cladding material hardness includes the following steps: (1) Obtain a data set and divide the data into a training set and a test set; (2) Using nn.Module as the parent class, build a fully connected neural network; (3) Import the training set into the fully connected layer neural network and train the fully connected layer neural network; (4) The test set is imported into the trained fully connected layer neural network, and the fully connected layer neural network is evaluated and tuned to obtain the fully connected layer neural network of the cladding material hardness.

[0028] In the present invention, the data set is derived from publicly available literature and experiments, and the specific acquisition method is a conventional method. The data set of the present invention is collected from a public search platform using "laser cladding" and "coating performance" as keywords. The fully connected layer neural network includes an input layer, 7 hidden layers and an output layer connected in sequence, and the specific construction method is a conventional method.

[0029] Preferably, a ReLU activation function is added after the input layer and each hidden layer; a Sigmoid activation function is added after the output layer; the test set is imported into the trained fully connected layer neural network, the predicted hardness is calculated, the predicted hardness and the actual hardness are imported into the loss function, the loss function value is calculated, and the fully connected layer neural network is updated based on the loss function value. Further preferably, the weights of the fully connected layer neural network are updated using the gradient descent method, and the weights corresponding to the minimum value of the loss function are saved. The gradient descent method is a conventional technique and does not affect the understanding of the present invention by those skilled in the art.

[0030] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. The specific data acquisition, input / output operations and algorithms, and laser cladding process are conventional technologies. Embodiment 1

[0031] A method for predicting hardness of laser cladding powder ratio, comprising the following steps: (1) The hardness of laser cladding under different powder ratios was obtained from public literature and public experiments, with a total of 840 samples, of which 756 samples were used as training sets and 84 samples were used as test sets; (2) Using nn.Module as the parent class, connect the input layer, seven hidden layers, and the output layer in sequence, add a ReLU activation function after the input layer and each hidden layer, and add a Sigmoid activation function after the output layer to obtain a fully connected neural network. (3) Import the training set into the fully connected layer neural network and train the fully connected layer neural network; (4) Import the test set into the trained fully connected layer neural network, calculate the predicted hardness, import the predicted hardness and actual hardness into the loss function, calculate the loss function value, perform gradient calculation on the loss function value, update the weight of the fully connected layer neural network, save the weight corresponding to the minimum value of the loss function, and obtain the fully connected layer neural network of the cladding material hardness; (5) The laser cladding powder ratio to be predicted is input into the fully connected layer neural network of the cladding material hardness through the powder ratio hardness prediction software 1.0 to predict its hardness.

[0032] The present invention inputs the ratio into the fully connected layer neural network of the cladding material hardness through the powder ratio hardness prediction software 1.0. The powder ratio hardness prediction software 1.0 contains all the main elements and also has an input box for extremely trace elements, such as Figure 2 shown.

[0033] Record the loss value at the end of each training round (epoch) and draw a graph of the loss function, such as Figure 3As shown in the figure, in the early stage of training, the loss value drops rapidly, then gradually slows down, tends to be stable and close to zero, without large fluctuations, indicating that the fully connected layer neural network performs well.

[0034] The conventional method for evaluating the performance of the prediction classification network is used, and the R² value is used as the evaluation standard to detect the accuracy of the fully connected layer neural network of the cladding material hardness of the present invention. 2 The closer the value is to 1, the greater the proportion of the regression sum of squares to the total sum of squares. The closer the regression line is to each observation point, the better the regression fit is, which means the better the prediction effect of the network is. Figure 4 As shown, the R obtained by running the fully connected layer neural network of the cladding material hardness of the present invention is 2 value, approximately 0.98. Embodiment 2

[0035] A storage medium stores a program executable by a processor, wherein the program is used to execute the laser cladding powder proportion hardness prediction method of embodiment 1. Embodiment 3

[0036] An electronic device includes a processor and a memory, wherein the memory stores a program for executing the laser cladding powder proportion hardness prediction method of embodiment 1.

[0037] Verification experiment 1 Take Ni60 powder, including 60.32wt% Ni, 0.85wt% C, 4.29wt% Si, 3.41wt% B, 14.3wt% Fe, 16.83wt% Cr, select the substrate, debug the laser cladding equipment (including laser, powder feeder, focusing mirror and other components, all of which are existing products), use the synchronous powder feeding method for laser cladding, the laser beam is irradiated to the surface of the substrate, and the powder feeder starts to feed powder at the same time. The Ni60 powder is quickly melted under the action of the laser beam and forms a metallurgical bond with the surface of the substrate, such as Figure 5 The sample shown is the sample after laser cladding on the substrate.

[0038] Use a metallographic cutting machine (existing product) to cut the laser cladding samples to obtain hardness test samples, such as Figure 6 As shown, nine different points were taken from the longitudinal section surface of the cladding layer of the hardness test sample, and the surface hardness data were measured by Vickers hardness tester (existing equipment HV-1000Z), which were 720.0HV1, 753.0HV1, 734.4HV1, 741.8HV1, 706.0HV1, 753.0HV1, 741.8HV1, 720.0HV1, 713.0HV1, and the average value was 731.44HV1, as shown in Figure 7 Shown is the metallographic diagram of the hardness test sample after the hardness is measured.

[0039] The ratio is input into the powder ratio hardness prediction software 1.0 of Example 1, and the output hardness is 744.4175 HV1. Figure 8 As shown, the difference between the actual measured hardness and the average value is 1.77%.

[0040] Verification Experiment 2 Take 316L powder and add 25wt% Del40 powder to mix, including 30.03wt% Ni, 0.08wt% C, 1.4wt% Si, 0.5wt% B, 51.57wt% Fe, 14.21wt% Cr, 1.88wt% Mo, 0.4wt% Mn, select the substrate, debug the laser cladding equipment (including laser, powder feeder, focusing mirror and other components, all of which are existing products), use the synchronous powder feeding method for laser cladding, the laser beam is irradiated to the surface of the substrate, and the powder feeder starts to feed powder at the same time. The 316L-25wt% Del40 powder is quickly melted under the action of the laser beam and forms a metallurgical bond with the surface of the substrate.

[0041] The laser cladding samples were cut using a metallographic cutting machine (existing product) to obtain hardness test samples. Nine different points were taken from the longitudinal section of the cladding layer of the hardness test samples and the surface hardness data were measured using a Vickers hardness tester (existing equipment HV-1000Z). The results were 206.2HV 0.5 、202.8HV 0.5 、198.6HV 0.5 、207.4HV 0.5 、219.5HV 0.5 、208.5HV 0.5 、192.7HV 0.5 、206.2HV 0.5 、212.1HV 0.5 , take the average value 206HV 0.5 .

[0042] The ratio is input into the powder ratio hardness prediction software 1.0 of Example 1, and the output hardness is 194.59827HV1. Fig. 9 As shown, the difference between the actual measured hardness and the average value is 5.53%.

[0043] According to production practice, hardness difference within 8% will not affect the use of the product, and more than 10% is a defective product. Application Experiment

[0044] The method for additive manufacturing using the laser cladding powder ratio hardness prediction method of Example 1 is used to take the rod material on which the surface of the additive cladding is to be clad for hardness testing. The average hardness is 251HV. 0.5The amount of powder raw materials is input into the fully connected neural network of the cladding material hardness through the powder ratio hardness prediction software 1.0. According to the formula of predicted hardness, the required ratio is obtained (the predicted hardness is 268.41742HV 0.5 ), mix the powders according to the ratio, and clad the mixed powder on the surface of the rod by laser cladding equipment according to the conventional method (refer to Verification Experiment 1), complete the additive manufacturing, and obtain the clad rod.

[0045] The tensile tests of the unclad bars and clad bars were carried out using an electronic universal testing machine (current equipment Instron 5982), such as Fig.10 As shown in the figure, the failure load images of the unclad bar and the clad bar are shown. The failure load of the unclad bar is about 5% larger than that of the clad bar, which will not affect the application performance of the product. Fig.11 As shown, the state diagram of the unclad rod and the clad rod after tensile test.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the hardness of laser cladding powder ratio, characterized in that: The laser cladding powder ratio to be predicted is input into the fully connected layer neural network of the cladding material hardness to predict its hardness.

2. The method for predicting hardness of laser cladding powder ratio according to claim 1, characterized in that: The construction of the fully connected layer neural network for cladding material hardness includes the following steps: (1) Obtain a data set and divide the data into a training set and a test set; (2) Using nn.Module as the parent class, build a fully connected neural network; (3) Import the training set into the fully connected layer neural network and train the fully connected layer neural network; (4) The test set is imported into the trained fully connected layer neural network, and the fully connected layer neural network is evaluated and tuned to obtain the fully connected layer neural network of the cladding material hardness.

3. The method for predicting hardness of laser cladding powder ratio according to claim 2, characterized in that: The datasets are sourced from publicly available literature and experiments.

4. The method for predicting hardness of laser cladding powder ratio according to claim 2, characterized in that: The fully connected neural network includes an input layer, 7 hidden layers and an output layer connected in sequence; a ReLU activation function is added after the input layer and each hidden layer; and a Sigmoid activation function is added after the output layer.

5. The method for predicting hardness of laser cladding powder ratio according to claim 2, characterized in that: The test set is imported into the trained fully connected layer neural network, the predicted hardness is calculated, the predicted hardness and the actual hardness are imported into the loss function, the loss function value is calculated, and based on the loss function value, the weights of the fully connected layer neural network are updated using the gradient descent method, and the weight corresponding to the minimum value of the loss function is saved.

6. A storage medium storing a program executable by a processor, characterized in that: The program is used to execute the laser cladding powder ratio hardness prediction method described in claim 1.

7. An electronic device, comprising a processor and a memory, characterized in that: The memory stores a program for executing the laser cladding powder ratio hardness prediction method according to claim 1.

8. A method for additive manufacturing using a laser cladding powder ratio hardness prediction method, characterized in that: The laser cladding powder ratio to be predicted is input into the fully connected layer neural network of the cladding material hardness to predict its hardness. The laser cladding powder ratio is obtained according to the prediction result. The laser cladding powder is configured according to the laser cladding powder ratio, and then laser cladding is performed on the substrate to realize additive manufacturing.

9. Application of the method for predicting the ratio hardness of laser cladding powders as described in claim 1 in predicting the ratio hardness of laser cladding powders.

10. Application of the laser cladding powder ratio hardness prediction method according to claim 1 in additive manufacturing.

Images