Method for predicting surface roughness of titanium alloy based on Fe3O4 nanoparticle reinforced cutting fluid

Abstract

The invention provides a titanium alloy milling surface roughness prediction method based on Fe3O4 nano-particle reinforced cutting fluid. The titanium alloy milling surface roughness prediction method comprises the following steps: preparing the cutting fluid containing Fe3O4 nano-particles; establishing a milling force, milling noise and surface roughness acquisition system, and carrying out TC4 titanium alloy wet milling tests under different process parameters; drawing a change curve of milling force, noise and surface roughness test data, and comparing the performance difference before and after addition of the Fe3O4 nanoparticles; and constructing a prediction model based on a convolutional neural network and a relevance vector machine to realize accurate prediction of the surface roughness. The cutting fluid has the advantages that the Fe3O4 nanoparticles are added into the cutting fluid, so that the milling force and noise are remarkably reduced, the surface roughness is improved, and intelligent prediction of the surface quality can be effectively realized by combining experimental testing with a machine learning model.

Description

Technical Field

[0001] This invention relates to a method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid, belonging to the field of machining technology. Background Technology

[0002] In the milling of titanium alloys, cutting fluids are typically used to improve lubrication and cooling conditions in the cutting zone, thereby reducing tool wear, milling forces, milling noise, and improving workpiece surface quality. However, conventional cutting fluids have limited capabilities in conventional lubrication and heat transfer, making it difficult to fully meet the requirements of high-quality machining of titanium alloys. In recent years, with the development of nanotechnology, nanoparticles have been introduced into cutting fluids, forming nano-cutting fluids. These cutting fluids, with their excellent lubrication properties, exhibit advantages such as reduced wear, noise reduction, and improved surface quality during machining, and have gradually attracted research attention. However, there are few quantitative research results on the improvement of milling performance by existing nano-cutting fluids, and even fewer empirical formulas or predictive models related to surface roughness, making it impossible to achieve intelligent prediction of machined surface quality. Summary of the Invention

[0003] The technical problem to be solved by this invention is to overcome the problems of milling force and noise in the existing technology of titanium alloy processing, and thus provide a method for enhancing the surface roughness of titanium alloy based on Fe3O4 nanoparticles. By introducing a certain concentration of Fe3O4 nanoparticles into the cutting fluid, the milling force and milling noise can be effectively reduced, and the surface roughness of the machined surface can be improved. At the same time, this invention also provides a roughness prediction and comparison method, which combines an algorithm model to predict the surface roughness after milling, thereby realizing the prediction of the surface quality of the machined surface.

[0004] This invention provides a method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid, comprising the following steps: S1. Prepare a cutting fluid containing Fe3O4 nanoparticles. Mix the cutting fluid with deionized water to prepare a coolant. Add Fe3O4 nanoparticles and a thixotropic agent to the coolant and stir to obtain a composite nano-cutting fluid. S2. Build a test and acquisition system for milling force, milling noise and surface roughness, and use cutting fluid and composite nano cutting fluid to mill the workpiece respectively. The test and acquisition system collects the milling force, milling noise and surface roughness during the milling process. S3. Plot the change curves of milling force, noise and surface roughness test data, and compare the changes of milling force, milling noise and surface roughness before and after adding Fe3O4 nanoparticles. S4. Use a convolutional neural network combined with a correlation vector machine algorithm (CNN-RVM) to predict the surface roughness after milling with added Fe3O4 nanoparticles.

[0005] This invention introduces Fe3O4 nanoparticles into cutting fluid and applies it to the milling of TC4 titanium alloy. The changes in milling force, milling noise, and surface roughness under conditions of adding and not adding Fe3O4 nanoparticles were compared and analyzed, verifying the effectiveness of Fe3O4 nanoparticle cutting fluid in improving machining performance. Simultaneously, this invention combines an algorithm to predict the surface roughness after milling with Fe3O4 nanoparticles, providing a theoretical basis and technical reference for the application of nanoparticle cutting fluids in titanium alloy machining.

[0006] The method of using Fe3O4 nanoparticle cutting fluid to improve milling performance not only enhances the machining performance of titanium alloy milling, but also provides a feasible technical path for the application and promotion of nanoparticle cutting fluid.

[0007] The following is a further optimized technical solution of the present invention: In step S1, Castrol Syntilo 9930C fully synthetic cutting fluid was prepared with deionized water at a volume ratio of 1:20 to form a coolant. 1g of Fe3O4 nanoparticles and 1ml of SRE-4410TD rheological thixotropic agent were added to 500 mL of the coolant and stirred to obtain a light black composite nano-cutting fluid.

[0008] In step S2, the test acquisition system consists of a milling force signal acquisition section, a milling noise signal acquisition section, and a surface roughness acquisition section. The milling force signal acquisition section includes a charge amplifier, a piezoelectric sensor, and a triaxial milling force tester. The milling noise signal acquisition section includes a precision sound level meter, a high-speed data acquisition instrument, and a computer component. The surface roughness acquisition section includes a surface roughness measuring instrument. R a The surface roughness was measured using a YJD-950C surface roughness measuring instrument.

[0009] In step S2, the milling force feature value and milling noise feature value are extracted using a testing and acquisition system for milling force, milling noise, and surface roughness. The extracted milling force feature value includes... F x , F y , F z The extracted milling noise feature value is the root mean square value of the sound pressure level. L .

[0010] In step S4, the specific operation of predicting the surface roughness after milling using a convolutional neural network combined with a support vector machine algorithm is as follows: S4.1 Data preprocessing: Divide the data collected during the milling process into training set and test set, and normalize the input features and output target of the convolutional neural network. The normalization formula is: (17) in, x The original data, x min , x max These are the minimum and maximum values ​​of the original data, respectively. y The data is after normalization; S4.2 Data format conversion: Convert the normalized input features into a four-dimensional tensor format acceptable to convolutional neural networks; S4.3 Construct a convolutional neural network. A deep convolutional neural network includes convolutional layers, batch normalization layers, activation layers, and max pooling layers. Two fully connected layers and a regression output layer are connected after the deep convolutional neural network to complete the regression task. S4.4 Training of Convolutional Neural Network Model: The Adam optimizer is used to train the convolutional neural network model. The maximum number of training epochs is set to 80, the initial learning rate is 0.01, and L2 regularization is used. After training, features are extracted from the intermediate pooling layers as depth representations. S4.5, Relevant Vector Machine Model Training: Using features extracted by a convolutional neural network as input, a relevant vector machine regression model is trained based on a linear kernel function to predict the test set. S4.6 Result Restoration and Evaluation: The prediction results output by the relevant vector machine model are inversely normalized to the original numerical range, and the error index is calculated to evaluate the model performance.

[0011] In step S4.6, the inverse normalization formula is: (18) in, x The original data, y For the normalized data, x min , x max These are the minimum and maximum values ​​of the original data, respectively. y max , y min These are the upper and lower limits of the target area, respectively.

[0012] Furthermore, the error indices include mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2The mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are calculated using the following formulas. 2 , (19) in, n y is the sample size. i For the true value, This is a predicted value; (20) (twenty one) (twenty two) in, This is the average of the true values.

[0013] This invention prepares a cutting fluid containing Fe3O4 nanoparticles; establishes a system for collecting milling force, milling noise, and surface roughness data; and conducts wet milling experiments on TC4 titanium alloy under different process parameters. It plots the variation curves of the experimental data to compare the performance differences before and after the addition of Fe3O4 nanoparticles; and constructs a prediction model based on convolutional neural networks and correlation vector machines to achieve accurate prediction of surface roughness. The advantages of this invention are that adding Fe3O4 nanoparticles to the cutting fluid significantly reduces milling force and noise, improves surface roughness, and the combination of experimental testing with machine learning models effectively enables intelligent prediction of surface quality.

[0014] This invention significantly improves the lubrication performance of cutting fluid by introducing Fe3O4 nanoparticles, thereby reducing milling force and noise, improving the surface roughness of titanium alloys, and enhancing machining quality. Furthermore, this invention combines an algorithmic model to predict the surface roughness after milling, achieving non-destructive measurement of machined surface quality. The method is simple and has good engineering application value and promising prospects for widespread application. Attached Figure Description

[0015] Figure 1 The milling force, milling noise and surface roughness measurement system built for this invention.

[0016] Figure 2 Adding nanoparticles to the milling force before and after the invention F x Comparison chart.

[0017] Figure 3 Adding nanoparticles to the milling force before and after the invention F y Comparison chart.

[0018] Figure 4 Adding nanoparticles to the milling force before and after the inventionF z Comparison chart.

[0019] Figure 5 Adding nanoparticles to milling noise before and after this invention L Comparison chart.

[0020] Figure 6 Surface roughness before and after adding nanoparticles in this invention R a Comparison chart.

[0021] Figure 7 This invention uses the CNN-RVM algorithm to predict surface roughness. R a A comparison chart of predicted and measured values.

[0022] Figure 8 This is a flowchart of the present invention. Detailed Implementation

[0023] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings: This embodiment is implemented under the premise of the technical solution of the present invention, and provides detailed implementation methods and specific operation processes, but the protection scope of the present invention is not limited to the following embodiments.

[0024] A method for predicting the surface roughness of titanium alloys based on cutting fluid reinforced with Fe3O4 nanoparticles, such as... Figure 8 As shown, it includes the following steps: S1. Prepare a cutting fluid containing Fe3O4 nanoparticles. Mix the cutting fluid with deionized water to prepare a coolant. Add Fe3O4 nanoparticles and a thixotropic agent to the coolant and stir to obtain a composite nano-cutting fluid.

[0025] Castrol Syntilo 9930C fully synthetic cutting fluid was mixed with deionized water at a volume ratio of 1:20 to prepare a coolant. 1g of 50nm Fe3O4 particles from Shanghai Yaoyi Alloy Materials Co., Ltd. and 1ml of SRE-4410TD rheological thixotropic agent from Guangzhou Dongfugui Chemical Raw Materials Co., Ltd. were added to 500mL of the coolant. After stirring, a light black composite nano-cutting fluid was obtained.

[0026] S2. Build a test and acquisition system for milling force, milling noise and surface roughness, and use cutting fluid and composite nano cutting fluid to mill the workpiece respectively. The test and acquisition system collects the milling force, milling noise and surface roughness during the milling process.

[0027] like Figure 1As shown, the test acquisition system consists of a milling force signal acquisition section, a milling noise signal acquisition section, and a surface roughness acquisition section. The milling force signal acquisition section includes a charge amplifier, a piezoelectric sensor, and a triaxial milling force tester. The milling noise signal acquisition section includes a precision sound level meter, a high-speed data acquisition instrument, WS-AV acoustic measurement and analysis software, and computer components. The surface roughness acquisition section includes a surface roughness measuring instrument and a surface roughness... R a The surface roughness was measured using a YJD-950C surface roughness measuring instrument.

[0028] A testing and acquisition system for milling force, milling noise, and surface roughness was used to extract milling force feature values ​​and milling noise feature values. The extracted milling force feature values ​​include... F x , F y , F z The extracted milling noise feature value is the root mean square value of the sound pressure level. L .

[0029] S3. Plot the change curves of test data such as milling force, noise and surface roughness, and compare the changes in milling force, milling noise and surface roughness before and after adding Fe3O4 nanoparticles. The test proves that the cutting fluid with added nanoparticles can effectively improve milling force, noise and surface roughness.

[0030] S4. Use a convolutional neural network combined with a correlation vector machine algorithm (CNN-RVM) to predict the surface roughness after milling with added Fe3O4 nanoparticles.

[0031] The principle of combining convolutional neural networks with support vector machines is as follows: Convolutional Neural Networks (CNNs) are a special type of feedforward deep learning model with outstanding feature extraction and nonlinear mapping capabilities. They can directly extract salient features from raw data, avoiding the cumbersome process of feature extraction relying on complex mathematical operations in traditional methods. A typical structure consists of an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer.

[0032] (1) Propagation from the input layer to the convolutional layer (1) in, g ( x ) is a non-linear activation function; l This represents the current network layer number. This represents the convolution operation; a l-1 For input to the upper layer; a l This indicates the output of this layer;W l These are the weights of the neurons in this layer; b l For bias; (2) Pooling layer (2) in, ψ ( x ) is the pooling function; (3) Fully connected layer (3) (4) Output layer (4) in, y To output the predicted value; f ( x ) is the output layer mapping function.

[0033] Relevance Vector Machine (RVM) is a regression and classification method based on sparse Bayesian learning. This method assumes that the output can be obtained as a linear combination of input features mapped by a kernel function. It introduces a zero-mean Gaussian prior and sets hyperparameters for each weight parameter to achieve automatic sparsity. Through marginal likelihood maximization (evidence approximation), the hyperparameters are iteratively updated, and irrelevant weights gradually shrink to zero, retaining only a few representative training samples, called the relevance vector.

[0034] (1) Basic regression model (5) in, y (x) is the input x The predicted value; ø( x ) is a nonlinear basis function; w This is the weight vector.

[0035] The observed values ​​contain Gaussian noise: (6) in, Let be the observed target value of the i-th training sample; This is the i-th training sample; These are assumed to be independent samples from a Gaussian noise process; ε i ~ N (0, σ 2 ), N (0, σ 2 () has a mean of 0 and a variance of σ 2The normal distribution of the is also known as the Gaussian distribution; (2) Likelihood function training set t =[ t 1,…, t N ] T Conditional probability: (7) It consists of all training samples; σ 2 For variance; N This represents the number of training samples; (3) Weight prior distribution (8) in, α i For hyperparameters, each α i Control weight w i The accuracy, α i →∞ w i →0; (4) Automatic correlation determination Through optimization α Automatically identify relevant vectors, if α i →∞, corresponding basis functions ϕ i ( x ) was pruned (contributing no to prediction); (5) Derivation of the posterior distribution The weighted posterior distribution is a Gaussian distribution. (9) (10) in, m The mean; ∑ Let A be the covariance matrix, and let A be a diagonal matrix, then A = diag( α ), β = σ -2 ; (6) Marginal likelihood maximization Hyperparameters are estimated by maximizing the marginal likelihood. α and β : (11) (12) For noise accuracy, It is a scalar function; (7) Hyperparameter update formula For each α i Take the derivative and set it to 0:

[0036] in, ϒ i For the effective degrees of freedom of the weights, For the matrix ∑ diagonal line of the first i Number, For vector m, the first... i The square of each element For the first i Hyperparameters α The updated value; Noise accuracy β renew: (14) (8) Prediction For new input x The predicted distribution is: (15) The prediction variance consists of two parts: (16) in, β -1 This is due to data noise uncertainty.

[0037] The specific steps for predicting the surface roughness after milling with the addition of Fe3O4 nanoparticles using a convolutional neural network combined with a support vector machine algorithm are as follows: S4.1 Data preprocessing: Divide the data collected during the milling process into training set and test set, and normalize the input features and output target of the convolutional neural network. The normalization formula is (17) in, x The original data, x min , x max These are the minimum and maximum values ​​of the original data, respectively. y The data is after normalization; S4.2 Data format conversion: Convert the normalized input features into a four-dimensional tensor format acceptable to convolutional neural networks; S4.3 Construct a convolutional neural network. A deep convolutional neural network includes convolutional layers, batch normalization layers, activation layers, and max pooling layers. Two fully connected layers and a regression output layer are connected after the deep convolutional neural network to complete the regression task. S4.4 Training of Convolutional Neural Network Model: The Adam optimizer is used to train the convolutional neural network model. The maximum number of training epochs is set to 80, the initial learning rate is 0.01, and L2 regularization is used. After training, features are extracted from the intermediate pooling layers as depth representations. S4.5, Relevant Vector Machine Model Training: Using features extracted by a convolutional neural network as input, a relevant vector machine regression model is trained based on a linear kernel function to predict the test set. S4.6 Result Restoration and Evaluation: The prediction results output by the relevant vector machine model are inversely normalized to the original numerical range, and the error index is calculated to evaluate the model performance.

[0038] The inverse normalization formula is (18) in, x The original data, y For the normalized data, x min , x max These are the minimum and maximum values ​​of the original data, respectively. y max , y min These are the upper and lower limits of the target area, respectively.

[0039] The error metrics include mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2 The mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are calculated using the following formulas. 2 , (19) in, n y is the sample size. i For the true value, This is a predicted value; (20) (twenty one) (twenty two) in, This is the average of the true values. Example 1

[0040] In the TC4 titanium alloy milling test, carbide cutting tools were used, and the workpiece was a block-shaped titanium alloy sample with dimensions of 100mm × 100mm × 40mm. Castrol Syntilo 9930C fully synthetic cutting fluid was selected and prepared as a coolant with deionized water at a ratio of 1:20. 1g of Fe3O4 nanoparticles and 1ml of SRE-4410TD thixotropic agent were added to 500mL of the solution, and after stirring, a light black composite nano-cutting fluid was obtained. In the milling test, three main parameters were considered: spindle speed, feed rate, and depth of cut. Spindle speed... n The design includes four levels: 700 r / min, 850 r / min, 1000 r / min, and 1150 r / min; feed rate... v f The design includes four speed levels: 8 mm / min, 10 mm / min, 12 mm / min, and 14 mm / min; milling depth... a p Four levels were designed: 1 mm, 2 mm, 3 mm, and 4 mm. The specific experimental plan is shown in Table 1.

[0041] Table 1 Milling Test Scheme

[0042] Milling tests were conducted according to the scheme shown in Table 1. The characteristic value of the milling force obtained by milling with Fe3O4 nanoparticle cutting fluid was... F x1 , F y1 and F z1 The characteristic value of milling noise is L 1. Surface roughness is R a1 The characteristic value of milling force obtained by milling without adding Fe3O4 nanoparticle cutting fluid is... F x2 , F y2 and F z3 The characteristic value of milling noise is L 2. Surface roughness is R a2 As shown in Tables 2 and 3.

[0043] Table 2. Experimental data after milling with added Fe3O4 nanoparticles No. <![CDATA[ F x1 ]]> <![CDATA[ F y1 ]]> <![CDATA[ F z1 ]]> <![CDATA[ L 1]]> <![CDATA[ R a1 ]]> <![CDATA[ R z1 ]]> <![CDATA[ R q1 ]]> 1 95.955 112.819 42.050 75.4 0.360 2.518 0.448 2 148.668 128.944 38.380 78.1 0.450 3.166 0.588 3 164.578 137.794 38.506 77.4 0.362 2.573 0.454 4 145.385 72.727 23.914 81.3 0.613 4.115 0.762 5 321.467 160.718 38.328 82.0 0.647 4.498 0.809 6 201.367 139.460 29.990 81.3 0.824 6.136 1.117 7 218.995 128.310 29.198 81.0 1.992 11.779 2.491 8 42.074 41.112 15.425 78.0 0.346 2.368 0.430 9 95.683 49.178 17.934 77.0 0.360 2.385 0.440 10 29.873 35.930 13.733 70.4 0.364 2.346 0.444 11 231.724 110.171 23.284 77.9 0.572 3.546 0.697 12 32.301 38.955 12.359 56.3 0.576 3.412 0.708 13 36.809 57.951 8.837 69.8 0.544 3.176 0.665 14 36.340 57.400 9.232 65.9 0.378 2.371 0.471 15 35.931 55.184 9.494 66.0 0.498 2.865 0.625 16 39.419 59.588 8.984 53.2 0.431 2.893 0.533 17 254.858 173.724 30.555 81.6 0.670 4.500 0.833 18 323.776 205.724 40.201 84.5 1.596 9.017 1.941 19 296.211 180.073 37.060 83.2 1.238 6.347 1.486 20 178.118 149.260 34.555 80.4 2.110 14.871 3.202 21 49.657 68.638 30.748 68.8 0.424 2.723 0.526 22 44.755 61.337 27.510 69.9 0.366 2.279 0.449 23 40.507 57.985 25.273 69.7 0.333 2.088 0.410 24 35.854 53.449 22.853 70.3 0.279 1.828 0.344 25 42.064 53.097 23.556 68.7 0.236 1.719 0.299 26 39.791 51.693 22.606 69.8 0.209 1.672 0.265 27 36.238 48.121 20.185 72.0 0.265 2.260 0.352 28 32.989 46.529 18.785 71.1 0.260 1.956 0.344 29 29.253 32.617 13.503 68.9 0.278 1.916 0.344 30 25.913 29.425 10.143 68.5 0.297 2.345 0.415 31 22.267 26.616 8.469 61.4 0.267 1.943 0.344 32 24.395 29.839 10.966 65.6 0.227 3.137 0.351 33 208.584 149.815 31.597 62.8 0.609 3.985 0.770 34 134.659 121.302 34.122 77.1 0.559 3.258 0.691 35 57.583 85.270 34.328 70.4 0.963 5.971 1.271 36 52.991 78.801 31.372 70.4 0.293 1.960 0.367 37 50.668 74.706 31.178 68.4 0.426 2.590 0.551 38 45.934 68.644 28.405 68.9 0.392 2.789 0.506 39 40.902 61.757 25.572 70.0 0.428 2.779 0.547 40 37.948 49.422 21.467 70.6 0.304 2.086 0.381 41 35.591 48.167 21.097 69.3 0.368 2.515 0.465 42 31.800 41.127 17.579 69.2 0.430 2.699 0.535 43 30.409 42.935 18.151 70.8 0.399 2.712 0.516 44 29.476 41.297 16.313 70.5 0.361 2.502 0.460 45 23.682 29.273 9.888 68.1 0.403 2.693 0.501 46 19.788 27.411 8.492 68.9 0.309 2.049 0.383 47 19.929 25.908 7.872 69.6 0.372 2.324 0.463 48 20.049 24.713 8.164 70.5 0.305 2.175 0.384 49 248.082 167.846 31.344 79.5 1.077 6.146 1.359 50 130.349 128.021 35.759 78.5 1.322 7.676 1.678 51 55.557 90.381 32.764 70.8 1.595 9.899 2.074 52 43.337 64.650 25.581 70.8 0.787 4.684 1.015 53 40.054 56.806 23.689 68.7 0.471 2.987 0.603 54 39.706 55.052 23.106 69.3 0.529 3.414 0.679 55 40.430 57.794 23.268 70.4 0.550 3.362 0.695 56 36.684 55.281 21.521 70.9 0.396 2.666 0.508 57 31.635 43.657 17.829 68.5 0.445 2.815 0.568 58 30.747 44.794 17.579 69.1 0.466 2.841 0.585 59 29.013 43.370 16.574 69.6 0.513 3.412 0.663 60 32.279 46.955 18.330 70.4 0.382 2.729 0.488 61 25.096 32.557 12.498 68.3 0.447 2.905 0.593 62 22.893 29.130 10.621 68.9 0.490 3.196 0.629 63 19.917 23.865 7.599 68.9 0.381 2.549 0.472 64 18.883 23.818 7.401 70.2 0.314 2.277 0.395 Table 3. Experimental data after milling without the addition of Fe3O4 nanoparticles No. <![CDATA[ F x2 ]]> <![CDATA[ F y2 ]]> <![CDATA[ F z2 ]]> <![CDATA[ L 2]]> <![CDATA[ R a2 ]]> <![CDATA[ R z2 ]]> <![CDATA[ R q2 ]]> 1 396.760 298.791 48.925 88.3 1.087 6.548 1.364 2 417.453 235.778 55.334 89.4 0.787 4.728 0.971 3 442.135 256.430 58.359 89.4 1.252 6.663 1.507 4 165.908 100.074 28.786 78.7 0.898 5.546 1.099 5 89.723 89.222 23.858 74.2 0.507 3.118 0.616 6 116.512 91.567 23.310 75.8 0.337 2.189 0.419 7 177.347 116.128 25.928 78.7 0.454 3.065 0.570 8 47.638 55.448 19.575 72.2 0.396 2.427 0.495 9 42.509 54.828 19.804 69.6 0.244 1.494 0.299 10 38.957 50.440 18.721 70 0.247 1.658 0.307 11 35.210 45.421 17.012 70.7 0.262 1.727 0.325 12 32.869 42.742 16.217 71.3 0.314 1.964 0.388 13 27.646 34.096 13.383 69.1 0.248 1.577 0.305 14 24.536 30.108 12.121 70.1 0.452 2.729 0.572 15 22.507 27.574 10.832 70.3 0.296 1.926 0.379 16 20.420 25.338 9.846 71.3 0.300 2.146 0.432 17 480.011 372.356 71.535 90 1.248 18.946 1.505 18 405.471 289.723 56.465 89.4 1.314 7.180 1.604 19 396.911 285.366 53.385 88.3 1.205 7.776 1.568 20 152.481 100.684 21.869 78.4 0.763 4.642 0.949 21 51.625 67.800 17.929 69.8 0.212 1.489 0.263 22 316.827 172.220 31.350 86.3 0.578 3.803 0.721 23 40.495 53.062 15.221 70.6 0.300 1.969 0.377 24 84.528 59.808 15.294 74.5 0.345 2.131 0.429 25 41.742 55.271 15.046 69.3 0.317 2.125 0.396 26 71.898 36.069 63.066 70.2 0.288 1.822 0.358 27 76.454 42.000 71.113 72 0.308 1.993 0.380 28 41.137 11.213 32.276 71.4 0.353 2.107 0.438 29 24.522 31.437 10.171 68.4 0.181 1.522 0.240 30 22.112 28.276 9.262 69.3 0.230 1.278 0.285 31 20.226 25.613 8.360 69.9 0.244 1.754 0.308 32 20.565 25.424 8.006 70 0.267 1.830 0.331 33 496.270 425.337 73.075 89.4 1.059 6.055 1.301 34 408.431 270.314 48.518 88.9 1.309 7.409 1.631 35 420.137 291.240 51.216 87 1.434 8.178 1.761 36 292.417 205.243 34.714 83.1 1.904 9.547 2.278 37 52.128 72.344 15.329 70.1 0.358 2.548 0.452 38 302.185 180.297 31.389 86.5 0.593 3.597 0.740 39 235.848 155.714 26.473 83.2 1.058 6.112 1.290 40 58.183 62.364 12.366 76.8 0.576 3.415 0.676 41 37.028 51.408 12.936 69.7 0.458 3.005 0.580 42 34.062 47.615 12.639 69.6 0.402 2.689 0.510 43 31.911 45.508 12.043 70.5 0.418 2.639 0.519 44 30.313 42.044 11.415 74.8 0.390 2.412 0.493 45 19.143 25.129 8.323 68.4 0.293 1.953 0.369 46 17.196 22.963 7.705 69.4 0.233 1.498 0.290 47 15.859 20.944 7.152 70.3 0.295 1.722 0.366 48 15.437 19.802 6.812 69.8 0.353 2.180 0.451 49 395.430 366.990 62.686 89.5 1.222 7.210 1.523 50 298.040 231.866 39.829 85.9 1.352 7.870 1.658 51 249.596 215.335 34.974 82.9 1.533 8.992 1.924 52 239.352 194.911 30.535 82.5 1.872 10.911 2.378 53 55.038 75.400 14.394 72.2 0.631 3.411 0.787 54 53.783 71.116 14.592 73.2 1.029 6.183 1.247 55 54.932 71.116 14.156 74.4 1.150 6.836 1.523 56 43.930 62.698 12.783 71.6 0.972 5.462 1.248 57 39.283 55.308 11.496 68.3 0.521 3.232 0.684 58 38.580 54.256 9.400 69.7 0.454 2.739 0.563 59 33.128 46.007 10.578 69.8 0.418 2.654 0.542 60 45.899 51.449 9.516 72 0.492 2.934 0.652 61 24.290 29.799 11.189 68 0.215 1.646 0.274 62 20.525 25.228 9.730 68.9 0.275 1.702 0.343 63 19.439 23.723 9.007 70.1 0.297 1.848 0.366 64 18.683 21.709 8.025 70.7 0.293 1.940 0.367 Based on the test data in Tables 2 and 3, plot the milling force. F x1 and F x2 , F y1 and F y2 , F z1 and F z2 Data comparison curves, such as Figure 2 , Figure 3 and Figure 4 As shown. After adding Fe3O4 nanoparticles, F x1 , F y1 and F z1 These figures represent reductions of approximately 40.32%, 31.97%, and 10.76%, respectively. Among them, x direction and y The milling force decreased most significantly in the cutting direction, indicating that the Fe3O4 nanoparticles formed a more effective lubrication in the cutting zone, which helped to reduce friction and resistance, thereby reducing milling force and improving stability.

[0044] Based on the test data in Tables 2 and 3, plot the milling force. L 1 and L 2. Data comparison curves, such as Figure 5 As shown, it also shows a good improvement trend in noise reduction. After adding Fe3O4 nanoparticles, the overall noise level was reduced by about 5.39%, indicating that Fe3O4 nanoparticles can reduce the noise caused by friction during milling and improve milling noise.

[0045] Based on the test data in Tables 2 and 3, plot the milling force. R a1 and R a2 Data comparison curves, such as Figure 6 As shown. Surface roughness R a1 The average decrease was approximately 11.56%. This indicates that the addition of Fe3O4 nanoparticles not only improved the operating conditions of the milling process but also reduced friction between the tool and the workpiece, decreased surface scratches, and thus improved the quality of the machined surface, effectively reducing surface roughness. R a .

[0046] The experimental data was divided into training and testing sets and input into the CNN-RVM model. The specific parameters for the MATLAB code are as follows: Data preprocessing: The data is divided into a training set (90%) and a test set (10%), and the input features and output targets are normalized.

[0047] (2) Data format conversion: Convert the normalized input features into a four-dimensional tensor format acceptable to CNN.

[0048] (3) Constructing a CNN network: Design a deep convolutional neural network containing convolutional layers, batch normalization layers, activation layers and max pooling layers, and then connect two fully connected layers and a regression output layer to complete the regression task.

[0049] (4) CNN model training: The Adam optimizer was used, with a maximum training epoch of 80, an initial learning rate of 0.01, and L2 regularization. After training, features were extracted from the intermediate pooling layers as deep representations.

[0050] (5) RVM model training: Using the features extracted by CNN as input, a related vector machine regression model is trained based on the linear kernel function to predict the test set.

[0051] (6) Result Restoration and Evaluation: The predicted results are inversely normalized to the original numerical range, and error indices (mean square error MSE, mean absolute error MAE, root mean square error RMSE, coefficient of determination R) are calculated. 2 To evaluate model performance.

[0052] Based on the above code parameters, the prediction results are shown in Table 4, and the evaluation metrics of the prediction model are shown in Table 5.

[0053] Table 4 Comparison of Predicted and Measured Values Predicted value Measured value 1 1.03532 1.077 2 0.46577 0.45 3 0.37141 0.399 4 0.49266 0.428 5 0.72808 0.647 6 0.45671 0.309 Table 5. Results of Predictive Model Evaluation Indicators Evaluation indicators <![CDATA[ R 2 ]]> MSE RMSE MAE result 91.2245% 0.0059 0.0767 0.0631 In summary, the method of this invention exhibits high accuracy and stability in prediction performance. Test results show that both the mean square error and root mean square error are at low levels, indicating a small deviation between the predicted and actual values, and effective control of the overall error. The low mean absolute error demonstrates the stability and reliability of the prediction process. Furthermore, the determination coefficient... R 2 The accuracy rate reached 91.22%, indicating that the prediction results of this method have a high degree of consistency with the measured data.

[0054] This invention relates to a method for predicting and comparing the surface roughness of titanium alloy milling based on Fe3O4 nanoparticle-reinforced cutting fluid. Unlike traditional cutting fluids with limited lubrication effects, this invention introduces Fe3O4 nanoparticles into the cutting fluid, creating a more effective lubrication environment in the cutting zone. Experimental results show that the addition of Fe3O4 nanoparticles significantly reduces the triaxial milling forces. x , y The milling force decreased most significantly in the direction of addition of nanoparticles. Simultaneously, the overall noise level decreased. Furthermore, the average surface roughness of the workpiece decreased by approximately 11.56%, indicating that this method improves the surface quality of the machined surface. Finally, the surface roughness after milling with added nanoparticles was predicted based on CNN-RVM. R a The prediction accuracy is relatively high.

[0055] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any transformations or substitutions that can be conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid, characterized in that, Includes the following steps: S1. Prepare a cutting fluid containing Fe3O4 nanoparticles. Mix the cutting fluid with deionized water to prepare a coolant. Add Fe3O4 nanoparticles and a thixotropic agent to the coolant and stir to obtain a composite nano-cutting fluid. S2. Build a test and acquisition system for milling force, milling noise and surface roughness, and use cutting fluid and composite nano cutting fluid to mill the workpiece respectively. The test and acquisition system collects the milling force, milling noise and surface roughness during the milling process. S3. Plot the change curves of milling force, noise and surface roughness test data, and compare the changes of milling force, milling noise and surface roughness before and after adding Fe3O4 nanoparticles. S4. Use a convolutional neural network combined with a correlation vector machine algorithm to predict the surface roughness after adding Fe3O4 nanoparticles during milling.

2. The method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid according to claim 1, characterized in that, In step S1, Castrol Syntilo 9930C fully synthetic cutting fluid was prepared with deionized water at a volume ratio of 1:20 to form a coolant. 1g of Fe3O4 nanoparticles and 1ml of SRE-4410TD rheological thixotropic agent were added to 500 mL of the coolant and stirred to obtain a light black composite nano-cutting fluid.

3. The method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid according to claim 1, characterized in that, In step S2, the test acquisition system consists of a milling force signal acquisition part, a milling noise signal acquisition part, and a surface roughness acquisition part. The milling force signal acquisition part includes a charge amplifier, a piezoelectric sensor, and a triaxial milling force tester. The milling noise signal acquisition section includes a precision sound level meter, a high-speed data acquisition instrument, and a computer component; the surface roughness acquisition section includes a surface roughness measuring instrument.

4. The method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid according to claim 3, characterized in that, In step S2, the milling force feature value and milling noise feature value are extracted using a testing and acquisition system for milling force, milling noise, and surface roughness. The extracted milling force feature value includes... F x , F y , F z The extracted milling noise feature value is the root mean square value of the sound pressure level. L .

5. The method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid according to claim 1, characterized in that, In step S4, the specific operation of predicting the surface roughness after milling using a convolutional neural network combined with a support vector machine algorithm is as follows: S4.1 Data preprocessing: Divide the data collected during the milling process into training set and test set, and normalize the input features and output target of the convolutional neural network. S4.2 Data format conversion: Convert the normalized input features into a four-dimensional tensor format acceptable to convolutional neural networks; S4.3 Construct a convolutional neural network. A deep convolutional neural network includes convolutional layers, batch normalization layers, activation layers, and max pooling layers. Two fully connected layers and a regression output layer are connected after the deep convolutional neural network to complete the regression task. S4.4 Training of Convolutional Neural Network Model: The Adam optimizer is used to train the convolutional neural network model. After training, features are extracted from the intermediate pooling layers as depth representations. S4.5, Relevant Vector Machine Model Training: Using features extracted by a convolutional neural network as input, a relevant vector machine regression model is trained based on a linear kernel function to predict the test set. S4.6 Result Restoration and Evaluation: The prediction results output by the relevant vector machine model are inversely normalized to the original numerical range, and the error index is calculated to evaluate the model performance.

6. The method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid according to claim 5, characterized in that, In step S4.6, the inverse normalization formula is: (18) in, x This is the original data. y For the normalized data, x min , x max These are the minimum and maximum values ​​of the original data, respectively. y max , y min These are the upper and lower limits of the target area, respectively.

7. The method for predicting the surface roughness of titanium alloys based on Fe3O4 nanoparticle-reinforced cutting fluid according to claim 5, characterized in that, The error metrics include mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2 The mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) are calculated using the following formulas. 2 , （19） in, n For the sample size, y i For the true value, This is a predicted value; （20） （21） （22） in, This is the average of the true values.

Images