Method for predicting surface quality of titanium alloy milled by micro-texture coating ball-end milling cutter based on vibration behavior
By constructing a prediction model based on vibration behavior and combining it with the MIC-BP neural network and WAA-BiLSTM-SVM, the problems of rapid tool wear and poor surface quality in titanium alloy milling were solved, and high-precision surface quality prediction and tool performance improvement were achieved.
Patent Information
- Application Number
- CN202510574894.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-06
- Publication Date
- 2025-09-16
AI Technical Summary
In the existing technology, titanium alloys are difficult to process. Traditional milling easily leads to rapid tool wear and poor surface quality. Micro-textured coated ball-end milling cutters can improve milling performance, but there is a lack of systematic prediction methods, and research on the correlation between vibration behavior and surface quality is insufficient.
Through experimental design and data acquisition, the milling vibration signals, force, temperature and wear are analyzed. The MIC-BP neural network and WAA-BiLSTM-SVM model are used to construct a prediction model based on vibration behavior. The correlation between vibration and milling performance is comprehensively studied, the key features are screened out, and a workpiece quality Ra prediction model is constructed.
The accuracy and generalization ability of titanium alloy milling surface quality prediction were improved. The surface roughness Ra value of the micro-textured AlSiTiN coated tool was improved by more than 20%, and the tool life was increased by more than 30%. The R2 of the model training set reached 0.942, and the R2 of the test set was 0.835, which was significantly better than the traditional method.
Smart Images
Figure CN120654526A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of mechanical processing technology, and in particular to a method for predicting the surface quality of titanium alloy milled by a micro-texture coated ball end milling cutter based on vibration behavior. The method is applicable to high-end manufacturing fields such as aerospace and medical equipment. Background Art
[0002] Titanium alloys have advantages such as high strength and good corrosion resistance and are widely used in high-end manufacturing fields such as aerospace and medical devices. However, titanium alloys are difficult to process, and traditional milling processes are prone to problems such as rapid tool wear and poor surface quality. Micro-textured coated ball end mills can improve milling performance to a certain extent, but accurately predicting the surface quality of titanium alloys milled with them remains a challenge. Vibration behavior is closely related to surface quality during the milling process, so prediction based on vibration behavior is of great significance.
[0003] Although extensive research has focused on vibration behavior and its impact on machining processes, most existing studies focus on vibration monitoring and control, while relatively few have delved into the complex relationship between vibration behavior and milling performance. Furthermore, predicting the surface quality of workpiece milling surfaces by analyzing the characteristic changes in milling vibration in the time-frequency domain lacks theoretical support and experimental evidence. Systematic research on the vibration behavior of difficult-to-machine materials such as titanium alloys, especially when combined with novel tool technologies (such as micro-textured coatings), is even more limited.
[0004] In summary, titanium alloys are difficult to machine, and conventional milling can lead to rapid tool wear and poor surface quality. Micro-textured ball-end milling cutters can improve milling performance, but existing research on the correlation between vibration behavior and surface quality is insufficient, lacking a systematic prediction method. Summary of the Invention
[0005] The present invention aims to solve the problems of great difficulty in processing titanium alloys and the fact that traditional milling easily leads to rapid tool wear and poor surface quality. Micro-texture coated ball end mills can improve milling performance, but the existing technology has insufficient research on the correlation between vibration behavior and surface quality and lacks a systematic prediction method. Therefore, a method for predicting the surface quality of titanium alloys milled by micro-texture coated ball end mills based on vibration behavior is proposed.
[0006] The technical solution adopted by the present invention to solve the above problems is:
[0007] The method of predicting the surface quality of titanium alloy milled by a micro-textured coating ball-end milling cutter based on vibration behavior according to the present invention comprises the following steps:
[0008] Step 1: Experimental design and data collection, including:
[0009] Step 1.1 Prepare multiple experimental tools, including micro-textured AlSiTiN coated tools, AlSiTiN coated tools, micro-textured tools, and untreated tools. The micro-textured AlSiTiN coated tool is the main analysis tool. Set up multiple tools for the experiment to avoid accidental influences, and select the one with stable performance for subsequent analysis.
[0010] Step 1.2 Determine cutting parameters, such as cutting speed v c =120m / min, feed per tooth f z =0.08mm / z, cutting depth a p =0.5mm, cutting width a e =0.5mm, spindle speed n = 3910 rpm, effective cutting radius R1 = 4.86, and tool feed f = 391. The workpiece's length and width were milled in a single pass, each 150mm long and 100mm wide. A total of 200 passes were made on one workpiece face, for a total milling length of 30m. During the milling process, every two passes constituted a signal acquisition cycle, resulting in 100 data points collected, which were then broken down into 200 data points for later analysis.
[0011] Step 1.3: Set up the experimental platform. This includes using a vise to hold a TC4 titanium alloy square material at a 15-degree angle as the milling material, attaching a Chengke CT1010SLFP triaxial accelerometer to the side of the workpiece to measure vibration data, using a Kistler rotary dynamometer to collect milling force data, using a thermometer with HIKMICRO Studio software to monitor milling temperature changes, using an industrial camera with Image-Pro Plus 6.0 software to observe tool wear, and using a TR-200 surface roughness tester to measure the machined surface roughness.
[0012] Step 2: Vibration amplitude characteristic analysis, including:
[0013] Step 2.1 Comparison of milling vibration signal waveforms: Collect vibration signals from different tools milling titanium alloy, analyze the original signal waveforms, and determine the relationship between the vibration amplitudes in different directions during the milling process. It can be seen that the vibration waveform range in the y direction is the most significant and has the highest amplitude. Compare the vibration signal waveforms of each tool in the y direction and observe the influence of tools such as micro-textured AlSiTiN coated milling cutters on the vibration amplitude.
[0014] Step 2.2 Study on the changing law of time-frequency characteristics of milling vibration: Obtain the time-frequency characteristic images of y-direction vibration of each milling cutter when milling titanium alloy, analyze the influence of micro-texture AlSiTiN coating on the time-frequency energy amplitude, compare the changes in the time-frequency energy amplitude characteristics of milling vibration of each milling cutter, and determine the performance of different tools in vibration control and their influence on processing stability.
[0015] Step 3: Study on the correlation between vibration and milling performance, including:
[0016] Step 3.1 Study on the correlation between milling vibration and milling force
[0017] Clarify the corresponding relationship between milling force and milling vibration direction, compare the milling force change curves of each milling cutter, analyze the milling force change law in stages, calculate the standard deviation of the milling force and the average value of the resultant force of each milling cutter, and evaluate the stability and size of the milling force.
[0018] The MIC-BP neural network is used to explore the correlation between the various directional milling forces and the various directional milling vibrations. The prediction effect is judged by calculating the feature importance and RMSE value, and the correlation between the milling resultant force and the various directional milling vibrations is determined. The y-direction vibration feature with the best fitting effect for the milling resultant force is screened out, and the correlation between the y-direction milling vibration and the milling resultant force is analyzed, including the Pearson correlation, Spearman correlation coefficient and Kendall-Tao correlation coefficient.
[0019] Step 3.2 Study on the correlation between milling vibration and milling temperature
[0020] The milling temperature change curves of each milling cutter were compared, the mean and standard deviation of the milling temperature were calculated, the milling temperature change trend and stability of different tools were analyzed, and the influence of the micro-textured AlSiTiN coating structure on the milling temperature was determined.
[0021] The MIC-BP neural network is used to calculate the correlation contribution and RMSE value between the milling vibration in each direction and the milling temperature. The features with the best fitting effect on the milling temperature in the x-direction vibration are screened out. The correlation between the x-direction milling vibration and the milling temperature is analyzed, including the Pearson correlation, Spearman correlation coefficient and Kendall-Tao correlation coefficient.
[0022] Step 3.3 Study on the correlation between milling vibration and tool wear
[0023] In the experiment, an industrial camera was used to capture tool wear images, and Image-Pro Plus 6.0 software was used to identify and measure them. The neighborhood averaging method was used to expand the tool wear data. The data curves before and after expansion were compared to ensure that data expansion did not affect the curve trend.
[0024] Compare the wear curves of various milling cutters, analyze the change rules of tool wear stages, calculate the mean and standard deviation of tool wear, and determine the wear resistance of different tools. Calculate the correlation contribution of all-directional milling vibration and tool wear value through MIC-BP neural network.
[0025] The RMSE value was used to screen out the features with the best fitting effect on tool wear in the x-axis vibration, and the correlation between the x-axis milling vibration and tool wear was analyzed, including the Pearson correlation coefficient, Spearman correlation coefficient and Kendall-Tao correlation coefficient.
[0026] Step 4: Surface processing quality prediction, including:
[0027] Step 4.1 Comparison of the quality of titanium alloy milling by various milling cutters:
[0028] When measuring milling quality, three measurement points were taken at the same level to calculate the average value. Ra, Rq, and Rz were selected as quality characterization values. The quality change curves of titanium alloy milling with each milling cutter were analyzed. The mean and standard deviation of milling quality of different tools were compared to determine the processing quality advantages of micro-textured AlSiTiN coated milling cutters, such as the average milling quality improvement ratio of Ra value and the increase ratio of tool usage.
[0029] Step 4.2 Study on milling quality prediction of micro-textured AlSiTiN coated milling cutter based on vibration characteristics:
[0030] Vibration features were screened based on the MIC-BP neural network. The milling data of two micro-textured AlSiTiN coated ball-end milling cutters with smaller errors in the experiment were selected as training and test sets, respectively. The contribution of vibration features to the workpiece quality Ra was analyzed, and the directions of milling vibration features considered when constructing the workpiece quality Ra prediction model were determined. The features in the x- and y-direction vibrations that best contributed to the workpiece quality prediction model were screened.
[0031] 4.3 Constructing a workpiece processing quality Ra prediction model based on WAA-BiLSTM-SVM:
[0032] The weighted average algorithm WAA is used to optimize the three parameters of BiLSTM, set the optimization parameter range, and automatically select the SVM hyperparameter combination based on the Bayesian optimizer. The prediction results of the model training set and test set are analyzed, including R 2 The prediction and generalization capabilities of the model are evaluated by using the fitness curve and the loss iteration change curve to analyze the optimization effect of the WAA algorithm.
[0033] The WAA-BiLSTM-SVM model is compared with traditional prediction models (Bayesian self-optimizing SVM and Bayesian optimized random forest RF) to compare the R 2 The values and RMSE values reflect the advantages of the model of the present invention.
[0034] The beneficial effects of the present invention are:
[0035] 1. Comprehensive analysis of the correlation between vibration behavior and milling performance:
[0036] Through systematic experimental design and data analysis, the present invention comprehensively considers the relationship between vibration behavior and milling performance such as milling force, milling temperature, and tool wear, providing a multi-dimensional basis for accurately predicting surface quality.
[0037] 2. Using the WAA-BiLSTM-SVM model, it has high prediction accuracy and strong generalization ability:
[0038] The prediction model constructed based on the weighted average optimization algorithm to optimize the bidirectional long short-term memory neural network combined with the support vector (WAA-BiLSTM-SVM) effectively improves the prediction accuracy, can better capture the long-term dependencies and complex nonlinear patterns in the data, and provides strong support for process optimization and quality control in the titanium alloy milling process.
[0039] 3. The surface roughness Ra value of the micro-textured AlSiTiN coated tool used in the present invention is improved by no less than 20% compared to the untreated tool; and the tool service life is increased by no less than 30%.
[0040] 4. The experiment of this invention shows that the model training set R 2 Up to 0.942, test set R 2 The predicted value of the milling surface quality of titanium alloy is 0.835, which is significantly better than the traditional method, and can improve the prediction accuracy of the surface quality of titanium alloy milling and provide a basis for optimizing the machining process. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 The tool and workpiece models used in the test of the present invention;
[0042] Figure 2 It is a milling test platform;
[0043] Figure 3 This is the vibration signal waveform of the untreated ball-end milling cutter when milling titanium alloy;
[0044] Figure 4 The WAA-BiLSTM-SVM model architecture and process;
[0045] Figure 5 This is the WAA-BiLSTM-SVM model training and verification effect diagram;
[0046] Figure 6 Iterative process for optimizing WAA;
[0047] Figure 7 Comparison of test results of each model. DETAILED DESCRIPTION
[0048] The present embodiment describes a method for predicting the surface quality of titanium alloy milled with a micro-textured coating ball end mill based on vibration behavior, which improves the accuracy of the prediction of the surface quality of titanium alloy milling and provides a basis for optimizing the machining process. The method is implemented by the following steps:
[0049] Step 1: Select the test material, set the cutting parameters, and build the milling test platform. The tool and workpiece model used in the test are as follows: Figure 1 As shown in Figure 2, the milling test platform is as follows: Figure 2 shown.
[0050] Step 2: Collect vibration signals of four types of tools, including micro-textured AlSiTiN coated milling cutter, AlSiTiN coated milling cutter, micro-textured milling cutter, and untreated milling cutter, when milling titanium alloy.
[0051] Step 3: Analyze the unprocessed vibration signal waveform. It can be seen that the main vibration contribution during milling comes from the y direction. The unprocessed ball end milling cutter milling titanium alloy vibration signal waveform is as follows: Figure 3 shown.
[0052] Step 4: Preprocess the raw vibration signal using variational mode decomposition (VMD) to remove high-frequency noise. Short-time Fourier transform (STFT) and continuous wavelet transform (CWT) were used to perform time-frequency analysis on the vibration signal. The time-frequency characteristics of the y-axis vibrations of different tools during titanium alloy milling were obtained. The vibration signal waveforms and time-frequency energy amplitude variations of the micro-textured AlSiTiN-coated milling cutter, the AlSiTiN-coated milling cutter, the micro-textured milling cutter, and the untreated milling cutter were compared and analyzed.
[0053] Step 5: Use the MIC-BP neural network to calculate the correlation between milling vibration and milling force, milling temperature, and tool wear. Calculate the contribution and RMSE of the correlation between each direction of milling vibration and the milling force, milling temperature, and tool wear, and select the vibration characteristics that best fit each performance indicator.
[0054] Step 6: Based on the selected vibration features, a WAA-BiLSTM-SVM model is constructed to predict the workpiece surface roughness Ra. The model architecture and process are as follows: Figure 4 shown.
[0055] IWAA assigns dynamic weights to 17 time-frequency features based on the MIC-BP contribution value;
[0056] II. The BiLSTM network contains a two-layer structure with 64 hidden units to capture bidirectional temporal dependencies;
[0057] III The SVM classifier uses the RBF kernel to map the BiLSTM output to the surface roughness level and is optimized by grid search.
[0058] Step 7: The evaluation index mainly uses the determination coefficient R 2 Value and root mean square error RMSE value, R 2 The value is shown in Formula 1, and the RMSE value is shown in Formula 2.
[0059]
[0060] Among them, y i is the actual value, is the predicted value, is the average of the actual values.
[0061]
[0062] Where n is the number of samples, y i is the actual value, is the predicted value.
[0063] Step 8: From Figure 5 The comparison chart of the training set results shows that the training set predicts R 2 The value is 0.942, indicating that there is a high linear correlation between the model's predictions on the training set and the true values, and the model can capture the trend of the data well; the RMSE value is 0.0159, indicating that the model training predictions are accurate, the fluctuation between the predicted values and the true values is small, and the prediction accuracy is high. 2 The value is 0.835, and the RMSE value is 0.0265. Although slightly worse than the training set performance, it still shows that the model has good predictive ability on the test set. Because the training set and test set of the model are both complete sets of data, that is, the training set to test set ratio is 1:1, the data surface of the model test set can also be well explained on unseen data, showing high generalization ability.
[0064] Step 9: Figure 6 The figure shows the optimization process of the BiLSTM network hyperparameters using the WAA algorithm. In the first few iterations, the fitness value drops rapidly, indicating that the model has found a good solution early on. There is a small drop around the 30th iteration, indicating that the algorithm has escaped the previous local optimum and found a better solution.
[0065] Step 10: Evaluate the model's prediction accuracy by comparing the model's predicted values with the actual measured surface roughness values. Use the coefficient of determination R 2 The value and the root mean square error RMSE value are used as evaluation indicators, R 2 The closer the value is to 1, the higher the linear correlation between the model's prediction and the true value on the training set and test set; the smaller the RMSE value, the higher the accuracy of the model's prediction and the smaller the fluctuation between the predicted value and the true value. 2 The value and RMSE value verify the advantages of this model. Figure 7 Shown are the R values of the three prediction models. 2 The values are compared with the RMSE values.
[0066] Step 11: The trained WAA-BiLSTM-SVM model was applied to the actual milling of titanium alloy. Vibration signals were collected in real time. After signal processing and feature extraction, the signals were input into the model to predict the workpiece surface roughness Ra. Based on the predicted results, milling parameters such as cutting speed and feed rate were adjusted in a timely manner to ensure surface quality.
[0067] The specific algorithm of the WAA-BiLSTM-SVM model in step 6 is as follows:
[0068] S1 Data Preprocessing
[0069] Preprocess the collected multivariate data, including cleaning, standardization, feature engineering, etc., to prepare a dataset suitable for model training.
[0070] S2 data analysis and data set division
[0071] Set parameters: define the training set ratio as 0.5, the output dimension as 1, and obtain the number of samples and input feature dimension.
[0072] Shuffle and divide the data set: Use the randperm function to shuffle the order of the data set, calculate the number of training set samples based on the proportion of the training set, and divide the training set and test set.
[0073] S3 data normalization processing
[0074] Input data normalization: Use the mapminmax function to normalize the input data of the training set and test set, map the data to the [0,1] interval, and save the normalization parameters for the normalization of the test set data.
[0075] Output data normalization: Perform similar normalization on the output data of the training set and test set, saving the parameters.
[0076] S4 data preprocessing and format conversion
[0077] Data tiling: Flattens the input data of the training and test sets into data of a specific dimension and transposes the output data.
[0078] Format conversion: The flattened data is further converted into a format suitable for model input.
[0079] S5 WAA optimization algorithm parameter settings
[0080] Use the Weighted Average Algorithm (WAA) to optimize BiLSTM hyperparameters, such as the number of hidden layer nodes, regularization coefficient, and learning rate. WAA optimizes the weighted average position of the population, balancing global exploration and local exploitation to improve model performance.
[0081] Set optimization parameters: Set the parameters of the weighted average algorithm (WAA):
[0082] Ⅰ The number of search agents is 5;
[0083] ⅡThe maximum number of iterations is 3;
[0084] III The number of optimized parameters is 3;
[0085] The lower bound of parameter IV is [1e-4, 5, 1e-5];
[0086] The upper bound of the V parameter is [1e-3, 100, 1e-3].
[0087] The WAA algorithm uses an iterative search process to continuously update the agent's position and find the optimal hyperparameter combination. The optimal hyperparameter values are extracted from the optimization results and used to train the BiLSTM model.
[0088] S6 defines the fitness function fical
[0089] The evalin function is used to obtain the required training data and parameters from the MATLAB workspace. These data and parameters include training input data, training target data, normalization parameters of output data, original training target data, number of input features, and output dimension.
[0090] Ⅰ Set the optimal number of hidden layer nodes (5, 10);
[0091] Ⅱ The best initial learning rate (1e-4, 1e-3);
[0092] ⅢOptimal L2 regularization coefficient (1e-5, 1e-3).
[0093] Run the WAA algorithm to obtain the optimal fitness value, optimal parameter position and fitness curve, and extract the optimal number of hidden layer nodes, optimal initial learning rate and optimal L2 regularization coefficient.
[0094] S7 model construction and training
[0095] Build the model structure: Use MATLAB's Deep Learning Toolbox to build a model that includes an input layer, a bidirectional long short-term memory network (BiLSTM) layer, an activation layer, a dropout layer, a fully connected layer, and a regression layer.
[0096] Set training parameters: Use the trainingOptions function to set training parameters, including using the Adam gradient descent algorithm, the maximum number of iterations, the initial learning rate, the learning rate descent strategy, the L2 regularization coefficient, shuffling the dataset, and turning off detailed output.
[0097] Ⅰ The optimization algorithm selects Adam gradient descent algorithm;
[0098] ⅡThe maximum number of iterations is 400;
[0099] ⅢThe initial learning rate is 0.01;
[0100] IV Learning rate reduction factor 0.5;
[0101] Ⅴ The learning rate adopts a segmented decrease strategy.
[0102] After 350 iterations, the learning rate is reduced to 0.1 times the original value, and L2 regularization is applied. The dataset is shuffled during each training, and the training process information is not displayed.
[0103] Train the network: Call the trainNetwork function, use the training set data to train the model, and obtain the trained network and training process information.
[0104] Visualization of S8 model training process
[0105] Iterative loss change curve: Extract training loss data from bestInfo, smooth it, and convert it into polar coordinate data to draw an iterative loss change curve to show the changing trend of loss during training.
[0106] RMSE iterative change curve: Extract training RMSE data and draw the RMSE iterative change curve to intuitively reflect the error changes during model training.
[0107] S8 feature extraction and model conversion
[0108] Feature extraction: Use the activations function to extract the features of the specified BiLSTM layer from the trained network to obtain the feature data of the training set and test set.
[0109] Data type conversion: Convert the extracted feature data into double-precision type to prepare for subsequent support vector machine (SVM) model training.
[0110] S9 SVM model training and simulation testing
[0111] Create an SVM model: Use the fitrsvm function to create an SVM model, setting options such as standardization, kernel function, and automatic optimization of hyperparameters.
[0112] Ⅰ The normalization parameter is set to true;
[0113] II Kernel function is Gaussian kernel function;
[0114] Ⅲ Automatic hyperparameter optimization is set to auto, so that the MATLAB fitrsvm function automatically optimizes the hyperparameters of the SVM. This includes:
[0115] ①BoxConstraint(c): controls the penalty intensity of the error term.
[0116] ②KernelScale(g): The bandwidth parameter of the Gaussian kernel function, which affects the influence range of a single training sample.
[0117] ③Epsilon: In ε-SVM, it defines the boundary of training samples that are not strictly classified as positive or negative.
[0118] Simulation test: Use the trained SVM model to predict the training set and test set to obtain the prediction results.
[0119] S10 data denormalization and model evaluation
[0120] Denormalization: Use the previously saved normalization parameters to denormalize the prediction results to obtain the actual prediction value.
[0121] Calculate evaluation indicators: Calculate various evaluation indicators of training sets and test sets, such as root mean square error RMSE, determination coefficient R 2 , mean square error MSE, residual prediction residual RPD, mean absolute error MAE, mean deviation error MBE, mean absolute percentage error MAPE, etc., to comprehensively evaluate the model performance.
[0122] S11 Results Visualization
[0123] Plotting results of training and test sets: Plot comparison charts of the actual and predicted values of the training and test sets, error charts, linear fit charts, etc. to intuitively display the model prediction effect.
[0124] Comprehensive evaluation visualization: Draw fitness curves, radar charts, compass charts, etc. to display model performance and evaluation indicators from different angles for easy analysis and comparison.
[0125] S12 result display and comparison
[0126] Print evaluation metrics: Print the evaluation metrics of the training set and test set on the command line for easy viewing and analysis.
[0127] Compare algorithm errors: Organize the evaluation indicators of the training set and test set into a table to compare the errors of the models on different data sets.
[0128] It should be noted that in the above embodiments, as long as the technical solutions are not contradictory, they can be permuted and combined. Those skilled in the art can exhaust all possibilities based on the mathematical knowledge of permutations and combinations. Therefore, the present invention will no longer describe the technical solutions after permutations and combinations one by one, but it should be understood that the technical solutions after permutations and combinations have been disclosed by the present invention.
[0129] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as a preferred embodiment as above, it is not intended to limit the present invention. Any technician familiar with the present profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical content disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement of the above embodiments made according to the technical essence of the present invention, within the spirit and principles of the present invention, without departing from the content of the technical solution of the present invention, shall still fall within the scope of protection of the technical solution of the present invention.
Claims
1. A method for predicting the surface quality of titanium alloy milled by a micro-textured coating ball-end milling cutter based on vibration behavior, characterized in that: The following steps are involved: Step 1: Experimental design and data collection, including: Step 1.1, select the test materials, Step 1.2, set cutting parameters, Step 1.3: Build a milling test platform: Step 2: Analysis of vibration amplitude characteristics, including: Step 2.1: Comparison of milling vibration signal waveforms: Step 2.2: Study on the variation of time-frequency characteristics of milling vibration: Step 3: Study the correlation between vibration and milling performance, including: Step 3.1: Study on the correlation between milling vibration and milling force Step 3.2: Study on the correlation between milling vibration and milling temperature Step 3.3: Study on the correlation between milling vibration and tool wear Step 4: Surface processing quality prediction, including: Step 4.
1. Comparison of the milling quality of titanium alloy by various milling cutters: Step 4.2: Study on milling quality prediction of micro-textured AlSiTiN coated milling cutter based on vibration characteristics: Step 4.3: Construct a workpiece processing quality Ra prediction model based on WAA-BiLSTM-SVM.
2. A method for predicting the surface quality of titanium alloy milled by a micro-textured coating ball-end milling cutter based on vibration behavior, characterized in that: In step 1.1, multiple experimental tools are prepared, including micro-textured AlSiTiN coated tools, AlSiTiN coated tools, micro-textured tools, and untreated tools. The micro-textured AlSiTiN coated tools are the main analysis tools. Multiple tools are set for the experiment to avoid accidental effects, and one tool with stable performance is selected for subsequent analysis. In step 1.2, set the cutting parameters: cutting speed v c =120m / min, feed per tooth f z =0.08mm / z, cutting depth a p =0.5mm, cutting width a e =0.5mm, spindle speed n=3910 rpm, effective cutting radius R1=4.86, tool feed f=391; The workpiece can be milled in a single pass with a length of 150 mm and a width of 100 mm; Each workpiece surface was milled 200 times in total, with a total milling length of 30m; During the milling process, every two cutting passes constitute a signal acquisition cycle, and a total of 100 data points are collected, which are then split into 200 data points for later analysis. In step 1.3: Build an experimental platform, use TC4 titanium alloy square material as the milling material, and install it at a 15° angle; use a three-axis accelerometer attached to the side of the workpiece to measure vibration data; use a rotary dynamometer to collect milling force data; use a thermometer to monitor milling temperature changes; use an industrial camera to observe tool wear conditions; and use a surface roughness tester to measure the roughness of the machined surface.
3. A method for predicting the surface quality of titanium alloy milled by a micro-textured coating ball-end milling cutter based on vibration behavior, characterized in that: In step 2.1, the vibration signals of titanium alloy milling with different tools are collected, the original signal waveforms are analyzed, and it is determined that the vibration waveform range in the y direction is the most significant; In step 2.2, the time-frequency characteristic images of the y-axis vibration of each milling cutter when milling titanium alloy are obtained, the influence of the micro-textured AlSiTiN coating on the time-frequency energy amplitude is analyzed, and the changes in the time-frequency energy amplitude characteristics of the milling vibration of each milling cutter are compared to determine the performance of different tools in vibration control and their influence on processing stability.
4. A method for predicting the surface quality of titanium alloy milled by a micro-textured coating ball-end milling cutter based on vibration behavior, characterized in that: In step 3.1, the corresponding relationship between the milling force and the milling vibration direction is clarified, the milling force change curves of each milling cutter are compared, the milling force change law is analyzed in stages, the standard deviation and average value of the milling force of each milling cutter are calculated, and the stability and size of the milling force are evaluated; The MIC-BP neural network was used to explore the correlation between all-directional milling force components and all-directional milling vibration. The prediction effect was judged by calculating the feature importance and RMSE value. The correlation between the milling resultant force and all-directional milling vibration was determined. The y-direction vibration feature with the best fitting effect for the milling resultant force was screened out. The correlation between the y-direction milling vibration and the milling resultant force was analyzed, including the Pearson correlation, Spearman correlation coefficient, and Kendall-Tao correlation coefficient. In step 3.2, the milling temperature change curves of each milling cutter are compared, the mean and standard deviation of the milling temperature are calculated, the milling temperature change trend and stability of different cutters are analyzed, and the influence of the micro-textured AlSiTiN coating structure on the milling temperature is determined; The MIC-BP neural network was used to calculate the correlation contribution and RMSE value between milling vibration in all directions and milling temperature. The features with the best fitting effect on milling temperature in x-axis vibration were screened out. The correlation between x-axis milling vibration and milling temperature was analyzed, including Pearson correlation, Spearman correlation coefficient and Kendall-Tao correlation coefficient. In step 3.3, the tool wear images were captured by an industrial camera, and the Image-Pro Plus 6.0 software was used to identify and measure the wear data. The neighborhood averaging method was used to expand the tool wear data. The data curves before and after expansion were compared to ensure that the data expansion did not affect the curve trend. The tool wear change curves of each milling cutter were compared, the change law of tool wear stages was analyzed, the mean and standard deviation of tool wear were calculated, and the wear resistance of different tools was determined; the correlation contribution and RMSE value of the milling vibration in all directions and tool wear values were calculated through the MIC-BP neural network, and the features with the best fitting effect for tool wear in the x-direction vibration were screened out, and the correlation between the x-direction milling vibration and tool wear was analyzed, including the Pearson correlation coefficient, Spearman correlation coefficient and Kendall-Tao correlation coefficient.
5. A method for predicting the surface quality of titanium alloy milled by a micro-textured coating ball-end milling cutter based on vibration behavior, characterized in that: In step 4.1, when measuring the milling quality, three measurement points are taken at the same level to obtain an average value, and Ra, Rq, and Rz are selected as quality characterization values. The quality change curves of each milling cutter when milling titanium alloy are analyzed, and the mean and standard deviation of the milling quality of different tools are compared to determine the processing quality advantages of the micro-textured AlSiTiN coated milling cutter, such as the average milling quality improvement ratio of Ra value and the improvement ratio of tool use; In step 4.2, vibration features were screened based on the MIC-BP neural network. Milling data from two micro-textured AlSiTiN-coated ball-end milling cutters with relatively small errors in the experiment were selected as training and test sets, respectively. The contribution of vibration features to the workpiece quality Ra was analyzed, and the directions of milling vibration features considered when constructing the workpiece quality Ra prediction model were determined. The features in the x- and y-direction vibrations that contributed most to the workpiece quality prediction model were screened. In step 4.3, the weighted average algorithm WAA is used to optimize the three parameters of BiLSTM, set the optimization parameter range, and automatically select the SVM hyperparameter combination based on the Bayesian optimizer; analyze the prediction results of the model training set and test set, including R 2 The value and RMSE value are used to evaluate the prediction ability and generalization ability of the model, and the optimization effect of the WAA algorithm is analyzed through the fitness curve and loss iteration change curve; Compare the WAA-BiLSTM-SVM model with the traditional prediction model and compare the R 2 The values and RMSE values reflect the advantages of the model of the present invention.