Milling surface roughness prediction method based on heterogeneous temperature data fusion

Abstract

The invention belongs to the technical field of numerical control machining, and particularly relates to a milling surface roughness prediction method based on heterogeneous temperature data fusion. A roughness prediction model used in the method comprises a 1D multi-scale residual convolutional neural network, a bidirectional long-short-term memory neural network, a 2D multi-scale residual convolutional neural network, a normalization layer, a projection layer and a KAN network. The preprocessed maximum temperature sequence and average temperature sequence pass through a 1D multi-scale residual convolutional neural network to obtain maximum temperature sequence features and average temperature sequence features; the maximum temperature sequence feature and the average temperature sequence feature pass through a bidirectional long-short-term memory neural network to obtain a maximum temperature long-short-term dependency feature and an average temperature long-short-term dependency feature; the preprocessed thermal imaging image passes through a 2D multi-scale residual convolutional neural network to obtain a thermal imaging feature map; and performing normalization, projection and splicing on the maximum temperature long and short term dependency feature, the average temperature long and short term dependency feature, the thermal imaging feature map, the process parameters and the tool profile tolerance, and then performing KAN network mapping to obtain the surface roughness. The problems of insufficient spatial information of a temperature sequence and time continuity of a thermal image are solved, more comprehensive temperature information is provided, and the prediction precision is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of numerical control machining, and specifically relates to a method for predicting the surface roughness of milling machining based on the fusion of heterogeneous temperature data. Background Art

[0002] As an important numerical control machining method, milling machining is widely used in the processing of various products due to its high efficiency and productivity. In milling machining, surface roughness is one of the most common and important indicators to characterize the surface quality of workpieces. Therefore, accurately predicting surface roughness is of great significance for optimizing machining parameters, improving production efficiency, and ensuring product quality. Milling temperature refers to the temperature generated by the friction between the cutting tool and the workpiece during the milling machining process, which has an impact on both the machining process and results, and is directly related to the surface quality of the workpiece, machining accuracy, tool life, machining efficiency, etc. In particular, the temperature change in the milling area will directly affect the surface roughness of the workpiece. Therefore, the surface roughness can be predicted by using temperature.

[0003] Traditional surface roughness prediction mainly relies on empirical formulas and regression analysis, etc., and it is difficult to accurately reflect the dynamic changes and non-linear relationships in the actual machining process. When the working conditions change, prediction may not be possible. Therefore, there are problems such as poor robustness and poor universality. With the improvement of computing power and the progress of machine learning technology, researchers have begun to explore using data-driven methods to improve the prediction of surface roughness. In particular, in recent years, deep learning technology has demonstrated excellent capabilities in dealing with complex pattern recognition tasks, providing a new solution for surface roughness prediction. However, existing surface roughness prediction methods based on deep learning often rely on temperature data from a single source and are difficult to comprehensively reflect the complex effects of temperature on the machining process. Therefore, the present invention proposes a method for predicting the surface roughness of milling machining based on the fusion of heterogeneous temperature data, and realizes the accurate prediction of surface roughness through the fusion of heterogeneous temperature data and the combination of process parameters. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the technical problem to be solved by the present invention is to provide a method for predicting the surface roughness of milling machining based on the fusion of heterogeneous temperature data.

[0005] The present invention adopts the following technical solutions to solve the above technical problems:

[0006] A method for predicting the surface roughness of milling machining based on the fusion of heterogeneous temperature data, characterized in that the method comprises the following steps:

[0007] Step 1: Milling the workpiece under different process parameters, collecting the thermal imaging images, maximum temperature sequence, and average temperature sequence during the milling machining process; after milling is completed, measuring the surface roughness of the workpiece;

[0008] Step 2: Preprocess the thermal imaging map, maximum temperature sequence, and average temperature sequence to construct a dataset;

[0009] Step 3: Construct a roughness prediction model, including a 1D multi-scale residual convolutional neural network, a bidirectional long short-term memory neural network, a 2D multi-scale residual convolutional neural network, a normalization layer, a projection layer, and a KAN network;

[0010] The preprocessed maximum temperature sequence and average temperature sequence pass through the 1D multi-scale residual convolutional neural network to obtain the maximum temperature sequence features and average temperature sequence features; the maximum temperature sequence features and average temperature sequence features pass through the bidirectional long short-term memory neural network to obtain the maximum temperature long short-term dependence features and average temperature long short-term dependence features; the preprocessed thermal imaging map passes through the 2D multi-scale residual convolutional neural network to obtain a thermal imaging feature map; after normalizing the maximum temperature long short-term dependence features, average temperature long short-term dependence features, thermal imaging feature map, process parameters, and tool profile, projection is performed to obtain the maximum temperature projection features, average temperature projection features, thermal imaging projection features, process projection features, and tool profile projection features; after splicing all the projection features, they are mapped through the KAN network to obtain the surface roughness;

[0011] Step 4: Use the dataset to train the roughness prediction model, and use the trained roughness prediction model for predicting the surface roughness of milling machining.

[0012] Further, the 1D multi-scale residual convolutional neural network includes a multi-scale one-dimensional convolutional module. The output features of the current one-dimensional convolutional module are connected with the input features through residual connection to obtain one-dimensional residual features. The one-dimensional residual features undergo a pooling operation to obtain the input features of the next one-dimensional convolutional module; the input features of the one-dimensional convolutional module sequentially pass through one-dimensional convolution, batch normalization, and activation operations to obtain the output features of the module;

[0013] The 2D multi-scale residual convolutional neural network includes a multi-scale two-dimensional convolutional module. The output features of the current two-dimensional convolutional module are connected with the input features through residual connection to obtain two-dimensional residual features. The two-dimensional residual features undergo a pooling operation to obtain the input features of the next two-dimensional convolutional module; the input features of the two-dimensional convolutional module sequentially pass through two-dimensional convolution, batch normalization, and activation operations to obtain the output features of the module.

[0014] Further, the maximum temperature sequence refers to the time sequence formed by the maximum temperature in the machining area at each sampling moment, and the average temperature sequence refers to the time sequence formed by the average temperature in the machining area at each sampling moment.

[0015] Furthermore, the preprocessing of the thermal imaging map includes denoising, alignment, cropping, grayscale conversion, and normalization, while the preprocessing of the maximum temperature sequence and the average temperature sequence includes padding and normalization.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) The temperature in milling is an important factor affecting the surface roughness of products. Traditional surface roughness prediction methods often rely on temperature data from a single source, only focusing on the time-domain or spatial features of the temperature data. The temperature sequence reflects the dynamic change of temperature in the time domain but lacks spatial information, while the thermal imaging map reflects the spatial distribution of temperature but lacks temporal continuity. Therefore, the present invention fuses two types of heterogeneous data, namely the temperature sequence and the thermal imaging map. The two complement each other to solve the problem of insufficient spatial information in the temperature sequence and the temporal continuity problem of the thermal imaging map, providing more comprehensive temperature information, thereby improving the prediction accuracy. The fusion of heterogeneous temperature data improves the robustness and stability of the model, reduces the noise interference of a single data source, and can alleviate the decrease in prediction accuracy caused by sensor errors or data quality problems. It can also enhance the generalization ability of the model, enabling it to adapt to different working conditions and environments and improving the wide range of applications.

[0018] (2) The roughness prediction model uses a 1D multi-scale residual convolutional neural network to capture the temperature sequence features at different scales. The temperature sequence features pass through a bidirectional long short-term memory neural network, which uses the information of the past and future of the temperature sequence to enable the network to better understand the relationship between the information at the current moment and the past and future information, capturing the long-term and short-term dependencies of the temperature sequence, thereby obtaining more comprehensive temperature sequence features; a 2D multi-scale residual convolutional neural network is used to extract the features of the thermal imaging map. Finally, the KAN network maps the feature dimension to the predicted surface roughness. The whole process realizes the non-linear mapping between heterogeneous temperature data and surface roughness, achieving accurate prediction of surface roughness. The 1D multi-scale residual convolutional neural network and the 2D multi-scale residual convolutional neural network avoid the problem of gradient disappearance through residual connections, improving the training efficiency and performance of the model. The KAN network fuses and predicts data of different modalities. The KAN network can handle complex non-linear relationships and provide a smooth activation function, thus more accurately capturing the subtle and complex changes of multi-modal data during the processing, improving the prediction accuracy. The adaptive network and regularization mechanism of KAN enhance the generalization ability of the model, avoiding overfitting while improving the robustness, and having unique advantages for processing continuous data and complex feature interactions.

[0019] (3) To obtain a rich dataset, an optimal Latin hypercube design is adopted for the milling experiments to ensure uniform distribution of process parameters, facilitating the acquisition of more abundant thermal images, maximum temperature sequences, and average temperature sequences under different process parameters, ensuring that the model can learn more information and helping to improve the prediction accuracy. Description of the Drawings

[0020] Figure 1 It is the overall flowchart of the present invention;

[0021] Figure 2 It is the structural schematic diagram of the roughness prediction model of the present invention;

[0022] Figure 3 It is the prediction result diagram of the milling surface roughness of the present invention. Detailed Embodiment

[0023] Specific embodiments are given below in conjunction with the drawings. The specific embodiments are only used to introduce the technical solutions of the present invention in detail and do not limit the protection scope of this application.

[0024] The present invention provides a method for predicting the milling surface roughness based on heterogeneous temperature data fusion (hereinafter referred to as the method, see Figures 1 to 3 ) and includes the following steps:

[0025] Step 1: The workpiece is milled under different process parameters, and a thermal imager is used to collect thermal images, maximum temperature sequences, and average temperature sequences during the milling process; after milling, a 3D topography scanner is used to measure the surface roughness of the workpiece as a label;

[0026] To ensure uniform parameter distribution, the main process parameters affecting the surface roughness, including feed rate, cutting depth, spindle speed, etc., are used as variables, and an optimal Latin hypercube method is adopted to design the milling experiments. The size of the workpiece in this embodiment is 20×70×70mm. The surface of the workpiece is initially milled flat to ensure that the actual cutting depth during data collection is the cutting depth designed in the experiment. The 70×70mm surface of the workpiece is used as the machining surface, and this machining surface is evenly divided into multiple machining areas. Each machining area is milled with different process parameters and the milling paths are the same.

[0027] A plurality of sampling points are evenly set in the machining area. The maximum temperature sequence refers to the time sequence formed by the maximum temperature in the machining area at each sampling moment, and the average temperature sequence refers to the time sequence formed by the average temperature in the machining area at each sampling moment.

[0028] Step 2: Preprocess the thermal images, maximum temperature sequences, and average temperature sequences to construct a dataset;

[0029] Preprocessing of the thermal imaging map: First, denoise the thermal imaging map to remove random noise in the thermal imaging map and improve the image quality; then, align and crop the denoised thermal imaging map according to the position and size of the processing area to ensure that all thermal imaging maps maintain the same positional relationship, facilitating the capture of the temperature change trend during the processing; perform grayscale processing on the cropped thermal imaging map, which can reduce the consumption of computing resources and improve the stability and generalization ability of the model; finally, perform normalization processing on the grayscale processed thermal imaging map to unify the pixels to the same range.

[0030] Preprocessing of the maximum temperature sequence and the average temperature sequence includes padding and normalization; since the feed rates of different process parameters are different, resulting in different lengths of the temperature sequences, the Newton cooling law is used to pad the temperature sequences so that the lengths of all temperature sequences are the same; the Newton cooling law describes the rate of heat transfer between an object and its surrounding environment, which is the law followed when an object with a temperature higher than the surrounding environment transfers heat to the surrounding medium and gradually cools. When there is a temperature difference between the object's surface and the surrounding environment, the heat dissipated per unit time per unit area is proportional to the temperature difference, and the mathematical expression is:

[0031]

[0032] In the formula, T and T env are the temperatures of the object and the surrounding environment at time t, respectively; k is the cooling coefficient, which depends on the physical properties of the object and its contact conditions with the environment.

[0033] Step 3: Construct a roughness prediction model, including a 1D multi-scale residual convolutional neural network, a bidirectional long short-term memory neural network, a 2D multi-scale residual convolutional neural network, a normalization layer, a projection layer, and a KAN network;

[0034] The preprocessed maximum temperature sequence and average temperature sequence are used to extract features through a 1D multi-scale residual convolutional neural network to obtain the maximum temperature sequence features and average temperature sequence features; the maximum temperature sequence features and average temperature sequence features are used to capture the long-term dependencies in both directions through a bidirectional long short-term memory neural network to obtain the maximum temperature long-term dependency features and average temperature long-term dependency features; the preprocessed thermal imaging map is used to extract features through a 2D multi-scale residual convolutional neural network to obtain a thermal imaging feature map; after the maximum temperature long-term dependency features, average temperature long-term dependency features, thermal imaging feature map, process parameters, and tool profile are processed by the normalization layer, they are then projected through the projection layer to map all features into the same space, obtaining the maximum temperature projection features, average temperature projection features, thermal imaging projection features, process projection features, and tool profile projection features; after these projection features are concatenated, they are then mapped through the KAN network to obtain the surface roughness.

[0035] The 1D multi-scale residual convolutional neural network includes a multi-scale one-dimensional convolutional module. The output feature of the current one-dimensional convolutional module is connected with the input feature by residual connection to obtain a one-dimensional residual feature, and the one-dimensional residual feature undergoes a pooling operation to obtain the input feature of the next one-dimensional convolutional module. The one-dimensional convolutional module uses one-dimensional convolutional operation to extract features from the input. The extracted features undergo batch normalization operation to ensure the stability of the training process, and the batch-normalized features undergo ReLU activation operation to introduce non-linearity to enhance the expression ability of the model. The multi-scale convolutional module uses convolutional kernels of different sizes to extract multi-scale features of the temperature sequence. The size of the convolutional kernel determines the receptive field size of the convolutional operation. Using a smaller convolutional kernel can capture local details, while a larger convolutional kernel has a larger receptive field and can capture dependency information within a larger range. The expression of the one-dimensional convolutional module is:

[0036] yi = ReLU(BatchNorm(Conv1d(xi))) (2)

[0037] In the formula, x i , y i represent the input and output features of the i-th convolutional module. ReLU(·) represents the ReLU activation operation, BatchNorm(·) represents the batch normalization operation, and Conv1d(·) represents the one-dimensional convolutional operation;

[0038] The residual connection and pooling operation are expressed as:

[0039] xi +1 = Pool(Conv1d(xi) + yi) (3)

[0040] In the formula, x i+1 represents the input feature of the (i + 1)-th convolutional module, and Pool(·) represents the pooling operation;

[0041] The bidirectional long short-term memory neural network is extended based on the long short-term memory neural network (LSTM). It captures information from both the forward and backward directions. Through bidirectional processing, the model can fully utilize the bidirectional temporal dependencies in the sequence. The core of LSTM lies in the memory cell state and the gating mechanism, including the forget gate, input gate, and output gate. The gating mechanism uses the Sigmoid activation function to control the degree of information flow. The processing process of LSTM is expressed as:

[0042] The forget gate is used to control the information to be forgotten, and is expressed as:

[0043] f t = σ(W f · [h t-1 , x t + bf ) (4)

[0044] Where, f t represents the output of the forget gate at time step t, σ represents the Sigmoid activation function, W f , b f represent the weight matrix and bias term of the forget gate at time step t, h t-1 represents the hidden state at time step t - 1, x t represents the input at time step t;

[0045] The input gate is used to control how much information of the current input will be written into the memory cell, expressed as:

[0046] i t = σ(W i · [h t-1 , x t + b i ) (5)

[0047] Where, i t represents the output of the input gate at time step t, W i , b i represent the weight matrix and bias term of the input gate at time step t;

[0048] Generate the candidate memory cell state:

[0049]

[0050] Where, represents the candidate memory state at time step t, tanh(·) represents the tanh activation function, W C , b C represent the weight matrix and bias term of the candidate memory;

[0051] According to the candidate memory cell state and the memory cell state of the previous time step, update the memory cell state:

[0052]

[0053] Where, C t , C t-1 represent the memory cell states at time steps t and t - 1, * represents the convolution operation;

[0054] The output gate controls the output of the current time step:

[0055] o t = σ(W o · [h t-1 , x t + b o ) (8)

[0056] Wherein, o t represents the output of the output gate at time step t, and W o , b o represent the weight matrix and bias term of the output gate;

[0057] Hidden state:

[0058] h t = o t ·tanh(C t ) (9)

[0059] Wherein, h t represents the hidden state at time step t, which contains the long-term and short-term memory information of the current time step. At the same time, tanh activation is performed to compress the value into the range of (-1, 1) to control the information flow;

[0060] The output feature of the bidirectional LSTM is composed of the hidden states output by two LSTMs spliced together. Then there is:

[0061] h t bi = Concate(h t forward , h t backward ) (10)

[0062] Wherein, htbi represents the output feature of the bidirectional LSTM, and Concate(·) represents the splicing operation. h t forward is the hidden state output by the LSTM processed from front to back, and h t backward is the hidden state output by the LSTM processed from back to front.

[0063] The 2D multi-scale residual convolutional neural network includes a multi-scale two-dimensional convolutional module. The output feature of the current two-dimensional convolutional module is connected with the input feature by residual connection to obtain a two-dimensional residual feature. The two-dimensional residual feature undergoes a pooling operation to obtain the input feature of the next two-dimensional convolutional module; the two-dimensional convolutional module uses two-dimensional convolutional operation to extract features from the input, and the extracted features undergo batch normalization and ReLU activation operation to obtain the output feature of the two-dimensional convolutional module.

[0064] Since traditional multi-layer perceptrons often have difficulty dealing with high-dimensional data, and the KAN network changes the fixed activation function to a learnable activation function, so that each weight parameter in the KAN network can be replaced by a single-variable function, and these single-variable functions are parameterized in the form of spline functions, thus providing extremely high flexibility and using fewer parameters to simulate complex functions, enhancing the interpretability of the model.

[0065] Step 4: Use the dataset to train the surface roughness prediction model, and use the trained surface roughness prediction model for predicting the surface roughness of milling operations.

[0066] Figure 3 This is the comparison chart of the prediction results of the method of the present invention, where the mean square error MSE is 0.026, the mean absolute error MAE is 0.0498, and the coefficient of determination R 2 is 0.9636. The mean square error MSE of the prediction results using only the temperature series (including the maximum and average temperature series) is 0.0785, the mean absolute error MAE is 0.1430, and the coefficient of determination R 2 is 0.8366. The mean square error MSE of the prediction results using only the thermal imaging is 0.1014, the mean absolute error MAE is 0.2259, and the coefficient of determination R 2 is 0.8074. It can be seen that the mean square error, mean absolute error, and coefficient of determination of the present invention are all significantly better than the prediction results using only the temperature series and thermal imaging. This is because of the heterogeneous data fusion of the temperature series and thermal imaging, and the two complement each other to solve the problem of insufficient spatial information of the temperature series and the problem of temporal continuity of the thermal imaging, which can provide more comprehensive temperature information. Therefore, the prediction accuracy is improved, and the robustness and generalization ability of the model are enhanced.

[0067] Matters not described in the present invention are applicable to the prior art.

Claims

1. A method for predicting surface roughness of milling machining based on heterogeneous temperature data fusion, characterized in that: The method comprises the following steps: Step 1: Perform milling on the workpiece under different process parameters, collect thermal images, maximum temperature sequence and average temperature sequence during the milling process; after milling, measure the surface roughness of the workpiece; Step 2: Preprocess the thermal images, maximum temperature series and average temperature series to construct a data set; Step 3: Construct a roughness prediction model, including 1D multi-scale residual convolutional neural network, bidirectional long short-term memory neural network, 2D multi-scale residual convolutional neural network, normalization layer, projection layer and KAN network; The preprocessed maximum temperature sequence and average temperature sequence are subjected to 1D multi-scale residual convolutional neural network to obtain the maximum temperature sequence features and average temperature sequence features; the maximum temperature sequence features and average temperature sequence features are subjected to bidirectional long short-term memory neural network to obtain the maximum temperature long-term and short-term dependence features and the average temperature long-term and short-term dependence features; the preprocessed thermal imaging image is subjected to 2D multi-scale residual convolutional neural network to obtain the thermal imaging feature map; the maximum temperature long-term and short-term dependence features, the average temperature long-term and short-term dependence features, the thermal imaging feature map, the process parameters and the tool profile are normalized and then projected to obtain the maximum temperature projection features, the average temperature projection features, the thermal imaging projection features, the process projection features and the tool profile projection features; after splicing all the projection features, the surface roughness is obtained through KAN network mapping; Step 4: Use the data set to train the roughness prediction model, and use the trained roughness prediction model for milling surface roughness prediction.

2. The method for predicting surface roughness of milling machining based on heterogeneous temperature data fusion according to claim 1 is characterized in that: The 1D multi-scale residual convolutional neural network includes a multi-scale one-dimensional convolutional module, the output feature of the current one-dimensional convolutional module is residually connected with the input feature to obtain a one-dimensional residual feature, and the one-dimensional residual feature is pooled to obtain the input feature of the next one-dimensional convolutional module; The input features of the one-dimensional convolution module are sequentially subjected to one-dimensional convolution, batch normalization, and activation operations to obtain the output features of the module; The 2D multi-scale residual convolutional neural network includes a multi-scale two-dimensional convolution module. The output features of the current two-dimensional convolution module are residually connected with the input features to obtain two-dimensional residual features. The two-dimensional residual features are pooled to obtain the input features of the next two-dimensional convolution module; the input features of the two-dimensional convolution module are sequentially subjected to two-dimensional convolution, batch normalization and activation operations to obtain the output features of the module.

3. The method for predicting surface roughness of milling machining based on heterogeneous temperature data fusion according to claim 1 or 2, characterized in that: The maximum temperature sequence refers to the time series in which the maximum temperature in the processing area is formed at each sampling moment, and the average temperature sequence refers to the time series in which the average temperature in the processing area is formed at each sampling moment.

4. The method for predicting surface roughness of milling machining based on heterogeneous temperature data fusion according to claim 1, characterized in that: The preprocessing of thermal images includes denoising, alignment, cropping, grayscale and normalization, and the preprocessing of maximum temperature series and average temperature series includes padding and normalization.

Images