Processing surface roughness prediction method and system based on deep learning

Abstract

The invention discloses a deep learning-based machining surface roughness prediction method and system, solves the problem of how to use a sensor as much as possible while improving machining surface roughness prediction, and belongs to the technical field of robot machining. A mutual cross fusion prediction model is constructed based on a deep learning network, and the roughness of the machined surface is predicted by using the internal joint torque and external vibration signals of the robot; the prediction model comprises a feature dimension reduction module, a multi-channel feature enhancement module and a mutual fusion module which are connected in sequence; the feature dimension reduction module performs dimension reduction on input to obtain low-dimensional features of each channel, and the multi-channel feature enhancement module realizes dynamic interaction among different channel features for the low-dimensional features of each channel to obtain multi-channel torque enhancement features and multi-channel vibration enhancement features; and the mutual fusion module carries out bidirectional fusion on the multi-channel torque enhancement characteristics and the multi-channel vibration enhancement characteristics to obtain fusion characteristics, and the roughness of the machined surface is predicted according to the fusion characteristics.

Description

Technical Field

[0001] The present invention relates to a method and system for predicting machining surface roughness based on deep learning, and belongs to the technical field of robotic machining. Background Art

[0002] Robots are increasingly being used in machining tasks due to their high flexibility and low cost. However, due to their low stiffness, robots are prone to significant vibration during machining, which can adversely affect surface roughness. Roughness is a key parameter for evaluating machined surface topography. There is a significant correlation between surface roughness and machining parameters. Notably, due to the nonlinear nature of robot performance, robots can exhibit varying levels of surface roughness under the same machining parameters in machining applications, making the prediction of robot-machined surface roughness extremely challenging.

[0003] Traditional surface roughness prediction models for machining take into account the influence of various physical factors such as tool runout, vibration, etc. However, the factors affecting surface roughness are complex, and there are many mechanisms that are difficult to describe and model. For this reason, a data-driven method based on signal characteristics is used in the task of predicting surface roughness for machining. Patent publication number CN119618146A discloses a surface roughness monitoring method with an embedded physical prediction model. By inputting theoretical cutting force, current signal, and machining parameters to train the surface roughness prediction model, the surface roughness value of the machined surface of the workpiece is calculated in real time. Patent publication number CN119260492A proposes an on-machine prediction method and device for roughness of the cylindrical grinding process. By obtaining the grinding image in any process, it is input into the improved Swin Transformer model for prediction and the prediction result is obtained. The above method realizes the prediction of machining surface roughness by establishing an end-to-end module. However, the robot machining process is particularly complex, so the data-driven method is particularly suitable for such complex tasks.

[0004] Data-driven approaches rely on inputs that effectively reflect data features. For robotic machining tasks, vibration signals can be used to reflect machining status information. To obtain further complementary and comprehensive information, more sensors are needed to achieve complementary feature information, but adding more sensors increases costs. Summary of the Invention

[0005] In order to solve the problem of how to improve the prediction of machining surface roughness while minimizing the use of sensors, the present invention provides a machining surface roughness prediction method and system based on deep learning.

[0006] A method for predicting machining surface roughness based on deep learning of the present invention comprises:

[0007] S1. Construct a data set. The input data is the robot's internal joint torque and external vibration signal. The output data is the machining surface roughness corresponding to the input data.

[0008] S2. Constructing a mutual cross-fusion prediction model based on a deep learning network, wherein the mutual cross-fusion prediction model includes a channel-by-channel feature dimension reduction module, a multi-channel feature enhancement module, and a mutual fusion module;

[0009] The robot's internal six-channel joint torque and external three-channel vibration signals are input into a channel-based feature dimensionality reduction module. The feature dimensionality reduction module performs feature dimensionality reduction on the internal six-channel joint torque and external three-channel vibration signals to obtain low-dimensional features of each channel, and inputs them into a parallel multi-channel feature enhancement module. The multi-channel feature enhancement module performs dynamic interaction between different channel features on the low-dimensional features of each channel to obtain multi-channel torque enhancement features and multi-channel vibration enhancement features corresponding to the internal six-channel joint torque and external three-channel vibration signals, respectively. The features are input into a mutual fusion module, which performs bidirectional fusion on the input multi-channel torque enhancement features and multi-channel vibration enhancement features to obtain fusion features, and predicts the roughness of the processed surface based on the fusion features.

[0010] S3. Use the data set to train the mutual cross-fusion prediction model, and use the trained mutual cross-fusion prediction model to predict the roughness of the processed surface.

[0011] Preferably, the mutual fusion module is implemented based on a dual-multi-head cross-attention mechanism, and the cross-modal attention vector of the interaction between the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature is calculated. The obtained cross-modal attention vector is then fused with the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature to obtain a fusion feature.

[0012] Preferably, the method for calculating the cross-modal attention vector of the interaction between the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature includes:

[0013] The multi-channel torque enhancement feature and the multi-channel vibration enhancement feature are C1 and C2 respectively, and the corresponding cross-modal attention vectors are:

[0014]

[0015] W O Represents a trainable parameter matrix;

[0016] Concat() means concatenation. represents the scaled dot product attention feature obtained by the m-th attention head from the cross-modal features of C1 to C2, represents the scaled dot product attention feature obtained by the m-th attention head from the cross-modal features of C2 to C1, m = 1,...,h, h represents the number of attention heads; represents the trainable parameter matrix of the mth attention head, Softmax represents the activation function, which is used to convert similarity into probability distribution. are the features assigned to the query vector Query, key vector Key, and value vector Value, respectively, with subscripts i = 1, 2, d k Indicates the dimension of the key vector Key.

[0017] Preferably, the feature dimension reduction module performs feature dimension reduction on the robot's six internal joint torque channels and three external vibration channel signals, and a method for obtaining low-dimensional features of each channel includes:

[0018] According to the internal joint torque or external vibration signal of each channel of the robot, a data set X=[x1,x2,...,x n ], where the input feature x i , i = 1, 2, ..., n, n represents the number of input features in the dataset;

[0019] Construct a directed weighted graph based on the data set X

[0020] According to the directed weighted graph Construct an undirected weighted graph G;

[0021] The undirected weighted graph G is effectively projected, and the low-dimensional feature map of the undirected weighted graph is obtained by minimizing the cross entropy function. The low-dimensional features of the multi-channel internal joint torque or multi-channel external vibration signal are obtained according to the low-dimensional feature map.

[0022] As a preference, input feature x i satisfy:

[0023]

[0024] ρ i =min{d(x i , x ij )|1≤j≤k,d(x i , x ij )>0}

[0025] Among them, x ij Represents x i The jth nearest neighbor, d(x i , x ij ) represents x i and x ij The distance between i represents the local connectivity constraint based on the distance index, σi represents the local Riemann metric of each point, k represents x i The number of neighbors it has under the metric.

[0026] As a preference, construct a directed weighted graph based on the dataset X methods, including:

[0027] Construct a directed weighted graph based on the data set X V represents the simplified set of X, and E′ represents the directed edge E′={(x i ,x ij )|1≤i≤n,1≤j≤k}, the edge weight ω is set to:

[0028]

[0029] ρ i =min{d(x i , x ij )|1≤j≤k,d(x i , x ij )>0}.

[0030] As a preference, according to a directed weighted graph Construct an undirected weighted graph G:

[0031] Get a directed weighted graph The weighted adjacency matrix A of , and the symbolic metric B is obtained according to the weighted adjacency matrix A;

[0032]

[0033] in, represents the Hadamard product, the element A of the weighted adjacency matrix A ij Indicates that from x i to x j The probability of the existence of a directed edge, the element B of the symbolic metric B ij Represent the probability that at least one of the two directed edges exists, and construct an undirected weighted graph G based on the symbolic metric.

[0034] Preferably, the cross entropy function is:

[0035]

[0036] Among them, ω h (e) and ω l (e) represents two fuzzy sets, an undirected weighted graph and a low-dimensional feature graph, respectively, and e represents the weight of the edge.

[0037] Preferably, the multi-channel feature enhancement module includes two branch structures, one of which is configured with a 6×2 convolution kernel to achieve cross-channel feature enhancement among the six channels within the robot;

[0038] The other branch structure is configured with a 3×2 convolution kernel to achieve cross-channel feature enhancement between the three external channels of the robot.

[0039] The beneficial effects of the present invention are as follows: the mutual cross-fusion prediction model in the present invention integrates the internal joint torque signal and external vibration signal of the robot, realizes the interaction of signal features in the fusion process, and better improves the processing process state information. The mutual cross-fusion prediction model of the present invention reduces the dimension of features by channel to reduce the interference of redundant features. The features after dimensionality reduction are constructed into a dual-branch structure, and the dynamic interaction between different channel features is realized through the multi-channel feature enhancement module. The mutual cross-fusion prediction model of the present invention is based on a dual-multi-head cross-attention mechanism, which realizes the collaborative interaction of cross-modal information, thereby completing the two-way deep collaborative representation between the internal and external signal features of the robot in the fusion process. Finally, the features are segmented and aggregated to accurately predict the surface roughness of the robot processing. By combining the internal and external signals of the robot to predict the surface roughness of the robot processing, the prediction of the surface roughness of the robot processing under different postures and different processing parameters can be achieved, and at the same time, it has a very high prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the internal and external signal acquisition system during the robot processing process of the present invention;

[0041] Figure 2 Schematic diagram of the structure of the cross-fusion prediction model in the embodiment of the present invention;

[0042] Figure 3 Schematic diagram of the structure of the multi-channel feature enhancement module in an example of the present invention;

[0043] Figure 4 This is a schematic diagram of the structure of the mutually integrated modules in an embodiment of the present invention;

[0044] Figure 5 This is a roughness prediction effect diagram of the cross-fusion prediction models in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0047] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0048] To accurately predict the surface roughness of robot-processed surfaces under different machining postures and parameters, this embodiment provides a surface roughness prediction method based on internal and external fusion signals and deep learning. By combining the robot's internal and external signals to predict the surface roughness of robot-processed surfaces, it can predict the surface roughness of robot-processed surfaces under different postures and different machining parameters with high prediction accuracy. The surface roughness prediction method based on deep learning in this embodiment includes:

[0049] Step 1: Construct a data set. The input data is the robot's internal joint torque and external vibration signal. The output data is the machining surface roughness corresponding to the input data:

[0050] A dataset of machined surface roughness was obtained through robotic machining experiments with different workpiece placements and machining parameters. Six internal torque channels and three external vibration channels were simultaneously collected. Channel-by-channel, internal and external signal features of the robot were extracted in the time, frequency, and time-frequency domains.

[0051] Specifically, in order to realize the collection of internal and external signals of the robot, this embodiment obtains a processing surface roughness data set through robot processing experiments with different workpiece placement and processing parameters. This embodiment uses a KR500-MT industrial robot for processing experiments. The robot is equipped with an electric spindle for processing operations. The tool holder adopts HSK model HSK63A-SDC8-100L. During the processing, the joint torque signals of the six channels inside the robot and the vibration signals of the three channels outside are collected simultaneously. The robot joint torque signal is obtained through the interface in the robot controller, and a three-axis acceleration sensor is installed on the electric spindle side to measure the vibration signal.

[0052] The surface roughness of robot-machined surfaces is closely related to the robot's posture and machining parameters. With this in mind, experiments with different machining parameters were conducted at six different workpiece placement positions. The workpiece material was Al 6061, and the workpiece dimensions were 140 × 140 × 40 mm. With each workpiece positioned identically, the spindle speed, feed rate, and axial cutting depth were varied, and multiple sets of machining experiments with different parameters were conducted. The workpiece surface information was collected using the roughness measuring instrument TR200. After completing a machining task at a certain axial cutting depth, the surface roughness of the stable machining section was measured three times using the roughness measuring instrument. During the measurement process, the workpiece was not disassembled to ensure in-situ measurement of the machined surface roughness. Figure 2 This is a schematic diagram of the internal and external signal acquisition system during the robot processing process.

[0053] The robot's internal and external signals comprehensively characterize the machining state, and extracting discriminative features from continuous signals is a prerequisite for accurate prediction. This implementation extracts features through analysis in the time, frequency, and time-frequency domains. The extraction process for the robot's internal and external signals is channel-by-channel. The six channels of the robot's internal signals consist of six torque signals in the joint space, and the three channels of the robot's external signals consist of vibration signals in three directions in Cartesian space. Before signal processing begins, the entry and exit signal segments are truncated, and only the intermediate stable segments of the signal are extracted. Furthermore, the start and end time points of the interception are calculated based on the roughness meter's measurement length and signal sampling time. After processing, a total of 9×16 groups of time-domain features, 9×13 groups of frequency-domain features, and 9×16 groups of time-frequency signal features are extracted. A total of 6×45 groups of features are extracted from the internal signals, and 3×45 groups of features are extracted from the external signals.

[0054] Step 2: Build a cross-fusion prediction model based on deep learning modules, such as Figure 2 As shown, it includes a channel-by-channel feature dimension reduction module, a multi-channel feature enhancement module, and a mutual fusion module:

[0055] Step 21: The robot's six internal channel joint torques and three external channel vibration signals are input into a channel-based feature dimensionality reduction module. The feature dimensionality reduction module performs feature dimensionality reduction on the six internal channel joint torques and three external channel vibration signals to obtain low-dimensional features for each channel, and inputs the low-dimensional features into a parallel multi-channel feature enhancement module.

[0056] During robotic machining, a multi-sensor system is used to acquire signals, from which complementary machining state responses can be derived through feature extraction. However, the extracted sensor signal features exhibit complex coupling relationships, and the presence of too many redundant features may reduce the learning ability of the model. To address this issue, dimensionality reduction technology can reduce the data dimension while retaining inherent structural information, thereby reducing the redundancy between signal features. The feature dimensionality reduction module of this embodiment reduces redundant dimensions before feature fusion. It includes the following steps:

[0057] Step 211: construct a data set X=[x1, x2, ..., x n ], where the input feature x i , i = 1, 2, ..., n, n represents the number of input features in the dataset;

[0058] k represents the standard, so the set is for each x i The closest distance ρ i and the smoothing normalization factor σ i Calculated. Combined with the nearest neighbor descent method, it is expressed as: ρ i =min{d(x i , x ij )|1≤j≤k,d(x i , x ij )>0}

[0059] Among them, x ij Represents x i The jth nearest neighbor, d(x i , x ij ) represents x i and x ij The distance between i Represents a local connection constraint based on the distance index, ensuring that x i Connected to at least one edge of weight 1. k represents x i The number of neighbors it has under the metric.

[0060] Smoothing normalization factor σ i is the local Riemannian metric of the i-th input feature, satisfying:

[0061]

[0062] Step 212: Construct a directed weighted graph based on the data set X

[0063] On the basis of ensuring data connection, a directed weighted graph is constructed Where V represents the simplified set of X. E' represents the directed edge E'={(x i,x ij )|1≤i≤n,1≤j≤k}, the edge weight ω is set to:

[0064]

[0065] where ω((x i ,x ij )) represents the edge weight, which is the probability of a given edge existing. Directed weighted graph It is asymmetric, and there may be two edges of unequal weights between two points. This construction method is used to capture the topological structure of the data to form a local graph, and then the local graphs are spliced according to different data groups to form a global representation of the manifold.

[0066] Step 213: According to the directed weighted graph Construct an undirected weighted graph G:

[0067] Let A be The weighted adjacency matrix represents the relationship between nodes in the graph and takes into account the symbolic metric B;

[0068]

[0069] in represents the Hadamard (or point-wise) product, the element A of the weighted adjacency matrix A ij Indicates that from x i to x j The probability of the existence of a directed edge, the element B of the symbolic metric B ij represents the probability that at least one of the two directed edges exists. Therefore, the high-dimensional graph representing G forms an undirected weighted graph given by the symbolic metric B, which approximates the original manifold with assumed local connectivity and uniform distribution, capturing the underlying structure of the data.

[0070] Step 214: Effectively project the undirected weighted graph G and obtain a low-dimensional feature graph of the undirected weighted graph by minimizing the cross entropy function. Based on the low-dimensional feature graph, obtain the low-dimensional features of the internal six-channel joint torque or the external three-channel vibration signal:

[0071] According to the effective projection of graph G, the low-dimensional layout is calculated. This part optimizes the cross entropy between the weighted edges of the undirected weighted graph and the low-dimensional feature graph, so that the low-dimensional feature graph composed of these points is as close as possible to the undirected weighted graph. The cross entropy function can be expressed as:

[0072]

[0073] Where CE(ω h ,ω l ) can measure two probability distributions, ω h (e) and ω l(e) represents two fuzzy sets, an undirected weighted graph and a low-dimensional feature graph, ω h (e) and ω l The difference information between (e) represents the similarity between the high-dimensional space and the transformed dimensional space, and e represents the weight of the edge. This formula represents the calculation principle of the force-directed graph layout algorithm. The first term represents the attraction between the point spans, so the first term has a greater weight under high-dimensional conditions because the distance between the points should be as small as possible. The second term represents the repulsive force between the two ends, which should also be as small as possible. Finally, the optimal weight of the edge in the low-dimensional space is found by minimizing the cross entropy function. This process is combined with the stochastic gradient descent method to optimize the mapping and obtain a low-dimensional data representation, N D Finally, the low-dimensional features of the multi-channel internal joint torque or multi-channel external vibration signal are obtained based on the low-dimensional feature map.

[0074] Step 22: The multi-channel feature enhancement module dynamically interacts with the low-dimensional features of the corresponding channels to obtain multi-channel torque enhancement features and multi-channel vibration enhancement features corresponding to the joint torque signals of the six internal channels and the vibration signals of the three external channels, respectively, and simultaneously inputs them into the mutual fusion module;

[0075] Dimensionality reduction technology provides a low-dimensional, efficient and information-preserving feature representation. However, the channel-level processing mechanism has inherent limitations and cannot effectively utilize the dynamic correlation and potential complementary information between cross-channel features. To solve this problem, this embodiment proposes a parallel multi-channel feature enhancement module, which achieves the complementarity of feature information of different channels through a dynamic feature interaction mechanism based on cross-channel convolution kernels. The input is N obtained after dimensionality reduction. D The module is designed with cross-channel convolution kernels with differentiated perception areas to adapt to the heterogeneous characteristics of the robot's internal and external signal channels. Then, based on the spatial distribution characteristics of the internal and external signal channels, a parallel dual-branch convolution submodule is constructed to establish a nonlinear mapping relationship between channels in the depth dimension. The multi-channel feature enhancement module of this embodiment includes two branch structures, such as Figure 3 As shown in the figure, one branch of the structure is configured with a 6×2 convolution kernel to achieve cross-channel feature enhancement among the six channels within the robot; the other branch of the structure is configured with a 3×2 convolution kernel to achieve cross-channel feature enhancement among the three channels outside the robot. This enables dynamic interaction between features from different channels, allowing features to gradually evolve during training, thereby enhancing the expressive power of information.

[0076] Step 23: The mutual fusion module performs bidirectional fusion on the input multi-channel torque enhancement features and the multi-channel vibration enhancement features to obtain fusion features, and predicts the roughness of the machined surface based on the fusion features;

[0077] In order to achieve information coordination and interaction in the fusion process, the present invention implements a mutual fusion module based on a dual-head cross attention mechanism, which calculates the attention weights between two different sequences to simulate the intra-modal relationship. In this invention, the two sequences are represented as the robot internal signal features and external signal characteristics These features are assigned to the Query (Q), Key (K) and Value (V) components of the attention mechanism to achieve mutual fusion of features. The fusion process consists of two parts: (1) internal to external feature fusion (C1→C2): C2 is assigned to K and V, and C1 is assigned to Q. (2) external to internal feature fusion (C2→C1): C1 is assigned to K and V, and C2 is assigned to Q. By fusing these two parts to better complete the feature state information and realize the interaction between features, the mutual fusion module of this embodiment is implemented based on the dual multi-head cross attention mechanism, and the cross-modal attention vector of the interaction between the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature is calculated. The cross-modal attention vector obtained is then fused with the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature to obtain a fused feature. Specifically, the method for calculating the cross-modal attention vector of the interaction between the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature includes:

[0078] The multi-channel torque enhancement feature and the multi-channel vibration enhancement feature are C1 and C2 respectively, and the corresponding cross-modal attention vectors are:

[0079]

[0080] W O Represents a trainable parameter matrix;

[0081] Concat() means concatenation. represents the scaled dot product attention feature obtained by the m-th attention head from the cross-modal features of C1 to C2, represents the scaled dot product attention feature obtained by the m-th attention head from the cross-modal features of C2 to C1, m = 1,...,h, h represents the number of attention heads; represents the trainable parameter matrix of the mth attention head, Softmax represents the activation function, which is used to convert similarity into probability distribution. are the features assigned to the query vector Query, key vector Key, and value vector Value, respectively, with subscripts i = 1, 2, d k Indicates the dimension of the key vector Key.

[0082] Cross-modal attention vector for the output and This module chooses to add these two parts to the previous enhanced features, and then aggregates them together in the depth direction to form the final fusion features.

[0083] Step 3: Use the data set to train the mutual cross-fusion prediction model, and use the trained mutual cross-fusion prediction model to predict the roughness of the processed surface.

[0084] In step 3, the hyperparameters of the cross-fusion prediction model are selected during training: the number of channel-level dimensionality reduction is 9, the batch size is 6, and the parameter epochs is 300. The ratio of the training set to the test set is 8:2.

[0085] This embodiment also provides a machining surface roughness prediction system based on deep learning, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the above-mentioned machining surface roughness prediction method based on deep learning.

[0086] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely illustrative of the principles and applications of the invention. It should be understood that many modifications may be made to the illustrative embodiments, and that other arrangements may be devised, without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in ways other than those described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be employed in conjunction with other described embodiments.

Claims

1. A method for predicting machining surface roughness based on deep learning, characterized in that: include: S1. Construct a data set. The input data is the robot's internal joint torque and external vibration signal. The output data is the machining surface roughness corresponding to the input data. S2. Constructing a mutual cross-fusion prediction model based on a deep learning network, wherein the mutual cross-fusion prediction model includes a channel-by-channel feature dimension reduction module, a multi-channel feature enhancement module, and a mutual fusion module; The robot's internal six-channel joint torque and external three-channel vibration signals are input into a channel-based feature dimensionality reduction module. The feature dimensionality reduction module performs feature dimensionality reduction on the internal six-channel joint torque and external three-channel vibration signals to obtain low-dimensional features of each channel, and inputs them into a parallel multi-channel feature enhancement module. The multi-channel feature enhancement module performs dynamic interaction between different channel features on the low-dimensional features of each channel to obtain multi-channel torque enhancement features and multi-channel vibration enhancement features corresponding to the internal six-channel joint torque and external three-channel vibration signals, respectively. The features are input into a mutual fusion module, which performs bidirectional fusion on the input multi-channel torque enhancement features and multi-channel vibration enhancement features to obtain fusion features, and predicts the roughness of the processed surface based on the fusion features. S3. Use the data set to train the mutual cross-fusion prediction model, and use the trained mutual cross-fusion prediction model to predict the roughness of the processed surface.

2. The method for predicting machining surface roughness based on deep learning according to claim 1, characterized in that: The mutual fusion module is implemented based on a dual-multi-head cross-attention mechanism, which calculates the cross-modal attention vector of the interaction between the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature, and then fuses the obtained cross-modal attention vector with the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature to obtain a fusion feature.

3. The method for predicting machining surface roughness based on deep learning according to claim 2, characterized in that: The method for calculating the cross-modal attention vector of the interaction between the multi-channel torque enhancement feature and the multi-channel vibration enhancement feature includes: The multi-channel torque enhancement feature and the multi-channel vibration enhancement feature are C1 and C2 respectively, and the corresponding cross-modal attention vectors are: W O Represents a trainable parameter matrix; Concat() means concatenation. represents the scaled dot product attention feature obtained by the m-th attention head from the cross-modal features of C1 to C2, represents the scaled dot product attention feature obtained by the m-th attention head from the cross-modal features of C2 to C1, m = 1,...,h, h represents the number of attention heads; represents the trainable parameter matrix of the mth attention head, Softmax represents the activation function, which is used to convert similarity into probability distribution. are the features assigned to the query vector Query, key vector Key, and value vector Value, respectively, with subscripts i = 1, 2, d k Indicates the dimension of the key vector Key.

4. The method for predicting machining surface roughness based on deep learning according to claim 1, characterized in that: The feature dimension reduction module performs feature dimension reduction on the robot's six internal joint torque channels and three external vibration channel signals. The method for obtaining low-dimensional features of each channel includes: According to the internal joint torque or external vibration signal of each channel of the robot, a data set X=[x1,x2,...,x n ], where the input feature x i , i = 1, 2, ..., n, n represents the number of input features in the dataset; Construct a directed weighted graph based on the data set X According to the directed weighted graph Construct an undirected weighted graph G; The undirected weighted graph G is effectively projected, and the low-dimensional feature map of the undirected weighted graph is obtained by minimizing the cross entropy function. The low-dimensional features of the multi-channel internal joint torque or multi-channel external vibration signal are obtained according to the low-dimensional feature map.

5. The method for predicting machining surface roughness based on deep learning according to claim 4, characterized in that: Input feature x i satisfy: ρ i =min{d(x i ,x ij )|1≤j≤k,d(x i ,x ij )>0} Among them, x ij Represents x i The jth nearest neighbor, d(x i , x ij ) represents x i and x ij The distance between i represents the local connectivity constraint based on the distance index, σ i represents the local Riemann metric of each point, k represents x i The number of neighbors it has under the metric.

6. The method for predicting machining surface roughness based on deep learning according to claim 5, characterized in that: Construct a directed weighted graph based on the data set X methods, including: Construct a directed weighted graph based on the data set X V represents the simplified set of X, and E′ represents the directed edge E′={(x i ,x ij )|1≤i≤n,1≤j≤k}, the edge weight ω is set to: ρ i =min{d(x i ,x ij )|1≤j≤k,d(x i ,x ij )>0}。 7. The method for predicting machining surface roughness based on deep learning according to claim 4, characterized in that: According to the directed weighted graph Construct an undirected weighted graph G: Get a directed weighted graph The weighted adjacency matrix A of , and the symbolic metric B is obtained according to the weighted adjacency matrix A; in, represents the Hadamard product, the element A of the weighted adjacency matrix A ij Indicates that from x i to x j The probability of the existence of a directed edge, the element B of the symbolic metric B ij Represent the probability that at least one of the two directed edges exists, and construct an undirected weighted graph G based on the symbolic metric.

8. The method for predicting machining surface roughness based on deep learning according to claim 4, characterized in that: The cross entropy function is: Among them, ω h (e) and ω l (e) represents two fuzzy sets, an undirected weighted graph and a low-dimensional feature graph, respectively, and e represents the weight of the edge.

9. The method for predicting machining surface roughness based on deep learning according to claim 1, characterized in that: The multi-channel feature enhancement module consists of two branches, one of which is equipped with a 6×2 convolution kernel to achieve cross-channel feature enhancement among the six channels within the robot. The other branch structure is configured with a 3×2 convolution kernel to achieve cross-channel feature enhancement between the three external channels of the robot.

10. A device for predicting machining surface roughness based on deep learning, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that: The processor executes the computer program to implement the steps of the method for predicting machining surface roughness based on deep learning according to any one of claims 1 to 9.

Images