Precision control method and device for die steel processing

Abstract

The invention discloses a precision control method and device for die steel machining, and relates to the technical field of intelligent control. The method comprises the following steps: acquiring a monitoring signal during die steel processing in real time; the monitoring signal comprises a cutting force signal and a vibration signal; preprocessing the monitoring signal to obtain a time-frequency graph group; taking the time-frequency graph group as input of a pre-trained tool wear evaluation model to obtain a tool wear state; and executing a preset scheduling strategy according to the tool wear state. The cutter abrasion state is fed back in real time through the cutting force signal and the vibration signal, when the abrasion degree of the cutter reaches a certain critical value, measures such as cutting parameter adjustment or cutter replacement can be taken in advance, machining errors caused by excessive abrasion of the cutter are avoided, and therefore the machining precision is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a precision control method and device for die steel processing. Background Art

[0002] The machining of mold steel is a high-precision, demanding manufacturing process. Especially in the production of precision molds, tool wear directly impacts machining quality and production efficiency. Mold steel's high hardness and complex cutting characteristics make it susceptible to tool wear during machining.

[0003] In traditional mold steel machining, machining parameters, including cutting speed, feed rate, and depth of cut, are typically pre-set by technicians based on experience. These parameters are typically set according to general standards based on material properties and machining requirements, and are not dynamically adjusted throughout the production process. While these preset machining parameters may meet machining requirements in some cases, in actual production, factors such as the complexity of mold steel, tool wear, and varying operating conditions make it difficult to ensure machining stability and accuracy with these static machining parameter settings. Furthermore, tool wear often only becomes apparent to technicians when wear reaches a certain level. At this point, the tool has already lost its optimal cutting ability, leading to errors during machining and even damage to the mold or workpiece. Because traditional methods lack real-time feedback on wear dynamics, timely adjustment of machining parameters or tool replacement is impossible, making it difficult to effectively guarantee machining accuracy and surface quality. Summary of the Invention

[0004] The purpose of the present invention is to solve the problem mentioned in the above background technology that the processing accuracy is difficult to be effectively guaranteed, and to propose a precision control method and device for die steel processing.

[0005] A first aspect of the present invention provides a precision control method for die steel processing, the method comprising: Acquire monitoring signals during die steel processing in real time; the monitoring signals include cutting force signals and vibration signals; Preprocessing the monitoring signal to obtain a time-frequency graph group; Using the time-frequency graph group as input to a pre-trained tool wear evaluation model to obtain a tool wear state; According to the tool wear status, a preset scheduling strategy is executed; the scheduling strategy includes adjusting processing parameters and replacing tools.

[0006] Optionally, preprocessing the monitoring signal to obtain a time-frequency graph group includes: Denoising the target signal to obtain a first valid signal; the target signal is any one of the cutting force signal and the vibration signal; Performing a two-dimensional representation of the first effective signal using a continuous wavelet transform to obtain a target time-frequency graph; The target time-frequency graphs of the multiple monitoring signals are arranged in sequence to obtain a time-frequency graph group; the time-frequency graph group includes a force signal time-frequency graph and a vibration signal time-frequency graph.

[0007] Optionally, the denoising the target signal to obtain the first valid signal includes: The sparrow search algorithm is used to optimize the key parameters of variational mode decomposition with the minimum mean envelope entropy as the fitness function, and the modal decomposition number K and penalty factor α are obtained. Performing variational modal decomposition on the target signal according to the modal decomposition number K and the penalty factor α to obtain multiple modal components; Calculate the variance contribution rate of each modal component, retain the modal components with variance contribution rate greater than the preset threshold, and discard the others; The signal is reconstructed according to the retained modal components to obtain the first effective signal after denoising.

[0008] Optionally, the tool wear evaluation model includes an information fusion network, an attention network, a convolutional downsampling network, and a classification network; wherein: The information fusion network is configured to use two parallel convolution branches to perform convolution processing on the force signal time-frequency map and the vibration signal time-frequency map respectively to obtain a first feature map and a second feature map; and to concatenate the first feature map and the second feature map in the channel dimension to obtain a fused feature map; The attention network is used to process the fused feature map using a channel attention mechanism to obtain a third feature map; and to process the third feature map using a spatial attention mechanism to obtain an attention-enhanced feature map; The convolutional downsampling network is used to downsample the feature map using a cascaded convolutional layer and a pooling layer to obtain a fourth feature map; The classification network is used to flatten the fourth feature map to obtain a one-dimensional feature vector, and use a fully connected network and softmax as a classifier to process the one-dimensional feature vector to obtain the tool wear status.

[0009] Optionally, the calculation process of the attention network includes: , Where X is the input feature map; and Respectively represent global average pooling and global maximum pooling in the spatial dimension; and Represents two fully connected layers; sigmoid is a nonlinear normalization function; and They represent global average pooling and global maximum pooling in the channel dimension respectively; concat represents the splicing operation; Conv represents the convolution operation; 、 、 and It is a temporary variable during the operation; Y is the output of the attention network.

[0010] A second aspect of the present invention provides a precision control device for die steel processing, the device comprising: A sensor monitoring module is used to obtain monitoring signals during mold steel processing in real time; the monitoring signals include cutting force signals and vibration signals; A preprocessing module is used to preprocess the monitoring signal to obtain a time-frequency graph group; a wear evaluation module, configured to use the time-frequency graph group as input to a pre-trained tool wear evaluation model to obtain a tool wear state; The action control module is used to execute a preset scheduling strategy according to the wear status of the tool; the scheduling strategy includes adjusting processing parameters and replacing tools.

[0011] Optionally, the preprocessing module includes: A denoising module, configured to denoise a target signal to obtain a first valid signal; the target signal is either a cutting force signal or a vibration signal; a two-dimensional characterization module, configured to perform two-dimensional characterization on the first effective signal using continuous wavelet transform to obtain a target time-frequency graph; The image stacking module is used to arrange the target time-frequency graphs of multiple monitoring signals in sequence to obtain a time-frequency graph group; the time-frequency graph group includes a force signal time-frequency graph and a vibration signal time-frequency graph.

[0012] Optionally, the denoising module includes: a decomposition optimization module for optimizing key parameters of variational modal decomposition using a sparrow search algorithm and taking minimum mean envelope entropy as a fitness function to obtain a modal decomposition number K and a penalty factor α; performing variational modal decomposition on the target signal according to the modal decomposition number K and the penalty factor α to obtain multiple modal components; The modal screening module is used to calculate the variance contribution rate of each modal component and retain the modal components with variance contribution rates greater than a preset threshold, otherwise discard them; The signal reconstruction module is used to reconstruct the signal according to the retained modal components to obtain a first effective signal after denoising.

[0013] Optionally, the tool wear evaluation model includes an information fusion network, an attention network, a convolutional downsampling network, and a classification network; wherein: The information fusion network is configured to use two parallel convolution branches to perform convolution processing on the force signal time-frequency map and the vibration signal time-frequency map respectively to obtain a first feature map and a second feature map; and to concatenate the first feature map and the second feature map in the channel dimension to obtain a fused feature map; The attention network is used to process the fused feature map using a channel attention mechanism to obtain a third feature map; and to process the third feature map using a spatial attention mechanism to obtain an attention-enhanced feature map; The convolutional downsampling network is used to downsample the feature map using a cascaded convolutional layer and a pooling layer to obtain a fourth feature map; The classification network is used to flatten the fourth feature map to obtain a one-dimensional feature vector, and use a fully connected network and softmax as a classifier to process the one-dimensional feature vector to obtain the tool wear status.

[0014] Optionally, the calculation process of the attention network includes: , Where X is the input feature map; and Respectively represent global average pooling and global maximum pooling in the spatial dimension; and Represents two fully connected layers; sigmoid is a nonlinear normalization function; and They represent global average pooling and global maximum pooling in the channel dimension respectively; concat represents the splicing operation; Conv represents the convolution operation; 、 、 and It is a temporary variable during the operation; Y is the output of the attention network.

[0015] Beneficial effects of the present invention: The present invention proposes a precision control method for mold steel processing, which includes: acquiring monitoring signals during mold steel processing in real time; the monitoring signals include cutting force signals and vibration signals; preprocessing the monitoring signals to obtain a time-frequency graph group; using the time-frequency graph group as input of a pre-trained tool wear evaluation model to obtain the tool wear status; and executing a preset scheduling strategy based on the tool wear status; the scheduling strategy includes adjusting processing parameters and replacing tools.

[0016] The tool wear status is fed back in real time through cutting force signals and vibration signals. When the tool wear reaches a certain critical value, measures can be taken in advance, such as adjusting cutting parameters or replacing the tool, to avoid machining errors caused by excessive tool wear, thereby ensuring machining accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a precision control method for die steel processing is provided for an embodiment of the present invention; Figure 2 A network architecture diagram of a tool wear evaluation model is provided for an embodiment of the present invention; Figure 3 A schematic structural diagram of a precision control device for die steel processing is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0018] 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 creative efforts are within the scope of protection of the present invention.

[0019] The embodiment of the present invention provides a precision control method for die steel processing. Figure 1 , Figure 1 A flowchart of a precision control method for die steel processing provided by an embodiment of the present invention. The method comprises the following steps: S101, obtaining monitoring signals during mold steel processing in real time.

[0020] S102: Preprocess the monitoring signal to obtain a time-frequency graph group.

[0021] S103 , using the time-frequency graph group as input to a pre-trained tool wear evaluation model to obtain a tool wear state.

[0022] S104: Execute a preset scheduling strategy based on the tool wear status.

[0023] Among them, the monitoring signals include cutting force signals and vibration signals; the scheduling strategies include adjusting processing parameters and replacing tools.

[0024] Based on an embodiment of the present invention, a precision control method for mold steel processing is provided. The tool wear status is fed back in real time through cutting force signals and vibration signals. When the degree of tool wear reaches a certain critical value, measures can be taken in advance, such as adjusting cutting parameters or replacing the tool, to avoid processing errors caused by excessive tool wear, thereby ensuring processing accuracy.

[0025] In one implementation, the tool wear state includes initial wear, moderate wear, and severe wear. The scheduling strategy can be: if the tool is in the initial wear state, no adjustment is made. If the tool is in the moderate wear state, the cutting speed and feed rate can be reduced according to a preset ratio. For example, the cutting speed is reduced by 5% and the feed rate is reduced by 2%. If the tool is in a severe wear state, the tool is replaced. The tool wear state will have a direct impact on the processing quality. By adopting this strategy, different control measures can be adopted in different wear stages to reduce precision problems caused by excessive tool wear during the processing process, such as inaccurate workpiece dimensions or excessive surface roughness, to ensure that the processing quality is always in the best state.

[0026] In one embodiment, step S102 includes: Step 1: De-noise the target signal to obtain a first valid signal.

[0027] Step 2: Use continuous wavelet transform to perform two-dimensional characterization on the first effective signal to obtain the target time-frequency diagram.

[0028] Step 3: Arrange the target time-frequency graphs of the multiple monitoring signals in order to obtain a time-frequency graph group.

[0029] The target signal is either a cutting force signal or a vibration signal. The time-frequency graph group includes a force signal time-frequency graph and a vibration signal time-frequency graph.

[0030] In one implementation, denoising can improve signal clarity due to various environmental noise, equipment interference, and other factors during machining, facilitating subsequent analysis and judgment. The denoised signal is closer to the actual machining state, improving the accuracy of tool wear detection.

[0031] The variational mode decomposition (VMD) method can be used to denoise the target signal and obtain the first valid signal. Specifically: Using the sparrow search algorithm and the minimum mean envelope entropy as the fitness function, the key parameters of variational modal decomposition are optimized to obtain the modal decomposition number K and penalty factor α. Based on the modal decomposition number K and penalty factor α, variational modal decomposition is performed on the target signal to obtain multiple modal components.

[0032] Calculate the variance contribution rate of each modal component, retain the modal components whose variance contribution rate is greater than the preset threshold, and discard the others. Specifically, the preset threshold can be set to 1%, and the modal components less than the threshold are regarded as noise and discarded. The calculation formula is: , Where u represents the modal component, the subscript i (or m) represents the i-th (or m-th) modal component, j is the index number; N is the length of the modal component; represents the mean of the i-th modal component; represents the variance of the i-th modal component; K is the number of modal components; represents the variance contribution rate of the i-th modal component.

[0033] The signal is reconstructed according to the retained modal components to obtain a denoised signal, which is recorded as the first valid signal.

[0034] VMD decomposes complex signals into multiple modal components with specific frequency content. This method helps capture the distinct frequency characteristics of the signal while effectively isolating noise. This precise decomposition separates the useful components of the noise from the signal, resulting in higher-quality denoised signals. The sparrow search algorithm, an optimization method based on swarm intelligence, can find the optimal solution over a large range. By optimizing key VMD parameters using this algorithm, we can avoid the local optimum that can occur in traditional methods and ensure optimal signal decomposition.

[0035] In one embodiment, see Figure 2 , Figure 2 A network architecture diagram of a tool wear evaluation model is provided for an embodiment of the present invention. The tool wear evaluation model includes an information fusion network, an attention network, a convolutional downsampling network, and a classification network; wherein: The information fusion network uses two parallel convolution branches to convolve the force signal time-frequency map FD and the vibration signal time-frequency map VD, respectively, to obtain the first and second feature maps. The first and second feature maps are then concatenated along the channel dimension to obtain a fused feature map. Specifically, each convolution branch uses a 3×3 convolutional layer (Conv), and each convolution layer uses a Relu activation function for nonlinear transformation.

[0036] The attention network is used to process the fused feature map using the channel attention mechanism to obtain the third feature map; the spatial attention mechanism is used to process the third feature map to obtain the attention-enhanced feature map. Specifically, the calculation process of the attention network (AM) includes: , Where X is the input feature map; and Respectively represent global average pooling and global maximum pooling in the spatial dimension; and Represents two fully connected layers; sigmoid is a nonlinear normalization function; the symbol Indicates weighted product of feature maps; and They represent global average pooling and global maximum pooling in the channel dimension respectively; concat represents the splicing operation; Conv represents the convolution operation; 、 、 and It is a temporary variable during the operation; Y is the output of the attention network.

[0037] The convolutional downsampling network downsamples the feature map using a cascade of convolutional and pooling layers to obtain the fourth feature map. Specifically, three convolutional and pooling layers can be used for downsampling. Each convolutional and pooling layer consists of a 3×3 convolutional layer (Conv) with a ReLU activation function and a MaxPool layer with a pooling stride of 2 and a pooling window size of 3×3.

[0038] The classification network flattens the fourth feature map to obtain a one-dimensional feature vector. This vector is then processed using a fully connected network (FCN) and a softmax function as a classifier to determine the tool wear status. Specifically, a multi-layer perceptron (MLP) with two hidden layers can be used to process the one-dimensional feature vector. A softmax function is used at the output layer to output the state classification probability. The state category with the highest probability value is output.

[0039] By fusing multi-source information through deep learning technology and performing feature extraction, the complexity and uncertainty of manually designed features are reduced and the accuracy of state recognition is improved.

[0040] The embodiment of the present invention provides a precision control device for die steel processing. Figure 3 , Figure 3 This is a schematic diagram of a precision control device for die steel processing provided by an embodiment of the present invention. The device includes: The sensor monitoring module is used to obtain monitoring signals during mold steel processing in real time.

[0041] The preprocessing module is used to preprocess the monitoring signal to obtain a time-frequency graph group.

[0042] The wear evaluation module is used to take the time-frequency graph group as the input of the pre-trained tool wear evaluation model to obtain the tool wear status.

[0043] The motion control module is used to execute the preset scheduling strategy according to the tool wear status.

[0044] Among them, the monitoring signals include cutting force signals and vibration signals; the scheduling strategies include adjusting processing parameters and replacing tools.

[0045] A precision control device for mold steel processing provided by an embodiment of the present invention provides real-time feedback on the tool wear status through cutting force signals and vibration signals. When the degree of tool wear reaches a certain critical value, measures can be taken in advance, such as adjusting cutting parameters or replacing the tool, to avoid processing errors caused by excessive tool wear, thereby ensuring processing accuracy.

[0046] In one embodiment, the pre-processing module includes: The denoising module is used to denoise the target signal to obtain a first valid signal. The target signal is either a cutting force signal or a vibration signal.

[0047] The two-dimensional characterization module is used to perform two-dimensional characterization on the first effective signal by using continuous wavelet transform to obtain a target time-frequency diagram.

[0048] The image stacking module is used to arrange the target time-frequency graphs of multiple monitoring signals in sequence to obtain a time-frequency graph group; the time-frequency graph group includes a force signal time-frequency graph and a vibration signal time-frequency graph.

[0049] In one embodiment, the denoising module includes: The decomposition optimization module uses a sparrow search algorithm with minimum mean envelope entropy as the fitness function to optimize the key parameters of variational modal decomposition (VMD) to obtain the modal decomposition number K and penalty factor α. Based on the modal decomposition number K and penalty factor α, the target signal is subjected to variational modal decomposition to obtain multiple modal components.

[0050] The modal screening module is used to calculate the variance contribution rate of each modal component and retain the modal components with variance contribution rates greater than a preset threshold, otherwise they are discarded.

[0051] The signal reconstruction module is used to reconstruct the signal according to the retained modal components to obtain a first effective signal after denoising.

[0052] It should be noted that, in this document, terms such as "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements that are inherent to such process, method, article or apparatus.

[0053] The embodiments of the present invention are described in detail above, but the contents described are only preferred embodiments of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A precision control method for die steel processing, characterized in that: The method comprises: Acquire monitoring signals during die steel processing in real time; the monitoring signals include cutting force signals and vibration signals; Preprocessing the monitoring signal to obtain a time-frequency graph group; Using the time-frequency graph group as input to a pre-trained tool wear evaluation model to obtain a tool wear state; According to the tool wear status, a preset scheduling strategy is executed; the scheduling strategy includes adjusting processing parameters and replacing tools.

2. The precision control method for die steel processing according to claim 1, characterized in that: The preprocessing of the monitoring signal to obtain the time-frequency graph group includes: Denoising the target signal to obtain a first valid signal; the target signal is any one of the cutting force signal and the vibration signal; Performing a two-dimensional representation of the first effective signal using a continuous wavelet transform to obtain a target time-frequency graph; The target time-frequency graphs of the multiple monitoring signals are arranged in sequence to obtain a time-frequency graph group; the time-frequency graph group includes a force signal time-frequency graph and a vibration signal time-frequency graph.

3. The precision control method for die steel processing according to claim 2, characterized in that: Denoising the target signal to obtain the first valid signal includes: The sparrow search algorithm is used to optimize the key parameters of variational mode decomposition with the minimum mean envelope entropy as the fitness function, and the modal decomposition number K and penalty factor α are obtained. Performing variational modal decomposition on the target signal according to the modal decomposition number K and the penalty factor α to obtain multiple modal components; Calculate the variance contribution rate of each modal component, retain the modal components with variance contribution rate greater than the preset threshold, and discard the others; The signal is reconstructed according to the retained modal components to obtain the first effective signal after denoising.

4. The precision control method for die steel processing according to claim 1, characterized in that: The tool wear evaluation model includes an information fusion network, an attention network, a convolutional downsampling network and a classification network; wherein: The information fusion network is configured to use two parallel convolution branches to perform convolution processing on the force signal time-frequency map and the vibration signal time-frequency map respectively to obtain a first feature map and a second feature map; and to concatenate the first feature map and the second feature map in the channel dimension to obtain a fused feature map; The attention network is used to process the fused feature map using a channel attention mechanism to obtain a third feature map; and to process the third feature map using a spatial attention mechanism to obtain an attention-enhanced feature map; The convolutional downsampling network is used to downsample the feature map using a cascaded convolutional layer and a pooling layer to obtain a fourth feature map; The classification network is used to flatten the fourth feature map to obtain a one-dimensional feature vector, and use a fully connected network and softmax as a classifier to process the one-dimensional feature vector to obtain the tool wear status.

5. The precision control method for die steel processing according to claim 4, characterized in that: The calculation process of the attention network includes: ， Where X is the input feature map; and Respectively represent global average pooling and global maximum pooling in the spatial dimension; and Represents two fully connected layers; sigmoid is a nonlinear normalization function; and They represent global average pooling and global maximum pooling in the channel dimension respectively; concat represents the splicing operation; Conv represents the convolution operation; 、 、 and It is a temporary variable during the operation; Y is the output of the attention network.

6. A precision control device for die steel processing, characterized in that: The device comprises: A sensor monitoring module is used to obtain monitoring signals during mold steel processing in real time; the monitoring signals include cutting force signals and vibration signals; A preprocessing module is used to preprocess the monitoring signal to obtain a time-frequency graph group; a wear evaluation module, configured to use the time-frequency graph group as input to a pre-trained tool wear evaluation model to obtain a tool wear state; The action control module is used to execute a preset scheduling strategy according to the tool wear status; the scheduling strategy includes adjusting processing parameters and replacing tools.

7. The precision control device for die steel processing according to claim 6, characterized in that: The pre-processing module comprises: A denoising module, configured to denoise a target signal to obtain a first valid signal; the target signal is either a cutting force signal or a vibration signal; a two-dimensional characterization module, configured to perform two-dimensional characterization on the first effective signal using continuous wavelet transform to obtain a target time-frequency graph; The image stacking module is used to arrange the target time-frequency graphs of multiple monitoring signals in sequence to obtain a time-frequency graph group; the time-frequency graph group includes a force signal time-frequency graph and a vibration signal time-frequency graph.

8. The precision control device for die steel processing according to claim 7, characterized in that: The denoising module includes: a decomposition optimization module for optimizing key parameters of variational modal decomposition using a sparrow search algorithm and taking minimum mean envelope entropy as a fitness function to obtain a modal decomposition number K and a penalty factor α; performing variational modal decomposition on the target signal according to the modal decomposition number K and the penalty factor α to obtain multiple modal components; The modal screening module is used to calculate the variance contribution rate of each modal component and retain the modal components with variance contribution rates greater than a preset threshold, otherwise discard them; The signal reconstruction module is used to reconstruct the signal according to the retained modal components to obtain a first effective signal after denoising.

9. The precision control device for die steel processing according to claim 6, characterized in that: The tool wear evaluation model includes an information fusion network, an attention network, a convolutional downsampling network and a classification network; wherein: The information fusion network is configured to use two parallel convolution branches to perform convolution processing on the force signal time-frequency map and the vibration signal time-frequency map respectively to obtain a first feature map and a second feature map; and to concatenate the first feature map and the second feature map in the channel dimension to obtain a fused feature map; The attention network is used to process the fused feature map using a channel attention mechanism to obtain a third feature map; and to process the third feature map using a spatial attention mechanism to obtain an attention-enhanced feature map; The convolutional downsampling network is used to downsample the feature map using a cascaded convolutional layer and a pooling layer to obtain a fourth feature map; The classification network is used to flatten the fourth feature map to obtain a one-dimensional feature vector, and use a fully connected network and softmax as a classifier to process the one-dimensional feature vector to obtain the tool wear status.

10. The precision control device for die steel processing according to claim 9, characterized in that: The calculation process of the attention network includes: ， Where X is the input feature map; and Respectively represent global average pooling and global maximum pooling in the spatial dimension; and Represents two fully connected layers; sigmoid is a nonlinear normalization function; and They represent global average pooling and global maximum pooling in the channel dimension respectively; concat represents the splicing operation; Conv represents the convolution operation; 、 、 and It is a temporary variable during the operation; Y is the output of the attention network.

Images