Intelligent motor rotor hot forging high-pressure die wear prediction method and system

Abstract

The invention discloses an intelligent motor rotor hot-forging high-pressure die wear prediction method and system. The method comprises the steps of image acquisition, image preprocessing, filtering processing, image reconstruction, wear prediction model establishment and motor rotor hot-forging high-pressure die wear prediction. The invention belongs to the field of image processing, and particularly relates to an intelligent motor rotor hot forging high-pressure die wear prediction method and system. According to the scheme, gray level distribution statistics is introduced to dynamically adjust a mapping range; designing double Gaussian weights of a spatial domain and an intensity domain, and applying logarithm domain illumination component estimation; the wear prediction accuracy of the hot-forging high-pressure die is improved; performing independent prediction on each sub-band of the wavelet through the pre-branch to realize early detection of the crack initiation stage; the backbone network focuses on the overall wear level and depth, and gives consideration to both macroscopic and microcosmic properties; distinguishing a micro crack signal in the initiation stage by defining segment-level classification loss, and defining a continuity penalty term; and the final mold wear prediction reliability is ensured.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an intelligent method and system for predicting wear of a motor rotor hot forging high-pressure die. Background Art

[0002] Methods for predicting high-pressure die wear during hot forging of motor rotors use online or offline monitoring and analysis of the die surface condition to predict the remaining die life and degree of wear during the hot forging process. However, typical methods for predicting high-pressure die wear during hot forging of motor rotors are unable to adapt to local brightness variations, resulting in poor accuracy in die wear prediction due to noise interference, poor sensitivity to noise, and the loss of microcrack details. These methods also lack dedicated monitoring of the crack initiation stage, leading to insufficient reliability in die wear prediction. Summary of the Invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent motor rotor hot forging high-pressure die wear prediction method and system. In view of the problem that the general motor rotor hot forging high-pressure die wear prediction method cannot adapt to the brightness changes in local areas, the surface microcracks are improperly identified due to noise interference, and the final die wear prediction accuracy is poor, this solution adopts the characteristics of non-uniform illumination and strong thermal radiation on the surface of hot forged workpieces, introduces grayscale distribution statistics to dynamically adjust the mapping range, quantizes and eliminates radiation gradients, and ensures that the contrast between microcracks and pits is globally uniform; designs dual Gaussian weights in spatial domain and intensity domain to achieve fine suppression of dust, oil and speckle noise; on this basis, applies logarithmic domain illumination component estimation and correction factors to remove residual non-uniform illumination artifacts, while protecting the crack and peeling edge features from being blurred; for micro For the high-frequency sub-bands of crack and notch details, a soft threshold re-estimation based on data statistics is proposed to adaptively eliminate the noise coefficient and amplify the real crack and notch signals at the same time; thereby improving the accuracy of hot forging high-pressure die wear prediction; in view of the fact that the general motor rotor hot forging high-pressure die wear prediction method is sensitive to noise, easy to lose micro-crack details, lacks special supervision of the crack initiation stage, and thus leads to insufficient reliability of die wear prediction, this scheme uses pre-branching to independently predict on each sub-band of the wavelet to achieve early detection of the crack initiation stage; the backbone network focuses on the overall wear level and depth, taking into account both macro and micro; by defining segment-level classification loss, tiny crack signals are distinguished in the initiation stage, and continuity penalty terms are defined to prevent prediction regression due to occasional noise, ensuring the monotonicity of wear level as the segment progresses; thereby ensuring the reliability of the final die wear prediction.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent motor rotor hot forging high-pressure die wear prediction method, the method comprising the following steps:

[0005] Step S1: image acquisition;

[0006] Step S2: image preprocessing;

[0007] Step S3: filtering process;

[0008] Step S4: image reconstruction;

[0009] Step S5: establishing a wear prediction model;

[0010] Step S6: Predicting the wear of the motor rotor hot forging high-pressure die.

[0011] Furthermore, in step S1, the image acquisition is to acquire historical motor rotor hot forging high-pressure die surface images and mark the die wear level; the die wear level is used as a data label.

[0012] Furthermore, in step S2, the image preprocessing is to preprocess the image using gray value mapping and adaptive contrast enhancement.

[0013] Furthermore, in step S3, the filtering process is to perform filtering process on the pre-processed image through dual-domain weighted filtering.

[0014] Furthermore, in step S4, the image reconstruction is based on wavelet decomposition and combined with high-frequency threshold re-evaluation, and the low-frequency component is reconstructed through soft threshold processing.

[0015] Furthermore, in step S5, the wear prediction model is established based on a deep learning model, using the original hot forging surface image and the reconstructed hot forging surface image to establish the wear prediction model; specifically, the following steps are included:

[0016] Step S51: Overall network structure design; including: pre-branch and image branch, which first perform a layer of ordinary convolution, batch normalization and ReLU activation on the original hot forging surface image to extract low-level texture features; pre-branch performs wavelet decomposition on the reconstructed hot forging surface image and independently predicts the category for each decomposition segment; signal branch, which flattens the reconstructed hot forging surface image and then passes it through a layer of full connection, batch normalization and ReLU to obtain a 128-dimensional working condition feature vector; feature fusion, which splices the image branch output with the signal branch output and sends it to the backbone network; the backbone network consists of K inverse residuals + SE + weighted mixed pooling, which gradually extracts deep wear features and compresses the feature map size; the output layer is generated through 1×1 convolution, obtains the probability of each wear level through Softmax, and outputs the wear depth;

[0017] Step S52: reverse residual; perform the following operations on the intermediate feature map X on the image branch: channel expansion; depthwise separable convolution; SE recalibration; compression and residual connection;

[0018] Step S53: weighted mixed pooling; replacing single maximum and average pooling by weighted mixed pooling;

[0019] Step S54: loss function design; the loss function consists of three parts: segment-level classification loss, continuity penalty term, and prediction loss. The segment-level classification loss and continuity penalty term are used to make early predictions of the reconstructed image, and the prediction loss is used to measure the model prediction results; including:

[0020] Step S541 defines the segment-level classification loss , perform contrastive cross entropy on each local segment after wavelet decomposition of the reconstructed hot forging surface image;

[0021] Step S542: Define continuity penalty term ;

[0022] Step S543: Define the total loss.

[0023] Furthermore, in step S6, the motor rotor hot forging high-pressure die wear prediction is to collect the motor rotor hot forging high-pressure die surface image in real time, and input it into the wear prediction model after image preprocessing, filtering processing and image reconstruction processing, and use the die wear level output by the model as the prediction result.

[0024] The present invention provides an intelligent motor rotor hot forging high-pressure die wear prediction system, which includes an image acquisition module, an image preprocessing module, a filtering processing module, an image reconstruction module, a wear prediction model establishment module and a motor rotor hot forging high-pressure die wear prediction module;

[0025] The image acquisition module acquires a surface image of the motor rotor hot forging high-pressure die and marks the wear level of the die;

[0026] The image preprocessing module preprocesses the image using gray value mapping and adaptive contrast enhancement;

[0027] The filtering processing module performs filtering processing on the pre-processed image through dual-domain weighted filtering;

[0028] The image reconstruction module reconstructs the low-frequency components through soft threshold processing based on wavelet decomposition and high-frequency threshold re-estimation;

[0029] The wear prediction model establishment module establishes a wear prediction model based on the pre-branch, the image branch and the signal branch;

[0030] The motor rotor hot forging high-pressure die wear prediction module predicts die wear based on real-time collected surface images of the motor rotor hot forging high-pressure die.

[0031] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0032] (1) Aiming at the problem that the general prediction method of high-pressure die wear for hot forging of motor rotors cannot adapt to the brightness changes in local areas, and the surface microcracks are improperly identified due to noise interference, resulting in poor accuracy of die wear prediction. This scheme introduces grayscale distribution statistics to dynamically adjust the mapping range, quantify and eliminate radiation gradients, and ensure that the contrast between microcracks and pits is globally uniform, targeting the characteristics of non-uniform illumination and strong thermal radiation on the surface of hot forging workpieces; designs dual Gaussian weights in the spatial domain and intensity domain to achieve fine suppression of dust, oil and speckle noise; on this basis, applies logarithmic domain illumination component estimation and correction factors to remove residual non-uniform illumination artifacts, while protecting the crack and spalling edge features from being blurred; for the high-frequency sub-bands of microcracks and notch details, proposes a soft threshold re-estimation based on data statistics to adaptively eliminate noise coefficients and amplify real crack and notch signals; thereby improving the accuracy of hot forging high-pressure die wear prediction.

[0033] (2) Aiming at the problem that the general motor rotor hot forging high-pressure die wear prediction method is sensitive to noise, easy to lose microcrack details, lacks special supervision of the crack initiation stage, and thus leads to insufficient reliability of die wear prediction, this scheme uses pre-branching to independently predict on each sub-band of the wavelet to achieve early detection of the crack initiation stage; the backbone network focuses on the overall wear level and depth, taking into account both macro and micro; by defining segment-level classification loss, it distinguishes small crack signals in the initiation stage, and defines continuity penalty terms to prevent prediction regression due to occasional noise, ensuring the monotonicity of wear level as the segment progresses; thus ensuring the reliability of the final die wear prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic flow chart of an intelligent method for predicting wear of a motor rotor hot forging high-pressure die provided by the present invention;

[0035] Figure 2 This is a schematic diagram of an intelligent motor rotor hot forging high-pressure die wear prediction system provided by the present invention;

[0036] Figure 3 is a flow chart of step S5.

[0037] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0039] In the description of the present invention, it should be understood that terms such as "up", "down", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0040] Example 1, see Figure 1 The present invention provides an intelligent motor rotor hot forging high-pressure die wear prediction method, which includes the following steps:

[0041] Step S1: Image acquisition: Acquire a surface image of the motor rotor hot forging high-pressure die and mark the wear level of the die;

[0042] Step S2: Image preprocessing: grayscale value mapping and adaptive contrast enhancement are used to preprocess the image;

[0043] Step S3: filtering processing: filtering the pre-processed image by dual-domain weighted filtering;

[0044] Step S4: Image reconstruction: Based on wavelet decomposition, combined with high-frequency threshold re-estimation, the low-frequency component is reconstructed through soft threshold processing;

[0045] Step S5: establishing a wear prediction model; establishing the wear prediction model based on the pre-branch, the image branch, and the signal branch;

[0046] Step S6: Predicting the wear of the motor rotor hot forging high-pressure die; performing die wear prediction on the real-time collected surface image of the motor rotor hot forging high-pressure die.

[0047] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the mold wear level is used as a data label; the mold wear level is normal, light wear, moderate wear and severe wear.

[0048] Example 3, see Figure 1This embodiment is based on the above embodiment. In step S2, image preprocessing is to balance the brightness offset caused by thermal radiation and uneven light source, and enhance the contrast of small surface wear depressions and cracks. Gray value mapping and adaptive contrast enhancement are used to preprocess the image. It is expressed as: ;

[0049] ; ; Wherein, PDF(q) is the relative frequency of the gray value q in the hot forging surface image; is the number of pixels whose grayscale value of the hot forging surface image is equal to q; N is the total number of pixels in the image; CDF(q) is the value from the darkest grayscale value To the cumulative distribution of the current gray value q; is the new grayscale value after mapping; is the brightest gray value; u is the gray value index.

[0050] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the filtering process is to suppress the dust, oil and speckle noise generated during the hot forging process. Traditional filtering is easy to blur the edge features. While suppressing the noise, the edges of cracks and peeling are retained to the maximum extent, and the non-uniform illumination is eliminated to make the reflective features of the wear area more prominent. The pre-processed image is filtered by dual-domain weighted filtering. It is expressed as: ;

[0051] ;

[0052] ;in, is the grayscale value of the filtered hot forging surface image at (o, b); The standard deviation is A two-dimensional Gaussian kernel is used to estimate the local illumination component; is the correction factor; is the target pixel coordinate; are the pixel coordinates within the neighborhood; is the grayscale value after correction; is the grayscale value after preprocessing; and are the Gaussian standard deviation in the spatial domain and the Gaussian standard deviation in the intensity domain, respectively; is the joint weight.

[0053] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the image reconstruction is that the low-frequency component reveals the overall deformation and pit depth, and the high-frequency component enhances the details of microcracks and scratches. Based on wavelet decomposition and combined with high-frequency threshold re-estimation, the low-frequency component is reconstructed through soft threshold processing, which is expressed as: ; High-frequency threshold reestimation, expressed as: ; ; The refactoring is: ;in, and They are low-frequency sub-band and high-frequency sub-band respectively; is the soft threshold processing value; sign(·) is the sign function; is the threshold; is the estimated standard deviation of the high-frequency coefficient; is the total number of high-frequency coefficients; is the inverse wavelet transform; is the reconstructed pixel value.

[0054] By performing the above operations, the general motor rotor hot forging high-pressure die wear prediction method has the problem that it cannot adapt to the brightness changes in local areas, the surface microcracks are improperly identified due to noise interference, and the final die wear prediction accuracy is poor. In view of the characteristics of non-uniform illumination and strong thermal radiation on the surface of hot forged workpieces, this scheme introduces grayscale distribution statistics to dynamically adjust the mapping range, quantify and eliminate radiation gradients, and ensure that the contrast between microcracks and pits is globally uniform; design dual Gaussian weights in the spatial domain and intensity domain to achieve fine suppression of dust, oil and speckle noise; on this basis, apply logarithmic domain illumination component estimation and correction factors to remove residual non-uniform illumination artifacts, while protecting the crack and spalling edge features from being blurred; for the high-frequency sub-bands of microcracks and scratch details, a soft threshold reestimation based on data statistics is proposed to adaptively eliminate the noise coefficient and amplify the real crack and scratch signals, thereby improving the accuracy of hot forging high-pressure die wear prediction.

[0055] Example 6, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S5, the wear prediction model is established based on a deep learning model, and the wear prediction model is established using the original hot forging surface image and the reconstructed hot forging surface image. Specifically, the following steps are included:

[0056] Step S51: Overall network structure design; including: pre-branch and image branch, which first perform a layer of ordinary convolution, batch normalization and ReLU activation on the original hot forging surface image to extract low-level texture features; pre-branch performs wavelet decomposition on the reconstructed hot forging surface image and independently predicts the category for each decomposition segment; signal branch, which flattens the reconstructed hot forging surface image and then passes it through a layer of full connection, batch normalization and ReLU to obtain a 128-dimensional working condition feature vector; feature fusion, which splices the image branch output with the signal branch output and sends it to the backbone network; the backbone network consists of several improved inverse residuals + SE + weighted mixed pooling, which gradually extracts deep wear features and compresses the feature map size; the output layer is generated through 1×1 convolution, obtains the probability of each wear level through Softmax, and outputs the wear depth;

[0057] Step S52: reverse residual; the dust, oil and high temperature radiation noise on the die surface at the hot forging site are complex. Through deep separable convolution, the filter is first separated in space and channel, and then SE focuses on the weak crack channel. The overall effect is better at suppressing environmental noise and highlighting crack signals. The intermediate feature map X on the image branch is operated in sequence: channel expansion, expressed as: ;Wherein, U is the channel expansion output; is the ReLU activation function; It is batch normalization processing; is the expanded convolution kernel; Is the channel from Expand to ; is the expansion coefficient; depthwise separable convolution, expressed as: ;Where V is the depth-wise separable convolution output; is the intra-channel convolution; s is the convolution kernel size; SE recalibration is expressed as: ; ; ;in, is the global average pooling result; is the scaling factor; H and W are the height and width of the image respectively, and i and j are the corresponding coordinate indices; is the value of the c-th channel at coordinates; and are two fully connected layers in SE; z is a vector consisting of the global average pooling results corresponding to all channels; is the V of the cth channel; is the result of SE recalibration; compressed and residual connected, expressed as: ; If s=1 and ,but ; Among them, compressed output; is the compressed convolution kernel; It is the restored channel; is the residual connection result;

[0058] Step S53: Weighted hybrid pooling. The crack is extremely subtle in its incipient stage and can be easily smoothed out if only avg pooling is used. If only max is used, the weak crack response may be below the noise threshold. Through learnable fusion weights, both the crack extreme value and the overall information of the wear area are flexibly considered. Hybrid pooling adaptively adjusts the weight of the two in a learnable manner to ensure that the crack signal is neither overly smoothed nor overly jittered. Weighted hybrid pooling replaces single maximum and average pooling, expressed as: ; ; ;and and Updated together with the network parameters by gradient descent; where, and are maximum pooling and average pooling respectively; v is the vector of all activation values in the sliding window; is the activation value of the nth position in the pooling window; W1 is the index set of all positions in the pooling window; is the weighted output; and is the learnable fusion weight;

[0059] Step S54: loss function design; the loss function consists of three parts: segment-level classification loss, continuity penalty term, and prediction loss. The segment-level classification loss and continuity penalty term are used to make early predictions of the reconstructed image, and the prediction loss is used to measure the model prediction results; including:

[0060] Step S541: Define segment-level classification loss , the comparative cross entropy is performed on each local segment after wavelet decomposition of the reconstructed hot forging surface image to detect the crack initiation stage earlier, which is expressed as: ; Where i is the sample index; j is the wavelet segment index of the sample segmentation; is the local segment length; is the starting and ending position of the jth wavelet segment of the i-th sample after wavelet decomposition; is the wavelet segment interval of the reconstructed hot forging surface image; It is feature extraction; is the network output; are network parameters; is the jth wavelet segment of the i-th sample; Is for , predict the probability that it belongs to category n; and are the mean and variance of the Gaussian distribution of category n; Corresponding category ; is a normalization operation;

[0061] Step S542: Define continuity penalty term , if the score of the adjacent segment decreases, a hinge penalty is applied to avoid wear level prediction regression due to noise or occasional defects; it is expressed as: ; ;in, It is the prediction result after softmax normalization; It is the prediction result without softmax normalization; is the j-1th wavelet segment of the i-th sample; is the length correction factor; are the starting and ending positions of the jth small dial segment of the i-th sample; g is the measurement parameter;

[0062] Step S543: Define total loss, total loss Expressed as: ;in, and is the loss weight parameter; It is the cross entropy loss between the final backbone network prediction category and the true category.

[0063] By performing the above operations, the general motor rotor hot forging high-pressure die wear prediction method is sensitive to noise, prone to losing microcrack details, and lacks special supervision of the crack initiation stage, which leads to insufficient reliability of die wear prediction. This scheme uses pre-branching to independently predict on each wavelet sub-band to achieve early detection in the crack initiation stage; the backbone network focuses on the overall wear level and depth, taking into account both macro and micro; by defining segment-level classification loss, tiny crack signals are distinguished in the initiation stage, and a continuity penalty term is defined to prevent prediction regression due to occasional noise, ensuring the monotonicity of wear level as the segment progresses; thereby ensuring the reliability of the final die wear prediction.

[0064] Example 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the wear prediction of the motor rotor hot forging high-pressure die is to collect the surface image of the motor rotor hot forging high-pressure die in real time, and input it into the wear prediction model after image preprocessing, filtering processing and image reconstruction processing, and use the die wear level output by the model as the prediction result; if it is severe wear, early warning processing is performed.

[0065] Example 8, see Figure 2This embodiment is based on the above embodiment. The present invention provides an intelligent motor rotor hot forging high-pressure die wear prediction system, including an image acquisition module, an image preprocessing module, a filtering processing module, an image reconstruction module, a wear prediction model establishment module and a motor rotor hot forging high-pressure die wear prediction module;

[0066] The image acquisition module acquires a surface image of the motor rotor hot forging high-pressure die and marks the wear level of the die;

[0067] The image preprocessing module preprocesses the image using gray value mapping and adaptive contrast enhancement;

[0068] The filtering processing module performs filtering processing on the pre-processed image through dual-domain weighted filtering;

[0069] The image reconstruction module reconstructs the low-frequency components through soft threshold processing based on wavelet decomposition and high-frequency threshold re-estimation;

[0070] The wear prediction model establishment module establishes a wear prediction model based on the pre-branch, the image branch and the signal branch;

[0071] The motor rotor hot forging high-pressure die wear prediction module predicts die wear based on real-time collected surface images of the motor rotor hot forging high-pressure die.

[0072] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0073] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0074] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. An intelligent method for predicting wear of high-pressure dies for hot forging of motor rotors, characterized by: The method comprises the following steps: Step S1: Image acquisition: Acquire a surface image of the motor rotor hot forging high-pressure die and mark the wear level of the die; Step S2: image preprocessing; Step S3: filtering process; Step S4: image reconstruction; Step S5: establishing a wear prediction model; Step S6: Predicting the wear of the motor rotor hot forging high-pressure die.

2. The intelligent motor rotor hot forging high-pressure die wear prediction method according to claim 1 is characterized by: In step S5, the wear prediction model is established based on a deep learning model, and the wear prediction model is established using the original hot forging surface image and the reconstructed hot forging surface image; The specific steps include: Step S51: Overall network structure design; including: pre-branch and image branch, which first perform a layer of ordinary convolution, batch normalization and ReLU activation on the original hot forging surface image to extract low-level texture features; pre-branch performs wavelet decomposition on the reconstructed hot forging surface image and independently predicts the category for each decomposition segment; signal branch, which flattens the reconstructed hot forging surface image and then passes it through a layer of full connection, batch normalization and ReLU to obtain a 128-dimensional working condition feature vector; feature fusion, which splices the image branch output with the signal branch output and sends it to the backbone network; the backbone network consists of K inverse residuals + SE + weighted mixed pooling, which gradually extracts deep wear features and compresses the feature map size; the output layer is generated through 1×1 convolution, obtains the probability of each wear level through Softmax, and outputs the wear depth; Step S52: reverse residual; perform the following operations on the intermediate feature map X on the image branch: channel expansion; depthwise separable convolution; SE recalibration; compression and residual connection; Step S53: weighted mixed pooling; replacing single maximum and average pooling by weighted mixed pooling; Step S54: loss function design.

3. The intelligent motor rotor hot forging high-pressure die wear prediction method according to claim 2 is characterized by: In step S5, the loss function is designed to include three parts: segment-level classification loss, continuity penalty term, and prediction loss. The segment-level classification loss and continuity penalty term are used to make early predictions of the reconstructed image, and the prediction loss is used to measure the prediction results of the model. include: Step S541 defines the segment-level classification loss , perform contrastive cross entropy on each local segment after wavelet decomposition of the reconstructed hot forging surface image; Step S542: Define continuity penalty term ; Step S543: Define the total loss.

4. The intelligent motor rotor hot forging high-pressure die wear prediction method according to claim 3 is characterized by: In step S2, the image preprocessing is to preprocess the image using gray value mapping and adaptive contrast enhancement.

5. The intelligent motor rotor hot forging high-pressure die wear prediction method according to claim 4 is characterized by: In step S3, the filtering process is to perform filtering on the pre-processed image by dual-domain weighted filtering.

6. The intelligent motor rotor hot forging high-pressure die wear prediction method according to claim 5, characterized in that: In step S4, the image reconstruction is based on wavelet decomposition and combined with high-frequency threshold re-estimation, and the low-frequency component is reconstructed through soft threshold processing.

7. The intelligent motor rotor hot forging high-pressure die wear prediction method according to claim 6, characterized in that: In step S6, the motor rotor hot forging high-pressure die wear prediction is to collect the motor rotor hot forging high-pressure die surface image in real time, and input it into the wear prediction model after image preprocessing, filtering processing and image reconstruction processing, and use the die wear level output by the model as the prediction result.

8. The intelligent motor rotor hot forging high-pressure die wear prediction method according to claim 7, characterized in that: In step S1 , the image acquisition uses the mold wear level as a data label.

9. An intelligent motor rotor hot forging high-pressure die wear prediction system, used to implement the intelligent motor rotor hot forging high-pressure die wear prediction method according to any one of claims 1 to 8, characterized in that: It includes image acquisition module, image preprocessing module, filtering processing module, image reconstruction module, wear prediction model establishment module and motor rotor hot forging high-pressure die wear prediction module; The image acquisition module acquires a surface image of the motor rotor hot forging high-pressure die and marks the wear level of the die; The image preprocessing module preprocesses the image using gray value mapping and adaptive contrast enhancement; The filtering processing module performs filtering processing on the pre-processed image through dual-domain weighted filtering; The image reconstruction module reconstructs the low-frequency components through soft threshold processing based on wavelet decomposition and high-frequency threshold re-estimation; The wear prediction model establishment module establishes a wear prediction model based on the pre-branch, the image branch and the signal branch; The motor rotor hot forging high-pressure die wear prediction module predicts die wear based on real-time collected surface images of the motor rotor hot forging high-pressure die.

Images