Deep learning algorithm-based stamping equipment key and heavy part production quality prediction method
By constructing a deep learning algorithm model based on CNN-BiLSTM, the problem of low efficiency in production quality inspection of traditional stamping equipment is solved, efficient quality prediction is achieved, yield and production efficiency are improved, and scrap rate is reduced.
Patent Information
- Application Number
- CN202510379859.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-28
- Publication Date
- 2025-08-15
AI Technical Summary
Traditional stamping equipment production quality inspection methods are inefficient, have poor real-time performance, high cost and serious material waste, and cannot effectively predict production quality, resulting in high rework and scrap rate.
Using deep learning algorithm based on CNN-BiLSTM, a production quality prediction model for stamping equipment is constructed, and through data preprocessing, model training and feature extraction, accurate prediction of the production quality of stamping equipment is achieved.
It improves the yield rate of stamping parts, reduces the scrap rate, optimizes the production process, reduces material waste, and enhances market competitiveness.
Smart Images
Figure CN120494143A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of quality inspection of stamping products, and in particular to a method for predicting the production quality of key parts of stamping equipment based on a deep learning algorithm. Background Art
[0002] As the manufacturing industry evolves, the requirements for the production quality of critical components in stamping equipment continue to increase. High-quality critical components not only ensure the normal operation of the machine and extend the service life of critical components during use, but also economically reduce the additional expenses of repair or rebuilding. However, the traditional quality inspection process, which first produces the actual components and then tests their appearance and performance, still wastes a lot of money. Therefore, in the manufacturing industry, predicting the production quality of critical components is a trend of the times.
[0003] With the rapid development of digitalization and intelligentization, the production quality inspection of critical parts has shifted from actual measurement to prediction, with extremely high requirements for prediction accuracy. Traditional inspection methods suffer from low efficiency, poor real-time performance, large scale, high cost, and material waste. They cannot avoid rework and scrap caused by quality issues, which seriously restricts production efficiency. Therefore, the production quality prediction of critical parts of stamping equipment based on deep learning algorithms is needed to improve the yield and consistency of critical parts, reduce the scrap rate, and provide high-quality stamping parts for subsequent processing and assembly, thereby ensuring the quality and performance of the entire product. Summary of the Invention
[0004] The technical problem to be solved by the present invention is to provide a production quality prediction method for key parts of stamping equipment based on a deep learning algorithm, construct an algorithm model based on CNN-BiLSTM (Convolutional Neural Networks-Bidirectional Long Short Term Memory, convolutional neural network-bidirectional long short-term memory network), realize data prediction, improve the yield of stamping parts, reduce the scrap rate, and improve the quality and performance of the entire production structure.
[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0006] The method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm includes the following steps:
[0007] S1. Obtain stamping equipment parameters and production product data from the stamping production workshop and pre-process the data.
[0008] S2. Build a CNN-BiLSTM model and train the model to obtain a trained CNN-BiLSTM model.
[0009] S3. Input the preprocessed data in step S1 into the trained CNN-BiLSTM model to obtain quality prediction data and complete the prediction of the production quality of key parts of the stamping equipment.
[0010] Furthermore, in step S1, direct observation, optical measurement, and laser scanning are used to obtain surface quality data for daily product quality inspection, precision manufacturing, and precision parts inspection under the conditions of rapid preliminary inspection, medium precision requirements, and high-precision three-dimensional data acquisition, respectively.
[0011] The mechanical properties data of metal materials are obtained by using tensile test, impact test and hardness test methods under static load axial tension, dynamic load and rapid evaluation respectively.
[0012] Direct measurement, optical measurement, and laser measurement methods are used to obtain specific dimensional data for daily product quality inspection, precision manufacturing, and precision component inspection under the conditions of rapid preliminary inspection, medium precision requirements, and high-precision three-dimensional data acquisition, respectively.
[0013] Furthermore, surface quality data includes surface roughness, surface defects, and surface cleanliness; mechanical property data includes tensile strength, elongation, impact toughness, and hardness; specific dimensional data includes length, width, height, diameter, wall thickness, and fitting dimensions.
[0014] Furthermore, in step S1, data preprocessing includes the following:
[0015] Data with a difference of more than three times the standard deviation from the data mean is defined as abnormal data. After recording abnormal data and missing data as error data, the mode is used to fill in the data; then the data is cleaned, normalized, and serialized.
[0016] Furthermore, in step S2, the CNN-BiLSTM model includes a CNN layer, a BiLSTM layer, and a fully connected layer connected in sequence.
[0017] Among them, the CNN layer includes convolution layer, activation function and pooling layer, and the BiLSTM layer includes forward LSTM and backward LSTM.
[0018] The preprocessed data in step S1 is input into the CNN-BiLSTM model to obtain quality prediction data. This data is compared with the true label of the historical data. After calculating the loss, back propagation is performed iteratively until the tolerance value is less than 0.5%, and the trained CNN-BiLSTM model is obtained.
[0019] Furthermore, the quality prediction data obtained includes the following:
[0020] The preprocessed data in step S1 is input into the CNN-BiLSTM model. After being processed by the convolutional layer in the CNN, the local features of the equipment vibration waveform and pressure time series are extracted, and a weighted ratio is obtained. A nonlinear activation function is used to obtain a high-dimensional feature map with spatial perception. This feature map is passed through the pooling layer for feature fusion processing to obtain multi-scale features of the equipment wear status.
[0021] The multi-scale features are transmitted to the BiLSTM layer. After the forward LSTM is used to capture the device performance attenuation trend, long-term device and production data features are obtained. After the backward LSTM is used to identify sudden abnormal signals, instantaneous abnormal features are obtained. The gated attention mechanism is used to dynamically fuse the long-term device and production data features with the instantaneous abnormal features to obtain the spatiotemporal feature vector associated with device parameters and product quality.
[0022] The spatiotemporal feature vector is transmitted to the fully connected layer, and after feature splicing, dimensionality reduction, regularization, and nonlinear mapping processing, the predicted equipment failure type probability and product quality information are obtained.
[0023] Furthermore, the present invention also proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements the steps of the method for predicting the production quality of key parts of stamping equipment based on deep learning algorithms.
[0024] Furthermore, the present invention also proposes a computer-readable storage medium, which stores a computer program. When the computer program is run by a processor, it executes the method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm.
[0025] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:
[0026] 1. The present invention targets the data on the production quality of key parts of stamping equipment, combines it with the technical field of deep learning algorithms, collects a large amount of historical data, and covers a very broad range of fields, providing a huge data structure for the prediction of the production quality of key parts of stamping equipment, which is beneficial to the model training process.
[0027] 2. Based on convolutional neural networks and bidirectional long short-term memory networks, this paper establishes a complete CNN-BiLSTM model, which provides a specific prediction process for the production quality prediction of key parts of stamping equipment, and provides a more rigorous prediction process for model training and evaluation by extracting features and capturing bidirectional information.
[0028] 3. The present invention is beneficial for improving quality management efficiency, optimizing production processes, reducing the risk of equipment failure, reducing material waste, enhancing market competitiveness, and reducing scrap rates. It is also applicable to a variety of manufacturing production processes and increases economic benefits.
[0029] 4. This invention breaks away from the constraints of traditional quality inspection, uses deep learning algorithms, reflects the development of digital intelligence in the manufacturing industry, and is suitable for physical workshops and large-scale stamping production lines that use digital twin systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 It is an overall implementation flow chart of the present invention. DETAILED DESCRIPTION
[0031] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.
[0032] To achieve the above objectives, the present invention proposes a method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm, such as Figure 1 The specific steps are as follows:
[0033] S1. Obtain stamping equipment parameters and product data from the stamping production workshop and pre-process the data to provide valid data for subsequent input into the model, reducing the limitation of low model operation efficiency caused by excessive data volume; the specific content is:
[0034] Direct observation, optical measurement, and laser scanning are used to obtain surface quality data for routine product quality inspection, precision manufacturing, and precision component inspection, respectively, under the conditions of rapid preliminary inspection, medium-precision requirements, and high-precision 3D data acquisition. Surface quality data includes surface roughness, surface defects, and surface cleanliness.
[0035] The mechanical properties of metal materials are obtained by using tensile testing, impact testing, and hardness testing methods under static axial tension, dynamic loading, and rapid evaluation. Mechanical properties include tensile strength, elongation, impact toughness, and hardness.
[0036] Direct measurement, optical measurement, and laser measurement are used to obtain specific dimensional data for daily product quality inspection, precision manufacturing, and precision component inspection, respectively, for rapid preliminary inspection, medium-precision requirements, and high-precision 3D data acquisition. Specific dimensional data includes length, width, height, diameter, wall thickness, and fit dimensions.
[0037] Data with a difference of more than three times the standard deviation from the data mean is defined as abnormal data. After recording abnormal data and missing data as error data, the mode is used to fill in the data; then the data is cleaned, normalized, and serialized.
[0038] S2. Build a CNN-BiLSTM model and train it to obtain a trained CNN-BiLSTM model. The specific contents are as follows:
[0039] The CNN-BiLSTM model consists of sequentially connected CNN layers, BiLSTM layers, and fully connected layers.
[0040] Among them, the CNN layer includes convolution layer, activation function and pooling layer, and the BiLSTM layer includes forward LSTM and backward LSTM.
[0041] The preprocessed data in step S1 is input into the CNN-BiLSTM model to obtain quality prediction data. This data is compared with the true label of the historical data. After calculating the loss, back propagation is performed iteratively until the tolerance value is less than 0.5%, and the trained CNN-BiLSTM model is obtained.
[0042] S3. Input the pre-processed data in step S1 into the trained CNN-BiLSTM model to obtain quality prediction data and complete the prediction of the production quality of key parts of the stamping equipment; the specific content is:
[0043] The preprocessed data in step S1 is input into the trained CNN-BiLSTM model. After being processed by the convolutional layer in the CNN, the local features of the equipment vibration waveform and pressure time series are extracted, and a weighted ratio is obtained. A nonlinear activation function is used to adapt to the complexity of large-scale data, reduce the data dimension and the amount of calculation, and obtain a high-dimensional feature map with spatial perception. This feature map is passed through the pooling layer and multi-receptive field feature fusion processing is performed to retain important features and improve the robustness of the features, thereby obtaining multi-scale features including the short-term and long-term wear status of the equipment.
[0044] The multi-scale features are transmitted to the BiLSTM layer. After the forward LSTM is used to capture the device performance attenuation trend, long-term device and production data features are obtained. After the backward LSTM is used to identify sudden abnormal signals, instantaneous abnormal features are obtained. The gated attention mechanism is used to dynamically fuse the long-term device and production data features with the instantaneous abnormal features to obtain the spatiotemporal feature vector associated with device parameters and product quality.
[0045] The spatiotemporal feature vector is transmitted to the fully connected layer, and after feature splicing, dimensionality reduction, regularization, and nonlinear mapping processing, the predicted equipment failure type probability and product quality information are obtained.
[0046] S4. Upload all the data during the operation to the cloud platform, and integrate computing, storage and other functions through the network platform to facilitate operators to query the required information and training data to achieve the desired purpose.
[0047] In this embodiment, taking the production of a vehicle shell as an example, the collected data is shown in Table 1.
[0048] Table 1 Data collected based on vehicle shell production
[0049] Serial number Label type 1 surface roughness float 2 Crack length float 3 Scratch length float 4 Surface cleanliness float 5 tensile strength float 6 Yield strength float 7 Elongation float 8 hardness float 9 length float 10 width float 11 high float 12 diameter float 13 wall thickness float 14 Matching size float … … …
[0050] As shown in Table 1, surface quality data include surface roughness, surface cleanliness, and surface defects such as cracks, scratches, rust, and inclusions; mechanical property data include tensile strength, yield strength, elongation, impact toughness, and hardness; specific dimensional data include length, width, height, diameter, wall thickness, and fitting dimensions.
[0051] By recording data from the entire process of vehicle shell production, the actual production data of different production times and different stamping equipment are the pillars of production quality prediction. The prediction of production quality can only be obtained through multiple readings, processing, and cyclic training based on these huge actual data.
[0052] In the convolution layer of the CNN layer, by processing the data, it can accurately determine whether there is crack data in the input vehicle shell surface data, and automatically retain the position and size features.
[0053] For example, if the time step data of the stamping process is input, such as pressure and temperature within 10ms, the forward LSTM can capture the characteristics of various data in the input data, and the backward LSTM reverses from the stamping end time. Through bidirectional feature fusion, the obtained data is fused with the corresponding data recorded at the same time by the forward LSTM, and finally the product prediction data after the vehicle shell is stamped is output.
[0054] An embodiment of the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable by the processor. It should be noted that when the processor executes the computer program, it corresponds to the specific steps of the method provided in the embodiment of the present invention and has the corresponding functional modules and beneficial effects of the method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present invention.
[0055] The present invention also provides a computer-readable storage medium storing a computer program. It should be noted that when executed by a processor, the computer program corresponds to the specific steps of the method provided in the present invention and has the corresponding functional modules and beneficial effects. For technical details not fully described in this embodiment, please refer to the method provided in the present invention.
[0056] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.
Claims
1. A method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm, characterized in that: include: S1. Obtain stamping equipment parameters and production product data from the stamping production workshop and pre-process the data; S2. Build a CNN-BiLSTM model and train the model to obtain a trained CNN-BiLSTM model; S3. Input the preprocessed data in step S1 into the trained CNN-BiLSTM model to obtain quality prediction data and complete the prediction of the production quality of key parts of the stamping equipment.
2. The method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm according to claim 1 is characterized in that: In step S1, surface quality data of daily product quality inspection, precision manufacturing, and precision parts inspection are obtained using direct observation, optical measurement, and laser scanning methods under the conditions of rapid preliminary inspection, medium precision requirements, and high-precision three-dimensional data acquisition, respectively; Use tensile test, impact test and hardness test methods to obtain mechanical properties data of metal materials under static load axial tension, dynamic load and rapid evaluation respectively; Direct measurement, optical measurement, and laser measurement methods are used to obtain specific dimensional data for daily product quality inspection, precision manufacturing, and precision component inspection under the conditions of rapid preliminary inspection, medium precision requirements, and high-precision three-dimensional data acquisition, respectively.
3. The method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm according to claim 2 is characterized in that: Surface quality data includes surface roughness, surface defects, and surface cleanliness; mechanical property data includes tensile strength, elongation, impact toughness, and hardness; specific dimension data includes length, width, height, diameter, wall thickness, and fitting dimensions.
4. The method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm according to claim 1 is characterized in that: In step S1, data preprocessing includes the following: Data with a difference of more than three standard deviations from the mean of the data were defined as abnormal data. After recording abnormal data and missing data as error data, mode filling was used; The data is then cleaned, normalized, and serialized.
5. The method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm according to claim 1 is characterized in that: In step S2, the CNN-BiLSTM model includes a CNN layer, a BiLSTM layer, and a fully connected layer connected in sequence; Among them, the CNN layer includes convolution layer, activation function and pooling layer, and the BiLSTM layer includes forward LSTM and backward LSTM; The preprocessed data in step S1 is input into the CNN-BiLSTM model to obtain quality prediction data. This data is compared with the true label of the historical data. After calculating the loss, back propagation is performed iteratively until the tolerance value is less than 0.5%, and the trained CNN-BiLSTM model is obtained.
6. The method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm according to claim 5 is characterized in that: The quality prediction data includes the following: The data preprocessed in step S1 is input into the CNN-BiLSTM model. After being processed by the convolutional layer in the CNN, the local features of the equipment vibration waveform and pressure time series are extracted and the weighted ratio is obtained. A nonlinear activation function is used to obtain a high-dimensional feature map with spatial perception. This feature map is passed through the pooling layer for feature fusion processing to obtain multi-scale features of the equipment wear status. The multi-scale features are transferred to the BiLSTM layer. After the forward LSTM captures the device performance attenuation trend, long-term device and production data features are obtained. After the backward LSTM identifies sudden abnormal signals, instantaneous abnormal features are obtained. The gated attention mechanism is used to dynamically fuse the long-term device and production data features with the instantaneous abnormal features to obtain the spatiotemporal feature vector associated with device parameters and product quality. The spatiotemporal feature vector is transmitted to the fully connected layer, and after feature splicing, dimensionality reduction, regularization, and nonlinear mapping processing, the predicted equipment failure type probability and product quality information are obtained.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method for predicting the production quality of key parts of stamping equipment based on deep learning algorithm as described in any one of claims 1 to 6 are implemented.
8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting the production quality of key parts of stamping equipment based on a deep learning algorithm according to any one of claims 1 to 6 is executed.