A VSR inspection method for turbocharger industry based on deep learning principle
Through deep learning-based methods, an automated VSR curve inspection system is established, which solves the problems of low efficiency and poor accuracy of traditional methods, and achieves efficient and accurate quality control to adapt to the needs of turbochargers under complex operating conditions.
Patent Information
- Application Number
- CN202411820677.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-11
- Publication Date
- 2025-04-01
AI Technical Summary
The VSR inspection method of the traditional turbocharger industry relies on manual observation and empirical judgment, resulting in low detection efficiency and poor accuracy, and is unable to adapt to the needs of turbochargers under complex operating conditions.
Using the VSR inspection method based on the principle of deep learning, a large number of qualified and unqualified VSR curve data are collected and marked, and convolutional neural network models are created, trained and evaluated to achieve automated curve inspection and quality control.
It improves the efficiency and accuracy of VSR curve inspection, can quickly process large amounts of data, realize real-time quality inspection, meet large-scale production needs, and reduce labor costs and training costs.
Smart Images
Figure CN119740096A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of turbocharging, and more specifically, to a VSR inspection method for the turbocharging industry based on deep learning principles. Background Art
[0002] During the manufacturing process, residual stress will be generated inside the turbocharger parts (such as impellers, shafts, housings, etc.) after forging, casting, welding, machining and other processes. These residual stresses will cause deformation and cracking of the parts during subsequent use, affecting the performance and service life of the turbocharger.
[0003] The VSR process releases residual stress by subjecting parts to vibration excitation and utilizing the microscopic plastic deformation characteristics of the material. During the vibration process, crystal defects such as dislocations inside the material move and rearrange, causing the residual stress to relax. Generally speaking, parameters such as vibration frequency, amplitude, and vibration time are adjusted according to factors such as the material, shape, size, and magnitude of residual stress of the parts. VSR helps to improve the fatigue resistance of parts. The presence of residual stress accelerates the initiation and propagation of fatigue cracks. By eliminating residual stress through VSR, the risk of fatigue cracks can be reduced, the service life of parts can be extended, and the turbocharger can be ensured to work reliably under complex operating conditions (such as high temperature, high-speed rotation, frequent start and stop, etc.).
[0004] Traditionally, the judgment of VSR test result curve in the turbocharger industry relies entirely on the inspector's visual observation and experience, which has problems such as low detection efficiency, poor accuracy, and insufficient adaptability to turbochargers under complex working conditions.
[0005] In order to solve the above problems, a technical solution is now provided. Summary of the invention
[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a VSR inspection method for the turbocharger industry based on the principle of deep learning to solve the problems raised in the above-mentioned background technology.
[0007] To achieve the above object, the present invention provides the following technical solutions: A VSR inspection method for the turbocharger industry based on deep learning principles includes the following steps: Step S100, collecting curve data from a large number of turbocharged VSR processing processes that have been verified as qualified, and marking the collected qualified curve samples; Step S200, collecting various types of VSR curve samples that lead to failure, and classifying and labeling the failed curve samples according to their failure reasons; Step S300, normalizing all collected curve data; Step S400, creating a convolutional neural network model and training the convolutional neural network model; Step S500, evaluating and applying the trained convolutional neural network model.
[0008] In a preferred embodiment, step S100 specifically includes the following contents: Install an acceleration sensor on the VSR equipment to collect the curves of vibration frequency, amplitude, and acceleration parameters changing over time; Strain gauges are attached to the surface of turbocharger parts and connected to strain gauges to record the changes in stress and strain during the VSR process and obtain stress-strain curves; Determine the sampling frequency and sampling time length of the data to ensure that all collected data have the same resolution and length in the time dimension; The collected data are integrated and the collected qualified curve samples are marked as qualified.
[0009] In a preferred embodiment, step S200 specifically includes the following contents: For turbochargers that have just completed VSR treatment but are found to be unqualified in preliminary inspection, curve data is immediately obtained from the data recording module of the VSR equipment; By changing the key parameters of the VSR equipment to simulate the situation that leads to failure, during the simulation process, record the parameter value and corresponding curve data of each adjustment, and establish the association between parameter adjustment and curve change; Develop a detailed classification of non-conformance reasons based on turbocharger quality standards and previous failure analysis experience; Conduct a detailed analysis on each unqualified curve sample, and based on the analysis results, mark each unqualified curve sample according to the predetermined classification standards, and record the category of unqualified cause of each sample.
[0010] In a preferred embodiment, step S300 specifically includes the following contents: The calculation formula is: in, represents the normalized data points, represents the original data points, Represents the minimum value of this feature in the data set, Indicates the maximum value of this feature in the data set; Normalize the data to Within the range.
[0012] In a preferred embodiment, step S400 specifically includes the following contents: Determine the parameters of the convolutional layer, including kernel size, stride, and number; Add pooling layers to reduce data dimensionality; Construct a fully connected layer and map the features extracted by the convolutional layer to the output layer. The number of nodes in the output layer is determined according to the classification category.
[0013] In a preferred embodiment, step S400 specifically further includes the following contents: Prepare a data set and divide it into a training set, a validation set, and a test set according to a certain ratio. The ratio is 70%-80% for training, 10%-15% for validation, and 10%-15% for testing. The model is trained multiple times using the training set. In each iteration, a batch of data is input into the model, the loss function value is calculated, and the parameters of the model are updated through the optimizer. After each training cycle, the validation set is used to evaluate the performance of the model and the hyperparameters of the model are adjusted based on the results of the validation set.
[0014] In a preferred embodiment, step S500 specifically includes the following contents: Select appropriate performance evaluation indicators based on the specific task type; Use the test set to perform a final evaluation on the trained model to obtain the performance indicators of the model on unseen data.
[0015] In a preferred embodiment, step S500 specifically further includes the following contents: The trained model is used to automatically inspect the VSR curve of mass-produced turbochargers. When new VSR curve data is input into the model, the model outputs its predicted category and corresponding probability value. The turbocharger quality is controlled based on the output of the model, and the causes of unqualified products are further analyzed. At the same time, new production data is continuously collected, and the model is regularly updated and optimized to adapt to changes in the production process.
[0016] The technical effects and advantages of the VSR inspection method for turbocharger industry based on deep learning principle of the present invention are as follows: 1. The trained model is used to automatically inspect the VSR curves of mass-produced turbochargers. Compared with the traditional manual inspection method, the model can quickly process a large amount of curve data and output the predicted category and the corresponding probability value. This not only greatly improves the inspection efficiency and saves labor costs, but also enables real-time and rapid quality inspection to meet the needs of large-scale production; 2. It can automatically inspect and judge the VSR curve morphology of mass production, greatly improve the accuracy, consistency and efficiency of the inspection results, and reduce the cost of training inspectors. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 This is a flow chart of a VSR inspection method for the turbocharger industry based on the deep learning principle of the present invention. DETAILED DESCRIPTION
[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention. Example
[0019] Figure 1 The present invention provides a VSR inspection method for the turbocharger industry based on the deep learning principle, which specifically includes the following steps: Step S100, collecting curve data from a large number of turbocharged VSR processing processes that have been verified as qualified, and marking the collected qualified curve samples; Step S200, collecting various types of VSR curve samples that lead to failure, and classifying and labeling the failed curve samples according to their failure reasons; Step S300, normalizing all collected curve data; Step S400, creating a convolutional neural network model and training the convolutional neural network model; Step S500, evaluating and applying the trained convolutional neural network model.
[0020] Step S100 specifically includes the following contents: Install an acceleration sensor on the VSR equipment to collect the curves of vibration frequency, amplitude, and acceleration parameters changing over time; Strain gauges are attached to the surface of turbocharger parts and connected to strain gauges to record the changes in stress and strain during the VSR process and obtain stress-strain curves; Determine the sampling frequency and sampling time length of the data to ensure that all collected data have the same resolution and length in the time dimension; The collected data are integrated and the collected qualified curve samples are marked as qualified.
[0021] It should be noted that by installing acceleration sensors on VSR equipment to collect the curves of vibration frequency, amplitude, and acceleration parameters over time, the vibration information in the VSR process can be fully captured. These vibration parameters are crucial for evaluating the effectiveness and stability of the VSR process. For example, the stability of the vibration frequency can reflect whether the working state of the VSR equipment is normal, and the changes in amplitude and acceleration are closely related to the elimination of internal stress in components.
[0022] At the same time, by using strain gauges and strain gauges to record stress-strain curves, the changes in stress inside parts can be directly obtained. This helps to understand how residual stress is gradually eliminated over time during the VSR process and whether the expected stress elimination effect has been achieved. This multi-parameter acquisition method can monitor the VSR process from different angles and provide richer data support for subsequent quality assessment.
[0023] Determine a unified data sampling frequency and sampling time length so that all collected data have the same resolution and length in the time dimension. This measure has many advantages. First, during the data analysis phase, a consistent data format facilitates data alignment and comparison. For example, when analyzing VSR curves of different batches or different parts, the same time resolution can ensure comparability between data points and avoid analysis errors caused by inconsistent data sampling.
[0024] Secondly, uniform data length helps to build a standardized data set. In machine learning or statistical analysis, the format and length of data usually need to be consistent, which makes it easy to input data into the model for training or analysis, improving the efficiency and accuracy of data processing.
[0025] Step S200 specifically includes the following contents: For turbochargers that have just completed VSR treatment but are found to be unqualified in preliminary inspection, curve data is immediately obtained from the data recording module of the VSR equipment; By changing the key parameters of the VSR equipment to simulate the situation that leads to failure, during the simulation process, record the parameter value and corresponding curve data of each adjustment, and establish the association between parameter adjustment and curve change; Develop a detailed classification of non-conformance reasons based on turbocharger quality standards and previous failure analysis experience; Conduct a detailed analysis on each unqualified curve sample, and based on the analysis results, mark each unqualified curve sample according to the predetermined classification standards, and record the category of unqualified cause of each sample.
[0026] It should be noted that for turbochargers that have just completed VSR treatment but failed the initial inspection, obtaining curve data immediately can ensure the authenticity and timeliness of the data. These data directly reflect the problems that occurred in the actual production process without the interference of additional factors, providing the most direct basis for the subsequent analysis of the reasons for failure. By analyzing these data, the links that may lead to failure can be quickly located, such as abnormal vibration of the VSR equipment during the treatment process, insufficient stress release and other problems.
[0027] Each unqualified curve sample is analyzed in detail and labeled according to the classification standards, and the category of unqualified reasons is recorded, so that each sample has clear "diagnostic information". In the subsequent quality improvement, process optimization and model training process, targeted measures can be taken according to this labeled information. For example, in model training, different types of unqualified samples can be learned as different categories, so that the model can better identify various unqualified patterns, thereby improving the accuracy and practicality of the model; in terms of process improvement, accurate labeling can help engineers quickly locate the process links that need to be adjusted and improve the efficiency of problem solving.
[0028] Step S300 specifically includes the following contents: The calculation formula is: in, represents the normalized data points, represents the original data points, Represents the minimum value of this feature in the data set, Indicates the maximum value of this feature in the data set; Normalize the data to Within the range.
[0030] It should be noted that normalized data helps improve the training efficiency and accuracy of the model. For many machine learning algorithms, the scale differences of data features will affect the convergence speed and performance of the model. If normalization is not performed, the large values of certain features may dominate optimization algorithms such as gradient descent, making it difficult to learn effective information of other features during model training. Normalized data can make the model training process more stable and efficient, allowing the model to better focus on the inherent patterns and laws of the data, thereby improving the generalization ability and prediction accuracy of the model.
[0031] Step S400 specifically includes the following contents: Determine the parameters of the convolutional layer, including kernel size, stride, and number; Add pooling layers to reduce data dimensionality; Construct a fully connected layer and map the features extracted by the convolutional layer to the output layer. The number of nodes in the output layer is determined according to the classification category.
[0032] It should be noted that the fully connected layer can integrate the various local features extracted by the convolution layer and classify the data as a whole. It converts the feature map output by the convolution layer into a form suitable for classification tasks, and maps these features to each category node of the output layer by learning the complex relationship between the features. For example, in the binary classification task of judging whether the turbocharger VSR curve is qualified or not, the fully connected layer can comprehensively judge whether the turbocharger meets the quality standards based on the previously extracted curve features.
[0033] Step S400 specifically includes the following contents: Prepare a data set and divide it into a training set, a validation set, and a test set according to a certain ratio. The ratio is 70% for training, 15% for validation, and 15% for testing. The model is trained multiple times using the training set. In each iteration, a batch of data is input into the model, the loss function value is calculated, and the parameters of the model are updated through the optimizer. After each training cycle, the validation set is used to evaluate the performance of the model and the hyperparameters of the model are adjusted based on the results of the validation set.
[0034] It should be noted that this division method helps the model learn the general rules of the data instead of just memorizing the samples in the training set. Through evaluation on the validation set and the test set, the model can better adapt to different data distributions and sample changes, thereby improving its generalization ability, so that the model can make accurate judgments when faced with new turbocharger VSR curve data that has not participated in the training.
[0035] At the same time, reasonable division of the data set also increases the stability of model performance evaluation. Different data divisions may cause some fluctuations in model evaluation indicators, but by following a certain division ratio range, this fluctuation can be reduced to a certain extent, making the evaluation results more reliable.
[0036] After each training cycle, the validation set is used to evaluate the model performance, which can promptly detect whether the model is overfitting. If the performance of the model on the validation set is no longer improved or even decreases, appropriate measures can be taken, such as reducing the learning rate, increasing the regularization term, or adjusting hyperparameters such as the network structure. This allows the model to maintain good generalization capabilities during the training process and avoids the model overfitting the training data and performing poorly in actual applications.
[0037] Step S500 specifically includes the following contents: Select appropriate performance evaluation indicators based on the specific task type; Use the test set to perform a final evaluation on the trained model to obtain the performance indicators of the model on unseen data.
[0038] Step S500 specifically includes the following contents: The trained model is used to automatically inspect the VSR curve of mass-produced turbochargers. When new VSR curve data is input into the model, the model outputs its predicted category and corresponding probability value. The turbocharger quality is controlled based on the output of the model, and the causes of unqualified products are further analyzed. At the same time, new production data is continuously collected, and the model is regularly updated and optimized to adapt to changes in the production process.
[0039] It should be noted that using the trained model to automatically inspect the VSR curve of mass-produced turbochargers can greatly improve inspection efficiency. Compared with traditional manual inspection methods, the model can quickly process a large amount of curve data and output the predicted category and corresponding probability value. This not only saves labor costs, but also enables real-time and rapid quality inspection to meet the needs of large-scale production. For example, on the production line, the model can instantly analyze the curve data of each turbocharger processed by VSR and quickly determine whether it is qualified or not.
[0040] Quality control is performed based on the output results of the model. Further analysis of the causes of unqualified products can help companies accurately locate problems in the production process. The probability value output by the model can also be used to make a preliminary judgment on the degree of unqualified products. For example, unqualified products with low probability values may have more serious problems and need to be paid special attention. This precise quality control method can effectively reduce the defective rate and improve the overall quality of the product.
[0041] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.
[0042] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or may be transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.
[0043] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0044] In the several embodiments provided in the present application, it should be understood that the disclosed methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.
[0045] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0046] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
[0047] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.
[0048] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0049] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A VSR inspection method for turbocharger industry based on deep learning principle, characterized in that: The steps include: Step S100, collecting curve data from a large number of turbocharged VSR processing processes that have been verified as qualified, and marking the collected qualified curve samples; Step S200, collecting various types of VSR curve samples that lead to failure, and classifying and labeling the failed curve samples according to their failure reasons; Step S300, normalizing all collected curve data; Step S400, creating a convolutional neural network model and training the convolutional neural network model; Step S500, evaluating and applying the trained convolutional neural network model.
2. The VSR inspection method for turbocharger industry based on deep learning principle according to claim 1 is characterized by: Step S100 specifically includes the following contents: Install an acceleration sensor on the VSR equipment to collect the curves of vibration frequency, amplitude, and acceleration parameters changing over time; Strain gauges are attached to the surface of turbocharger parts and connected to strain gauges to record the changes in stress and strain during the VSR process and obtain stress-strain curves; Determine the sampling frequency and sampling time length of the data to ensure that all collected data have the same resolution and length in the time dimension; The collected data are integrated and the collected qualified curve samples are marked as qualified.
3. The VSR inspection method for turbocharger industry based on deep learning principle according to claim 1 is characterized by: Step S200 specifically includes the following contents: For turbochargers that have just completed VSR treatment but are found to be unqualified in preliminary inspection, curve data is immediately obtained from the data recording module of the VSR equipment; By changing the key parameters of the VSR equipment to simulate the situation that leads to failure, during the simulation process, record the parameter value and corresponding curve data of each adjustment, and establish the association between parameter adjustment and curve change; Develop a detailed classification of non-conformance reasons based on turbocharger quality standards and previous failure analysis experience; Conduct a detailed analysis on each unqualified curve sample, and based on the analysis results, mark each unqualified curve sample according to the predetermined classification standards, and record the category of unqualified cause of each sample.
4. The VSR inspection method for turbocharger industry based on deep learning principle according to claim 1 is characterized by: Step S300 specifically includes the following contents: The calculation formula is: ； in, represents the normalized data points, represents the original data points, Represents the minimum value of this feature in the data set, Indicates the maximum value of this feature in the data set; Normalize the data to Within the range.
5. The VSR inspection method for turbocharger industry based on deep learning principle according to claim 1 is characterized by: Step S400 specifically includes the following contents: Determine the parameters of the convolutional layer, including kernel size, stride, and number; Add pooling layers to reduce data dimensionality; Construct a fully connected layer and map the features extracted by the convolutional layer to the output layer. The number of nodes in the output layer is determined according to the classification category.
6. A VSR inspection method for turbocharger industry based on deep learning principle according to claim 5, characterized in that: Step S400 specifically includes the following contents: Prepare a data set and divide it into a training set, a validation set, and a test set according to a certain ratio. The ratio is 70%-80% for training, 10%-15% for validation, and 10%-15% for testing. The model is trained multiple times using the training set. In each iteration, a batch of data is input into the model, the loss function value is calculated, and the parameters of the model are updated through the optimizer. After each training cycle, the validation set is used to evaluate the performance of the model and the hyperparameters of the model are adjusted based on the results of the validation set.
7. The VSR inspection method for turbocharger industry based on deep learning principle according to claim 6 is characterized by: Step S500 specifically includes the following contents: Select appropriate performance evaluation indicators based on the specific task type; Use the test set to perform a final evaluation on the trained model to obtain the performance indicators of the model on unseen data.
8. The VSR inspection method for turbocharger industry based on deep learning principle according to claim 7 is characterized by: Step S500 specifically includes the following contents: The trained model is used to automatically inspect the VSR curve of mass-produced turbochargers. When new VSR curve data is input into the model, the model outputs its predicted category and corresponding probability value. The turbocharger quality is controlled based on the output of the model, and the causes of unqualified products are further analyzed. At the same time, new production data is continuously collected, and the model is regularly updated and optimized to adapt to changes in the production process.