Identification apparatus and method for contact fatigue failure characteristic vibration signal

Abstract

The present invention belongs to the technical field of contact fatigue testing. Disclosed are an identification apparatus and method for a contact fatigue failure characteristic vibration signal. The identification apparatus comprises a contact fatigue tester, wherein a microscopic observation module is fixed above an observation port of the contact fatigue tester, and the microscopic observation module comprises a stepper electric motor, a vertical guide rail, and a CCD camera provided with a microscope lens, which are sequentially connected. The method comprises: establishing a mapping relationship between spalling pit density characteristic information and a vibration signal by means of an LSTM algorithm; and on the basis of a vibration signal characteristic and the established mapping relationship, predetermining a contact fatigue failure. According to the identification apparatus and method for a contact fatigue failure characteristic vibration signal in the present invention, a specimen does not need to be repeatedly disassembled, in-situ measurement is achieved, errors caused by repeated installation are reduced, and the test efficiency is improved; and a mapping relationship between a spalling pit density and a vibration signal is established, and on the basis of the mapping relationship, a contact fatigue failure can be predetermined.

Description

A device and method for identifying characteristic vibration signals of contact fatigue failure Technical Field

[0001] The present invention relates to the technical field of contact fatigue experiments, and in particular to a device and method for identifying contact fatigue failure characteristic vibration signals. Background Art

[0002] Key components are the core of high-end equipment, subject to cyclic loads. Fatigue failure is the primary mode of failure, with severe consequences. Core transmission parts such as gears and bearings are typical key components, and their failure mode is contact fatigue, manifesting as surface spalling. Different vibration signals of transmission components correspond to different spalling pit densities on the component surface, and different spalling pit densities correspond to different contact fatigue failure mechanisms. By tracking the evolution of spalling pits and conducting mechanistic research, it is ultimately possible to guide improvements in the processing and treatment of key transmission components.

[0003] Existing research on the mechanism of contact fatigue involves installing a contact fatigue specimen on a contact fatigue testing machine, setting the loading force on the specimen, and detecting the vibration signal during operation through a sensor. When the vibration signal exceeds a threshold, the specimen is determined to have fatigue failure. The specimen is then disassembled and the surface peeling pits are measured offline.

[0004] The vibration signals detected by the contact fatigue tester can reveal the experimental state of the specimen, but they cannot reflect the density of the contact fatigue spalling pits. To determine the density evolution of the spalling pits, the specimen must be disassembled and placed on a surface topography instrument for offline testing. After testing, it must be reinstalled on the contact fatigue tester to continue the experiment. This will cause the specimen's rotation center to change due to reassembly, and the specimen's rotation center cannot be consistent throughout the experiment, causing experimental deviations. Repeated disassembly and assembly also affects test efficiency.

[0005] On the other hand, the existing technology cannot reflect the evolution law of spalling pit density in contact fatigue experiments of different durations. At the same time, it has not been able to establish a correlation between spalling pit density and vibration signals, and it is even impossible to predict contact fatigue failure.

[0006] Summary of the Invention

[0007] The purpose of the present invention is to provide a contact fatigue failure characteristic vibration signal identification device and method, which solves the installation error caused by repeated disassembly and assembly of the test piece in the existing contact fatigue testing machine, which makes the rotation center of the test piece inconsistent during the entire test process and causes experimental errors.

[0008] To achieve the above-mentioned objectives, the present invention provides a device and method for identifying characteristic vibration signals of contact fatigue failure. The device includes a contact fatigue testing machine, which is characterized in that a microscopic observation module is fixed above the observation port of the contact fatigue testing machine, and the microscopic observation module includes a stepper motor, a vertical guide rail and a CCD camera with a microscope lens installed, which are connected in sequence.

[0009] A method for identifying characteristic vibration signals of contact fatigue failure is based on the above-mentioned device for identifying characteristic vibration signals of contact fatigue failure, and the specific steps are as follows:

[0010] S1. After the contact fatigue testing machine stops, open the observation port and the microscopic observation module, adjust the microscopic observation module to perform in-situ measurement, and obtain a large-field microscopic morphology image of the specimen surface;

[0011] S2. The contact fatigue testing machine executes step S1 every hour until the test piece fails due to fatigue, and collects the microscopic topography image obtained by in-situ measurement and the idler vibration time series signal data within 5 minutes before the contact fatigue testing machine stops;

[0012] S3, calculating the number and density of spalling pits in each microscopic topography image obtained in step S2;

[0013] S4. Using the idler vibration time series signal collected in step S2 as input data for model training, and the spalling pit density calculated from the microscopic topography image corresponding to the vibration time series signal calculated in step S3 as a label for model training, a deep learning dataset is constructed;

[0014] S5. Using the deep learning dataset obtained in step S4, construct an LSTM Regression model that establishes an effective mapping relationship between the vibration time series signal features and the spalling pit density feature information;

[0015] S6. Apply the LSTM Regression model obtained in step S5 to automatically identify the vibration time series signal characteristics and predict the corresponding spalling pit density.

[0016] Preferably, the adjusting of the microscopic observation module to perform in-situ measurement in step S1 specifically includes: stopping the idler wheel of the contact fatigue testing machine from rotating and separating from the specimen, stopping the input of lubricating fluid, opening the observation port of the contact fatigue testing machine, sliding the CCD camera along the vertical guide rail by a stepper motor, allowing the microscope lens to penetrate into the interior of the contact fatigue testing machine from the observation port, and turning on the coaxial light source; using an autofocus algorithm to focus the microscope lens on the specimen surface; and using a field of view stitching algorithm to obtain a large-field microscopic morphology image of the specimen surface.

[0017] Preferably, the method for calculating the number and density of spalling pits in step S3 comprises the following steps:

[0018] S3-1. Use the Nnet++ semantic segmentation neural network to segment and locate the color information of the spalling pits in the collected microscopic images, and calculate the number of spalling pits in a single microscopic image;

[0019] S3-2. Calculate the actual length of each pixel value based on the magnification of the microscope and the resolution of the measured microscopic image, and map the actual length of the pixel value to each pixel point of the microscopic image to calculate the actual area of ​​the microscopic image;

[0020] S3-3. Use the following formula to calculate the spalling pit density:

[0021] Where ρ is the spalling pit density, n is the number of spalling pits in a single microscopic topography image, and v is the actual area of ​​the microscopic topography image.

[0022] Preferably, 75% of the data in the deep learning dataset in step S4 is used as a training dataset, and the remaining 25% is used as a validation dataset.

[0023] Preferably, the method for constructing the LSTM Regression model in step S5 includes the following steps:

[0024] S5-1. Connect the output layer of the LSTM long short-term memory neural network to the FCNN fully connected neural network, use the LSTM long short-term memory neural network to extract features from the vibration time series signal, and use the FCNN fully connected neural network to learn the potential mapping relationship between the vibration time series signal features and the spalling pit density feature information, and build an LSTM regression model.

[0025] S5-2. Use the training dataset to train the LSTM Regression model, allowing it to gradually learn the relationship between vibration signal features and spalling pit density features. Then, evaluate the LSTM Regression model on the validation dataset.

[0026] S5-3. Repeat step S5-2 until an LSTM Regression model is obtained that establishes an effective mapping relationship between the vibration signal features and the spalling pit density feature information.

[0027] Therefore, the present invention adopts the above-mentioned contact fatigue failure characteristic vibration signal identification device and method, which has the following technical effects:

[0028] (1) In-situ measurement of the micromorphology of the specimen surface to obtain the spalling pit density, avoiding the installation error caused by repeated disassembly and assembly of the specimen in the existing technology, and the experimental error caused by inconsistent rotation center of the specimen during the entire test process. At the same time, in-situ measurement does not require repeated disassembly and assembly of the specimen, thereby improving test efficiency;

[0029] (2) Based on the advantage of in-situ measurement, the density of contact fatigue spalling pits under different time periods is detected. Through the LSTM algorithm, the correlation between the vibration signal and the spalling pit density on the specimen surface is established. Finally, the spalling pit density can be predicted by the vibration signal, providing an efficient test tool for the study of contact fatigue failure mechanism.

[0030] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and those skilled in the art can, without inventive effort, derive other implementation drawings based on the provided drawings.

[0032] FIG1 is a schematic structural diagram of a device for identifying characteristic vibration signals of contact fatigue failure according to an embodiment of the present invention;

[0033] FIG2 is a flow chart of a method for identifying characteristic vibration signals of contact fatigue failure according to a second embodiment;

[0034] FIG3 is a microscopic topography image obtained by in-situ measurement in Example 2, wherein part (a) is a microscopic topography image selected on the first day, part (b) is a microscopic topography image selected on the second day, and part (c) is a microscopic topography image selected on the third day;

[0035] FIG4 is the vibration time series signal data collected on the third day of Example 2;

[0036] FIG5 is an effect diagram of the spalling pit region segmented by the Nnet++ semantic segmentation neural network in Example 2, wherein part (a) is an effect diagram of the spalling pit region segmented by the microscopic topography image selected on the first day; part (b) is an effect diagram of the spalling pit region segmented by the microscopic topography image selected on the first day; and part (c) is an effect diagram of the spalling pit region segmented by the microscopic topography image selected on the first day;

[0037] FIG6 is a schematic diagram of the LSTM Regression model structure of Example 2.

[0038] Reference numerals: 1. stepping motor; 2. vertical guide rail; 3. CCD camera; 4. microscope lens. DETAILED DESCRIPTION

[0039] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0040] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0041] Example 1

[0042] As shown in FIG1 , a device for identifying characteristic vibration signals of contact fatigue failure includes adding a microscopic observation module to the observation port of an existing contact fatigue testing machine.

[0043] The microscopic observation module includes a stepping motor 1, a vertical guide rail 2, a CCD camera 3 and a microscope lens 4 which are connected in sequence. The microscope lens 4 is in deep contact with the interior of the fatigue testing machine through an observation port.

[0044] Example 2

[0045] As shown in FIG2 , a method for identifying characteristic vibration signals of contact fatigue failure is performed based on a device for identifying characteristic vibration signals of contact fatigue failure in Example 1. The specific steps include:

[0046] S1. The specimen used in this example is a cylindrical specimen with a length of 110 mm and a diameter of 10 mm. After the idler pulley of the contact fatigue testing machine stops rotating and separates from the specimen, the lubricant supply is discontinued. The top observation port of the contact fatigue testing machine is opened, CCD camera 3 is lowered, and the coaxial light source is turned on to prepare for in-situ surface micromorphology measurement and spalling pit density extraction.

[0047] S2. Use an autofocus algorithm to make the microscope lens 4 focus on the surface of the test piece.

[0048] S3. Use the field of view stitching algorithm to obtain a large-field microscopic morphology image of the specimen surface.

[0049] S4. The contact fatigue testing machine executes steps S1-S3 every hour of operation to perform regular spalling pit density detection until the specimen fails due to fatigue.

[0050] The specific test time is: 8:00 to 18:00 from July 7 to July 9, 2023.

[0051] The microscopic morphology images obtained by in-situ measurement and the idler vibration time series signal data within 5 minutes before the contact fatigue testing machine stopped were collected.

[0052] Multiple microtopography images were collected, so one in-situ measurement image was selected for each of the first, second, and third days of testing, as shown in Figure 3. Part (a) is the microtopography image selected on the first day, part (b) is the microtopography image selected on the second day, and part (c) is the microtopography image selected on the third day. The vibration time series signal data corresponding to the obtained microtopography images is shown in Table 1. Because the amplitude signal contains 900 data points detected within 5 minutes, it cannot be fully displayed in the table. Therefore, only a portion of the data is shown in each group. Group (a) in Table 1 corresponds to part (a) in Figure 3, group (b) corresponds to part (b) in Figure 3, and group (c) corresponds to part (c) in Figure 3. The complete vibration time series signal data collected on the third day is shown in Figure 4.

[0053] S5. Apply the Nnet++ semantic segmentation neural network to segment and locate the color information of the spalling pits in the collected microtopography images, and calculate the number of spalling pits in each microtopography image. The results of the segmented spalling pit area are shown in Figure 5, where (a) is the result of the segmentation of the spalling pit area in the microtopography image selected on the first day; (b) is the result of the segmentation of the spalling pit area in the microtopography image selected on the first day; and (c) is the result of the segmentation of the spalling pit area in the microtopography image selected on the first day.

[0054] The actual length of each pixel value is calculated based on the magnification of the microscope and the measured image resolution, and then the actual length of the pixel is mapped to each pixel point in the image to calculate the actual area of ​​the entire image.

[0055] The spall pit density was calculated using the following formula:

[0056] Where ρ is the spalling pit density, n is the number of spalling pits in a single microscopic topography image, and v is the actual area of ​​the microscopic topography image.

[0057] The calculation results are as follows: the spalling pit density of group (a) is 32.80%, the spalling pit density of group (b) is 33.14%, and the spalling pit density of group (c) is 29.68%.

[0058] S6. Use the vibration time series signal with a sequence length of 900 collected in S4 as the input data for model training, and the spalling pit density calculated from the microscopic morphology image corresponding to the vibration time series signal as the label for model training to construct a deep learning dataset.

[0059] To ensure the model has sufficient data to learn during training and an independent dataset to validate its performance, the entire dataset is randomly divided, with 75% of the data used as the training dataset and the remaining 25% as the validation dataset. This division of the training and validation datasets is intended to ensure the effectiveness and generalization ability of the model, providing a foundation for subsequent model training and evaluation.

[0060] S7. Construct an LSTM Regression model to process the association between vibration time series signals and spalling pit density feature information.

[0061] The LSTM Regression model consists of two main parts: the LSTM long short-term memory neural network and the FCNN fully connected neural network.

[0062] The features of vibration time series signals are extracted through LSTM long short-term memory neural network. LSTM is a recurrent neural network with memory ability, which can effectively capture the long-term dependencies in vibration time series data.

[0063] The potential mapping relationship between vibration signal features and spalling pit density characteristics is learned through the FCNN fully connected neural network. A fully connected neural network is a deep learning model that can learn complex nonlinear mapping relationships between different features.

[0064] The LSTM regression model is trained using the training dataset, gradually learning the relationship between vibration signal features and spalling pit density. The model is then evaluated on the validation dataset to verify the effectiveness of the training. This process is iterative, requiring multiple training and validation cycles to continuously improve model performance. The best model is saved in each iteration, ultimately resulting in an LSTM regression model that effectively maps vibration signal features to spalling pit density information. Figure 6 shows a schematic diagram of the LSTM regression model structure.

[0065] S8. Using a data-driven approach, we apply an LSTM regression model to establish a mapping between spalling pit density characteristics and vibration signal features. The trained model can then be used to quickly predict the corresponding spalling pit density based on the vibration signal characteristics, allowing us to predict contact fatigue failure based on these characteristics.

[0066] The vibration time series signals obtained in steps S1-S4 were applied as input data to the LSTM Regression model. The spalling pit density predicted by the LSTM Regression model was: 32.27% for group (a), 32.95% for group (b), and 29.45% for group (c), which were relatively close to the actual spalling pit density.

[0067] Table 1 Part of the vibration time series signal data collected in Example 1

[0068] Therefore, the present invention adopts the above-mentioned contact fatigue failure characteristic vibration signal identification device and method to measure the microscopic morphology of the specimen surface on-site to obtain the spalling pit density, avoiding the installation error caused by repeated disassembly and assembly of the specimen in the prior art, and the experimental error caused by inconsistent rotation center of the specimen during the entire test process. At the same time, on-site measurement does not require repeated disassembly and assembly of the specimen, thereby improving the test efficiency; based on the advantages of on-site measurement, the detection of contact fatigue spalling pit density under different time lengths is realized, and through the LSTM algorithm, the correlation between the vibration signal and the spalling pit density on the specimen surface is established. Finally, the spalling pit density can be predicted by the vibration signal, providing an efficient test tool for the study of contact fatigue failure mechanism.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A vibration signal recognition device for contact fatigue failure characteristics, including a contact fatigue testing machine, characterized in that: A microscopic observation module is fixed above the observation port of the contact fatigue testing machine. The microscopic observation module includes a stepping motor, a vertical guide rail, and a CCD camera equipped with a microscopic lens, which are connected in sequence.

2. A method for identifying vibration signals of contact fatigue failure characteristics, characterized in that, Based on the contact fatigue failure characteristic vibration signal recognition device described in Claim 1, the specific steps are as follows: S1. After the contact fatigue testing machine stops rotating, open the observation port and the microscopic observation module, adjust the microscopic observation module for in-situ measurement, and obtain a large-field microscopic morphology image of the specimen surface. S2. Every hour the contact fatigue testing machine runs, execute Step S1 until the specimen fails due to fatigue, and collect the microscopic morphology images obtained from the in-situ measurement and the idler vibration time-series signal data within the last 5 minutes before the contact fatigue testing machine stops rotating. S3. Calculate the number and density of spalling pits in each microscopic morphology image obtained in Step S2. S4. Use the idler vibration time-series signal collected in Step S2 as the input data for model training, and the spalling pit density calculated from the microscopic morphology image corresponding to the vibration time-series signal calculated in Step S3 as the label for model training to construct a deep learning data set. S5. Use the deep learning data set obtained in Step S4 to construct an LSTM Regression regression model that establishes an effective mapping relationship between the vibration time-series signal features and the spalling pit density feature information. S6. Apply the LSTM Regression regression model obtained in Step S5 to automatically identify the vibration time-series signal features and predict the corresponding spalling pit density.

3. The vibration signal recognition method for contact fatigue failure characteristics according to claim 2, wherein The specific adjustment of the microscopic observation module for in-situ measurement in Step S1 specifically includes: After the idler of the contact fatigue testing machine stops rotating and is separated from the specimen, stop inputting the lubricating fluid, open the observation port of the contact fatigue testing machine, make the CCD camera slide along the vertical guide rail through the stepping motor, make the microscopic lens penetrate into the contact fatigue testing machine from the observation port, and turn on the coaxial light source; use the autofocus algorithm to focus the microscopic lens on the specimen surface; use the field-of-view stitching algorithm to obtain a large-field microscopic morphology image of the specimen surface.

4. The vibration signal recognition device and method for contact fatigue failure characteristics according to claim 2, characterized in that The specific steps for calculating the number and density of spalling pits in Step S3 are as follows: S3-1. Apply the Nnet++ semantic segmentation neural network to segment and locate the color information of the spalling pits in the collected microscopic morphology image, and calculate the number of spalling pits in a single microscopic morphology image. S3-2. Calculate the actual length of each pixel value according to the magnification of the microscope and the resolution of the measured microscopic morphology image, and map the actual length of the pixel value to each pixel point of the microscopic morphology image to calculate the actual area of the microscopic morphology image. S3-3. Calculate the spalling pit density using the following formula: Where ρ is the spalling pit density, n is the number of spalling pits in a single microscopic morphology image, and v is the actual area of the microscopic morphology image.

5. The vibration signal recognition device and method for contact fatigue failure characteristics according to claim 1, characterized in that: 75% of the data in the deep learning data set in Step S4 is used as the training data set, and the remaining 25% is used as the validation data set.

6. The vibration signal recognition device and method for contact fatigue failure characteristics according to claim 5, characterized in that The method for constructing the LSTM Regression regression model in Step S5 includes the following steps: S5-1. Connect the output layer of the LSTM (Long Short-Term Memory) neural network to the FCNN (Fully Connected Neural Network). Use the LSTM neural network to extract features from the vibration time series signal, and use the FCNN to learn the potential mapping relationship between the vibration time series signal features and the spalling pit density feature information, and construct an LSTM Regression model; S5-2. Use the training data set to train the LSTM Regression model, enabling the LSTM Regression model to gradually learn the relationship between the vibration signal features and the spalling pit density features, and then evaluate the LSTM Regression model on the validation data set; S5-3. Repeat step S5-2 until an LSTM Regression model that establishes an effective mapping relationship between the vibration signal features and the spalling pit density feature information is obtained.

Images