A working condition following gear wear monitoring and prediction method based on vibration signals

Abstract

The present application relates to a kind of working condition follow-up gear wear monitoring and prediction method based on vibration signal, belong to gear transmission technical field, solve the problem that different working conditions and different wear states of different positions of gear are not considered in the prior art to gear wear monitoring and prediction and the influence, including: S1: dynamic working condition division based on multi-source sensor fusion, establish working condition Map chart, real-time working condition Map chart update;S2: establish Archard wear correction model, calculate the wear of gear;S3: establish the nonlinear vibration model of gear considering the influence of wear and clearance, the vibration signal of gear system is solved;S4: establish CNN-LSTM hybrid wear diagnosis model, training;S5: real-time acquisition and processing obtain the vibration signal of gear system, input CNN-LSTM hybrid wear diagnosis model, obtain the current wear state probability distribution of gear, and the wear development trend prediction of gear.

Description

Technical Field

[0001] This invention relates to the field of gear transmission technology, and specifically to a method for monitoring and predicting gear wear based on vibration signals under operating conditions. Background Technology

[0002] With the increasing demands for power density and reliability in transmission systems from fields such as new energy vehicles, wind power, and aerospace, gear wear monitoring under service conditions faces severe challenges due to factors such as sudden load changes, speed fluctuations, and temperature variations. Visual inspection of gear wear upon unpacking is costly and time-consuming, highlighting the urgent need for a more efficient and accurate method for monitoring gear wear.

[0003] Traditional vibration signal-based monitoring methods (such as Fourier spectrum analysis) perform well under fixed operating conditions, but under varying operating conditions, the non-stationary and nonlinear characteristics of the signal, the strong time-varying nature of gear meshing frequency and load, cause traditional monitoring methods to fail. Specifically, this manifests as feature extraction failure, insufficient model generalization ability, and failure to consider the effects of multi-physics coupling.

[0004] The ISO 10816 standard uses the effective value (RMS) and peak factor of the vibration signal as wear assessment indicators. It decomposes the vibration signal into different frequency components using methods such as Fourier transform and wavelet transform, extracting key features (such as meshing frequency sidebands and energy entropy) for diagnosis. However, this method relies solely on a fixed diagnostic threshold, failing to consider the nonlinear modulation of the vibration signal by microscopic wear on the tooth surface. Furthermore, it cannot handle nonlinear vibration signal characteristics under varying operating conditions, leading to deviations in the extraction of relevant vibration signal features.

[0005] Gear damage recognition models based on self-learning neural networks (such as CNN and LSTM) can classify different wear levels by training classifiers using time-domain and frequency-domain features. However, this model relies on a large amount of labeled data, and the time synchronization and spatiotemporal alignment algorithms for multi-source data under varying operating conditions are complex and have poor real-time performance. Data distribution differences and threshold shifts also lead to poor generalization ability. Furthermore, the model does not establish a dynamic mapping relationship between wear and multi-physics fields, resulting in inaccurate diagnostic results.

[0006] Chinese invention patent application, publication number CN109871652A, entitled "A method for predicting gear wear based on dynamic meshing force," discloses a method for predicting gear wear by calculating the dynamic meshing force and dynamic load distribution coefficient of a gear pair using a gear rotor system dynamic model. However, this method does not consider the different wear states at different positions of the gears caused by complex influencing factors under different working conditions, making it difficult to achieve accurate gear wear prediction.

[0007] Chinese invention patent application, publication number CN117634241A, entitled "A method for predicting the dynamic wear of gear pairs considering the micro-geometry of tooth surfaces," discloses a method for calculating and predicting the wear of gear teeth by establishing a contact analysis model that considers the micro-geometry of tooth surfaces and a dynamic load calculation model for gear pairs based on the finite element method. However, this method still does not consider the different wear states at different locations of the gear caused by complex influencing factors under different working conditions, resulting in inaccurate prediction results.

[0008] Therefore, there is a need in this field for improved gear wear monitoring and prediction models to provide more accurate gear wear detection and prediction under complex operating conditions. Summary of the Invention

[0009] This invention patent proposes a method for monitoring and predicting gear wear based on vibration signals under working conditions. It considers the effects of lubrication and temperature and realizes dynamic wear prediction based on the mapping relationship between vibration signals and wear degree.

[0010] A method for monitoring and predicting wear of gears based on vibration signals according to an embodiment of the present invention includes:

[0011] S1: Dynamic working condition division based on multi-source sensor fusion, establishing a working condition map, and updating the working condition map in real time. The working condition map includes regional grids for multiple working conditions.

[0012] S2: Establish an Arcard wear correction model based on thermo-elasto-fluidic lubrication theory, using the working condition map as input, to calculate the wear of the gears;

[0013] S3: Establish a nonlinear vibration model of the gear considering the effects of wear and clearance. Using the working condition map and the wear amount of the gear as input, obtain the vibration signal of the gear system under different wear conditions.

[0014] S4: Establish a CNN-LSTM hybrid wear diagnosis model and train it to obtain a trained CNN-LSTM hybrid wear diagnosis model;

[0015] S5: The vibration signal of the gear system, which is collected and processed in real time, is input into the trained CNN-LSTM hybrid wear diagnosis model to obtain the probability distribution of the current wear state of the gear and the predicted data of the wear development trend of the gear in the future time period. The monitoring and prediction results are provided to the gear system maintenance planning process.

[0016] Optionally, S1 specifically includes:

[0017] S11: Using speed and torque parameters as core clustering variables, the clustering results group similar operating states into the same operating condition region. Also using oil temperature parameters as auxiliary clustering variables, the clustering results group similar operating states into the same operating state. Construct an operating condition map with regional grids including multiple operating conditions, and the center point of each regional grid is used as the operating condition point of the region.

[0018] S12: Install sensors to collect gear signals in real time, including speed sensors, torque sensors and temperature sensors, to collect parameters such as gear speed, input torque and oil temperature in real time;

[0019] S13: The signals collected by the installed sensors are used to dynamically identify the regional grid boundaries of the working conditions through density feasibility analysis, thereby realizing the automatic identification of real-time working condition data and updating of the working condition map.

[0020] Optionally, in S11:

[0021] The working condition regions are divided by clustering results using speed and torque parameters as the core clustering variables, including low-speed light load, low-speed heavy load, high-speed light load, and high-speed heavy load working condition regions.

[0022] The operating condition regions are divided by clustering results using oil temperature as a clustering auxiliary variable, including low temperature, normal temperature and high temperature operating condition regions.

[0023] By combining the two working condition regions, a working condition map is obtained that includes a regional grid of multiple working conditions.

[0024] Optionally, S13 specifically includes:

[0025] Based on the regional grid of the working condition map, an adaptive neighborhood radius is dynamically adjusted according to the real-time collected gear oil temperature parameters.

[0026] Using an improved DBSCAN algorithm, through By combining the wear level with the minimum number of points, the minimum number of points in the neighborhood can be dynamically adjusted.

[0027] Set a threshold for the dwell time of a working condition. For working condition points whose dwell time in the working condition area is shorter than the threshold, they are considered as abnormal working condition switching and are removed to obtain dynamically optimized working condition points.

[0028] Based on the dynamically adjusted adaptive neighborhood radius, the dynamically optimized minimum number of neighborhood points, and the dynamically optimized working condition points, the boundary of the regional grid is translated and reconstructed to obtain an updated working condition map.

[0029] Optionally, S2 specifically includes:

[0030] S21: Based on the operating points determined in the operating condition map, solve the thermo-elasto-hydrodynamic lubrication model to obtain the oil film thickness, oil film pressure and oil film temperature parameters in the contact area.

[0031] S22: By combining the parameters of oil film thickness, a nonlinear mapping between friction coefficient and film thickness ratio is established to realize real-time correction of friction state and obtain real-time corrected film thickness ratio.

[0032] S23: By combining the parameters of oil film pressure and oil film temperature, a real-time mapping between lubricating oil viscosity and oil film pressure and oil film temperature is established to obtain real-time corrected lubricating oil parameters.

[0033] S24: Use the obtained film thickness ratio as the lubrication state discrimination parameter, establish the Archard wear correction model based on the discrimination result, and obtain the wear amount of the gear.

[0034] Optionally, S24 specifically includes:

[0035] When film thickness ratio When the lubrication conditions are good, the wear of the gear is calculated according to the classic Archard wear model:

[0036]

[0037] in, This represents the wear volume of the gear. The wear coefficient of the gear. For the normal load on the gear, This is the relative sliding distance. Material hardness;

[0038] When film thickness ratio When the condition is at boundary lubrication, the wear of the gear is calculated by multiplying the classic Archard wear model by a correction factor:

[0039]

[0040] Among them, the correction coefficient We obtain it from the following formula:

[0041]

[0042] in, , The wear factor function is related to the operating conditions. This indicates the current operating status.

[0043] Optionally, S3 specifically includes:

[0044] S31: Establish a time-varying meshing stiffness model that considers different wear depths along the tooth surface direction of the gear, using the working condition map and the wear amount of the gear as inputs, to obtain the time-varying meshing stiffness of the gear.

[0045] S32: Establish the mapping relationship between gear wear and static transmission error to obtain the static transmission error;

[0046] S33: Establish a nonlinear dynamic model of the gear considering translation and torsion, using the time-varying meshing stiffness of the gear obtained in S31 as input, and introduce static transmission error to obtain the vibration signal of the gear system under different wear conditions.

[0047] Optionally, S4 specifically includes:

[0048] S41: Build a CNN, stack convolutional layers, including the first layer convolution, the middle layer convolution and the deep layer convolution, and obtain and output a 128-dimensional feature vector compressed by global average pooling;

[0049] S42: Expand the 128-dimensional feature vector compressed by global average pooling into a sequence of length T in chronological order to obtain the input tensor;

[0050] S43: Establish a bidirectional LSTM architecture, with each layer containing 128 memory units. Use the input tensor obtained in S42 as input to obtain the temporal information of the past and future input tensors as the LSTM output.

[0051] S44: The LSTM output is mapped to 6 wear levels through a fully connected layer to output the wear state probability distribution of the gear, thus establishing a CNN-LSTM hybrid wear diagnosis model;

[0052] S45: Using the cross-entropy loss function and the Adam optimizer, a training strategy is formulated. The vibration signals of the gear system obtained in S3 are classified to obtain training data, which is then input into the CNN-LSTM hybrid wear diagnosis model for training, resulting in a trained CNN-LSTM hybrid wear diagnosis model.

[0053] Optionally, S5 specifically includes:

[0054] S51: The vibration signal of the gear system is obtained by collecting and processing the signal through the installed sensor. The vibration signal is then denoised by ensemble empirical mode decomposition to obtain the denoised vibration signal.

[0055] S52: An improved local mean decomposition combined with S-transform is used to convert the noise-reduced vibration signal into a time-frequency image, which is then input into a trained CNN-LSTM hybrid wear diagnosis model to obtain the probability distribution of the wear state of the gear, as well as the predicted data of the wear development trend of the gear in the future time period, and to plot the curve of the predicted future wear amount of the gear.

[0056] Compared with the prior art, the method for monitoring and predicting wear of gears based on vibration signals according to an embodiment of the present invention has at least the following beneficial effects.

[0057] 1) By establishing a working condition map and automatically identifying and updating the working condition map in real time, the limitations of a single parameter are overcome, the accuracy of working condition identification is improved, and the noise problem of switching between complex working conditions is solved. By introducing a working condition dwell time threshold, short-term abnormal switching working conditions are eliminated, and frequent oscillations of the map are avoided. It is especially suitable for industrial equipment with frequent start-stop.

[0058] 2) By establishing an Arcard wear correction model based on thermo-elasto-fluidic lubrication theory, dynamic discrimination of lubrication state is achieved, and wear amount is accurately quantified. Compared with the traditional Arcard wear model, which does not consider the influence of lubrication state on wear, this model distinguishes lubrication state in real time by film thickness ratio. Based on different judgment results, the wear amount is calculated by using the classic Arcard formula and by introducing a correction coefficient, respectively, to realize the quantification of wear aggravation effect under boundary lubrication state.

[0059] 3) Multi-physics coupling improves calculation accuracy. The oil film thickness, pressure and temperature parameters are solved by thermo-elastohydrodynamic lubrication model. A nonlinear mapping between friction coefficient and film thickness ratio is established. The wear coefficient is corrected by real-time mapping of lubricating oil viscosity-pressure-temperature. It is especially suitable for high-speed heavy-load conditions.

[0060] 4) A nonlinear vibration model of the gear considering the effects of wear and clearance is established. Through non-uniform wear modeling, the true meshing characteristics are restored. A time-varying meshing stiffness model is established along the tooth root to the tooth tip. The model uses coupled calculations of five parts, including bending stiffness and Hertzian contact stiffness, to address stiffness calculation errors caused by differences in wear depth. Simultaneously, the wear-clearance-vibration full-link mapping is considered. A static transmission error model is used to integrate long-period errors, short-period errors, and transmission errors caused by wear. The Runge-Kutta method is then used to solve the nonlinear dynamic equations, providing high-fidelity data support for subsequent diagnosis and wear prediction.

[0061] 5) Based on the CNN method, wear impact features are extracted through stacked convolutional kernels, achieving collaborative capture of time-frequency features and time-series trends. Using the bidirectional LSTM method, wear evolution trends are captured through memory units, enabling prediction of wear volume for the next 100,000 revolutions. Furthermore, the wear is divided into multiple levels using the Softmax activation function, achieving multi-level wear status output. Real-time wear distribution output provides quantitative decision-making basis for gearbox maintenance. Compared to traditional threshold diagnosis, wear prediction can be performed 1-2 operating cycles in advance.

[0062] 6) This invention achieves accurate judgment of gearbox wear faults through the whole-link technology innovation of dynamic division of working conditions, accurate wear calculation, high-fidelity vibration modeling and intelligent diagnosis and prediction. It avoids repeated opening and inspection, saves time and reduces costs. At the same time, it can provide early warning of possible sudden faults caused by wear, thereby reducing the risk of damage to transmission equipment. This method can effectively improve the safety of the transmission system. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly introduced below. The features and advantages of the present invention can be more clearly understood by referring to the accompanying drawings. The accompanying drawings are schematic and should not be construed as limiting the present invention in any way. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a flowchart of a method for monitoring and predicting gear wear based on vibration signals according to an embodiment of the present invention.

[0065] Figure 2 This is a flowchart of S1 in the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0066] Figure 3 This is a flowchart of S2 in the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0067] Figure 4 This is a flowchart of S3 in the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0068] Figure 5 This is a flowchart of S4 in the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0069] Figure 6This is a flowchart of S5 in the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0070] Figure 7 This is a schematic diagram of the working condition map established by S1 in the working condition following gear wear monitoring and prediction method based on vibration signal provided according to an embodiment of the present invention.

[0071] Figure 8 The flowchart for updating the working condition Map in S1 of the working condition following gear wear monitoring and prediction method based on vibration signal provided according to an embodiment of the present invention.

[0072] Figure 9 is a diagram of the vibration signal of a healthy gear obtained in an embodiment of the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0073] Figure 10 A diagram illustrating the gear level 1 wear vibration signal obtained in an embodiment of the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0074] Figure 11 The image shows a gear level 2 wear vibration signal obtained in an embodiment of the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0075] Figure 12 The image shows a gear level 3 wear vibration signal obtained in an embodiment of the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0076] Figure 13 The illustration shows a gear level 4 wear vibration signal obtained in an embodiment of the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0077] Figure 14 The image shows a gear level 5 wear vibration signal obtained in an embodiment of the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0078] Figure 15 The image shows the predicted wear amount and wear level obtained in an embodiment of the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention.

[0079] Figure 16The predicted time cloud map of the wear level is obtained in an embodiment of the vibration signal-based working condition following gear wear monitoring and prediction method provided according to an embodiment of the present invention. Detailed Implementation

[0080] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0081] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein. Therefore, the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0082] The following describes in detail, with reference to the accompanying drawings, a method for monitoring and predicting wear of gears based on vibration signals according to an embodiment of the present invention.

[0083] like Figures 1 to 6 As shown, a method for monitoring and predicting wear of gears based on vibration signals according to an embodiment of the present invention includes the following steps.

[0084] S1: Based on multi-source sensor fusion, dynamic operating condition division is implemented, an operating condition map is established, and the operating condition map is updated in real time. The operating condition map includes multiple regional grids representing different operating conditions, with the center point of each regional grid serving as the operating condition point representing that region. For example... Figure 2 As shown, S1 specifically includes the following steps.

[0085] S11: Key operating parameters such as speed, torque, and oil temperature are used as core clustering variables. The clustering results group similar operating states into the same operating condition region (e.g., low-speed light load, low-speed heavy load, high-speed light load, high-speed heavy load, etc.). Oil temperature is used as an auxiliary clustering variable, and the clustering results group similar operating states into the same operating state (e.g., low temperature, normal temperature, high temperature, etc.). A working condition map is constructed, including grids of multiple typical operating condition regions. The aforementioned similar operating states can be determined by the wear rate, which can be calculated in subsequent steps. With dynamic changes such as equipment aging and environmental changes, traditional fixed thresholds cannot be updated in real time, which can easily lead to misjudgment of operating conditions (e.g., under the same operating speed and torque, the wear rate of gears will change when the ambient temperature of the gears is different). The working condition map construction method using the above combined clustering method can overcome the lag of traditional fixed threshold division, better determine the working condition boundary, and ensure the rationality of the working condition map division.

[0086] like Figure 7As shown in the example, the operating condition map can be set up as follows: The horizontal axis of the operating condition map represents speed, and the vertical axis represents torque. The horizontal axis is divided into 10 intervals, each ranging from 0 to 20000 rpm, with each interval representing 2000 rpm. The vertical axis is also divided into 10 intervals, each ranging from 0 to 500 N / m, with each interval representing 50 N / m. The resulting intervals are then combined and colored differently. Figure 7 Different oil temperatures are represented by varying shades of color, resulting in 16 regional grids from A to L. The wear rate is similar within each regional grid, and the center point of the regional grid is taken as the operating point representing that region for calculation.

[0087] S12: Install sensors to acquire gear signals in real time. This can include speed sensors, torque sensors, and temperature sensors to collect parameters such as gear speed, input torque, and oil temperature. Installing multiple sensors overcomes the limitations of traditional single-parameter operating condition classification methods, addressing the problem of insufficient model generalization ability under varying operating conditions. The real-time acquired gear speed, input torque, and oil temperature parameters can be used for subsequent updates to the operating condition map.

[0088] S13: Signals are collected through installed sensors. An improved DBSCAN algorithm is used to dynamically identify the regional grid boundaries of the working conditions through density feasibility analysis, enabling automatic identification of real-time working condition data and updating of the working condition map. See also... Figure 8 The specific process of this step is as follows.

[0089] Specifically, based on the regional grid of the working condition map, the dynamically adjusted adaptive neighborhood radius is represented as follows:

[0090]

[0091] in, The radius of a region grid unit. For oil temperature sensitivity coefficient, The deviation between the current oil temperature and the historical average oil temperature. For each moment, the area grid unit is one area grid of the operating condition Map divided by S11; the current oil temperature is the gear oil temperature collected in real time by the temperature sensor.

[0092] Using an improved DBSCAN algorithm, through By adjusting the minimum number of points based on the wear level, dynamic optimization with the minimum number of points in the neighborhood is achieved:

[0093]

[0094] in, Base points The wear sensitivity coefficient, This corresponds to the wear level.

[0095] When the boundary points of a region grid for a certain operating condition are continuously exceeded by new data points a set threshold number of times (e.g., N times), the boundary translation and reconstruction of the region grid for that operating condition is triggered. The method for determining if a boundary point of the region grid is exceeded by new data points is that if the number of new data points exceeds the boundary point of the region grid, it is considered that the boundary point of the region grid has been exceeded by new data points. After the boundary reconstruction of the region grid in the operating condition map, density feasibility analysis can be used to feed back optimization parameters to the real-time data acquisition steps.

[0096] Furthermore, for complex operating conditions, a threshold for the dwell time of an operating condition can be set. If the dwell time in the operating condition area is shorter than the threshold, it is considered an abnormal operating condition switch, and the operating condition (operating point) is removed and not included in the calculation. This solves the problem of repeated switching of the operating condition map under complex operating conditions. As needed, the above-mentioned threshold for the dwell time of an operating condition can be set, for example, but not limited to, 5s, 10s, 12s, etc.

[0097] Based on the above-mentioned dynamically adjusted adaptive neighborhood radius and minimum number of neighborhood points, as well as the dynamic optimization of working condition points, dynamic reconstruction of working conditions is achieved, thereby enabling the updating of the working condition map.

[0098] This yields a real-time updated operating condition map, which serves as input for the following Arcard wear correction model based on thermo-elasto-hydrodynamic lubrication (EHL) theory and the gear nonlinear vibration model considering the effects of wear and clearance.

[0099] S2: Establish an Archard wear correction model based on the thermo-elasto-fluidic lubrication (EHL) theory, using the operating condition map as input, to calculate the wear of the gears. For example... Figure 3 As shown, S2 specifically includes the following steps.

[0100] S21: Based on the operating points determined in the operating condition map, solve the thermo-elastohydrodynamic (TEHL) model to obtain the oil film thickness, oil film pressure and oil film temperature parameters in the contact area.

[0101] S22: Based on the oil film thickness parameter, establish the friction coefficient versus film thickness ratio. The nonlinear mapping enables real-time correction of the friction state, resulting in a real-time corrected film thickness ratio. (Film thickness ratio) It can be obtained from the oil film thickness and the coefficient of friction.

[0102] S23: By combining the parameters of oil film pressure and oil film temperature, a real-time mapping between lubricating oil viscosity and oil film pressure and oil film temperature is established to realize the real-time correction of lubricating oil parameters and obtain the real-time corrected lubricating oil parameters.

[0103] S24: The obtained film thickness ratio As a parameter for judging lubrication status, an Archard wear correction model is established based on the judgment result to obtain the wear amount of the gear.

[0104] Specifically, when the film thickness ratio When the lubrication conditions are good, the wear of the gear is calculated according to the classic Archard wear model:

[0105]

[0106] in, This represents the wear volume of the gear. The wear coefficient of the gear. For the normal load on the gear, This is the relative sliding distance. The material hardness of the gear. The wear coefficient of the gear. The wear coefficient of the gear can be obtained based on the friction coefficient and the lubricating oil parameters obtained from S23, which are related to the friction coefficient and the lubricating oil parameters.

[0107] When film thickness ratio When the condition is defined as boundary lubrication, the lubrication effect is poor. The wear of the gear is calculated by multiplying the classic Archard wear model by a correction factor.

[0108]

[0109] The correction factor can be obtained from the following formula:

[0110]

[0111] in, , The wear factor function, which is related to operating conditions, can be calibrated experimentally. This indicates the current operating status.

[0112] The above yields the Archard wear correction model based on the thermo-elastohydrodynamic (EHL) theory.

[0113] S3: Establish a nonlinear vibration model for the gear considering the effects of wear and clearance. Using the working condition map and the wear amount of the gear as input, this model is used to obtain the vibration signal of the gear system under different wear conditions. For example... Figure 4 As shown, S3 specifically includes the following steps.

[0114] S31: Establish a time-varying meshing stiffness model that considers different wear depths along the tooth surface of the gear, using the working condition map and the wear amount of the gear as inputs, to obtain the time-varying meshing stiffness of the gear.

[0115] The wear rate varies at each position along the tooth root to the tooth tip of a gear. By calculating the bending stiffness, Hertzian contact stiffness, compressive stiffness, shear stiffness, and matrix deformation stiffness at each wear position, the time-varying meshing stiffness of the gear is obtained under different wear levels. This model solves the problem of inaccurate results caused by the assumption of uniform wear in traditional models.

[0116] S32: Establish a mapping relationship between gear wear and static transmission error to obtain the static transmission error. Gear wear causes changes in gear meshing clearance, resulting in changes in static transmission error. A mapping is established along the tooth root to the tooth tip between different wear depths and the static transmission error to obtain the static transmission error variation function. The calculation process for static transmission error (USTE) is as follows.

[0117] Establish long-period error:

[0118]

[0119] In the formula, For the rotational frequency of the driving gear, For the total tangential deviation, At time t. The total tangential deviation is the sum of the tangential deviations of all gears in the gear system.

[0120] Establish short-period error:

[0121]

[0122] In the formula, For biting frequency, This represents the combined tangential deviation of a single gear.

[0123] Establish random error:

[0124]

[0125] In the formula, These are random numbers that are uniformly distributed between [0,1].

[0126] Establish the transmission error caused by gear wear, where the transmission error caused by gear wear is the amount of wear on the driving gear. and the wear of the driven gear Projection along the line of engagement:

[0127]

[0128] in, This indicates the pressure angle of the gear.

[0129] Based on long period error Short-period error Random error Transmission error caused by gear wear The static propagation error (USTE) is obtained as follows:

[0130] .

[0131] Therefore, a static transmission error variation function is established, and the static transmission error variation can be obtained based on this static transmission error variation function.

[0132] S33: Establish a nonlinear dynamic model of the gear considering translation and torsion. Use the time-varying meshing stiffness of the gear obtained in S31 as input, and introduce the static transmission error obtained by the static transmission error variation function. Solve the differential equations of the gear nonlinear dynamic model using the Runge-Kutta method to obtain the vibration signal of the gear system under different wear conditions.

[0133] The Runge-Kutta method is used to solve the differential equations of the gear nonlinear dynamics model as follows:

[0134]

[0135] In the formula, For the quality matrix, Here is the damping matrix. Here is the stiffness matrix. For generalized coordinates, It is the gap function. It is the generalized force vector matrix.

[0136] The above stiffness matrix terms are obtained from the time-varying meshing stiffness of the gear based on S31, and the above generalized coordinate terms are obtained from the static transmission error and the displacement in the meshing line direction based on S32. The vibration acceleration can be obtained by taking the second derivative with respect to the displacement in the meshing line direction (generalized coordinates).

[0137] S4: Establish a CNN (Convolutional Neural Network)-LSTM (Long Short-Term Memory) hybrid wear diagnosis model to obtain the probability distribution of wear state of the gear system. Based on the vibration signal obtained in S3, train this CNN-LSTM hybrid wear diagnosis model to obtain a trained CNN-LSTM hybrid wear diagnosis model. For example... Figure 5 As shown, S4 specifically includes the following steps.

[0138] S41: Build a CNN (Convolutional Neural Network) by stacking convolutional layers, including the first, middle, and deep convolutional layers, to obtain and output a 128-dimensional feature vector compressed by global average pooling, which retains key information in the time and frequency domains. The specific process of stacking convolutional layers to build a CNN is as follows.

[0139] The first convolutional layer uses a 3×3 kernel (stride 1, padding 1) to extract the vibration signal of the weak impact in the early stage of wear. The number of convolutional kernels is set to 64, and the activation function is ReLU. This first convolutional layer can extract the vibration signal of the gear system obtained in S3 above.

[0140] The middle convolutional layers use 5×5 kernels (stride 2, padding 2) to capture more macroscopic structural features, such as harmonic distribution changes caused by increased wear. The number of kernels is increased to 128, and batch normalization is used to accelerate convergence. The input to the middle convolutional layers is the output of the first convolutional layer.

[0141] The deep convolutional layer uses a 7×7 kernel (stride 2, padding 3) to extract global features, with a total of 256 kernels. Dimensionality is reduced through max pooling (2×2 kernels). The output of this convolutional layer is then compressed into a 128-dimensional feature vector using global average pooling, preserving key information in the time-frequency domain and avoiding the overfitting risk associated with fully connected layers. The input to the deep convolutional layer is the output of the intermediate convolutional layer.

[0142] S42: Expand the 128-dimensional feature vectors extracted by CNN and compressed by global average pooling into a sequence of length T in chronological order, forming an input tensor of shape [time step, feature dimension].

[0143] S43: Establish a bidirectional LSTM architecture, with each layer containing 128 memory units. Using the input tensor obtained in S42 as input, the temporal information of the past and future input tensors is obtained as the LSTM output. This LSTM architecture can simultaneously capture the wear evolution trend of gears under different working conditions, obtain the temporal information of the past and future input tensors, and prevent overfitting through a dropout layer (dropout rate=0.2).

[0144] S44: The LSTM output is mapped to six wear levels (e.g., levels 0-5) via a fully connected layer. The activation function is Softmax, and the output shows the probability distribution of the gear's wear state. This probability distribution reflects the current wear condition of the gear. Specifically, the six wear levels are: Level 0 represents a normal, wear-free state; Levels 1-2 correspond to initial wear, characterized by minor scratches or slight material loss; Levels 3-4 represent intermediate wear, with significant changes in the tooth profile; and Level 5 indicates severe wear, potentially leading to major failures such as tooth surface peeling or tooth breakage. This clear level classification facilitates quick assessment of gear wear by maintenance personnel. The probability distribution of the wear state of the entire gear system can be derived from the probability distribution of the wear state of a single gear.

[0145] This leads to a CNN-LSTM hybrid wear diagnosis model, which is used to obtain the wear state probability distribution of the gear system based on the vibration signal of the gear system.

[0146] S45: Using the cross-entropy loss function and the Adam optimizer, a training strategy is formulated to classify the vibration signals of the gear system obtained from the gear nonlinear dynamics model established in S3, and use them as training data input to train the CNN-LSTM hybrid wear diagnosis model, thus obtaining the trained CNN-LSTM hybrid wear diagnosis model.

[0147] S5: The vibration signals of the gear system, acquired and processed in real time, are input into the trained CNN-LSTM hybrid wear diagnosis model to obtain the probability distribution of the current wear state of the gears and the predicted wear development trend data over a future period. This data is then provided to the gear system maintenance planning process as a basis for maintenance personnel to formulate maintenance strategies. The vibration signals of the gear system are obtained by executing S1 to S3 above. Figure 6 As shown, S5 specifically includes the following steps.

[0148] S51: Vibration signals of the gear system are obtained by collecting and processing information through installed sensors. Ensemble Empirical Mode Decomposition (EEMD) is used for noise reduction to obtain the noise-reduced vibration signal. The noise reduction process includes adaptively decomposing the vibration signal into multiple intrinsic mode functions (IMFs), removing noise-dominant IMF components, and suppressing the interference of background noise on gear meshing characteristics and wear impact characteristics.

[0149] S52: An improved Local Mean Decomposition (LMD) combined with S-Transform (LMD-S) is used to convert the denoised vibration signal into a time-frequency image. This converted time-frequency image is then used as input to a trained CNN-LSTM hybrid wear diagnosis model to obtain the probability distribution of gear wear state. Leveraging the LSTM's ability to capture long-term dependencies in temporal features, the model outputs predicted wear trend data for the gear over a future time period and plots a curve showing the future wear variation of the gear, for example, but not limited to, one operating cycle (100,000 revolutions per cycle). This helps maintenance personnel plan maintenance strategies in advance and avoid sudden failures. Specifically, LMD adaptively decomposes the multi-component signal, and S-Transform maintains phase consistency in the time-frequency domain.

[0150] Example 1

[0151] To better understand the present invention, an embodiment of a vibration signal-based method for monitoring and predicting wear of gears according to an embodiment of the present invention is described below. In this embodiment, wear monitoring and prediction are performed on a set of gear systems.

[0152] Figures 9 to 14 In this embodiment, after real-time sensor information acquisition and processing of a gear system using the method provided in the above implementation, vibration signals of the gear at six wear levels, from normal to severe wear, are obtained. Figure 15 The graph shows the predicted wear amount and wear level as a function of the operating cycle. Figure 16 This is a time-varying graph representing the wear level over the predicted operating cycle.

[0153] As can be seen from the figure, by Figures 9 to 14 It can be seen that the vibration peak value reflects the change in the degree of wear. From healthy to level 5 wear, the peak value of the time-domain vibration signal gradually increases with the increase of the wear level, which intuitively verifies the positive correlation between wear and vibration response and reflects the model's ability to capture the time-domain characteristics of different wear states.

[0154] Depend on Figure 15 It can be seen that the model can achieve time-series fitting of the wear prediction curve. The prediction curve of wear change with the operating cycle shows the model's ability to quantitatively predict the wear development trend, providing a visual reference for the future wear growth of gears and supporting the early formulation of maintenance strategies.

[0155] Depend on Figure 16 It enables visualization of wear status time cloud map, and intuitively presents the wear level distribution of different operating cycles through color gradient, clearly showing the evolution process of wear from the initial stage to the severe stage, which facilitates quick identification of the current wear stage and future trends, and improves the intuitiveness of operation and maintenance decision making.

[0156] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0157] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0158] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A vibration signal based operating condition following gear wear monitoring and prediction method, characterized in that, Comprise: S1: dynamic working condition division based on multi-source sensor fusion, establish working condition Map figure, and real-time working condition Map figure update, working condition Map figure includes multiple working condition area grid; S2: establish Archard wear correction model based on thermal elastohydrodynamic lubrication theory, take working condition Map figure as input, calculate gear wear; S3: establish gear nonlinear vibration model considering wear and gap influence, take working condition Map figure and gear wear as input, solve gear system vibration signal under different wear state; S4: establish CNN-LSTM hybrid wear diagnosis model, and train, get trained CNN-LSTM hybrid wear diagnosis model; S5: the vibration signal of gear system collected and processed in real time is input into the trained CNN-LSTM hybrid wear diagnosis model, the current wear state probability distribution of gear and the wear development trend prediction data of gear in a future time period are obtained as monitoring and prediction results and are provided to gear system maintenance planning process; Wherein, S2 specifically comprises: S21: according to the working condition point determined in working condition Map figure, solve thermal elastohydrodynamic lubrication model, obtain contact area oil film thickness, oil film pressure and oil film temperature parameters; S22: combined with the parameters of oil film thickness, establish nonlinear mapping of friction coefficient and film thickness ratio, realize real-time correction of friction state, and obtain real-time corrected film thickness ratio; S23: combined with the parameters of oil film pressure and oil film temperature, establish real-time mapping of lubricating oil viscosity, oil film pressure and oil film temperature, and obtain real-time corrected lubricating oil parameters; S24: take the obtained film thickness ratio as the lubrication state discrimination parameter, establish Archard wear correction model according to the discrimination result, and obtain the wear of gear; Wherein, S24 specifically comprises: When the film thickness ratio is greater than 3, it indicates that the lubrication condition is good, and the wear of the gear is calculated according to the classic Archard wear model: wherein, is the wear volume of the gear, is the wear coefficient of the gear, is the normal load of the gear, is the relative sliding distance, is the material hardness; When the film thickness ratio When the film thickness ratio When the film thickness ratio When the film thickness ratio When the film thickness ratio When the film thickness ratio When the film thickness ratio When the film thickness ratio When the wherein the correction factor is obtained from the equation: wherein , is a wear factor function related to the operating condition, denotes the current operating condition.

2. The vibration signal based operating condition following gear wear monitoring and prediction method of claim 1, wherein, S1 specifically comprises: S11: take the parameters of rotating speed and torque as clustering core variable, and the clustering result will be the same running state into the same working condition area, and take the parameter of oil temperature as clustering auxiliary variable, and the clustering result will be the same running state into the same running state, construct working condition Map figure including multiple working condition area grid, and the center point of each area grid is taken as the working condition point of the area; S12: install sensors for real-time acquisition of gear signals, including installation of rotating speed sensor, torque sensor and temperature sensor, real-time acquisition of rotating speed, input torque and oil temperature parameters of gear; S13: through the installed sensors, the collected signals are obtained, the improved DBSCAN algorithm is adopted, the dynamic identification of working condition area grid boundary is realized through density feasibility analysis, and the automatic identification of real-time working condition data and working condition Map figure update are realized.

3. A vibration signal based working condition following gear wear monitoring and prediction method according to claim 2, characterized in that, S13 specifically comprises: The area grid of working condition Map figure is the basis, the dynamically adjusted adaptive neighborhood radius is obtained according to the real-time collected oil temperature parameters of gear; The improved DBSCAN algorithm is adopted to obtain the minimum point number of the neighborhood by The minimum point number of the neighborhood is dynamically adjusted by combining the wear grade adjustment. Set working condition residence time threshold, for the working condition point whose residence time in working condition area is shorter than working condition residence time threshold, it is regarded as abnormal working condition switching, the working condition point is excluded, and the dynamically optimized working condition point is obtained; Based on the dynamic adjustment of adaptive neighborhood radius and the dynamic optimization of the minimum number of neighborhood points, and the dynamic optimization of working condition points, the boundary translation and reconstruction of the regional grid are carried out to obtain the updated working condition Map.

4. The vibration signal based operating condition following gear wear monitoring and prediction method of claim 1, wherein, S3 specifically comprises: S31: a time-varying engagement stiffness model considering different wear depths along the tooth surface direction of the gear is established, and the working condition Map and the wear amount of the gear are used as inputs to obtain the time-varying engagement stiffness of the gear; S32: a mapping relationship between the wear amount of the gear and the static transmission error is established to obtain the static transmission error; S33: a gear nonlinear dynamics model considering translation-torsion is established, the time-varying engagement stiffness of the gear obtained in S31 is used as input, and the static transmission error is introduced to obtain the vibration signal of the gear system under different wear states.

5. The vibration signal based operating condition following gear wear monitoring and prediction method of claim 1, wherein, S4 specifically comprises: S41: a CNN is established, and a convolution layer stack is performed, including a first convolution, a middle convolution and a deep convolution, to obtain and output a 128-dimensional feature vector compressed by global average pooling; S42: the 128-dimensional feature vector compressed by global average pooling is unfolded in time sequence to obtain a sequence with a length of T to obtain an input tensor; S43: a bidirectional LSTM architecture is established, each layer contains 128 memory cells, and the input tensor obtained in S42 is used as input to obtain past and future input tensor time sequence information as LSTM output; S44: the LSTM output is mapped to 6 wear grades through a full connection layer to output the wear state probability distribution of the gear, and a CNN-LSTM hybrid wear diagnosis model is established; S45: a cross-entropy loss function is used, an Adam optimizer is used, a training strategy is formulated, the vibration signal of the gear system obtained in S3 is classified to obtain training data, and the training data is input into the CNN-LSTM hybrid wear diagnosis model for training to obtain a trained CNN-LSTM hybrid wear diagnosis model.

6. The vibration signal based operating condition following gear wear monitoring and prediction method of claim 2, wherein, S5 specifically comprises: S51: the vibration signal of the gear system is obtained by collecting signals through the installed sensor and processing, and the ensemble empirical mode decomposition is used to denoise the vibration signal to obtain a denoised vibration signal; S52: the denoised vibration signal is converted into a time-frequency image by using the improved local mean decomposition combined with the S transform, and the trained CNN-LSTM hybrid wear diagnosis model is input to obtain the wear state probability distribution of the gear and the wear development trend prediction data of the gear in a future time period, and a change curve of the future wear amount of the predicted gear is drawn.

Images