Digital twin imbalance fault diagnosis method and system for gear transmission system

Abstract

The invention belongs to the technical field of gear transmission systems, and particularly discloses a digital twinborn imbalance fault diagnosis method and system for a gear transmission system, and the method comprises the following steps: collecting basic parameters of the gear transmission system, building a dynamical model of the gear transmission system, obtaining twinborn signals under different fault types, and determining the twinborn imbalance fault of the gear transmission system. Calculating an adaptive weight coefficient; aligning impact peak values in time domains of the actually measured fault signal and the original twin fault signal after filtering, and performing time domain weighted reconstruction to obtain a fusion fault signal so as to obtain a large number of fusion fault signal samples with labels, and constructing a training set; training set samples are input into the fault diagnosis network, the feature extractor extracts the features of the signals, the classifier processes the extracted features, and the fault types of the actually measured samples are recognized. By adopting the technical scheme, the adaptive weight coefficient is obtained based on the twinborn signal and the actually measured signal to carry out time domain weighted fusion, data enhancement is realized, and the diagnosis accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of gear transmission systems and relates to a digital twin imbalance fault diagnosis method and system for a gear transmission system. Background Art

[0002] Gear transmissions are widely used in high-end equipment due to their precise transmission ratios, high efficiency, and stable operation. However, gearboxes, as an essential mechanical transmission system in various mechanical devices, are subject to long-term exposure to complex and harsh operating environments, leading to inevitable failures, resulting in unstable operation and even equipment downtime. Therefore, gearbox condition monitoring and fault diagnosis are essential to ensure efficient and stable operation of the equipment.

[0003] Currently, intelligent diagnostic algorithms based on deep learning rely on a large number of well-balanced training samples. However, the actual operating conditions of gearboxes are complex and changeable, and most of the time they operate normally, with low failure rates, making fault samples scarce and difficult to obtain.

[0004] In the case of imbalanced data training, the diagnostic model tends to be more inclined to identify the normal state of the majority class and ignore the fault state of the minority class, resulting in the model's lack of sufficient generalization on monitoring data outside the training data.

[0005] Although it is possible to simulate fault data that is difficult to obtain under actual operating conditions by establishing a dynamic model of the gear transmission system in the gearbox, some parameters cannot be accurately known when establishing the model and can only be determined based on prior knowledge. It is also difficult to simulate the noise during on-site collection, and the measured signal is also affected by measurement errors and attenuation along the transmission path. These factors lead to serious distribution differences between the dynamic model response and the actual measurement data.

[0006] In addition, most current studies use labeled samples obtained from dynamic model simulation as the source domain and unlabeled measured samples as the target domain, and further reduce the distribution difference between the two domains through domain adaptation. However, they do not fully utilize the small amount of label information in the measured fault samples. In addition, the training process is relatively complicated and requires adjusting the weight relationship between multiple loss functions, which makes it difficult for training to converge. Summary of the Invention

[0007] The purpose of the present invention is to provide a digital twin imbalance fault diagnosis method and system for a gear transmission system to improve the diagnostic accuracy and generalization capability.

[0008] In order to achieve the above objectives, the basic solution of the present invention is: a digital twin imbalance fault diagnosis method for a gear transmission system, comprising the following steps:

[0009] S1, collect the basic parameters of the gear transmission system, calculate the time-varying meshing stiffness excitation of the gears, and establish the dynamic model of the gear transmission system;

[0010] S2, based on the dynamic model established in step S1, optimizing the radial support stiffness and radial support damping, as well as the axial support stiffness and axial support damping in the dynamic model according to measured data, to achieve consistency between model calibration and virtual and real vibration response results;

[0011] Implant different types of local gear faults and simulate to obtain twin signals under different fault types;

[0012] S3, randomly sampling a small amount of gearbox measured fault signals collected on site to obtain segmented measured fault signal samples;

[0013] The gear fault twin signal sample and the measured signal sample obtained by simulation in step S2 are processed by low-pass filtering, and the sum of the amplitudes of the two signals at the characteristic frequencies in the frequency domain after filtering is calculated to obtain the adaptive weight coefficient ω;

[0014] S4, aligning the impulse peaks of the filtered measured fault signal and the original twin fault signal in the time domain through generalized cross-correlation phase correction;

[0015] The adaptive weight coefficient ω is used to perform weighted fusion on the original twin fault signal and the corrected measured fault signal in the time domain to obtain a fused fault signal.

[0016] S5, performing overlapping sampling on the fused fault signal to obtain a large number of labeled fused fault signal samples;

[0017] A large number of fused fault signal samples are mixed with a small number of labeled measured fault samples to form a fault sample training set, and a training set is constructed together with the measured normal state samples;

[0018] S6, input the training set samples in step S5 into the fault diagnosis network composed of a feature extractor and a classifier, the feature extractor extracts the features of the signal, and the classifier processes the extracted features to identify the fault type of the measured sample.

[0019] The working principle and beneficial effects of this basic solution are as follows: This technical solution establishes a dynamic model of the gear transmission system, and updates the dynamic model parameters by measuring the vibration acceleration signal under normal conditions, combined with the time domain root mean square value and the frequency domain spectrum center of gravity, to achieve consistency between model calibration and virtual and real vibration response results.

[0020] Based on the twin signals and the measured signals, the time-domain weighted adaptive weight coefficients are calculated. After the generalized cross-correlation phase correction, the time-domain signals are fused through adaptive weighting to expand the fault samples and achieve data enhancement.

[0021] Combining a large number of fused fault samples with a small number of labeled measured fault samples, the fault diagnosis network is trained to identify the types of faults in the measured samples. The method has high diagnostic accuracy and generalization ability.

[0022] Furthermore, the method for calculating the time-varying meshing stiffness excitation of the gear and establishing the dynamic model of the gear transmission system using the lumped parameter method is as follows:

[0023] S101, calculate the time-varying meshing stiffness excitation K of the helical gear pair in the gearbox during the meshing process m ,for:

[0024]

[0025] Among them, M is the number of gear pairs involved in meshing, N0 is the number of slices of the helical gear along the tooth width direction, represents the stiffness of the Jth slice of the Ith meshing tooth, γ = a, b, s, f, h represent the axial compression stiffness, bending stiffness, shear stiffness, matrix stiffness and Hertzian contact stiffness of the healthy helical gear, respectively, and the subscripts p and g represent the driving gear and the driven gear, respectively;

[0026] S102, based on time-varying mesh stiffness excitation K m , calculate the dynamic meshing force F in the gear transmission system m :

[0027]

[0028] Among them, δ m is the transmission error, C m is the meshing damping, is the first-order derivative of the transmission error;

[0029] S103, based on the dynamic meshing force F m , the lumped parameter method is used to establish the dynamic model of the helical gear transmission system, and its dynamic equation is expressed as:

[0030]

[0031] Among them, M p and M g Represent the mass of the driving wheel and the driven wheel respectively, J p and J g Represents the moment of inertia of the driving wheel and the driven wheel respectively, C xi ,C yi ,C zi Respectively represent the bearing support damping of the driving wheel and the driven wheel in the x, y, and z directions; K xi ,K yi ,K ziRespectively represent the bearing support stiffness of the driving wheel and the driven wheel in the x, y, and z directions, i = p, g; C bw is the torsional damping, K bw is the torsional stiffness, F m is the dynamic meshing force of gears, β b is the base circle helix angle, α n is the meshing angle of the gear pair, R bp R is the base circle radius of the driving wheel, gp is the base circle radius of the passive wheel, T0 is the input torque, T L is the load torque; x i Respectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the x direction; y i Respectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the y direction; z p Respectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the z direction; θ iz They represent the torsional angular acceleration, torsional angular velocity and torsional angular displacement of the driving wheel and the driven wheel in the z direction respectively.

[0032] Establishing a dynamic model of the gear transmission system is beneficial for data analysis.

[0033] Furthermore, based on the measured data, the bearing radial support stiffness and radial support damping, as well as the axial support stiffness and axial support damping in the dynamic model were optimized. Different types of local gear faults were implanted, and the method for simulating and obtaining twin signals under different fault types is as follows:

[0034] S201, based on the measured normal state signal and the simulated normal state signal, calculate the root mean square value rms of the two signals in the time domain and the spectral center of gravity f in the low frequency band of the frequency domain C ,for:

[0035]

[0036] Where N is the number of data points in the signal time domain, x j represents the jth data point in the signal time domain, K is the number of data points in the signal frequency domain, and f k represents the frequency of the kth data point in the signal frequency domain; s(k) represents the corresponding frequency amplitude;

[0037] S202, assuming that the support stiffness and damping of the driving wheel and the driven wheel in the x and y directions are the same, and the support stiffness and damping in the z direction are the same, simplify the dynamic model, that is, satisfy K xi =K yi =K0,Cxi =C yi =C0,K zp =K zg =K1,C zp =C zg = C1; i = p, g, where the subscripts p and g represent the driving wheel and the driven wheel respectively; C xi ,C yi ,C zi Respectively represent the bearing support damping of the driving wheel and the driven wheel in the x, y, and z directions; K xi ,K yi ,K zi Respectively represent the bearing support stiffness of the driving wheel and the driven wheel in the x, y, and z directions;

[0038] S203, based on the measured and simulated normal state signals in step S201, the parameters K1 = [K0, K1] and C = [C0, C1] in the dynamic model are optimized by the Beluga optimization algorithm (BWO). The objective function is expressed as:

[0039]

[0040] Among them, K0, K1 represent the radial support stiffness and axial support stiffness respectively, C0, C1 represent the radial support damping and axial support damping respectively, K1 represents the stiffness vector, C represents the damping vector, Obj represents the objective function of the optimization algorithm, rms s Represents the root mean square value of the simulation signal in the time domain, rms m represents the RMS value of the measured signal in the time domain, Represents the spectral center of the low-frequency band of the simulation signal frequency domain, Represents the spectral center of the low-frequency band of the measured signal in the frequency domain;

[0041] S204, in the digital twin model of the gear transmission system, by changing the time-varying meshing stiffness excitation K m , different types of local gear faults are implanted, including: tooth root cracks, tooth surface spalling, tooth surface wear, and tooth fracture, to obtain twin signals under different fault types.

[0042] The radial support stiffness and radial support damping, as well as the axial support stiffness and axial support damping in the dynamic model are optimized based on the measured data to achieve consistency between model calibration and virtual and real vibration response results.

[0043] Furthermore, in step S3, a small amount of gearbox measured fault signals collected on site are randomly sampled to obtain segmented measured fault signal samples, and the sum of the amplitudes at the characteristic frequencies in the frequency domain of the two signals after filtering is calculated to obtain the adaptive weight coefficient ω as follows:

[0044] S301, from a small number of fault measured signal segments x(t) = [x1, x2, ..., x m ] randomly selects the sampling starting point, intercepts the signal segment with a length of L, and obtains the measured fault signal sample x n (t) = [x e x e+1 ,…,x e+L-1 ], the sampling starting point must satisfy 1≤e≤m-L+1, expressed as:

[0045] e=randint(1,m-L+1)

[0046] Where e represents the sampling start index of the signal segment, m represents the length of the measured signal segment, n represents the number of samples, and randint(·) represents the random integer function;

[0047] The measured fault signal sample x n (t) and twin fault signal sample y n (t) The high-frequency noise in the measured fault signal is removed by the designed low-pass digital filter, and the frequency distribution of the two is ensured to be in the same range, and the filtered measured fault signal is obtained. and twin fault signals

[0048] S302, the measured fault signal after filtering in step S301 and twin fault signals Perform Fourier transform and calculate the two signals at the rotation frequency f r , meshing frequency f m The sum of the amplitudes of its higher harmonics is expressed as:

[0049]

[0050] Among them, A fr Indicates the amplitude at the rotation frequency, vf m Indicates the vth harmonic of the meshing frequency, A vfm represents the amplitude at the meshing frequency v harmonic, and A represents the sum of the amplitudes at the characteristic frequencies of the fault signal;

[0051] S303, based on the sum of the amplitudes of the two fault signal characteristic frequencies in step S302, calculate the adaptive weight coefficient ω, which is:

[0052]

[0053] Among them, A E and A S Represent the sum of the amplitudes of the measured fault signal and the twin fault signal at their characteristic frequencies.

[0054] Based on the twin signal and the measured signal, the time-domain weighted adaptive weight coefficient is calculated, which is convenient for use.

[0055] Furthermore, the filtered measured fault signal and the original twin fault signal are aligned with the impulse peaks in the time domain by generalized cross-correlation phase correction. The specific steps are as follows:

[0056] Calculate the filtered measured fault signal and the original twin fault signal y n The cross-correlation function of (t) is:

[0057]

[0058] Among them, R xy (·) represents the cross-correlation function of two signals, τ is the delay, and t represents time;

[0059] Based on the cross-correlation function, the time delay τ* corresponding to the maximum value is found, and the measured fault signal Perform translation to align the two signal impulse peaks. The expression is:

[0060]

[0061] Where τ* represents the time delay when the cross-correlation function reaches its maximum value, Indicates the measured fault signal after phase correction.

[0062] Performing generalized cross-correlation phase correction can effectively reduce energy loss and loss of fault characteristics during signal fusion.

[0063] Furthermore, the adaptive weight coefficient ω is used to perform weighted fusion on the original twin fault signal and the corrected measured fault signal in the time domain to obtain the fused fault signal Q n (t) is:

[0064]

[0065] Among them, Q n (t) represents the fused fault signal, n is the number of samples; A E and A S Represent the sum of the amplitudes of the measured fault signal and the twin fault signal at their characteristic frequencies.

[0066] After generalized cross-correlation phase correction, the time domain signals are fused through adaptive weighting to expand the fault samples for data enhancement.

[0067] Furthermore, the feature extractor is provided with a convolution block, which includes a convolution layer, a batch normalization layer and a pooling layer, and the expression is:

[0068]

[0069] Among them, H l is the input of the convolutional layer, H r is the output of the convolutional layer, num is the number of input feature maps, r is the number of output feature maps, that is, the number of convolution kernels; b is the bias, μ is H r The average value of H r The standard deviation of , γ1 is the scaling parameter, β1 is the translation parameter, G is the activation function, and max represents the maximum value operation; K lr is the weight of the rth convolution kernel for the lth input feature map, b r is the bias matrix of the rth convolution kernel, is the output of the convolutional layer, is the output after the batch normalization layer;

[0070] After the input passes through two convolution blocks, multi-scale feature extraction is performed with the help of convolution blocks with two convolution kernels of different sizes, and then spliced ​​from the channel dimension to obtain the splicing feature H, which is:

[0071] H=concat(H1,H2)

[0072] Among them, H1 is the output of the first convolution block, H2 is the output of the second convolution block, and concat(·) means splicing from the channel dimension;

[0073] After the concatenated feature H passes through a convolution block, the features of each channel are adaptively weighted through the attention weighted network:

[0074]

[0075] in, It is a vector composed of the mean of each channel of the splicing feature H; represents the mean of the chth channel of the concatenated feature H, where ch is the number of channels; B is the weight vector output by the attention weighted network; f a (·) represents the mapping relationship in the attention weighted network, α ch is the weight corresponding to each channel; H ω represents the weighted output feature map.

[0076] The feature extractor is used to extract the required features for subsequent use.

[0077] Furthermore, the classifier includes two fully connected layers. The classifier output is converted into a probability value of [0, 1] after passing through the Softmax function to realize the identification of the fault type. The expression is:

[0078] Y = Softmax(G(∑ω1x+b))

[0079]

[0080] Among them, x is the input, ω1 is the weight matrix, Y is the output, b is the bias, G is the activation function, R is the fault category, z u is the u-th output value of the fault diagnosis network.

[0081] Based on the classifier, fault classification diagnosis is realized with simple operation.

[0082] The present invention also provides a digital twin imbalance fault diagnosis system for a gear transmission system, comprising a data acquisition unit and a processing unit, wherein the data acquisition unit is used to acquire basic parameters of the gear transmission system, and the output end of the data acquisition unit is connected to the input end of the processing unit;

[0083] The processing unit executes the method of the present invention to achieve fault diagnosis of the gear transmission system.

[0084] This system combines a large number of fused fault samples with a small number of labeled measured fault samples to improve diagnostic accuracy and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 1 is a flow chart of the digital twin imbalance fault diagnosis method for a gear transmission system according to the present invention;

[0086] Figure 2 It is a structural schematic diagram of a helical gear transmission system of the gear transmission system digital twin imbalance fault diagnosis method of the present invention;

[0087] Figure 3 It is the spectrum diagram of the measured and twin fault signals before and after filtering of the digital twin imbalance fault diagnosis method of the gear transmission system of the present invention;

[0088] Figure 4 It is a time domain waveform diagram of the twin fault signal and the measured fault signal after phase correction in the digital twin imbalance fault diagnosis method of the gear transmission system of the present invention;

[0089] Figure 5 It is a time domain waveform diagram of the fusion signal and the measured signal of the digital twin imbalance fault diagnosis method of the gear transmission system of the present invention. DETAILED DESCRIPTION

[0090] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0091] In the description of the present invention, it should be understood that the terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0092] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal communication between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0093] The present invention discloses a digital twin imbalance fault diagnosis method for a gear transmission system, which is based on the imbalance data set of digital twin technology and data fusion, and comprehensively considers the root mean square value of the signal in the time domain and the center of gravity of the frequency domain spectrum to optimize the model parameters. Figure 1 As shown in Figure 1, the digital twin imbalance fault diagnosis method for the gear transmission system includes the following steps:

[0094] S1, collects the basic parameters of the gear transmission system, including gear module, number of teeth, tooth width, pressure angle, helix angle, input shaft speed, and input torque, calculates the time-varying mesh stiffness excitation of the gears, and establishes a dynamic model of the gear transmission system;

[0095] S2. Based on the dynamic model established in step S1, the radial support stiffness and radial support damping, as well as the axial support stiffness and axial support damping in the dynamic model are optimized according to the measured data to achieve consistency between the model calibration and the virtual and real vibration response results; a corresponding objective function is established, and the existing Beluga Optimization Algorithm (BWO) is used to optimize these two stiffnesses and damping. The specific objective function can be seen in S203.

[0096] Implant different types of local gear faults and simulate to obtain twin signals under different fault types;

[0097] S3, randomly sampling a small amount of gearbox measured fault signals collected on site to obtain segmented measured fault signal samples;

[0098] The gear fault twin signal samples simulated in step S2 and the measured signal samples are processed by low-pass filtering, and the sum of the amplitudes of the two signals at the characteristic frequencies in the frequency domain after filtering is calculated to obtain the adaptive weight coefficient ω, which is used to perform weighted fusion of the time domain signals;

[0099] S4, such as Figure 3 As shown in Figure 1, the filtered measured fault signal and the original twin fault signal are aligned with the impulse peaks in the time domain of the two signals through generalized cross-correlation phase correction;

[0100] like Figure 4 and Figure 5 As shown in the figure, the adaptive weight coefficient ω is used to perform weighted fusion of the original twin fault signal and the corrected measured fault signal in the time domain to obtain a fused fault signal; signal fusion data enhancement is performed based on generalized cross-correlation phase correction and adaptive weight coefficient to expand the fault sample.

[0101] S5, performing overlapping sampling on the fused fault signal to obtain a large number of labeled fused fault signal samples;

[0102] A large number of fused fault signal samples are mixed with a small number of labeled measured fault samples to form a fault sample training set, and a training set is constructed together with the measured normal state samples. During the actual operation of the gearbox, faults rarely occur, so the number of fault samples that can be obtained is relatively small, while the dynamic model can simulate a large number of signals.

[0103] A large number of fused fault signal samples are mixed with a small number of measured fault signal samples in a certain ratio to form a fault training set. Together with the measured normal samples, a balanced training set is constructed. This method fully utilizes the label information of the small number of measured fault signals while effectively increasing the number of training samples.

[0104] S6: Input the training set samples from step S5 into a fault diagnosis network consisting of a feature extractor and a classifier. The feature extractor extracts signal features, and the classifier processes the extracted features to identify the types of faults in the measured samples, including normal gears, tooth root cracks, tooth surface wear, tooth surface spalling, and gear tooth fracture. Simultaneously, a large number of fused samples and a small number of labeled measured fault samples are used to perform fault diagnosis, improving diagnostic accuracy and generalization capabilities.

[0105] In a preferred embodiment of the present invention, Figure 2 As shown in the figure, the method for calculating the time-varying meshing stiffness excitation of the gear and establishing the dynamic model of the gear transmission system using the lumped parameter method is:

[0106] S101, calculate the time-varying meshing stiffness excitation K of the helical gear pair in the gearbox (subway gearbox) during the meshing process m ,for:

[0107]

[0108] Among them, M is the number of gear teeth involved in meshing, N0 is the number of slices along the tooth width direction of the helical gear, represents the stiffness of the Jth slice of the Ith meshing tooth, γ = a, b, s, f, h represent the axial compression stiffness, bending stiffness, shear stiffness, matrix stiffness and Hertzian contact stiffness of the healthy helical gear, respectively, and the subscripts p and g represent the driving gear and the driven gear, respectively;

[0109] S102, based on time-varying mesh stiffness excitation K m , calculate the dynamic meshing force F in the gear transmission system m :

[0110]

[0111] Among them, δ m is the transmission error, C m is the meshing damping, is the first-order derivative of the transmission error;

[0112] S103, based on the dynamic meshing force F m , the lumped parameter method is used to establish the dynamic model of the helical gear transmission system, and its dynamic equation is expressed as:

[0113]

[0114] Among them, M p and M g Represent the mass of the driving wheel and the driven wheel respectively, J p and J g Represents the moment of inertia of the driving wheel and the driven wheel respectively, C xi ,C yi ,C zi Respectively represent the bearing support damping of the driving wheel and the driven wheel in the x, y, and z directions; K xi ,K yi ,K zi Respectively represent the bearing support stiffness of the driving wheel and the driven wheel in the x, y, and z directions, i = p, g; C bw is the torsional damping, K bw is the torsional stiffness, F m is the dynamic meshing force of gears, β b is the base circle helix angle, α n is the meshing angle of the gear pair, R bp R is the base circle radius of the driving wheel, gp is the base circle radius of the passive wheel, T0 is the input torque, T L is the load torque; x i Respectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the x direction; y iRespectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the y direction; z p Respectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the z direction; θ iz They represent the torsional angular acceleration, torsional angular velocity and torsional angular displacement of the driving wheel and the driven wheel in the z direction respectively.

[0115] In a preferred embodiment of the present invention, the radial support stiffness and radial support damping, as well as the axial support stiffness and axial support damping in the dynamic model are optimized based on measured data, different types of local gear faults are implanted, and the method for simulating and obtaining twin signals under different fault types is as follows:

[0116] S201, based on the gear transmission system dynamics model established in step S103 and the measured normal state gear vibration acceleration signal, simulate and obtain the normal state gear vibration acceleration signal under the same working conditions. Calculate the root mean square value rms of the measured normal state signal and the simulated normal state signal in the time domain and the spectral center of gravity f in the low frequency band of the frequency domain for the two signals. C ,for:

[0117]

[0118] Where N is the number of data points in the signal time domain, x j represents the jth data point in the signal time domain, k is the number of data points in the signal frequency domain, fk represents the frequency of the kth data point in the signal frequency domain; s(k) represents the corresponding frequency amplitude;

[0119] S202, assuming that the support stiffness and damping of the driving wheel and the driven wheel in the x and y directions are the same, and the support stiffness and damping in the z direction are the same, simplify the dynamic model, that is, satisfy K xi =K yi =K0,C xi =C yi =C0,K zp =K zg =K1,C zp =C zg = C1; i = p, g, where the subscripts p and g represent the driving wheel and the driven wheel respectively; C xi ,C yi ,C zi Respectively represent the bearing support damping of the driving wheel and the driven wheel in the x, y, and z directions; K xi ,K yi ,K zi Respectively represent the bearing support stiffness of the driving wheel and the driven wheel in the x, y, and z directions;

[0120] S203, based on the measured and simulated normal state signals in step S201, the parameters K1 = [K0, K1] and C = [C0, C1] in the dynamic model are optimized by the Beluga optimization algorithm (BWO). The objective function is expressed as:

[0121]

[0122] Among them, K0, K1 represent the radial support stiffness and axial support stiffness respectively, C0, C1 represent the radial support damping and axial support damping respectively, K1 represents the stiffness vector, C represents the damping vector, Obj represents the objective function of the optimization algorithm, rms s Represents the root mean square value of the simulation signal in the time domain, rms m represents the RMS value of the measured signal in the time domain, Represents the spectral center of the low-frequency band of the simulation signal frequency domain, Represents the spectral center of the low-frequency band of the measured signal in the frequency domain;

[0123] The time domain and frequency domain characteristics of the signal are taken into account at the same time. Different from the weighted summation form of the traditional objective function, the objective function adopts multiplication to achieve the coordinated optimization of multiple indicators, thereby improving the parameter optimization effect. Through multiple iterative optimizations, the optimal parameter K is obtained. * and C * , establish a digital twin model of the gear transmission system.

[0124] S204, in the digital twin model of the gear transmission system, by changing the time-varying meshing stiffness excitation K m , different types of local gear faults are implanted, including: tooth root cracks, tooth surface spalling, tooth surface wear, and tooth fracture, to obtain twin signals under different fault types.

[0125] In a preferred embodiment of the present invention, in step S3, a small amount of gearbox measured fault signals collected on-site are randomly sampled to obtain segmented measured fault signal samples, and the sum of the amplitudes at the characteristic frequencies in the frequency domain of the two signals after filtering is calculated to obtain the adaptive weight coefficient ω as follows:

[0126] S301, from a small number of fault measured signal segments x(t) = [x1, x2, ..., x m ] randomly selects the sampling starting point, intercepts the signal segment with a length of L, and obtains the measured fault signal sample x n (t) = [x e x e+1 ,…,x e+L-1 ], the sampling starting point must satisfy 1≤e≤m-L+1, expressed as:

[0127] e=randint(1,m-L+1)

[0128] Where e represents the sampling start index of the signal segment, m represents the length of the measured signal segment, n represents the number of samples, and randint(·) represents the random integer function;

[0129] The measured fault signal sample x n (t) and twin fault signal sample y n (t) The high-frequency noise in the measured fault signal is removed by the designed low-pass digital filter, and the frequency distribution of the two is ensured to be in the same range, and the filtered measured fault signal is obtained. and twin fault signals

[0130] S302, the measured fault signal after filtering in step S301 and twin fault signals Perform Fourier transform and calculate the two signals at the rotation frequency f r , meshing frequency f m The sum of the amplitudes of its higher harmonics is expressed as:

[0131]

[0132] Among them, A fr Indicates the amplitude at the rotation frequency, vf m Indicates the vth harmonic of the meshing frequency, A vfm represents the amplitude at the meshing frequency v harmonic, and A represents the sum of the amplitudes at the characteristic frequencies of the fault signal;

[0133] S303, based on the sum of the amplitudes of the two fault signal characteristic frequencies in step S302, calculate the adaptive weight coefficient ω, which is:

[0134]

[0135] Among them, A E and A S Represent the sum of the amplitudes of the measured fault signal and the twin fault signal at their characteristic frequencies.

[0136] In a preferred embodiment of the present invention, the filtered measured fault signal and the original twin fault signal are aligned with the impulse peaks in the time domain of the two signals through generalized cross-correlation phase correction. The specific steps are as follows:

[0137] Calculate the filtered measured fault signal and the original twin fault signal y n The cross-correlation function of (t) is:

[0138]

[0139] Among them, Rxy (·) represents the cross-correlation function of two signals, τ is the delay, and t represents time;

[0140] Based on the cross-correlation function, the time delay τ* corresponding to the maximum value is found, and the measured fault signal Perform translation to align the two signal impulse peaks. The expression is:

[0141]

[0142] Where τ* represents the time delay when the cross-correlation function reaches its maximum value, It represents the measured fault signal after phase correction; this effectively reduces the energy loss and the loss of fault characteristics during the signal fusion process.

[0143] In a preferred embodiment of the present invention, the adaptive weight coefficient ω is used to perform weighted fusion on the original twin fault signal and the corrected measured fault signal in the time domain to obtain the fused fault signal Q n (t) is:

[0144]

[0145] Among them, Q n (t) represents the fused fault signal, n is the number of samples; A E and A S Represent the sum of the amplitudes of the measured fault signal and the twin fault signal at their characteristic frequencies.

[0146] The method adaptively adjusts weight coefficients based on signal characteristics and performs signal fusion, minimizing the amplitude difference between the fused signal and the measured signal, highlighting the fault impact characteristics. Furthermore, the fused signal includes signal components from other components in the measured signal, further reducing the distribution difference between the fused signal and the measured signal.

[0147] In a preferred embodiment of the present invention, a convolution block is provided in the feature extractor, and the convolution block includes a convolution layer, a batch normalization layer, and a pooling layer, and the expression is:

[0148]

[0149] Among them, H i is the input of the convolutional layer, H r is the output of the convolutional layer, M is the number of input feature maps, r is the number of output feature maps, b is the bias, and μ is H r The average value of H r The standard deviation of , γ1 is the scaling parameter, β1 is the translation parameter, G is the activation function, and max represents the maximum value operation; K ir is the weight of the rth convolution kernel for the i-th input feature map, br is the bias matrix of the rth convolution kernel, is the output of the convolutional layer, is the output after the batch normalization layer;

[0150] After the input passes through two convolution blocks, multi-scale feature extraction is performed with the help of convolution blocks with two convolution kernels of different sizes, and then spliced ​​from the channel dimension to obtain the splicing feature H, which is:

[0151] H=concat(H1,H2)

[0152] Among them, H1 is the output of the first convolution block, H2 is the output of the second convolution block, and concat(·) means splicing from the channel dimension;

[0153] After the concatenated feature H passes through a convolution block, the features of each channel are adaptively weighted through the attention weighted network:

[0154]

[0155] in, It is a vector composed of the mean of each channel of the splicing feature H; represents the mean of the chth channel of the concatenated feature H, where ch is the number of channels; B is the weight vector output by the attention weighted network; f a (·) represents the mapping relationship in the attention weighted network, α ch is the weight corresponding to each channel; H ω represents the weighted output feature map.

[0156] In a preferred embodiment of the present invention, the classifier includes two fully connected layers. The classifier output is converted into a probability value of [0, 1] after passing through the Softmax function to realize the identification of the fault type. The expression is:

[0157] Y = Softmax(G(∑ω1x+b))

[0158]

[0159] Among them, x is the input, ω1 is the weight matrix, Y is the output, b is the bias, G is the activation function, R is the fault category, z u is the u-th output value of the fault diagnosis network.

[0160] The present invention also provides a digital twin imbalance fault diagnosis system for a gear transmission system, comprising a data acquisition unit and a processing unit. The data acquisition unit is used to collect basic parameters of the gear transmission system, and the output end of the data acquisition unit is connected to the input end of the processing unit.

[0161] The processing unit executes the method of the present invention to achieve fault diagnosis of the gear transmission system.

[0162] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0163] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A digital twin imbalance fault diagnosis method for a gear transmission system, characterized in that: The steps include: S1, collect the basic parameters of the gear transmission system, calculate the time-varying meshing stiffness excitation of the gears, and establish the dynamic model of the gear transmission system; S2, based on the dynamic model established in step S1, optimizing the radial support stiffness and radial support damping, as well as the axial support stiffness and axial support damping in the dynamic model according to measured data, to achieve consistency between model calibration and virtual and real vibration response results; Implant different types of local gear faults and simulate to obtain twin signals under different fault types; S3, randomly sampling the gearbox measured fault signal collected on site to obtain segmented measured fault signal samples; The gear fault twin signal sample and the measured signal sample obtained by simulation in step S2 are processed by low-pass filtering, and the sum of the amplitudes of the two signals at the characteristic frequencies in the frequency domain after filtering is calculated to obtain the adaptive weight coefficient ω; S4, aligning the impulse peaks of the filtered measured fault signal and the original twin fault signal in the time domain through generalized cross-correlation phase correction; The adaptive weight coefficient ω is used to perform weighted fusion on the original twin fault signal and the corrected measured fault signal in the time domain to obtain a fused fault signal. S5, performing overlapping sampling on the fused fault signal to obtain labeled fused fault signal samples; The fused fault signal samples are mixed with the labeled measured fault samples to form a fault sample training set, and the training set is constructed together with the measured normal state samples. S6, input the training set samples in step S5 into the fault diagnosis network composed of a feature extractor and a classifier, the feature extractor extracts the features of the signal, and the classifier processes the extracted features to identify the fault type of the measured sample.

2. The gear transmission system digital twin imbalance fault diagnosis method according to claim 1, characterized in that: The method for calculating the time-varying meshing stiffness excitation of gears and establishing the dynamic model of the gear transmission system using the lumped parameter method is as follows: S101, calculate the time-varying meshing stiffness excitation K of the helical gear pair in the gearbox during the meshing process m ,for: Among them, M is the number of gear teeth involved in meshing, N0 is the number of slices along the tooth width direction of the helical gear, represents the stiffness of the Jth slice of the Ith meshing tooth, γ = a, b, s, f, h represent the axial compression stiffness, bending stiffness, shear stiffness, matrix stiffness and Hertzian contact stiffness of the healthy helical gear, respectively, and the subscripts p and g represent the driving gear and the driven gear, respectively; S102, based on time-varying mesh stiffness excitation K m , calculate the dynamic meshing force F in the gear transmission system m : Among them, δ m is the transmission error, C m is the meshing damping, is the first-order derivative of the transmission error; S103, based on the dynamic meshing force F m , the lumped parameter method is used to establish the dynamic model of the helical gear transmission system, and its dynamic equation is expressed as: Among them, M p and M g Represent the mass of the driving wheel and the driven wheel respectively, J p and J g Represents the moment of inertia of the driving wheel and the driven wheel respectively, C xi ,C yi ,C zi Respectively represent the bearing support damping of the driving wheel and the driven wheel in the x, y, and z directions; K xi ,K yi ,K zi Respectively represent the bearing support stiffness of the driving wheel and the driven wheel in the x, y, and z directions, i = p, g; C bw is the torsional damping, K bw is the torsional stiffness, F m is the dynamic meshing force of gears, β b is the base circle helix angle, α n is the meshing angle of the gear pair, R bp R is the base circle radius of the driving wheel, gp is the base circle radius of the passive wheel, T0 is the input torque, T L is the load torque; x i Respectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the x direction; y i Respectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the y direction; z p Respectively represent the vibration acceleration, vibration velocity and vibration displacement of the driving wheel and the driven wheel in the z direction; θ iz They represent the torsional angular acceleration, torsional angular velocity and torsional angular displacement of the driving wheel and the driven wheel in the z direction respectively.

3. The gear transmission system digital twin imbalance fault diagnosis method according to claim 1, characterized in that: The following method is used to optimize the bearing radial support stiffness and radial support damping, as well as the axial support stiffness and axial support damping in the dynamic model based on measured data, implant different types of gear local faults, and simulate and obtain twin signals under different fault types: S201, based on the measured normal state signal and the simulated normal state signal, calculate the root mean square value rms of the two signals in the time domain and the spectral center of gravity f in the low frequency band of the frequency domain C ,for: Where N is the number of data points in the signal time domain, x j represents the jth data point in the signal time domain, K is the number of data points in the signal frequency domain, and f k represents the frequency of the kth data point in the signal frequency domain; s(k) represents the corresponding frequency amplitude; S202, assuming that the support stiffness and damping of the driving wheel and the driven wheel in the x and y directions are the same, and the support stiffness and damping in the z direction are the same, simplify the dynamic model, that is, satisfy K xi =K yi =K0,C xi =C yi =C0,K zp =K zg =K1,C zp =C zg = C1; i = p, g, where the subscripts p and g represent the driving wheel and the driven wheel respectively; C xi ,C yi ,C zi Respectively represent the bearing support damping of the driving wheel and the driven wheel in the x, y, and z directions; K xi ,K yi ,K zi Respectively represent the bearing support stiffness of the driving wheel and the driven wheel in the x, y, and z directions; S203, based on the measured and simulated normal state signals in step S201, the parameters K1 = [K0, K1] and C = [C0, C1] in the dynamic model are optimized by the Beluga optimization algorithm (BWO). The objective function is expressed as: Among them, K0, K1 represent the radial support stiffness and axial support stiffness respectively, C0, C1 represent the radial support damping and axial support damping respectively, K1 represents the stiffness vector, C represents the damping vector, Obj represents the objective function of the optimization algorithm, rms s Represents the root mean square value of the simulation signal in the time domain, rms m represents the root mean square value of the measured signal in the time domain, f c s Represents the spectral center of the low-frequency band of the simulation signal in the frequency domain, f c m Represents the spectral center of the low-frequency band of the measured signal in the frequency domain; S204, in the digital twin model of the gear transmission system, by changing the time-varying meshing stiffness excitation K m , different types of local gear faults are implanted, including: tooth root cracks, tooth surface spalling, tooth surface wear, and tooth fracture, to obtain twin signals under different fault types.

4. The gear transmission system digital twin imbalance fault diagnosis method according to claim 1, characterized in that: In step S3, a small amount of gearbox fault signals collected on site are randomly sampled to obtain segmented fault signal samples, and the sum of the amplitudes at the characteristic frequencies in the frequency domain of the two signals after filtering is calculated to obtain the adaptive weight coefficient ω as follows: S301, from a small number of fault measured signal segments x(t) = [x1, x2, ..., x m ] randomly selects the sampling starting point, intercepts the signal segment with a length of L, and obtains the measured fault signal sample x n (t) = [x e x e+1 ,…,x e+L-1 ], the sampling starting point must satisfy 1≤e≤m-L+1, expressed as: e=randint(1,m-L+1) Where e represents the sampling start index of the signal segment, m represents the length of the measured signal segment, n represents the number of samples, and randint(·) represents the random integer function; The measured fault signal sample x n (t) and twin fault signal sample y n (t) The high-frequency noise in the measured fault signal is removed by the designed low-pass digital filter, and the frequency distribution of the two is ensured to be in the same range, and the filtered measured fault signal is obtained. and twin fault signals S302, the measured fault signal after filtering in step S301 and twin fault signals Perform Fourier transform and calculate the two signals at the rotation frequency f r , meshing frequency f m The sum of the amplitudes of its higher harmonics is expressed as: Among them, A fr Indicates the amplitude at the rotation frequency, vf m Indicates the vth harmonic of the meshing frequency, A vfm represents the amplitude at the meshing frequency v harmonic, and A represents the sum of the amplitudes at the characteristic frequencies of the fault signal; S303, based on the sum of the amplitudes of the two fault signal characteristic frequencies in step S302, calculate the adaptive weight coefficient ω, which is: Among them, A E and A S Represent the sum of the amplitudes of the measured fault signal and the twin fault signal at their characteristic frequencies.

5. The gear transmission system digital twin imbalance fault diagnosis method according to claim 1, characterized in that: The filtered measured fault signal and the original twin fault signal are aligned with each other in the time domain by generalized cross-correlation phase correction. The specific steps are as follows: Calculate the filtered measured fault signal and the original twin fault signal y n The cross-correlation function of (t) is: Among them, R xy (·) represents the cross-correlation function of two signals, τ is the delay, and t represents time; Based on the cross-correlation function, the time delay τ* corresponding to the maximum value is found, and the measured fault signal Perform translation to align the two signal impulse peaks. The expression is: Where τ* represents the time delay when the cross-correlation function reaches its maximum value, Indicates the measured fault signal after phase correction.

6. The gear transmission system digital twin imbalance fault diagnosis method according to claim 5, characterized in that: The adaptive weight coefficient ω is used to perform weighted fusion of the original twin fault signal and the corrected measured fault signal in the time domain to obtain the fused fault signal Q n (t) is: Among them, Q n (t) represents the fused fault signal, n is the number of samples; A E and A S Represent the sum of the amplitudes of the measured fault signal and the twin fault signal at their characteristic frequencies.

7. The gear transmission system digital twin imbalance fault diagnosis method according to claim 1, characterized in that: The feature extractor is provided with a convolution block, which includes a convolution layer, a batch normalization layer and a pooling layer, and the expression is: Among them, H l is the input of the convolutional layer, H r is the output of the convolutional layer, num is the number of input feature maps, r is the number of output feature maps, that is, the number of convolution kernels; b is the bias, μ is H r The average value of H r The standard deviation of , γ1 is the scaling parameter, β1 is the translation parameter, G is the activation function, and max represents the maximum value operation; K lr is the weight of the rth convolution kernel for the lth input feature map, b r is the bias matrix of the rth convolution kernel, is the output of the convolutional layer, is the output after the batch normalization layer; After the input passes through two convolution blocks, multi-scale feature extraction is performed with the help of convolution blocks with two convolution kernels of different sizes, and then spliced ​​from the channel dimension to obtain the splicing feature H, which is: H=concat(H1,H2) Among them, H1 is the output of the first convolution block, H2 is the output of the second convolution block, and concat(·) means splicing from the channel dimension; After the concatenated feature H passes through a convolution block, the features of each channel are adaptively weighted through the attention weighted network: in, It is a vector composed of the mean of each channel of the splicing feature H; represents the mean of the chth channel of the concatenated feature H, where ch is the number of channels; B is the weight vector output by the attention weighted network; f a (·) represents the mapping relationship in the attention weighted network, α ch is the weight corresponding to each channel; H ω represents the weighted output feature map.

8. The gear transmission system digital twin imbalance fault diagnosis method according to claim 1, characterized in that: The classifier includes two fully connected layers. The classifier output is converted into a probability value of [0, 1] after passing through the Softmax function to identify the fault type. The expression is: Y = Softmax(G(∑ω1x+b)) Among them, x is the input, ω1 is the weight matrix, Y is the output, b is the bias, G is the activation function, R is the fault category, z u is the u-th output value of the fault diagnosis network.

9. A digital twin imbalance fault diagnosis system for a gear transmission system, characterized in that: It includes a data acquisition unit and a processing unit, wherein the data acquisition unit is used to acquire basic parameters of the gear transmission system, and the output end of the data acquisition unit is connected to the input end of the processing unit; The processing unit executes the method according to any one of claims 1 to 8 to implement fault diagnosis of the gear transmission system.

Images