Active control method of gearbox gear transmission noise and transmission device

Abstract

The application discloses a gearbox gear transmission noise active control method and a transmission device, and relates to the technical field of electric vehicle transmission devices, which comprises the following steps: extracting the howling frequency in the vibration signal of a gearbox shell to determine the corresponding meshing order, comparing the meshing order with the number of teeth of each gear pair in the gearbox to locate the gear pair of the howling source; performing frequency spectrum analysis on the vibration signal to determine the error type of the gear pair of the howling source and the howling degree quantization index; determining the control parameters of the harmonic current based on a control parameter mapping model according to the error type, the howling degree quantization index and current working condition parameters; and synthesizing the compensation harmonic current according to the control parameters, and superimposing the compensation harmonic current into the motor driving current to offset the meshing excitation of the gear pair of the howling source. The application solves the technical problem that the existing gearbox gear transmission noise active control technology lacks a full-process self-adaptive closed loop, which leads to the fact that the frequency and phase cannot be accurately and real-timely matched, and the consistency of the noise reduction effect is poor.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle transmission technology, specifically to an active control method and transmission device for gearbox gear transmission noise. Background Technology

[0002] During operation, the gear transmission system generates periodic meshing excitation forces due to transmission errors between the gear teeth during meshing. These excitation forces are transmitted to the gearbox housing via the gear shaft and bearings, causing the housing to vibrate and radiate noise. Among these noises, whine is the most prominent and easily perceived by the driver. Its frequency components are mainly the meshing frequency of the gear pair and its harmonics, exhibiting obvious order characteristics and a narrow-band spectrum.

[0003] However, existing active control technologies still have the following shortcomings: lack of precise location of howling sources; mismatch of fixed frequency compensation strategies when speed changes; neglect of electromechanical transmission phase delay leading to phase reversal failure; lack of automatic error type identification and howling degree quantification methods, with control parameters relying on manual experience to look up tables; and unclear compensation current injection levels, making it difficult to deploy and operate with low latency and high fidelity in existing motor controller architectures. Summary of the Invention

[0004] This invention provides an active control method and transmission device for gearbox gear transmission noise. It solves the technical problem that existing active control technologies for gearbox gear transmission noise lack a full-process adaptive closed-loop, resulting in inaccurate real-time matching of frequency and phase, and poor consistency in noise reduction effects.

[0005] In view of the above problems, the present invention provides an active control method for gearbox gear transmission noise, the method comprising: Vibration signals of the gearbox housing are collected by vibration sensors installed on the outer surface of the gearbox housing; The vibration signal is subjected to order analysis to extract the whistling frequency component in the vibration signal, the meshing order corresponding to the whistling frequency component is determined, and the meshing order is compared with the number of teeth of each gear pair in the gearbox to locate the gear pair from which the whistling originates. The vibration signal is subjected to spectral analysis, and based on the spectral analysis results, the error type and quantification index of the gear pair from which the whistling originates are determined. Based on a pre-trained control parameter mapping model, the control parameters of the harmonic current are determined according to the error type, the quantification index of the howling degree, and the current operating parameters. The control parameters include the injection order, injection amplitude, and injection phase. The compensation harmonic current is synthesized according to the control parameters and superimposed on the motor drive current to counteract the meshing excitation of the gear pair from which the howling originates.

[0006] The present invention also provides an active control transmission device for gearbox gear transmission noise, comprising: The vibration signal acquisition module is used to acquire vibration signals of the gearbox housing through vibration sensors installed on the outer surface of the gearbox housing; The whistling sound source localization module is used to perform order analysis on the vibration signal, extract the whistling frequency component in the vibration signal, determine the meshing order corresponding to the whistling frequency component, and locate the whistling source gear pair by comparing the meshing order with the number of teeth of each gear pair in the gearbox. The error identification and quantification module is used to perform spectrum analysis on the vibration signal and determine the error type and quantification index of the whistling source gear pair based on the spectrum analysis results. The control parameter decision module is used to determine the control parameters of the harmonic current based on the pre-trained control parameter mapping model, according to the error type, the quantification index of the howling degree, and the current operating condition parameters. The control parameters include the injection order, injection amplitude, and injection phase. The compensation current synthesis and injection module is used to synthesize compensation harmonic current according to the control parameters and superimpose the compensation harmonic current into the motor drive current to counteract the meshing excitation of the gear pair from which the howling originates.

[0007] One or more technical solutions provided in this invention have at least the following technical effects or advantages: First, this invention fundamentally eliminates the interference of real-time changes in vehicle speed or motor speed on frequency identification by converting absolute frequency into a constant meshing order for sound source localization. Simultaneously, by introducing a preset matching threshold tolerance mechanism in the tooth count comparison, it effectively compensates for minor calculation deviations caused by spectral resolution limitations and frequency drift, avoiding misjudgment or missed detection of howling sources.

[0008] Secondly, this invention utilizes the inherent mapping relationship between sideband amplitude distribution and gear physical error types, combined with a pre-trained error type classification model, to automatically identify specific error categories such as tooth profile angle error, tooth direction angle error, tooth pitch deviation, and local faults. Simultaneously, by constructing the sideband energy ratio as a quantitative indicator of howling intensity, this indicator is essentially decoupled from speed and load, remaining stable and monotonic across the entire operating range, and highly consistent with subjective human hearing, providing an objective and continuous evaluation metric for subsequent on-demand fine control.

[0009] Subsequently, this invention inputs the error type, the quantification index of the howling degree, and the real-time operating condition parameters into the pre-trained control parameter mapping model. The model synchronously outputs the optimal injection amplitude and phase, enabling the harmonic current compensation parameters to adaptively and dynamically match according to the current fault mode and operating status. This replaces the traditional manual calibration table lookup method, significantly reducing the calibration workload and improving the completeness of operating condition coverage.

[0010] Finally, this invention ensures that the compensation frequency always changes synchronously with the actual meshing frequency by using a dynamic calculation method that multiplies the injection order by the real-time rotational frequency, thus maintaining effective noise reduction across the entire speed range. By using feedforward compensation in the injection phase to compensate for the complete electromechanical transmission phase delay from the current command to the meshing point, it ensures that the compensation force is precisely out of phase with the excitation force at the meshing point, achieving the maximum cancellation effect. By completing the compensation harmonic current superposition in the innermost current loop of the motor controller, it fully utilizes the high bandwidth and fast response characteristics of the current loop to achieve low-latency, high-fidelity engineered injection execution.

[0011] In summary, this invention solves the technical problem that existing active noise control technologies for gear transmissions lack a full-process adaptive closed loop, resulting in inaccurate real-time matching of frequency and phase and poor consistency in noise reduction effects. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating an active control method for gear transmission noise in a gearbox, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an active control transmission device for gearbox gear transmission noise provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: Vibration signal acquisition module 11, howling sound source localization module 12, error identification and quantification module 13, control parameter decision module 14, compensation current synthesis and injection module 15. Detailed Implementation

[0014] This invention provides an active control method and transmission device for gearbox gear transmission noise, which specifically solves the technical problem that existing active control technologies for gearbox gear transmission noise lack a full-process adaptive closed loop, resulting in inaccurate real-time matching of frequency and phase and poor consistency of noise reduction effect.

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0016] Example 1, as Figure 1 As shown, the present invention provides an active control method for gearbox gear transmission noise, the method comprising: S100: Vibration signals of the gearbox housing are collected by a vibration sensor installed on the outer surface of the gearbox housing.

[0017] In the active control of whistling in a gearbox gear transmission system, the acquisition of the original vibration signal is the data source for the entire control method. However, the gearbox housing surface is simultaneously subjected to broadband vibrations generated by the meshing of multiple gear pairs, bearing rotational vibrations, and torsional vibrations from the engine input. These multiple vibration sources couple with each other, resulting in vibration signals from the outer surface of the housing exhibiting strong background noise and overlapping frequency bands. If the sensor placement is inappropriate, the whistling characteristic signal can easily be submerged by vibration noise along the structural transmission path, making it difficult to extract the sensitive signal that truly reflects the gear meshing state. Therefore, accurately acquiring the vibration signal of the gearbox housing is a crucial problem that needs to be solved in this step.

[0018] Vibration sensors are installed at predetermined measuring points on the outer surface of the gearbox housing. The predetermined measuring points are located at the outer positions of the housing corresponding to each bearing housing or in the housing projection area of ​​the gear pair meshing center line. The vibration sensors are piezoelectric accelerometers, which are fixed to the housing surface by stud connection, magnetic base or adhesive to ensure that there is a sufficiently high contact stiffness between the sensor and the housing, and to ensure high-fidelity transmission of vibration signals.

[0019] During data acquisition, the vibration sensor picks up the broadband time-domain vibration acceleration signal generated by the meshing excitation of the internal gear pairs in the gearbox housing. This signal is then subjected to charge amplification and anti-aliasing filtering by the signal conditioning circuit to filter out high-frequency noise and avoid spectral aliasing. The conditioned analog signal is then converted from analog to digital by the data acquisition card to generate a time-domain discrete vibration signal sequence that can be read by the digital signal processor. The sampling frequency is set according to the highest meshing frequency of the gearbox gears, and is at least 2.56 times that highest meshing frequency. The acquisition duration covers the complete change cycle of the gearbox from the initial speed to the target speed.

[0020] The process of acquiring vibration signals from the gearbox housing in this embodiment of the invention solves the technical problem of how to efficiently obtain the original vibration signal containing whistling characteristics from the outer surface of the housing during gearbox operation. Through reasonable location selection, firm installation, signal conditioning, and sufficient sampling, the authenticity, integrity, and high signal-to-noise ratio of the original vibration signal are ensured, providing a reliable data foundation for accurate extraction of whistling frequency components in subsequent order analysis.

[0021] S200: Perform order analysis on the vibration signal, extract the whistling frequency component from the vibration signal, determine the meshing order corresponding to the whistling frequency component, and compare the meshing order with the number of teeth of each gear pair in the gearbox to locate the gear pair from which the whistling originates.

[0022] In multi-stage gear transmission systems, due to the presence of multiple simultaneously rotating and coupled gear pairs, the vibration signal on the housing surface is often a mixed signal from multiple excitation sources. The problem to be solved in this step is how to efficiently, accurately, and with engineering tolerance extract the most harmful squealing component from the complex vibration signal containing strong background noise and multiple frequency components under real-time changing speed conditions, and precisely pinpoint its physical source to a specific pair of gears, thus providing a unique and reliable target for subsequent active noise reduction control.

[0023] Step S200 in the method provided in this embodiment of the invention includes: The vibration signal is transformed by time-frequency conversion to convert the time-domain vibration signal into a frequency-domain signal, thereby obtaining the spectrum of the vibration signal; Frequency components with energy amplitudes exceeding a preset energy threshold are extracted from the spectrum as candidate frequency components for whistling. The candidate frequency components for whistling are compared with the theoretical meshing frequencies of each gear pair in the gearbox. The candidate frequency component for whistling that matches any theoretical meshing frequency is determined as the whistling frequency component. The whistling frequency component with the highest energy amplitude is selected as the target whistling frequency component. The theoretical meshing frequency is equal to the number of teeth of the driving gear of each gear pair multiplied by the rotational frequency corresponding to the current rotational speed of the shaft where the driving gear of the gear pair is located. The ratio of the frequency value of the target squealing frequency component to the rotational frequency of the shaft where the driving gear of the gear pair that matches the theoretical meshing frequency of the target squealing frequency component is located is calculated to obtain the meshing order corresponding to the target squealing frequency component. The meshing order is compared one by one with the number of teeth of the driving gear in each gear pair in the gearbox, and the gear pair whose number of teeth of the driving gear matches the meshing order is identified as the gear pair from which the whistling sound originates.

[0024] In this embodiment of the invention, the vibration signal is first subjected to time-frequency transformation, converting the time-domain vibration signal into a frequency-domain signal to obtain the vibration signal spectrum. The time-domain vibration waveform collected by the vibration sensor, which varies with time, is mapped to the frequency dimension using a Fast Fourier Transform (FFT). The resulting spectrum plot shows the horizontal axis representing frequency and the vertical axis representing vibration energy amplitude. This processing aims to decompose the complex vibration signal, which is difficult to distinguish in the time domain, into a series of discrete frequency components, thereby facilitating the subsequent identification of periodic characteristic frequencies related to gear meshing from a frequency domain perspective. This provides the data basis for order analysis and howling frequency extraction.

[0025] Furthermore, to avoid misjudging background noise or low-energy random vibrations as howling, a preset energy threshold is introduced. Only frequency points with amplitudes exceeding the preset energy threshold will be considered as candidate frequencies for howling.

[0026] The preset energy threshold is set using the basis noise multiple method. The root mean square amplitude of the frequency region in the current vibration signal spectrum that clearly does not belong to any gear meshing frequency, its harmonics, or sidebands is calculated as the basis noise level. The preset energy threshold is then set as a specified multiple of the basis noise. The selection of the specified multiple should be combined with the sensor's signal-to-noise ratio (SNR) characteristics; 1.5 to 2.5 times is suitable for high SNR, and 3 to 5 times for low SNR, ensuring that the true howling signal is not missed without introducing excessive noise interference.

[0027] For example, in the current spectrum, the average amplitude measured in the high-frequency range of 2500Hz~3000Hz is 0.04g, and the signal-to-noise ratio of the sensor is calibrated as good, with a multiplier of 2.5. Then the preset energy threshold = 0.04g × 2.5 = 0.10g, and only frequency components with amplitudes exceeding 0.10g are marked as candidate frequency components for howling.

[0028] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0029] According to the principle of gear transmission, each gear pair will inevitably generate a theoretical meshing frequency related to its number of teeth and rotational speed during normal meshing. The candidate whistling frequencies are compared one-to-one with the theoretical meshing frequencies of all gear pairs. The candidate whistling frequency component that matches any theoretical meshing frequency is determined as the whistling frequency component, rather than random interference. When multiple candidate whistling frequencies that meet the matching conditions exist at the same time, the candidate whistling frequency with the highest amplitude often represents the main contributor to vibration energy and the most prominent noise source. Therefore, this candidate whistling frequency component with the highest amplitude is prioritized and used as the target whistling frequency component to ensure the targeted nature of active control.

[0030] It should be noted that during the actual operation of the gearbox, multiple gear pairs may simultaneously generate whistling noise, meaning that there are multiple whistling frequency components in the spectrum with energy amplitudes exceeding a preset energy threshold. In this embodiment of the invention, only the whistling frequency component with the highest energy amplitude is selected as the target whistling frequency component within the current control cycle, and subsequent error type determination, control parameter determination, and harmonic current compensation injection are performed. After the active control of the current control cycle is completed, order analysis is performed again in the next control cycle. After removing the locked highest energy whistling frequency component, the second highest energy whistling frequency component is identified from the remaining whistling candidate frequency components, and it is located and controlled according to the same process. This iterative cycle, through time-sharing polling, applies active suppression to each whistling source sequentially within multiple consecutive control cycles, thereby achieving comprehensive suppression of multiple whistling sources within the gearbox.

[0031] Furthermore, the ratio of the frequency value of the target squealing frequency component to the rotational frequency of the shaft where the driving gear of the gear pair corresponds to the theoretical meshing frequency that matches the target squealing frequency component is located is calculated to obtain the meshing order corresponding to the target squealing frequency component. According to gear dynamics theory, the quotient obtained by dividing the meshing frequency by the rotational frequency is numerically equal to the number of teeth of the driving gear of the gear pair. This process of calculating the meshing order normalizes the absolute frequency value, which originally varied with the rotational speed, into a constant ratio independent of the rotational speed. This meshing order is an inherent identifier of the gear pair, which does not change with the vehicle speed or motor speed, and is a key intermediate parameter for subsequent precise positioning of the gear pair.

[0032] Finally, the calculated meshing order is compared one by one with the number of teeth on the driving gear of each gear pair in the gearbox. The gear pair whose number of driving gear teeth matches the meshing order is identified as the source of the whistling. This process does not rely on external speed sensors or additional markers, but is entirely based on the characteristics of the signal itself and known mechanical parameters, and has extremely high reliability and engineering applicability.

[0033] The meshing order is compared one by one with the number of teeth of the driving gear in each gear pair in the gearbox. The gear pairs whose number of teeth of the driving gear matches the meshing order are identified as the gear pairs from which the whistling sound originates, including: Obtain the number of teeth on the driving gear of each gear pair in the gearbox and generate a list of the number of teeth on each gear pair; Calculate the difference between the meshing order and the number of teeth of each driving gear in the gear pair tooth count list, and determine the gear pair with the absolute value of the difference being less than a preset matching threshold as the source gear pair of the whistling.

[0034] Transmissions typically contain multi-stage gear drives. The number of teeth of all possible driving gears can be read in advance from the hardware design parameters of the transmission and constructed into an ordered or unordered list. This allows the comparison process to be completed quickly by simply calling the list of teeth, without having to traverse the complex gear meshing logic in real time, which significantly improves data processing efficiency.

[0035] In practical engineering applications, due to fluctuations in rotational speed, limitations in frequency resolution, or small cumulative errors in signal acquisition and calculation, the calculated target squeal order often cannot be exactly equal to the precise integer value of the number of teeth. Therefore, this embodiment of the invention introduces a preset matching threshold. The absolute value of the difference between the order and the number of teeth in the list is calculated, and it is determined whether this absolute value falls within the threshold range. If it falls within the threshold range, the match is considered successful. The setting of the matching threshold fully considers engineering error tolerance, effectively avoiding misjudgment or missed judgment of squeal sources due to small numerical deviations, and greatly enhancing the adaptability and robustness of the control method under complex automotive operating conditions.

[0036] The preset matching threshold is an adaptive threshold based on the speed fluctuation rate. The short-time standard deviation of the drive shaft speed signal is calculated in real time to characterize the current speed fluctuation level. The preset matching threshold = baseline threshold + speed fluctuation compensation amount. The baseline threshold is set based on FFT frequency resolution, and the speed fluctuation compensation amount is calculated by dividing the current speed standard deviation by the current average speed, multiplying by the target number of teeth, and then multiplying by a safety factor. The advantage of this method is that the threshold tightens when the speed is stable to ensure positioning accuracy, and automatically widens when the speed fluctuates drastically to ensure no missed judgments.

[0037] For example, if the FFT frequency resolution is 2Hz and the current rotational frequency is 50Hz, the relative error introduced by the frequency estimation is approximately 2 / 50 = 4%. For a gear pair with 30 teeth, the order calculation deviation is approximately 30 × 4% = 1.2, which is much larger than the ideal value. Therefore, the baseline threshold should not be too small. Considering the actual situation, the baseline threshold is set to 0.5. If the current speed standard deviation is 2.5 RPM and the average speed is 1500 RPM, then the volatility = 2.5 / 1500 ≈ 0.17%. The target number of teeth is 30, and the compensation amount = 30 × 0.17% × 1.2 ≈ 0.06. The final preset matching threshold = 0.5 + 0.06 = 0.56.

[0038] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0039] This invention completely eliminates the influence of vehicle speed or motor speed changes on frequency identification by converting absolute frequency into a constant meshing order for comparison, enabling the whistling source localization method to work stably across the entire speed range. By using a gear pair tooth count list and a preset matching threshold, it not only simplifies the comparison calculation but also effectively compensates for the insufficient spectral resolution of the Fast Fourier Transform (FFT) and the calculation errors caused by frequency drift, avoiding localization failure due to minor numerical deviations and greatly improving the robustness of identification under harsh working conditions.

[0040] S300: Perform spectrum analysis on the vibration signal, and determine the error type and quantification index of the gear pair from which the whistling originates based on the spectrum analysis results.

[0041] Current technologies still lack the ability to automatically identify the root causes of whistling and a continuous, normalized, and quantitative evaluation method for the severity of whistling. Traditional methods often only focus on the magnitude of vibration amplitude, failing to distinguish whether whistling is caused by tooth profile deviation or assembly eccentricity, and also failing to eliminate the interference of operating conditions such as speed and load on the evaluation scale. This results in subsequent active control relying on fixed strategies or experience-based lookup tables, lacking adaptability and specificity, and making it difficult to achieve optimal noise reduction effects.

[0042] Step S300 in the method provided in this embodiment of the invention includes: The vibration signal is subjected to spectral analysis to extract the amplitude of each order sideband on both sides of the meshing frequency of the gear pair from which the whistling originates, and the sideband amplitude sequence is formed by arranging them in order of order. The sideband amplitude sequence is input into a pre-trained error type classification model, which outputs the error type of the gear pair from which the howling originates. The energy ratio of the sidebands is obtained by calculating the ratio of the sum of the energy of each order sideband on both sides of the fundamental meshing frequency of the gear pair from which the whistling originates to the energy value at the fundamental meshing frequency. The sideband energy ratio is used as a quantitative indicator of the degree of howling from the gear pair from which the howling originates.

[0043] In this embodiment of the invention, a refined spectral analysis is first performed on the vibration signal of the located source gear pair causing the howling. According to amplitude modulation-frequency modulation theory, when there are manufacturing or installation errors in the gear, a series of sidebands with rotational frequencies will be generated on both sides of its meshing frequency. The amplitude distribution of these sidebands contains key information about the error type. The specific amplitudes of these sidebands are extracted from the spectrum and then arranged according to their order to form an ordered numerical sequence. This sequence, as a set of structured feature vectors, carries complete information about the vibration modulation of the gear pair, providing quantitative input for subsequent error pattern recognition. The obtained sideband amplitude sequence is input into a pre-trained error type classification model, which outputs the error type of the gear pair causing the howling, thereby achieving intelligent diagnosis of the physical root cause of the howling.

[0044] Furthermore, in order to quantify the severity of the squealing, this embodiment of the invention calculates the sum of the energy of each order of sidebands on both sides of the fundamental meshing frequency of the gear pair from which the squealing originates, and calculates the ratio of this sum of energy to the energy value at the fundamental meshing frequency. This ratio is used as the sideband energy ratio. The physical meaning of this ratio lies in reflecting the degree of dispersion of vibration energy on the modulation sidebands on both sides of the carrier frequency. When the gear meshing is perfect and error-free, the sideband energy is extremely low, and this ratio is close to 0. When there is a significant error in the gears leading to an enhanced vibration modulation effect, the sideband energy increases significantly, and this ratio increases accordingly. Therefore, this ratio is a relative quantity independent of rotational speed and load, and can stably reflect the current health status of the gear pair and the severity of the squealing.

[0045] Finally, the calculated sideband energy ratio is set as the quantification index of the whistling intensity of the gear pair from which the whistling originates. The quantification index of whistling intensity is basically decoupled from the speed and load, and can stably and objectively reflect the real-time severity level of the whistling noise across the entire operating range.

[0046] The pre-training process of the error type classification model includes: Multiple sets of historical calibration data are acquired. Each set of historical calibration data includes sideband amplitude sequence samples and corresponding error type labels. The error type labels include at least tooth profile angle error, tooth direction angle error, tooth pitch deviation, and local fault. An initial error type classification model is constructed, which includes a physical prior convolutional layer, a pooling layer, and a fully connected output layer. The initial weights of the convolution kernels of the physical prior convolutional layer are predetermined based on the sideband spectral structure of the gear fault vibration signal. The convolution kernels are comb filter type convolution kernels, and the passband positions of the comb filter type convolution kernels correspond to the frequency positions of each order of the sideband. The sideband amplitude sequence samples are input into the physical prior convolutional layer of the initial error type classification model for feature extraction to obtain a first feature map. The first feature map is then input into the pooling layer of the initial error type classification model for dimensionality reduction to obtain a second feature map. The second feature map is then flattened and input into the fully connected output layer of the initial error type classification model to output the predicted error type. Using the classification cross-entropy loss between the predicted error type and the error type label as the training objective, supervised training is performed on the initial error type classification model, and the network parameters of the initial error type classification model are updated until the verification convergence is obtained, thus obtaining the pre-trained error type classification model.

[0047] In the pre-training process of the error type classification model in this invention, the training dataset is first constructed. Multiple sets of historical calibration data are acquired, each set containing a sideband amplitude sequence sample and the corresponding error type label. It should be noted that the error type label covers at least the four typical error categories commonly found in gear transmission systems: tooth profile angle error, tooth direction angle error, tooth pitch deviation, and local faults. The sources of these data samples include, but are not limited to, actual vibration signals collected during gearbox bench tests, test data of calibration gear pairs with intentionally machined teeth with known errors, and simulated data generated through gear dynamics simulation models.

[0048] Secondly, the network architecture of the initial error type classification model is constructed. The initial error type classification model adopts a convolutional neural network structure, which includes a physical prior convolutional layer, a pooling layer, and a fully connected output layer from input to output. The physical prior convolutional layer is the core differentiating design of the entire model. Its initial weights are not randomly initialized, but are pre-determined based on the inherent sideband spectral structure of the gear fault vibration signal. Specifically, the convolutional kernel adopts a comb filter type convolutional kernel. The passband position of this type of convolutional kernel is set to correspond one-to-one with the frequency position of each order of the sideband, enabling the convolutional kernel to have the ability to selectively and sensitively extract the sideband amplitude distribution pattern during the initialization stage. This embeds the physical mechanism of gear transmission into the deep learning model, providing good prior guidance for subsequent feature learning.

[0049] Next, the forward propagation process of the model is executed. The sideband amplitude sequence samples are input into the physical prior convolutional layer of the initial error type classification model. The input sequence is convolved using a comb filter-type convolutional kernel in this layer to enhance the periodic modulation features of the sidebands, outputting a first feature map. Subsequently, the first feature map is input into a pooling layer for dimensionality reduction. Max pooling or average pooling operations are used to reduce the size of the feature map, reducing the number of computational parameters while retaining the main feature information, outputting a second feature map. Then, the second feature map is flattened into a one-dimensional feature vector, which is then input into a fully connected output layer. The fully connected output layer uses the Softmax activation function to output the predicted probability distribution of the sideband amplitude sequence samples corresponding to each preset error category, thereby obtaining the predicted error type.

[0050] Finally, supervised training of the model is performed. Using the classification cross-entropy loss between the predicted error type and the error type label as the training objective, the backpropagation algorithm and gradient descent optimizer are used to iteratively update all network parameters of the initial error type classification model. During training, historical calibration data is divided into training and validation sets. After each round of training, the model performance is evaluated using the validation set until the classification accuracy on the validation set converges and the loss value no longer decreases significantly. Training is then terminated, and the current model parameters are saved, resulting in the pre-trained error type classification model. This pre-trained model can then be deployed in an online active control system for real-time identification of the error type of the gear pair originating from the whistling sound.

[0051] The initial weights of the convolution kernels of the physical prior convolutional layer are predetermined based on the sideband spectral structure of the gear fault vibration signal, including: During the pre-training phase of the error type classification model, the preset number of teeth on the driving gear and the preset calibration speed of the driving shaft of the preset calibration gear pair are obtained. The preset number of teeth on the driving gear is the number of teeth with the highest frequency among the number of teeth of the gear pair corresponding to each sideband amplitude sequence sample in the multiple sets of historical calibration data. Based on the preset number of teeth of the driving gear and the preset rated rotational speed of the driving shaft, the meshing frequency and the sideband interval frequency are calculated, wherein the meshing frequency is equal to the preset number of teeth of the driving gear multiplied by the rotational frequency corresponding to the preset rated rotational speed of the driving shaft, and the sideband interval frequency is equal to the rotational frequency corresponding to the preset rated rotational speed of the driving shaft; The position of the kth order in each order of the sideband is determined as the frequency position of the meshing frequency plus or minus k times the sideband spacing frequency, where k is an integer greater than 0, and the value of k corresponds one-to-one with the order. The order range of the sideband is from 1 to the preset maximum order. Using the positions of each order of the sideband as the passband center, a comb filter template is constructed with a preset bandwidth. The gain of the comb filter template at each order position of the sideband is a preset first gain value, and the gain at other frequency positions is a preset second gain value, wherein the first gain value is greater than the second gain value. The comb filter template is used as the initial weights of the convolution kernel of the physical prior convolutional layer.

[0052] In this embodiment of the invention, before the pre-training stage of the error type classification model, a standardized reference gear pair parameter is first determined. The preset number of teeth on the driving gear and the preset calibration rotational speed of the driving shaft are obtained. The principle for determining the preset number of teeth on the driving gear is to statistically analyze the actual number of teeth on the gear pair corresponding to the amplitude sequence samples of each sideband in multiple sets of historical calibration data, and select the number of teeth with the highest frequency of occurrence as the preset value. This ensures that the passband position of the comb filter matches the frequency band distribution of most training samples, maximizing the coverage of prior knowledge.

[0053] Secondly, based on the determined preset number of teeth on the driving gear and the preset rated speed of the driving shaft, the theoretical meshing frequency and sideband interval frequency are calculated. According to the gear transmission principle, the meshing frequency is equal to the preset number of teeth on the driving gear multiplied by the rotational frequency corresponding to the preset rated speed of the driving shaft; the sideband interval frequency is equal to the rotational frequency corresponding to the preset rated speed of the driving shaft itself. These two frequency parameters are the basis for determining the passband position of the comb filter.

[0054] Next, the specific frequency positions of each order in the sideband are determined. For the k-th order in each sideband, for the left sideband, its frequency position is determined as the meshing frequency minus k times the sideband spacing frequency; for the right sideband, it is the meshing frequency plus k times the sideband spacing frequency. k is an integer greater than 0, and the value of k corresponds one-to-one with the order. For example, k=1 corresponds to the first-order sideband, k=2 corresponds to the second-order sideband, and so on. The order range of the sideband is set from 1 to the preset maximum order. The maximum order is determined based on the energy cumulative contribution rate truncation method of historical calibration data. During the model pre-training phase, the amplitude of each order in the sideband is statistically analyzed in multiple sets of historical calibration data, and the proportion of each order's energy to the total sideband energy is calculated. Then, starting from the first order, the values ​​are accumulated step by step. When the cumulative contribution rate first reaches or exceeds 95%, that order is determined as the preset maximum order. This method has a statistical basis and, once determined, can be used consistently throughout the pre-training and inference phases without requiring repeated online calculations.

[0055] For example, statistical analysis of the sideband amplitudes in 500 sets of historical calibration data shows that the average energy of the first-order sideband accounts for 58% of the total sideband energy; the second-order accounts for 22%, totaling 80%; the third-order accounts for 10%, totaling 90%; the fourth-order accounts for 4%, totaling 94%; and the fifth-order accounts for 2.5%, totaling 96.5%. The cumulative contribution rate first exceeds 95% in the fifth order. Therefore, the preset maximum order is set to 5, meaning the comb filter covers ±1 to ±5 times the turnaround frequency on both sides of the meshing frequency, for a total of 10 sideband passband positions.

[0056] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0057] Then, a comb filter template is constructed based on the aforementioned frequency positions. Using the order positions of the sidebands as passband centers, a comb filter template is constructed with a preset bandwidth. In the comb filter template, the gain at each order position of the sideband is set to a preset first gain value, and the gain at other non-sideband frequency positions is set to a preset second gain value, with the first gain value being greater than the second gain value. Preferably, the first gain value is set to 1, and the second gain value is set to 0, thereby forming an ideal comb filtering effect, enabling the convolution kernel to maximize the passage of sideband components and suppress unrelated frequency band components during initialization.

[0058] The preset bandwidth is set using an adaptive percentage method based on the current rotation frequency: preset bandwidth = current rotation frequency × β, where β is between 0.3 and 0.4, ensuring the bandwidth is not less than twice the FFT frequency resolution and not greater than 0.4 times the sideband spacing. This method automatically adjusts the bandwidth according to changes in the operating speed. At high speeds, the rotation frequency is large, and the bandwidth is appropriately widened to cover more spectral lines; at low speeds, the rotation frequency is small, and the bandwidth is automatically tightened to avoid passband overlap. This value β, determined during the pre-training phase, remains fixed during the inference phase.

[0059] For example, if the current frequency shift is 30Hz and β=0.35, then the preset bandwidth is 30×0.35=10.5Hz, rounded down to 10Hz. With an FFT frequency resolution of 3Hz, 10Hz can cover approximately 3.3 spectral lines, meeting the minimum requirement of covering 2-3 spectral lines. The spacing between adjacent sidebands is 30Hz, and 10Hz is less than 0.4 times the spacing, ensuring that adjacent passbands do not overlap.

[0060] It should be noted that the above values ​​are for illustrative purposes only and do not constitute a limitation on the present invention.

[0061] Finally, the comb filter template constructed above is directly used as the initial weights of the convolutional kernels of the physical prior convolutional layer. With this initialization scheme, the model has clear physical meaning and a reasonable feature extraction direction in the initial training stage. Subsequent training only requires fine-tuning the weights based on the prior to adapt to the subtle differences of different gearbox models and different error types. This significantly reduces the dependence of the error type classification model on large-scale labeled data and accelerates the training convergence speed.

[0062] This invention, by inputting the sideband amplitude sequence into a pre-trained error type classification model, can automatically, quickly, and accurately identify the physical root cause of gear squealing, providing a basis for maintenance decisions or the formulation of adaptive control strategies. The squealing intensity quantification index obtained in this invention is a continuous value, allowing subsequent harmonic current injection to no longer rely on discrete gear positions or empirical tables, but rather to dynamically and continuously adjust the injection amplitude and phase based on the real-time calculated squealing intensity.

[0063] S400: Based on a pre-trained control parameter mapping model, the control parameters of the harmonic current are determined according to the error type, the quantification index of the howling degree, and the current operating parameters. The control parameters include the injection order, injection amplitude, and injection phase.

[0064] Traditional active control methods lack a decision-making mechanism that deeply integrates multi-dimensional information such as error type, howling degree, and operating parameters. As a result, they have to rely on manual calibration tables or empirical formulas to determine control parameters. This makes it impossible for the control strategy to adaptively and dynamically adjust according to the actual fault mode and real-time operating status of the gear pair. Furthermore, the calibration workload is large, the operating conditions are not fully covered, and the response lag is significant.

[0065] Step S400 in the method provided in this embodiment of the invention includes: The number of teeth on the driving gear of the gear pair from which the whistling originates is taken as the injection order; Obtain the current speed and current torque, and use the current speed and current torque as the current operating condition parameters; The error type, the quantification index of the howling degree, and the current operating condition parameters are input into a pre-trained control parameter mapping model, and the control parameter mapping model outputs the normalized injection amplitude and the normalized injection phase. The normalized injection amplitude is restored to the injection amplitude by the inverse normalization process, and the normalized injection phase is restored to the injection phase by the inverse normalization process. The injection order, the injection amplitude, and the injection phase are used as control parameters for the harmonic current.

[0066] In this embodiment of the invention, the number of teeth on the driving gear of the gear pair from which the howling source was located in the aforementioned steps is directly used as the injection order of the harmonic current. According to the gear transmission sound vibration mechanism, the order of the howling noise is equal to the number of teeth on the driving gear. Therefore, the injection order of the compensating harmonic current should be consistent with this order in order to achieve accurate phase alignment and meshing excitation cancellation.

[0067] Secondly, the current motor speed and current output torque are read in real time through the motor controller or vehicle controller, and these current speed and current torque are used as the current operating condition parameters. Speed ​​and torque are the main external variables affecting the meshing state of the gear pair and the vibration transmission characteristics. Using them as model inputs allows the control parameters to be adaptively adjusted according to the actual operating conditions.

[0068] Next, the error type, quantification of howling intensity, and current operating parameters are input into the pre-trained control parameter mapping model. This model, trained offline, establishes a nonlinear mapping relationship between error type, howling intensity, operating parameters, and optimal harmonic injection parameters. After forward inference, the model outputs normalized injection amplitude and normalized injection phase. It's important to note that normalized output is used because the labeled data was normalized during model training; outputting normalized values ​​helps maintain numerical stability and accelerates the inference process.

[0069] Then, the normalized injection amplitude is restored to the injection amplitude with actual physical dimensions according to the inverse transformation formula during normalization. The injection amplitude is usually in amperes and represents the magnitude of the compensation harmonic current. At the same time, the normalized injection phase is restored to the actual injection phase according to the inverse transformation of normalization. The injection phase is usually in degrees and represents the phase offset of the compensation harmonic current relative to the fundamental current.

[0070] Finally, the injection order, injection amplitude, and injection phase determined in the above steps are used together as control parameters for the harmonic current. These three parameters constitute a complete set of compensation commands. The injection order determines the frequency reference of the compensated harmonic current, the injection amplitude determines the magnitude of the compensation force, and the injection phase determines whether the compensation direction is accurately aligned with the inverse phase point of the howling source. All three are indispensable and work together to ensure the effectiveness of active noise reduction.

[0071] The pre-training process of the control parameter mapping model includes: Multiple sets of historical control data are acquired. Each set of historical control data includes error type samples, howling degree samples, operating condition parameter samples, and corresponding optimal injection amplitude and optimal injection phase labels. The operating condition parameter samples include speed samples and torque samples. The optimal injection amplitude and the optimal injection phase label are respectively subjected to minimum-maximum normalization to obtain normalized optimal injection amplitude label and normalized optimal injection phase label; An initial control parameter mapping model is constructed based on a lightweight neural network, which includes an input layer, a hidden layer, and an output layer. Using the error type samples, the howling intensity samples, and the operating condition parameter samples as input features, and the normalized optimal injection amplitude label and the normalized optimal injection phase label as supervision labels, the initial control parameter mapping model is trained in a supervised manner. During the training process, the training objective is to minimize the mean square error between the normalized injection amplitude and normalized injection phase output by the initial control parameter mapping model and the normalized optimal injection amplitude label and normalized optimal injection phase label. The network parameters of the initial control parameter mapping model are updated until the verification convergence, thus obtaining the pre-trained control parameter mapping model.

[0072] In this embodiment of the invention, firstly, a training dataset for a control parameter mapping model is constructed. Multiple sets of historical control data are acquired, each set containing an error type sample, a squeal intensity sample, a set of operating condition parameter samples, and labels for the optimal injection amplitude and optimal injection phase corresponding to that set of inputs. Specifically, the operating condition parameter samples include speed samples and torque samples. In the calibration test, for each combination of error type and squeal intensity, under the corresponding speed and torque conditions, the injection amplitude and phase are swept while the vibration response is monitored in real time. The amplitude and phase corresponding to the minimum vibration response are determined as the optimal label values. This calibration process can be completed offline on a gearbox test bench to ensure the accuracy and reliability of the label data.

[0073] Secondly, the label data undergoes normalization preprocessing. Considering the significant differences in the physical value ranges of the injected amplitude and injected phase, directly training the model with the original dimensions would lead to an imbalance in the sensitivity of the loss function across different dimensions, affecting the model's convergence performance. Therefore, min-max normalization is performed on the optimal injected amplitude and optimal injected phase labels respectively, compressing their numerical range to the [0, 1] interval to obtain normalized optimal injected amplitude and optimal injected phase labels. Simultaneously, the minimum and maximum values ​​of each dimension are recorded to allow for inverse normalization restoration of the output results during online model inference.

[0074] Next, the network architecture of the initial control parameter mapping model is constructed. The initial control parameter mapping model is built based on a lightweight neural network, which includes an input layer, hidden layers, and an output layer. For example, the hidden layers adopt a 1-2 layer fully connected structure, with 32-128 neurons in each layer. The ReLU activation function is used to introduce non-linear expressive power. The purpose of the lightweight structure design is to ensure that the control parameter mapping model has sufficient mapping accuracy while keeping its parameter count within a very small range, facilitating efficient operation in the subsequent automotive embedded controller.

[0075] Then, the error type samples, the squealing intensity samples, and the operating condition parameter samples are used as input features, and the normalized optimal injection amplitude label and the normalized optimal injection phase label are used as supervision labels to perform supervised training on the initial control parameter mapping model. Specifically, the error type is a categorical variable and needs to be converted into a numerical vector through one-hot encoding before input; the squealing intensity is a continuous numerical value and is directly input; the speed and torque are also continuous numerical values ​​and are also directly input.

[0076] Finally, the training objective is to minimize the mean square error between the normalized injection amplitude and normalized injection phase outputs of the initial control parameter mapping model and the normalized optimal injection amplitude and normalized optimal injection phase labels. The network parameters of the initial control parameter mapping model are updated using backpropagation and a gradient descent optimizer. Historical control data is divided into training and validation sets. After each training round, the mean square error loss value on the validation set is calculated. When the validation loss no longer decreases significantly for several consecutive rounds and reaches the preset convergence condition, training is terminated and the current model parameters are saved, thus obtaining the pre-trained control parameter mapping model. This model can then be deployed in the online inference module of the active control system to quickly generate optimal harmonic injection parameters based on real-time input.

[0077] This invention implements an intelligent parameter decision-making mechanism that takes the root cause diagnosis results of whistling and real-time operating condition information as input, and outputs the optimal compensation parameters in real time through a lightweight neural network model. This enables the injection order, injection amplitude, and injection phase of the harmonic current to adaptively and accurately match according to the error type, whistling severity, and changes in speed and torque. While ensuring the optimal noise reduction effect, the model has extremely low parameter count, millisecond-level online inference latency, and does not require additional controller hardware costs. It has good mass production deployment and cross-operating condition generalization capabilities.

[0078] S500: Synthesize a compensation harmonic current according to the control parameters, and superimpose the compensation harmonic current into the motor drive current to counteract the meshing excitation of the gear pair from which the whistling originates.

[0079] In existing technologies, the compensation frequency cannot track the actual meshing frequency changes in real time under variable speed conditions, resulting in a severe degradation of noise reduction effect at non-calibrated speed points. Moreover, the phase delay introduced by the complete electromechanical transmission path from the current command to the meshing point is ignored, causing the compensation force to fail to form an accurate antiphase relationship with the excitation force at the meshing point, and the noise reduction effect deviates from the theoretical optimal value. The compensation signal injection level is unclear, and there is a lack of a systematic implementation plan for low-latency, high-fidelity injection in the existing motor controller architecture, making it difficult to effectively deploy active control in real vehicle embedded environments.

[0080] Step S500 in the method provided in this embodiment of the invention includes: Obtain the rotational frequency corresponding to the current rotational speed of the shaft where the driving gear of the gear pair from which the whistling originates is located, and multiply the injection order in the control parameters by the rotational frequency to obtain the frequency of the compensation harmonic current; Based on the injection amplitude and injection phase in the control parameters, a compensated harmonic current signal having the frequency, the injection amplitude, and the injection phase is generated; The compensation harmonic current signal is superimposed on the motor drive current, so that the motor outputs a compensation torque that is opposite in phase to the meshing excitation of the gear pair from which the howling originates, in order to counteract the meshing excitation.

[0081] In this embodiment of the invention, the frequency of the compensation harmonic current is first calculated. The current rotational speed of the shaft containing the driving gear of the gear pair originating from the whistling is acquired in real time, and the corresponding rotational frequency is calculated based on this current speed. The injection order in the control parameters is multiplied by the rotational frequency to obtain the frequency of the compensation harmonic current. According to the gear transmission principle, the meshing frequency is equal to the number of teeth of the driving gear multiplied by the rotational frequency of the shaft. Since the injection order is set to the number of teeth of the driving gear of the gear pair originating from the whistling, the physical meaning of this product is the actual meshing frequency of the gear pair under the current operating conditions. The compensation harmonic current is injected at this frequency to ensure that it is precisely aligned with the whistling excitation frequency in the frequency domain.

[0082] Secondly, based on the injection amplitude and injection phase in the control parameters, a waveform synthesis algorithm is used to generate a compensated harmonic current signal with the calculated frequency, specified injection amplitude, and specified injection phase. Preferably, this signal is a single-frequency sinusoidal current signal, the amplitude of which is determined by the injection amplitude, and its initial phase is determined by the injection phase. The injection phase is set such that after the compensated harmonic current signal is output by the motor electromagnetic torque and transmitted to the gear pair via the drive shaft, the dynamic force generated at the meshing point is 180° out of phase with the meshing excitation force of the gear pair itself, which is the source of the squealing. To achieve this phase alignment, when determining the injection phase, in addition to considering the phase reference output by the model, the electromagnetic-mechanical transmission phase delay from the motor stator current to the gear pair meshing point must also be taken into account. The transmission phase delay can be obtained in advance through bench calibration or system identification and is feedforward compensated in the final injection phase.

[0083] Finally, the synthesized compensated harmonic current signal is superimposed onto the motor drive current command through the current loop of the motor controller. This superposition operation is completed at the current loop level, that is, the original d / q axis current command output by the motor controller is algebraically summed with the compensated harmonic current command, and the synthesized total current command is pulse-width modulated to drive the motor. Under the excitation of the compensated harmonic current, the motor outputs a compensated torque with a phase opposite to the meshing excitation of the gear pair from which the squeal originates. The compensated torque is transmitted to the gear meshing point through the drive shaft, generating a dynamic force in the frequency domain with an amplitude similar to but a phase opposite to the original meshing excitation. The two forces cancel each other out after superposition, thereby effectively weakening the meshing excitation of the gear pair at the physical level, and ultimately achieving the active control effect of reducing the squeal noise radiated from the gearbox housing. Preferably, the superposition of the compensated harmonic current adopts an open-loop feedforward method to achieve a fast response based on the established accurate model; alternatively, a vibration sensor feedback closed loop can be added to the control loop to monitor the residual vibration after compensation in real time and dynamically fine-tune the injected amplitude and phase to further enhance the robustness and environmental adaptability of the control system.

[0084] This invention achieves real-time synthesis and dynamic superposition of compensating harmonic currents at the current loop level of the motor controller, enabling the motor output to produce a compensating torque that is out of phase with the meshing excitation. The compensation frequency is dynamically updated with the rotational speed, and the phase is ensured to be strictly out of phase at the meshing point after feedforward delay compensation, thus achieving a stable and optimal meshing excitation cancellation effect across the entire speed range. Furthermore, this solution requires no modification to the motor controller hardware, has good compatibility, and supports the introduction of feedback closed loops based on feedforward, further enhancing the long-term robustness and engineering reliability of the control system.

[0085] In summary, the embodiments of the present invention have at least the following technical effects: First, this invention fundamentally eliminates the interference of real-time changes in vehicle speed or motor speed on frequency identification by converting absolute frequency into a constant meshing order for sound source localization. Simultaneously, by introducing a preset matching threshold tolerance mechanism in the tooth count comparison, it effectively compensates for minor calculation deviations caused by spectral resolution limitations and frequency drift, avoiding misjudgment or missed detection of howling sources.

[0086] Secondly, this invention utilizes the inherent mapping relationship between sideband amplitude distribution and gear physical error types, combined with a pre-trained error type classification model, to automatically identify specific error categories such as tooth profile angle error, tooth direction angle error, tooth pitch deviation, and local faults. Simultaneously, by constructing the sideband energy ratio as a quantitative indicator of howling intensity, this indicator is essentially decoupled from speed and load, remaining stable and monotonic across the entire operating range, and highly consistent with subjective human hearing, providing an objective and continuous evaluation metric for subsequent on-demand fine control.

[0087] Subsequently, this invention inputs the error type, the quantification index of the howling degree, and the real-time operating condition parameters into the pre-trained control parameter mapping model. The model synchronously outputs the optimal injection amplitude and phase, enabling the harmonic current compensation parameters to adaptively and dynamically match according to the current fault mode and operating status. This replaces the traditional manual calibration table lookup method, significantly reducing the calibration workload and improving the completeness of operating condition coverage.

[0088] Finally, this invention ensures that the compensation frequency always changes synchronously with the actual meshing frequency by using a dynamic calculation method that multiplies the injection order by the real-time rotational frequency, thus maintaining effective noise reduction across the entire speed range. By using feedforward compensation in the injection phase to compensate for the complete electromechanical transmission phase delay from the current command to the meshing point, it ensures that the compensation force is precisely out of phase with the excitation force at the meshing point, achieving the maximum cancellation effect. By completing the compensation harmonic current superposition in the innermost current loop of the motor controller, it fully utilizes the high bandwidth and fast response characteristics of the current loop to achieve low-latency, high-fidelity engineered injection execution.

[0089] In summary, this invention solves the technical problem that existing active noise control technologies for gear transmissions lack a full-process adaptive closed loop, resulting in inaccurate real-time matching of frequency and phase and poor consistency in noise reduction effects.

[0090] Example 2, as Figure 2 As shown, based on the same inventive concept as the active control method for gear transmission noise provided in Embodiment 1, this embodiment of the invention also provides an active control transmission device for gear transmission noise, comprising: The vibration signal acquisition module 11 is used to acquire the vibration signal of the gearbox housing through a vibration sensor installed on the outer surface of the gearbox housing; The whistling sound source localization module 12 is used to perform order analysis on the vibration signal, extract the whistling frequency component in the vibration signal, determine the meshing order corresponding to the whistling frequency component, and locate the whistling source gear pair by comparing the meshing order with the number of teeth of each gear pair in the gearbox. Error identification and quantification module 13 is used to perform spectrum analysis on the vibration signal and determine the error type and quantification index of the whistling source gear pair based on the spectrum analysis results. The control parameter decision module 14 is used to determine the control parameters of the harmonic current based on the pre-trained control parameter mapping model, according to the error type, the quantification index of the howling degree, and the current operating condition parameters. The control parameters include the injection order, injection amplitude, and injection phase. The compensation current synthesis and injection module 15 is used to synthesize compensation harmonic current according to the control parameters and superimpose the compensation harmonic current into the motor drive current to counteract the meshing excitation of the gear pair from which the howling originates.

[0091] In one embodiment, the howling sound source localization module 12 is specifically used for: The vibration signal is transformed by time-frequency conversion to convert the time-domain vibration signal into a frequency-domain signal, thereby obtaining the spectrum of the vibration signal; Frequency components with energy amplitudes exceeding a preset energy threshold are extracted from the spectrum as candidate frequency components for whistling. The candidate frequency components for whistling are compared with the theoretical meshing frequencies of each gear pair in the gearbox. The candidate frequency component for whistling that matches any theoretical meshing frequency is determined as the whistling frequency component. The whistling frequency component with the highest energy amplitude is selected as the target whistling frequency component. The theoretical meshing frequency is equal to the number of teeth of the driving gear of each gear pair multiplied by the rotational frequency corresponding to the current rotational speed of the shaft where the driving gear of the gear pair is located. The ratio of the frequency value of the target squealing frequency component to the rotational frequency of the shaft where the driving gear of the gear pair that matches the theoretical meshing frequency of the target squealing frequency component is located is calculated to obtain the meshing order corresponding to the target squealing frequency component. The meshing order is compared one by one with the number of teeth of the driving gear in each gear pair in the gearbox, and the gear pair whose number of teeth of the driving gear matches the meshing order is identified as the gear pair from which the whistling sound originates.

[0092] The meshing order is compared one by one with the number of teeth of the driving gear in each gear pair in the gearbox. The gear pairs whose number of teeth of the driving gear matches the meshing order are identified as the gear pairs from which the whistling sound originates, including: Obtain the number of teeth on the driving gear of each gear pair in the gearbox and generate a list of the number of teeth on each gear pair; Calculate the difference between the meshing order and the number of teeth of each driving gear in the gear pair tooth count list, and determine the gear pair with the absolute value of the difference being less than a preset matching threshold as the source gear pair of the whistling.

[0093] In one embodiment, the error identification and quantization module 13 is specifically used for: The vibration signal is subjected to spectral analysis to extract the amplitude of each order sideband on both sides of the meshing frequency of the gear pair from which the whistling originates, and the sideband amplitude sequence is formed by arranging them in order of order. The sideband amplitude sequence is input into a pre-trained error type classification model, which outputs the error type of the gear pair from which the howling originates. The energy ratio of the sidebands is obtained by calculating the ratio of the sum of the energy of each order sideband on both sides of the fundamental meshing frequency of the gear pair from which the whistling originates to the energy value at the fundamental meshing frequency. The sideband energy ratio is used as a quantitative indicator of the degree of howling from the gear pair from which the howling originates.

[0094] The pre-training process of the error type classification model includes: Multiple sets of historical calibration data are acquired. Each set of historical calibration data includes sideband amplitude sequence samples and corresponding error type labels. The error type labels include at least tooth profile angle error, tooth direction angle error, tooth pitch deviation, and local fault. An initial error type classification model is constructed, which includes a physical prior convolutional layer, a pooling layer, and a fully connected output layer. The initial weights of the convolution kernels of the physical prior convolutional layer are predetermined based on the sideband spectral structure of the gear fault vibration signal. The convolution kernels are comb filter type convolution kernels, and the passband positions of the comb filter type convolution kernels correspond to the frequency positions of each order of the sideband. The sideband amplitude sequence samples are input into the physical prior convolutional layer of the initial error type classification model for feature extraction to obtain a first feature map. The first feature map is then input into the pooling layer of the initial error type classification model for dimensionality reduction to obtain a second feature map. The second feature map is then flattened and input into the fully connected output layer of the initial error type classification model to output the predicted error type. Using the classification cross-entropy loss between the predicted error type and the error type label as the training objective, supervised training is performed on the initial error type classification model, and the network parameters of the initial error type classification model are updated until the verification convergence is obtained, thus obtaining the pre-trained error type classification model.

[0095] The initial weights of the convolution kernels of the physical prior convolutional layer are predetermined based on the sideband spectral structure of the gear fault vibration signal, including: During the pre-training phase of the error type classification model, the preset number of teeth on the driving gear and the preset calibration speed of the driving shaft of the preset calibration gear pair are obtained. The preset number of teeth on the driving gear is the number of teeth with the highest frequency among the number of teeth of the gear pair corresponding to each sideband amplitude sequence sample in the multiple sets of historical calibration data. Based on the preset number of teeth of the driving gear and the preset rated rotational speed of the driving shaft, the meshing frequency and the sideband interval frequency are calculated, wherein the meshing frequency is equal to the preset number of teeth of the driving gear multiplied by the rotational frequency corresponding to the preset rated rotational speed of the driving shaft, and the sideband interval frequency is equal to the rotational frequency corresponding to the preset rated rotational speed of the driving shaft; The position of the kth order in each order of the sideband is determined as the frequency position of the meshing frequency plus or minus k times the sideband spacing frequency, where k is an integer greater than 0, and the value of k corresponds one-to-one with the order. The order range of the sideband is from 1 to the preset maximum order. Using the positions of each order of the sideband as the passband center, a comb filter template is constructed with a preset bandwidth. The gain of the comb filter template at each order position of the sideband is a preset first gain value, and the gain at other frequency positions is a preset second gain value, wherein the first gain value is greater than the second gain value. The comb filter template is used as the initial weights of the convolution kernel of the physical prior convolutional layer.

[0096] In one embodiment, the control parameter decision module 14 is specifically used for: The number of teeth on the driving gear of the gear pair from which the whistling originates is taken as the injection order; Obtain the current speed and current torque, and use the current speed and current torque as the current operating condition parameters; The error type, the quantification index of the howling degree, and the current operating condition parameters are input into a pre-trained control parameter mapping model, and the control parameter mapping model outputs the normalized injection amplitude and the normalized injection phase. The normalized injection amplitude is restored to the injection amplitude by the inverse normalization process, and the normalized injection phase is restored to the injection phase by the inverse normalization process. The injection order, the injection amplitude, and the injection phase are used as control parameters for the harmonic current.

[0097] The pre-training process of the control parameter mapping model includes: Multiple sets of historical control data are acquired. Each set of historical control data includes error type samples, howling degree samples, operating condition parameter samples, and corresponding optimal injection amplitude and optimal injection phase labels. The operating condition parameter samples include speed samples and torque samples. The optimal injection amplitude and the optimal injection phase label are respectively subjected to minimum-maximum normalization to obtain normalized optimal injection amplitude label and normalized optimal injection phase label; An initial control parameter mapping model is constructed based on a lightweight neural network, which includes an input layer, a hidden layer, and an output layer. Using the error type samples, the howling intensity samples, and the operating condition parameter samples as input features, and the normalized optimal injection amplitude label and the normalized optimal injection phase label as supervision labels, the initial control parameter mapping model is trained in a supervised manner. During the training process, the training objective is to minimize the mean square error between the normalized injection amplitude and normalized injection phase output by the initial control parameter mapping model and the normalized optimal injection amplitude label and normalized optimal injection phase label. The network parameters of the initial control parameter mapping model are updated until the verification convergence, thus obtaining the pre-trained control parameter mapping model.

[0098] In one embodiment, the compensation current synthesis and injection module 15 is specifically used for: Obtain the rotational frequency corresponding to the current rotational speed of the shaft where the driving gear of the gear pair from which the whistling originates is located, and multiply the injection order in the control parameters by the rotational frequency to obtain the frequency of the compensation harmonic current; Based on the injection amplitude and injection phase in the control parameters, a compensated harmonic current signal having the frequency, the injection amplitude, and the injection phase is generated; The compensation harmonic current signal is superimposed on the drive current of the motor, so that the motor outputs a compensation torque that is opposite in phase to the meshing excitation of the gear pair from which the howling originates, in order to counteract the meshing excitation.

[0099] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0101] This specification and accompanying drawings are merely illustrative examples of the invention and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its scope. Therefore, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is intended to include these modifications and modifications.

Claims

1. A method for actively controlling gear transmission noise in a gearbox, characterized in that, The method includes: Vibration signals of the gearbox housing are collected by vibration sensors installed on the outer surface of the gearbox housing; The vibration signal is subjected to order analysis to extract the whistling frequency component in the vibration signal, the meshing order corresponding to the whistling frequency component is determined, and the meshing order is compared with the number of teeth of each gear pair in the gearbox to locate the gear pair from which the whistling originates. The vibration signal is subjected to spectral analysis, and based on the spectral analysis results, the error type and quantification index of the gear pair from which the whistling originates are determined. Based on a pre-trained control parameter mapping model, the control parameters of the harmonic current are determined according to the error type, the quantification index of the howling degree, and the current operating parameters. The control parameters include the injection order, injection amplitude, and injection phase. The compensation harmonic current is synthesized according to the control parameters and superimposed on the motor drive current to counteract the meshing excitation of the gear pair from which the howling originates.

2. The active control method for gearbox gear transmission noise according to claim 1, characterized in that, The vibration signal is subjected to order analysis to extract the whistling frequency component, determine the meshing order corresponding to the whistling frequency component, and locate the source gear pair of the whistling by comparing the meshing order with the number of teeth of each gear pair in the gearbox. The vibration signal is transformed by time-frequency conversion to convert the time-domain vibration signal into a frequency-domain signal, thereby obtaining the spectrum of the vibration signal; Frequency components with energy amplitudes exceeding a preset energy threshold are extracted from the spectrum as candidate frequency components for whistling. The candidate frequency components for whistling are compared with the theoretical meshing frequencies of each gear pair in the gearbox. The candidate frequency component for whistling that matches any theoretical meshing frequency is determined as the whistling frequency component. The whistling frequency component with the highest energy amplitude is selected as the target whistling frequency component. The theoretical meshing frequency is equal to the number of teeth of the driving gear of each gear pair multiplied by the rotational frequency corresponding to the current rotational speed of the shaft where the driving gear of the gear pair is located. The ratio of the frequency value of the target squealing frequency component to the rotational frequency of the shaft where the driving gear of the gear pair that matches the theoretical meshing frequency of the target squealing frequency component is located is calculated to obtain the meshing order corresponding to the target squealing frequency component. The meshing order is compared one by one with the number of teeth of the driving gear in each gear pair in the gearbox, and the gear pair whose number of teeth of the driving gear matches the meshing order is identified as the gear pair from which the whistling sound originates.

3. The active control method for gearbox gear transmission noise according to claim 2, characterized in that, The meshing order is compared one by one with the number of teeth of the driving gear in each gear pair in the gearbox. The gear pairs whose number of teeth of the driving gear matches the meshing order are identified as the gear pairs from which the whistling sound originates, including: Obtain the number of teeth on the driving gear of each gear pair in the gearbox and generate a list of the number of teeth on each gear pair; Calculate the difference between the meshing order and the number of teeth of each driving gear in the gear pair tooth count list, and determine the gear pair with the absolute value of the difference being less than a preset matching threshold as the source gear pair of the whistling.

4. The active control method for gearbox gear transmission noise according to claim 1, characterized in that, The vibration signal is subjected to spectral analysis. Based on the spectral analysis results, the error type and quantification index of the gear pair from which the squealing originates are determined, including: The vibration signal is subjected to spectral analysis to extract the amplitude of each order sideband on both sides of the meshing frequency of the gear pair from which the whistling originates, and the sideband amplitude sequence is formed by arranging them in order of order. The sideband amplitude sequence is input into a pre-trained error type classification model, which outputs the error type of the gear pair from which the howling originates. The energy ratio of the sidebands is obtained by calculating the ratio of the sum of the energy of each order sideband on both sides of the fundamental meshing frequency of the gear pair from which the whistling originates to the energy value at the fundamental meshing frequency. The sideband energy ratio is used as a quantitative indicator of the degree of howling from the gear pair from which the howling originates.

5. The active control method for gearbox gear transmission noise according to claim 4, characterized in that, The pre-training process of the error type classification model includes: Multiple sets of historical calibration data are acquired. Each set of historical calibration data includes sideband amplitude sequence samples and corresponding error type labels. The error type labels include at least tooth profile angle error, tooth direction angle error, tooth pitch deviation, and local fault. An initial error type classification model is constructed, which includes a physical prior convolutional layer, a pooling layer, and a fully connected output layer. The initial weights of the convolution kernels of the physical prior convolutional layer are predetermined based on the sideband spectral structure of the gear fault vibration signal. The convolution kernels are comb filter type convolution kernels, and the passband positions of the comb filter type convolution kernels correspond to the frequency positions of each order of the sideband. The sideband amplitude sequence samples are input into the physical prior convolutional layer of the initial error type classification model for feature extraction to obtain a first feature map. The first feature map is then input into the pooling layer of the initial error type classification model for dimensionality reduction to obtain a second feature map. The second feature map is then flattened and input into the fully connected output layer of the initial error type classification model to output the predicted error type. Using the classification cross-entropy loss between the predicted error type and the error type label as the training objective, supervised training is performed on the initial error type classification model, and the network parameters of the initial error type classification model are updated until the verification convergence is obtained, thus obtaining the pre-trained error type classification model.

6. The active control method for gearbox gear transmission noise according to claim 5, characterized in that, The initial weights of the convolution kernels of the physical prior convolutional layer are predetermined based on the sideband spectral structure of the gear fault vibration signal, including: During the pre-training phase of the error type classification model, the preset number of teeth on the driving gear and the preset calibration speed of the driving shaft of the preset calibration gear pair are obtained. The preset number of teeth on the driving gear is the number of teeth with the highest frequency among the number of teeth of the gear pair corresponding to each sideband amplitude sequence sample in the multiple sets of historical calibration data. Based on the preset number of teeth of the driving gear and the preset rated rotational speed of the driving shaft, the meshing frequency and the sideband interval frequency are calculated, wherein the meshing frequency is equal to the preset number of teeth of the driving gear multiplied by the rotational frequency corresponding to the preset rated rotational speed of the driving shaft, and the sideband interval frequency is equal to the rotational frequency corresponding to the preset rated rotational speed of the driving shaft; The position of the kth order in each order of the sideband is determined as the frequency position of the meshing frequency plus or minus k times the sideband spacing frequency, where k is an integer greater than 0, and the value of k corresponds one-to-one with the order. The order range of the sideband is from 1 to the preset maximum order. Using the positions of each order of the sideband as the passband center, a comb filter template is constructed with a preset bandwidth. The gain of the comb filter template at each order position of the sideband is a preset first gain value, and the gain at other frequency positions is a preset second gain value, wherein the first gain value is greater than the second gain value. The comb filter template is used as the initial weights of the convolution kernel of the physical prior convolutional layer.

7. The active control method for gearbox gear transmission noise according to claim 1, characterized in that, Based on a pre-trained control parameter mapping model, the control parameters for the harmonic current are determined according to the error type, the quantification index of the howling intensity, and the current operating parameters, including: The number of teeth on the driving gear of the gear pair from which the whistling originates is taken as the injection order; Obtain the current speed and current torque, and use the current speed and current torque as the current operating condition parameters; The error type, the quantification index of the howling degree, and the current operating condition parameters are input into a pre-trained control parameter mapping model, and the control parameter mapping model outputs the normalized injection amplitude and the normalized injection phase. The normalized injection amplitude is restored to the injection amplitude by the inverse normalization process, and the normalized injection phase is restored to the injection phase by the inverse normalization process. The injection order, the injection amplitude, and the injection phase are used as control parameters for the harmonic current.

8. The active control method for gearbox gear transmission noise according to claim 7, characterized in that, The pre-training process of the control parameter mapping model includes: Multiple sets of historical control data are acquired. Each set of historical control data includes error type samples, howling degree samples, operating condition parameter samples, and corresponding optimal injection amplitude and optimal injection phase labels. The operating condition parameter samples include speed samples and torque samples. The optimal injection amplitude and the optimal injection phase label are respectively subjected to minimum-maximum normalization to obtain normalized optimal injection amplitude label and normalized optimal injection phase label; An initial control parameter mapping model is constructed based on a lightweight neural network, which includes an input layer, a hidden layer, and an output layer. Using the error type samples, the howling intensity samples, and the operating condition parameter samples as input features, and the normalized optimal injection amplitude label and the normalized optimal injection phase label as supervision labels, the initial control parameter mapping model is trained in a supervised manner. During the training process, the training objective is to minimize the mean square error between the normalized injection amplitude and normalized injection phase output by the initial control parameter mapping model and the normalized optimal injection amplitude label and normalized optimal injection phase label. The network parameters of the initial control parameter mapping model are updated until the verification convergence, thus obtaining the pre-trained control parameter mapping model.

9. The active control method for gearbox gear transmission noise according to claim 1, characterized in that, The control parameters are used to synthesize a compensation harmonic current, which is then superimposed on the motor drive current to counteract the meshing excitation of the gear pair from which the whistling originates. This includes: Obtain the rotational frequency corresponding to the current rotational speed of the shaft where the driving gear of the gear pair from which the whistling originates is located, and multiply the injection order in the control parameters by the rotational frequency to obtain the frequency of the compensation harmonic current; Based on the injection amplitude and injection phase in the control parameters, a compensated harmonic current signal having the frequency, the injection amplitude, and the injection phase is generated; The compensation harmonic current signal is superimposed on the motor drive current, so that the motor outputs a compensation torque that is opposite in phase to the meshing excitation of the gear pair from which the howling originates, in order to counteract the meshing excitation.

10. A transmission device for actively controlling gear transmission noise in a gearbox, characterized in that, An active control method for gearbox gear transmission noise according to any one of claims 1 to 9 includes: The vibration signal acquisition module is used to acquire vibration signals of the gearbox housing through vibration sensors installed on the outer surface of the gearbox housing; The whistling sound source localization module is used to perform order analysis on the vibration signal, extract the whistling frequency component in the vibration signal, determine the meshing order corresponding to the whistling frequency component, and locate the whistling source gear pair by comparing the meshing order with the number of teeth of each gear pair in the gearbox. The error identification and quantification module is used to perform spectrum analysis on the vibration signal and determine the error type and quantification index of the whistling source gear pair based on the spectrum analysis results. The control parameter decision module is used to determine the control parameters of the harmonic current based on the pre-trained control parameter mapping model, according to the error type, the quantification index of the howling degree, and the current operating condition parameters. The control parameters include the injection order, injection amplitude, and injection phase. The compensation current synthesis and injection module is used to synthesize compensation harmonic current according to the control parameters and superimpose the compensation harmonic current into the motor drive current to counteract the meshing excitation of the gear pair from which the howling originates.

Images