Sound signal denoising method and system based on notch filter and improved kalman filter

Abstract

This invention discloses a method and system for acoustic signal denoising based on notch filtering and improved Kalman filtering, relating to the field of acoustic signal processing technology. The proposed scheme includes acquiring the operating acoustic signal of a transmission device, dividing it into multiple processing windows according to a sliding time window, and characterizing the order of each processing window to obtain an order spectrum. Candidate order carriers are identified in the order spectrum, and order residuals are extracted from their preset order neighborhoods on both sides. By coupling the relative order distance, envelope change, and phase continuation relationship of the order residuals on both sides of the candidate order carrier in the order spectrum to form a residual closure degree, and based on this, a hollow notch kernel is constructed by dividing the carrier core suppression region and the side protection region. This solves the technical defect of existing notch processing, which determines the effective range only based on the energy concentration degree and easily weakens early fault modulation information attached to the carrier while weakening normal operating sound.

Description

Technical Field

[0001] This invention relates to the field of acoustic signal processing technology, and more specifically, this application relates to an acoustic signal noise reduction method and system based on notch filtering and improved Kalman filtering. Background Technology

[0002] During operation, transmission devices are usually accompanied by acoustic components such as gear meshing, shaft rotation, bearing contact, and structural transmission. Their operating sound signals have obvious periodicity and order correlation. Under normal fluctuations in load, speed, and assembly status, the narrow band component in the normal operating sound will drift slowly with the operating conditions. Early fault sounds often do not appear as independent and prominent abnormal sound sources, but are superimposed on the normal order components in the form of envelope fluctuations, phase disturbances, or changes in energy of adjacent orders.

[0003] Existing methods for noise reduction of transmission device sound signals typically first identify narrowband components with concentrated energy in the frequency or order domain, then weaken these components using notch filtering, and finally smooth the remaining signal using Kalman filtering or other state estimation methods. The applicability of this type of method is based on the premise that the narrowband components weakened by notch filtering mainly belong to regular operating noise, while the residual fluctuations in the subsequent estimation process mainly come from random noise. For scenarios with a single interference source or where fault characteristics are clearly separated from operating noise, the above processing method can achieve a certain noise reduction effect.

[0004] However, under normal operating conditions of transmission devices, normal order components and fault modulation components are often located in adjacent or even mutually dependent positions in the acoustic spectrum. There may also be a continuous relationship in envelope and phase between them. If the range of notch filtering is determined solely based on the degree of energy concentration, it is easy to weaken the subtle information related to the fault while weakening the normal operating sound. Furthermore, when the residual changes after notch filtering continue to enter the Kalman filter, they are easily treated as random fluctuations and participate in the noise statistical update, thereby further reducing the identifiability of the fault sound components. Therefore, the core technical problem that needs to be solved is: how to balance the suppression of normal order noise and the preservation of adjacent fault modulation information during the noise reduction process of the transmission device's acoustic signal, and avoid the continuous weakening of the fault sound structure during notch filtering and state estimation. Therefore, an acoustic signal noise reduction method and system based on notch filtering and improved Kalman filtering are proposed to solve this problem. Summary of the Invention

[0005] To address the aforementioned technical problems, this paper provides a method and system for acoustic signal noise reduction based on notch filtering and improved Kalman filtering. This technical solution solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In a first aspect, this application provides a method for acoustic signal noise reduction based on notch filtering and improved Kalman filtering, the method comprising:

[0008] The operating sound signal of the transmission device is acquired, divided into multiple processing windows according to the sliding time window, and the order of each processing window is characterized to obtain the order sound spectrum.

[0009] Candidate order carriers are identified in the order spectrum, and order residuals are extracted in the preset order neighborhoods on both sides of them. The residual closure degree is formed based on the relative order distance, envelope change and phase continuation relationship of the order residuals.

[0010] The carrier core suppression region and the side protection region are divided in the preset order neighborhood using the residual closure degree. For the order sampling points of the carrier core suppression region, the suppression filter coefficient is determined by the carrier energy concentration degree and the relative order distance. For the order sampling points of the side protection region, the protection filter coefficient is determined by the residual closure degree and the relative order distance. The suppression filter coefficient and the protection filter coefficient are arranged in order of order to form a hollow notch core.

[0011] The hollow notch core is used to weight the order acoustic spectrum and reconstruct the notch signal to obtain the notch signal. The notch differential residual is determined based on the difference between the running acoustic signal and the notch signal. The side protection residual is determined based on the retained acoustic signal of the side protection zone.

[0012] Using the acoustic signal after notch filtering as the observation sequence, Kalman prediction is performed by combining the state estimation of the initial or previous processing window and noise statistical parameters to obtain innovative residuals.

[0013] The differential residual of the notch filter, the side protection residual, and the innovation residual are mapped to the same order of representation. Homology is determined according to the relative order distance, envelope direction, and phase continuity relationship. Homology innovation residuals are defined as structural innovation quantities and form fault candidate components, while non-homology innovation residuals are defined as random innovation quantities.

[0014] The noise statistics parameters are updated using the random innovation quantity and the random noise in the notch-filtered acoustic signal is suppressed. The state estimation is corrected using the fault candidate component, and the noise-reduced fault acoustic signal is obtained.

[0015] Secondly, this application provides an acoustic signal noise reduction system based on notch filtering and improved Kalman filtering, for implementing the aforementioned acoustic signal noise reduction method based on notch filtering and improved Kalman filtering, including:

[0016] The order spectrum generation module is used to acquire the operating sound signal of the transmission device, divide it into multiple processing windows according to the sliding time window, and perform order characterization on each processing window to obtain the order spectrum.

[0017] The closure degree forming module is used to determine candidate order carriers in the order spectrum, extract order residuals in the preset order neighborhoods on both sides of the carriers, and form residual closure degree based on the relative order distance, envelope change and phase continuation relationship of the order residuals.

[0018] The notch core generation module is used to divide the carrier core suppression region and the side protection region within a preset order neighborhood using the residual closure degree. For the order sampling points of the carrier core suppression region, the suppression filter coefficient is determined by the carrier energy concentration degree and the relative order distance. For the order sampling points of the side protection region, the protection filter coefficient is determined by the residual closure degree and the relative order distance. The suppression filter coefficient and the protection filter coefficient are arranged in order of order to form a hollow notch core.

[0019] The notch residual determination module is used to use a hollow notch core to weight the order acoustic spectrum and reconstruct the notch, obtain the notch-before acoustic signal, determine the notch differential residual based on the difference between the running acoustic signal and the notch-before acoustic signal, and determine the side protection residual based on the retained acoustic signal of the side protection zone.

[0020] The Kalman prediction module is used to perform Kalman prediction based on the acoustic signal after notch filtering as the observation sequence, combined with the state estimation and noise statistics parameters of the initial or previous processing window, to obtain innovative residuals.

[0021] The same-origin discrimination module is used to map the notch differential residual, side protection residual and innovation residual to the same order of representation, and to perform same-origin discrimination according to the relative order distance, envelope direction and phase continuity relationship. The same-origin innovation residual is defined as structural innovation quantity and formed as fault candidate component, and the non-same-origin innovation residual is defined as random innovation quantity.

[0022] The noise reduction and extraction module is used to update the noise statistics parameters using random innovation quantities and suppress random noise in the notch-filtered acoustic signal. It also uses fault candidate components to correct the state estimation and obtain the noise-reduced fault acoustic signal.

[0023] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described acoustic signal noise reduction method based on notch filtering and improved Kalman filtering.

[0024] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned acoustic signal noise reduction method based on notch filtering and improved Kalman filtering.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] This application addresses the technical shortcomings of existing notch processing, which determines the effective range solely based on the degree of energy concentration and is prone to weakening early fault modulation information attached to the carrier while weakening normal operation sound. It achieves the identifiable preservation of low-energy but high-modulation-correlation-order segments and avoids the premature identification of subtle fault information as noise in the notch stage.

[0027] This application maps the notch differential residual, the side protection residual, and the innovation residual to the same order representation and performs homology discrimination according to the three dimensions of order proximity, envelope direction, and phase continuity. The innovation residual is split into structural innovation quantity and random innovation quantity, which are respectively entered into the state correction and noise statistics update channels. This solves the technical defect in the existing method that the residual changes after notch filtering are generally treated as random fluctuations and participate in noise statistics when entering Kalman filtering, which leads to the dilution of fault sound components by noise statistics in the state estimation stage. It realizes cross-validation between the two reference residuals on the notch side and the innovation residual on the Kalman side, and avoids the continuous weakening of the fault sound structure in the notch filtering and state estimation stages. Attached Figure Description

[0028] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Wherein:

[0029] Figure 1 This is a flowchart of the acoustic signal noise reduction method based on notch filtering and improved Kalman filtering proposed in this invention;

[0030] Figure 2 This is a block diagram of the acoustic signal noise reduction system based on notch filtering and improved Kalman filtering proposed in this invention. Detailed Implementation

[0031] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0032] Reference Figure 1 As shown, this application proposes an acoustic signal noise reduction method based on notch filtering and improved Kalman filtering, including:

[0033] The operating sound signal of the transmission device is acquired, divided into multiple processing windows according to the sliding time window, and the order of each processing window is characterized to obtain the order sound spectrum.

[0034] It should be noted that the acquisition of the operating sound signal is accomplished by acoustic sensors arranged on the surface of the transmission housing or in the near field. The sampling rate should be more than twice that of the frequency corresponding to the order of interest. The sliding time window is divided according to the set window length and overlap rate along the time axis to obtain multiple processing windows that are connected end to end. The order is represented by remapping the operating sound signal in each processing window from the time-frequency coordinate to the time-order coordinate by the synchronously acquired speed indication, so that the meshing order, shaft order and other components still stably occupy fixed order positions under the variable speed condition.

[0035] Candidate order carriers are identified in the order spectrum, and order residuals are extracted in the preset order neighborhoods on both sides of them. The residual closure degree is formed based on the relative order distance, envelope change and phase continuation relationship of the order residuals.

[0036] It should be noted that the process of determining the candidate order carrier and extracting the order residual is as follows:

[0037] Sub-step 1.1.1: Scan the energy distribution of each order sampling point along the order axis in the order spectrum, and register the order sampling points whose energy peaks are higher than the background energy baseline by a set multiple and whose peak positions correspond to the meshing fundamental frequency, shaft fundamental frequency or their harmonic positions as candidate order carriers. Each candidate order carrier corresponds to an order center position.

[0038] Sub-step 1.1.2: Using the order center position of the candidate order carrier as the center of symmetry, extend a preset order neighborhood along the positive and negative directions of the order axis. Within the order neighborhood, deduct the main lobe component of the energy of the candidate order carrier itself at each sampling point. Use the remaining energy after deduction and its phase trajectory as the order residuals on both sides of the candidate order carrier.

[0039] Sub-step 1.1.3: Take the order difference between the order sampling points corresponding to the order residuals on both sides of the candidate order carrier and the order center position of the candidate order carrier as the relative order distance, take the trajectory of the amplitude of each order residual changing with the order as the envelope change, and take the trajectory of the phase of each order residual changing with the order as the phase continuity relationship. The three together constitute the input quantity for forming the residual closure degree.

[0040] It should be noted that the width of the preset order neighborhood should cover the order span of the typical fault modulation sideband. Since the sidebands of gear meshing order are usually distributed at frequency intervals, the width of the preset order neighborhood is set to be no less than the order value corresponding to one frequency and no more than the order value corresponding to three frequencies. Its value range is from 0.5 to 3.5, with a typical value of 2.0.

[0041] The carrier core suppression region and the side protection region are divided in the preset order neighborhood using the residual closure degree. For the order sampling points of the carrier core suppression region, the suppression filter coefficient is determined by the carrier energy concentration degree and the relative order distance. For the order sampling points of the side protection region, the protection filter coefficient is determined by the residual closure degree and the relative order distance. The suppression filter coefficient and the protection filter coefficient are arranged in order of order to form a hollow notch core.

[0042] It should be noted that the process for dividing the carrier core suppression region and the side protection region, and determining the notch coefficient, is as follows:

[0043] Sub-step 1.2.1: Based on the order center position of the candidate order carrier, the order segments with residual closure less than the division threshold in the preset order neighborhood are classified into the carrier core suppression region, and the order segments with residual closure greater than the division threshold are classified into the side protection region, thus obtaining two types of regions that cover different order ranges respectively.

[0044] Sub-step 1.2.2: For each order sampling point within the carrier core suppression region, the ratio of the peak energy at the candidate order carrier to the average energy in its neighborhood is taken as the carrier energy concentration level. The suppression filter coefficient of the sampling point of that order is obtained by combining the relative order distance with the distance attenuation law;

[0045] For example, the expression for calculating the suppression filter coefficient is:

[0046] ;

[0047] in, The suppression filter coefficients are the k-th order sampling point within the carrier core suppression region.

[0048] This is the relative order distance from the sampling point of this order to the center position of the candidate order carrier order. For the degree of carrier energy concentration, The strength coefficient, This is the distance attenuation factor;

[0049] Sub-step 1.2.3: For each order of sampling point within the adjacent protected area, the residual closure of the neighborhood to which the sampling point belongs is used as the protection benchmark strength, and the protection benchmark strength is monotonically constrained and adjusted according to the relative order distance to obtain the protection filter coefficient. The value of the protection filter coefficient is monotonically non-increasing with the increase of the relative order distance, that is, the closer to the side position of the candidate order carrier, the milder the protection strength.

[0050] Sub-step 1.2.4: Connect the suppression filter coefficients of each order sampling point in the carrier core suppression region with the protection filter coefficients of each order sampling point in the side protection region along the order axis in order of position to form a hollow notch core that suppresses carrier energy at the center position and preserves adjacent order components at the periphery position.

[0051] Since the positions near the sidebands on both sides of the candidate subcarrier may carry early fault modulation information, the division threshold should not be too low. Its value ranges from 0.3 to 0.7, with a typical value of 0.5.

[0052] The hollow notch core is used to weight the order acoustic spectrum and reconstruct the notch signal to obtain the notch signal. The notch differential residual is determined based on the difference between the running acoustic signal and the notch signal. The side protection residual is determined based on the retained acoustic signal of the side protection zone.

[0053] It should be noted that the execution process of weighted notch filtering and residual extraction is as follows: the hollow notch kernel is multiplied with the order spectrum along the order axis point by point to obtain the weighted order spectrum, which is then reconstructed into the notch-filtered acoustic signal in the time domain through order-time inverse transform; the running acoustic signal and the notch-filtered acoustic signal are subtracted point by point along the time axis to obtain the notch differential residual; the order spectrum within the corresponding order range of the side protection zone is inversely transformed into a time domain signal separately to obtain the side protection residual; the notch differential residual carries the energy trajectory weakened by the notch filter, and the side protection residual carries the disturbance trajectory retained in the adjacent order.

[0054] Using the acoustic signal after notch filtering as the observation sequence, Kalman prediction is performed by combining the state estimation of the initial or previous processing window and noise statistical parameters to obtain innovative residuals.

[0055] It should be noted that the state estimation vector consists of the instantaneous amplitude and instantaneous rate of change of the acoustic signal after notching. The state transition matrix is ​​set according to the approximately uniform change relationship between adjacent sampling points. The observation matrix maps the state estimation vector to the instantaneous amplitude of the acoustic signal after notching. The noise statistics parameters include the process noise covariance and the observation noise covariance. The initial processing window uses empirical initial values, and the subsequent processing windows use the updated values ​​output after the previous processing window completes the last step of this embodiment. Kalman prediction takes the state estimation vector and noise statistics parameters as input and outputs the predicted observation value at the current time. The difference between the predicted observation value and the measured value of the acoustic signal after notching at the current time is used as the innovation residual.

[0056] The differential residual of the notch filter, the side protection residual, and the innovation residual are mapped to the same order of representation. Homology is determined according to the relative order distance, envelope direction, and phase continuity relationship. Homology innovation residuals are defined as structural innovation quantities and form fault candidate components, while non-homology innovation residuals are defined as random innovation quantities.

[0057] It should be noted that the notch differential residual and the side protection residual are two reference residuals generated by the preceding notch circuit, and the innovation residual is the deviation generated by the Kalman prediction. After the three are unified by order coordinates, the common source relationship is determined by sampling point by point. The results are divided into two categories: fault candidate components and random innovation quantities, which are sent to the subsequent state correction and noise update channels respectively.

[0058] The noise statistics parameters are updated using the random innovation quantity and the random noise in the notch-filtered acoustic signal is suppressed. The state estimation is corrected using the fault candidate component, and the noise-reduced fault acoustic signal is obtained.

[0059] It should be noted that the noise reduction output process is as follows:

[0060] Sub-step 1.3.1: Update the observed noise covariance by the random innovation quantity according to its statistical variance in the current processing window, and use the updated noise statistical parameters as the input parameters for the Kalman prediction in the next processing window;

[0061] Sub-step 1.3.2: Generate a random noise estimate based on the random innovation amount, subtract it from the notch-filtered acoustic signal, and obtain an intermediate acoustic signal with random disturbances removed;

[0062] Sub-step 1.3.3: Apply the fault candidate components to the state estimation vector according to their order position and amplitude to obtain the updated state estimate corrected by the fault modulation information. Map the updated state estimate to the time domain to obtain the fault modulation reconstruction signal. Superimpose the fault modulation reconstruction signal on the intermediate acoustic signal output in sub-step 1.3.2 to obtain the noise-reduced fault acoustic signal.

[0063] The output denoised fault sound signal includes the following fields: processing window number, start and end time, sampling rate, denoised waveform, list of fault candidate component order positions within the corresponding order range, and root mean square value of random noise estimation. For example, a specific output example is: processing window number = W037, start and end time = 12.40 seconds to 12.80 seconds, sampling rate = 51200Hz, list of fault candidate component order positions = [3.07, 3.13, 6.07], root mean square value of random noise estimation = 0.018.

[0064] Additionally, for abnormal situations such as instantaneous packet loss or saturation clipping that may occur during the acquisition of the acoustic signal, the identification condition is that the number of consecutive abnormal sampling points within a single processing window exceeds a set proportion. When the abnormal amount is within the tolerable range, bridging compensation is performed by linear interpolation of the preceding and following normal sampling points to maintain the continuity of the order representation of the current processing window. When the abnormal amount exceeds the tolerable range, the current processing window is marked as invalid and notch filtering and Kalman prediction are skipped. The state estimate and noise statistics of the previous processing window are deferred to the next valid processing window as is. The processing flow is restarted after the abnormality is resolved.

[0065] Furthermore, in this embodiment, the sliding time window, hollow notch kernel, and Kalman recursion together constitute the continuous operation carrier of the scheme. The sliding time window is an analysis unit that advances along the time axis with a fixed or adjustable step size. Its window length determines the order resolution, and its overlap rate determines the smoothness of the connection between adjacent processing windows. The window length ranges from 0.2 to 2.0 seconds, with a typical value of 0.5 seconds. The overlap rate ranges from 25% to 75%, with a typical value of 50%. The hollow notch kernel is a one-dimensional coefficient sequence arranged along the order axis. Its length is equal to the number of order sampling points in the preset order neighborhood. It only takes effect in the notch stage of each processing window and is not retained across windows. The Kalman recursion retains the state estimation vector and noise statistics parameters across processing windows. The output of the previous processing window is directly used as the input of the next processing window.

[0066] Through the above technical solution, this embodiment introduces residual closure degree in the order domain as the basis for dividing the carrier core suppression region and the side protection region, constructs a hollow notch core with central suppression and peripheral preservation, and performs homology discrimination on the innovative residual after Kalman prediction to separate the structural innovation quantity and the random innovation quantity. This achieves the parallel consideration of suppressing normal order noise and preserving the modulation information of adjacent faults, and provides a fault modulation channel that will not be diluted by random statistics for subsequent state correction and noise update.

[0067] In an optional embodiment, the residual closure is formed based on the relative order distance, envelope change, and phase continuation relationship of the order residuals, specifically including:

[0068] For the order residuals in the preset order neighborhood on both sides of the candidate order carrier, calculate the relative order distance from the order sampling point corresponding to each order residual to the candidate order carrier, and convert it into distance proximity according to the distance attenuation law.

[0069] It should be noted that the calculation process of distance proximity is as follows: subtract the order center position of the candidate order carrier from the order coordinates of the order sampling points corresponding to each order residual and take the absolute value to obtain the relative order distance (unit: order). Substitute the relative order distance into the exponential decay law to obtain the distance proximity with a value between 0 and 1.

[0070] For example, the expression for calculating proximity is:

[0071] ;

[0072] in, The proximity of the sampling point corresponding to the k-th order residual is given by the distance between the sampling points of that order. This represents the relative order distance between sampling points of this order. It is the attenuation rate coefficient, with a value ranging from 0.5 to 2.0, and a typical value of 1.0;

[0073] Extract the envelope amplitude trajectory of the order residuals on both sides of the candidate order carrier, compare the same direction of the envelope amplitude trajectories on both sides and judge the corresponding change trend to obtain the envelope correspondence degree.

[0074] It should be noted that the formation process of envelope correspondence is as follows: For the left and right order residuals of the candidate order carrier, the amplitude of each order sampling point is extracted along the order and connected to form the left envelope amplitude trajectory and the right envelope amplitude trajectory; the left and right envelope amplitude trajectories are aligned point by point along the order direction, and the same direction of the two in the rising and falling trends is compared, and the correspondence between the two in the peak and valley positions is compared; the same direction ratio and the peak-valley correspondence ratio are fused according to the preset weight to form the envelope correspondence with a value between 0 and 1.

[0075] The phase change trajectory of the order residuals on both sides of the candidate order carrier is extracted, and the continuity deviation of the phase change trajectory on both sides is determined along the order to obtain the phase continuity.

[0076] It should be noted that the formation process of phase continuity is as follows: for the left and right order residuals of the candidate order carrier, the phase of each order sampling point is extracted along the order sequence and connected to form the left phase change trajectory and the right phase change trajectory; the phase difference of each sampling point of the phase change trajectory on both sides is calculated along the order direction and its absolute value is taken to obtain the continuity deviation; the continuity deviation is reverse normalized to the phase continuity with a value between 0 and 1 according to the preset mapping method, that is, the smaller the continuity deviation, the closer the phase continuity is to 1.

[0077] When the phase continuity is lower than the preset continuity condition, the distance proximity and envelope correspondence are reduced by a preset attenuation ratio. When the envelope correspondence does not meet the preset correspondence condition, the phase continuity is reduced by a preset attenuation ratio to obtain the distance proximity, envelope correspondence and phase continuity after conditional coupling.

[0078] It should be noted that the execution process of conditional coupling adjustment is as follows: the phase continuity is compared with the preset continuity condition. When the phase continuity is lower than the preset continuity condition, the distance proximity and envelope correspondence are multiplied by the preset attenuation ratio to obtain the adjustment result; otherwise, the original value is maintained. The envelope correspondence is compared with the preset correspondence condition. When the envelope correspondence does not meet the preset correspondence condition, the phase continuity is multiplied by the preset attenuation ratio to obtain the adjustment result; otherwise, the original value is maintained. The distance proximity, envelope correspondence, and phase continuity after conditional coupling are output.

[0079] Since the purpose of conditional coupling is to weaken the influence of the other two factors when the physical relationship is not closed, the preset continuous condition and the preset corresponding condition should be set with relatively strict values. The preset continuous condition is set to 0.4 to 0.7, with a typical value of 0.55; the preset corresponding condition is set to 0.4 to 0.7, with a typical value of 0.55; and the preset attenuation ratio is set to 0.3 to 0.7, with a typical value of 0.5.

[0080] The distance proximity, envelope correspondence, and phase continuity after conditional coupling are combined according to a preset fusion method to synthesize the residual closure of the neighborhood corresponding to the candidate carrier of the next order.

[0081] It should be noted that the synthesis process of residual closure is as follows: the distance proximity, envelope correspondence and phase continuity after conditional coupling are synthesized into a single scalar according to a preset fusion method. The preset fusion method is weighted product or weighted geometric mean. The scalar obtained by fusion is the residual closure of the neighborhood corresponding to the candidate carrier of that order, and the value is between 0 and 1.

[0082] It should be noted that the distance proximity, envelope correspondence, and phase continuity are all normalized to be dimensionless and have the same value range. When the three are involved in the same fusion operation, the dimensionality is guaranteed. The residual closure obtained by fusion can be directly compared with the division threshold. The comparison result is used to divide the carrier core suppression region and the side protection region.

[0083] Through the above technical solution, this embodiment transforms the relative order distance, envelope change, and phase continuity relationship into three dimensionless indices: distance proximity, envelope correspondence, and phase continuity, respectively. By using a conditional coupling method, the contribution of the other two indices is weakened when any one of the relationships is not closed. The resulting residual closure degree has a clear physical meaning rather than being a simple numerical superposition. This provides a basis for subsequent region division to distinguish between two types of order segments: "low energy but high modulation correlation" and "low energy and no correlation".

[0084] In an optional embodiment, the suppression filter coefficients and the guard filter coefficients are arranged in order of order to form a hollow notch kernel, specifically including:

[0085] Within the carrier core suppression region, the suppression reference strength is determined based on the carrier energy concentration corresponding to each order sampling point, and the suppression reference strength is adjusted by distance attenuation in combination with the relative order distance from the sampling point to the candidate order carrier to obtain the suppression filter coefficient.

[0086] It should be noted that the process of forming the suppression filter coefficient is as follows: the ratio of the energy peak at the candidate order carrier to the average energy in its order neighborhood is taken as the carrier energy concentration degree. The carrier energy concentration degree is converted into a suppression reference strength with a value between 0 and 1 according to a preset mapping method. With the relative order distance as input, the suppression reference strength is scaled up and down by sampling point according to the distance attenuation law to obtain the suppression filter coefficient of each order sampling point in the carrier core suppression area.

[0087] Within the adjacent protection zone, the protection reference strength is determined based on the residual closure of the neighborhood of each order sampling point, and the protection reference strength is monotonically constrained and adjusted in conjunction with the relative order distance from the sampling point to the candidate order carrier to obtain the protection filter coefficient.

[0088] It should be noted that the formation process of the protection filter coefficient is as follows: the residual closure degree of the neighborhood of each order sampling point in the side protection zone is directly taken as the protection benchmark strength; the protection benchmark strength is adjusted point by point according to the monotonic constraint method with the relative order distance as input, so that the value of the protection filter coefficient is monotonically non-increasing with the increase of the relative order distance, thus obtaining the protection filter coefficient of each order sampling point in the side protection zone.

[0089] A transition order zone is defined at the junction of the carrier core suppression region and the side protection zone. For the order sampling points in the transition order zone, the residual closure of the corresponding order neighborhood is used as the fusion weight to weight the adjacent suppression filter coefficients and protection filter coefficients to obtain the transition filter coefficients.

[0090] It should be noted that the formation process of the transition filter coefficients is as follows: several order sampling points are taken on both sides of the adjacent boundary of the carrier core suppression region and the side protection region to form a transition order band; for each order sampling point in the transition order band, the residual closure of the order neighborhood to which the sampling point belongs is used as the fusion weight, and the suppression filter coefficients and protection filter coefficients at its adjacent positions are weighted and fused according to the weight. The higher the residual closure, the more the fusion result is biased towards the protection filter coefficient, and vice versa; the transition filter coefficients of each order sampling point in the transition order band are output.

[0091] Because a narrow transition band can cause abrupt changes in coefficients and lead to spectral leakage, while a wide band can encroach on the effective range of the carrier core suppression region or the side protection region, the width of the transition band is set to 0.05 to 0.2, with a typical value of 0.1.

[0092] The suppression filter coefficient, the transition filter coefficient, and the guard filter coefficient are connected sequentially according to their order to form a hollow notch core.

[0093] It should be noted that the connection process of the hollow notch core is as follows: the protection filter coefficient of the left side protection zone, the transition filter coefficient of the left transition order zone, the suppression filter coefficient of the carrier core suppression zone, the transition filter coefficient of the right transition order zone, and the protection filter coefficient of the right side protection zone are spliced ​​in sequence from left to right along the order axis to obtain a hollow notch core that presents a "retained at both ends, suppressed in the center, and smooth at the boundary" shape along the order axis.

[0094] It should be noted that the carrier energy concentration is obtained by the ratio of the energy peak at the candidate carrier in the order spectrum to the average energy in its order neighborhood. Together with the residual closure, it serves as the two inputs in this embodiment to support the values ​​of the suppression reference strength and the protection reference strength, so that the central position of the hollow notch core is driven by energy evidence and the peripheral position is driven by modulation relationship evidence. The two are connected numerically without abrupt changes through the transition order band.

[0095] Through the above technical solution, this embodiment introduces a transition order band and uses the residual closure degree as the fusion weight to perform a weighted transition between the suppression filter coefficient and the protection filter coefficient, so that the hollow notch kernel does not have a steep jump in the order domain, providing a filter structure for weighted notch reconstruction that will not generate pseudo signals due to abrupt changes in coefficients.

[0096] In an optional embodiment, the notch differential residual, side protection residual, and innovative residual are mapped to the same order of representation, and homology is determined according to relative order distance, envelope direction, and phase continuity relationship, specifically including:

[0097] The differential residual, side protection residual, and innovative residual are uniformly mapped to the same order representation according to the order coordinates of the current processing window, and the amplitude and phase of the three are normalized to obtain the differential residual component, protection residual component, and innovative residual component.

[0098] It should be noted that the unified order mapping and normalization process is as follows: using the order coordinates of the current processing window as a reference, time-order transformation is performed on the notch differential residual, side protection residual, and innovation residual respectively, so that the three are aligned along the same order axis; the amplitude of each of the three is divided by its maximum amplitude in the current processing window, and the phase of each of the three is subtracted from its phase reference at the candidate order carrier, to obtain the differential residual component, protection residual component, and innovation residual component after amplitude normalization and phase zeroing.

[0099] For each order sampling point in the innovative residual component, its order position is matched with the order neighborhood of the differential residual component and the protection residual component to obtain the neighborhood matching discrimination result.

[0100] It should be noted that the execution process of neighbor matching is as follows: for each order sampling point in the innovative residual component, scan along the order axis to see if there are significant order components in the differential residual component and the protection residual component within the set order tolerance range near the order sampling point; if there are, the neighbor matching discrimination result of the order sampling point is set as a hit and the order neighborhood to which it belongs is recorded; otherwise, it is set as a miss.

[0101] For the innovative residual component order sampling points that meet the neighbor matching discrimination results, the direction of their envelope amplitude change is compared with the direction of their envelope amplitude change in the same order neighboring protected residual component to obtain the envelope direction matching discrimination results.

[0102] It should be noted that the execution process of envelope direction matching is as follows: for the order sampling points of the innovative residual components whose neighboring matching results are hits, extract the direction of the change of the envelope amplitude with the order in the neighborhood of that order; extract the direction of the change of the envelope amplitude with the order of the protected residual components in the same neighborhood; and take whether the two directions are in the same direction as the envelope direction matching result, setting them as hit if they are in the same direction and as miss if they are in opposite directions.

[0103] For the innovative residual component order sampling points that meet the neighbor matching discrimination results, the phase direction along the order direction is continuously compared with the phase direction of the differential residual component or the guard residual component in the same order neighborhood to obtain the phase continuity matching discrimination results.

[0104] It should be noted that the execution process of phase continuation matching is as follows: for the order sampling points of the innovative residual components whose neighboring matching discrimination results are hits, extract their phase direction along the order direction; extract the phase direction of the differential residual components or guard residual components in the same order neighborhood along the order direction; take whether the phase difference between the two along the order direction is within the preset continuous criterion range as the phase continuation matching discrimination result, and set it as a hit if it falls within the range, otherwise set it as a miss.

[0105] When the neighbor matching discrimination result, the envelope direction matching discrimination result, and the phase continuation matching discrimination result all satisfy the preset homogeneity condition, the corresponding innovative residual component order sampling point is designated as a structural innovation quantity and forms a fault candidate component; otherwise, the corresponding innovative residual component order sampling point is designated as a random innovation quantity.

[0106] It should be noted that the preset co-origin condition is that the neighbor matching discrimination result, the envelope direction matching discrimination result, and the phase continuation matching discrimination result are all hits; the innovative residual component order sampling points that are hits in all three discrimination results, together with their order position, amplitude, and phase, are defined as structural innovation quantities, and are aggregated according to order neighborhood to form fault candidate components; the remaining order sampling points are defined as random innovation quantities;

[0107] Because a narrow order tolerance can lead to missed detection of homology, while a wide tolerance can introduce false homology, the order tolerance range for nearest neighbor matching is 0.05 to 0.3, with a typical value of 0.15; the phase difference range for the phase continuation criterion is 0.1 to 0.5 radians, with a typical value of 0.3 radians.

[0108] It should be noted that the differential residual component and the protective residual component originate from the notch filter element, corresponding to the energy trajectory weakened by the hollow notch filter core and the disturbance trajectory preserved on the periphery, respectively; the innovative residual component originates from the Kalman prediction element, corresponding to the deviation between the predicted observation value and the measured value; the three are cross-discriminated under a unified order coordinate system, so that the deviation on the Kalman side is no longer interpreted as random noise in isolation, but forms a traceable homology with the two reference residuals on the notch filter side.

[0109] Through the above technical solution, this embodiment maps the innovation residual and the two reference residuals on the notch side to the same order coordinates and then compares them item by item from three dimensions: order proximity, envelope direction and phase continuation. The innovation residual, which was originally treated as a single random disturbance, is split into fault candidate components and random innovation quantities. This ensures that subsequent noise statistics updates only absorb random components and state corrections only receive fault components, thus avoiding the dilution of fault modulation information by noise statistics in the state estimation stage.

[0110] In an optional embodiment, before dividing the processing window into multiple processing windows according to the sliding time window, a phase reference holding window mechanism based on operating condition fluctuations is further included, specifically including:

[0111] The operating sound signal is scanned for a short time according to the preset initial window. The main order drift, carrier phase change, and operating condition indication obtained from the synchronously acquired speed indication or load indication are extracted within the initial window. The operating condition fluctuation index is formed based on the main order drift, carrier phase change, and operating condition indication.

[0112] It should be noted that the formation process of the operating condition fluctuation index is as follows: A preset initial window is used as a short-time scanning unit to slide across the operating acoustic signal. Order characterization is performed on each initial window to obtain the instantaneous order acoustic spectrum. The displacement of the main order peak along the order axis between adjacent initial windows is extracted as the main order drift (unit: order). The phase difference at the main order peak position between adjacent initial windows is extracted as the carrier phase change (unit: radians). The operating condition indication is obtained from the standard deviation of the synchronously acquired speed indication or load indication within the initial window. The main order drift, carrier phase change, and operating condition indication are converted into dimensionless components according to their respective normalized references and then weighted and summed to obtain the operating condition fluctuation index.

[0113] When the operating condition fluctuation index is not higher than the preset stability threshold, the window length of the sliding time window is extended by the preset upper limit and the basic overlap rate is adopted. When the operating condition fluctuation index is higher than the preset stability threshold, the window length of the sliding time window is shortened by the preset lower limit and the overlap rate is increased, so as to obtain the window length and overlap rate configuration of the current processing window.

[0114] It should be noted that the execution process for configuring the window length and overlap rate is as follows: the operating condition fluctuation index is compared with the preset stable threshold; when the operating condition fluctuation index is not higher than the preset stable threshold, the window length of the sliding time window is set to the preset upper limit value, and the overlap rate is set to the basic overlap rate; when the operating condition fluctuation index is higher than the preset stable threshold, the window length of the sliding time window is set to the preset lower limit value, and the overlap rate is set to an increased value higher than the basic overlap rate; the window length and overlap rate configuration of the current processing window are output.

[0115] Since the upper and lower limits of the window length determine the trade-off between order resolution and time tracking capability, the preset upper limit of the window length is set to 1.0 to 2.0 seconds, with a typical value of 1.0 seconds; the preset lower limit of the window length is set to 0.2 to 0.5 seconds, with a typical value of 0.3 seconds; the basic overlap rate is set to 25% to 50%, with a typical value of 50%; the improved overlap rate is set to 60% to 80%, with a typical value of 75%; and the preset stability threshold is set to 0.2 to 0.5, with a typical value of 0.3.

[0116] The acoustic signal is segmented according to the window length and overlap rate of the current processing window, and the order representation of adjacent processing windows is phase-shifted and corrected using the carrier phase of adjacent processing windows in the overlapping interval as the alignment reference, so as to obtain the processing window sequence with phase reference alignment.

[0117] It should be noted that the phase shift correction process is as follows: the running acoustic signal is divided along the time axis according to the current window length and overlap rate to obtain a sequence of processing windows that are connected end to end; for each pair of adjacent processing windows, the phase at the candidate order carrier in the overlapping interval is extracted, and the offset of the next processing window relative to the previous processing window in that phase is calculated; this offset is used as the shift amount to perform overall shift correction on the order representation of the next processing window along the phase dimension; after processing each pair, a sequence of processing windows with phase reference alignment is obtained.

[0118] It should be noted that the applicable premise for phase shift correction is that the speed change between adjacent processing windows does not cross a significant shift or impact event; when the synchronously acquired speed indication changes abruptly between adjacent processing windows, the current phase alignment chain should be stopped, and the first processing window after the impact event should be used as the new phase reference starting point to re-establish the processing window sequence.

[0119] Through the above technical solution, this embodiment performs linkage adjustment on the window length and overlap rate of the sliding time window based on the operating condition fluctuation index, and performs phase shift correction on the order characterization of adjacent processing windows using the carrier phase in the overlapping interval as the alignment reference, thereby obtaining a processing window sequence with a comparable phase reference. This provides a phase reference that is not affected by operating condition fluctuations for the phase continuity discrimination and cross-window homology confidence assessment in the subsequent residual closure formation.

[0120] In an optional embodiment, after defining the homogeneous innovation residual as a structural innovation quantity and forming a fault candidate component, the method further includes a cross-window confirmation mechanism based on multi-window homogeneity confidence, specifically including:

[0121] The fault candidate components in the current processing window are matched one by one with the fault candidate components recorded in the previous processing windows according to their order position, envelope change trajectory and phase continuation direction, under the processing window sequence aligned with the phase reference, to obtain the multi-window homology confidence of the current fault candidate components.

[0122] It should be noted that the formation process of the multi-window homology confidence score is as follows: Under the processing window sequence with phase reference alignment, for each fault candidate component in the current processing window, candidates whose order position is within the preset order tolerance range are retrieved from the fault candidate component records of several previous processing windows; for each retrieved candidate, the order position deviation, envelope change trajectory similarity, and phase continuation direction deviation between the current fault candidate component and the candidate are compared respectively, and the three comparison results are combined into a single matching score according to a preset fusion method; the single matching scores of all candidates are weighted and summed according to window distance attenuation to obtain the multi-window homology confidence score of the current fault candidate component;

[0123] Since too few preceding processing windows for matching will result in insufficient confidence statistics, while too many will introduce operating condition drift interference, the number of preceding processing windows for matching ranges from 3 to 10, with a typical value of 5; the preset confirmation condition ranges from 0.6 to 0.85, with a typical value of 0.75; and the preset mismatch condition ranges from 0.2 to 0.4, with a typical value of 0.3.

[0124] When the confidence level of multi-window homology meets the preset confirmation conditions, the corresponding fault candidate component is marked as a valid fault candidate component.

[0125] When the confidence level of multi-window homology does not meet the preset confirmation condition but is not lower than the preset mismatch condition, the corresponding fault candidate component is placed into the observation queue and its order position and envelope change trajectory are retained.

[0126] When the confidence level of multi-window homology is consistently lower than the preset mismatch condition, the corresponding fault candidate component is converted into a random innovation quantity and included in the random innovation quantity set.

[0127] It should be noted that the execution process of the three types of branch processing is as follows: the multi-window homology confidence score is compared with the preset confirmation condition and the preset mismatch condition respectively; when the confidence score meets the preset confirmation condition, the corresponding fault candidate component is marked as a valid fault candidate component and sent to the subsequent state correction channel; when the confidence score is between the preset confirmation condition and the preset mismatch condition, the corresponding fault candidate component, its order position, and envelope change trajectory are written into the observation queue, waiting for the subsequent processing window to continue the judgment. The fault candidate components in the observation queue are recalculated for multi-window homology confidence score and the classification is updated after each new processing window is completed; when the confidence score is continuously lower than the preset mismatch condition in several consecutive processing windows, the corresponding fault candidate component is transferred from the structural innovation channel to the random innovation channel and included in the random innovation quantity set to participate in the noise statistics update.

[0128] It should also be noted that the observation queue is a circular buffer that stores candidate fault components to be judged in a first-in-first-out order. Its maximum capacity ranges from 20 to 100, with a typical value of 50. When the queue is full, the earliest candidate fault component that entered the queue and has not yet met the preset confirmation conditions is popped out and reclassified according to the three branches mentioned above based on its current confidence level. The number of consecutive processing windows for determining "continuously below the preset mismatch condition" ranges from 3 to 8, with a typical value of 5.

[0129] Through the above technical solution, this embodiment performs multi-dimensional matching between the fault candidate components obtained by single-window discrimination and the fault candidate components at the same position in the previous window under the processing window sequence aligned with the phase reference, and processes them separately according to three branches based on the multi-window homology confidence, thereby obtaining effective fault candidate components supported by mechanical continuity evidence, and providing the state correction channel with an input that is not contaminated by transient interference.

[0130] In an optional embodiment, after updating the noise statistics parameters using random innovation and suppressing random noise in the notch-filtered acoustic signal, and correcting the state estimation using fault candidate components to obtain the denoised fault acoustic signal, a write-back correction mechanism for the side protection boundary is also included, specifically including:

[0131] Extract the valid fault candidate components and their multi-window homology confidence scores from the current processing window output. Then, group the valid fault candidate components into the side protection zone of the corresponding candidate order carrier according to their order positions to obtain the order neighborhood label to be written back.

[0132] It should be noted that the formation process of the order neighborhood marker to be written back is as follows: extract all valid fault candidate components and their corresponding multi-window homology confidence scores from the output of the current processing window; for each valid fault candidate component, find the nearest candidate order carrier in the hollow notch kernel of the current processing window according to its order position, and collect the valid fault candidate component and its confidence score into the side protection zone of the candidate order carrier; register the collected side protection zone as the order neighborhood marker to be written back, and each marker includes the candidate order carrier number, order position interval, number of collected valid fault candidate components and average confidence score.

[0133] For the order neighborhood that is marked as a valid fault candidate component in several consecutive processing windows, during the construction of the hollow notch core in the next processing window, the order range of the corresponding side protection zone is expanded along the direction of increasing protection strength and the corresponding protection filter coefficient is adjusted.

[0134] For the order neighborhood where the confidence of multi-window homology continues to decrease in several consecutive processing windows, during the construction of the hollow notch kernel in the next processing window, the order range of the corresponding side protection zone is shrunk along the direction of decreasing protection strength and the corresponding protection filter coefficient is adjusted.

[0135] It should be noted that the write-back correction process is as follows: For each neighborhood to be written back, it is counted whether it has been continuously marked in the most recent processing windows; when it has been continuously marked and the average confidence is rising or stable, the expansion branch is entered: the boundary of the side protection zone corresponding to the neighborhood is expanded outward by a preset increment away from the candidate carrier, and the protection filter coefficients within the range of that order are increased by a preset increment; when it has been continuously marked but the average confidence decreases window by window, the contraction branch is entered: the boundary of the side protection zone corresponding to the neighborhood is contracted inward by a preset increment towards the candidate carrier, and the protection filter coefficients within the range of that order are decreased by a preset increment; the expansion branch and the contraction branch are mutually exclusive and are not executed simultaneously in the same neighborhood; the configuration of the side protection zone after expansion or contraction is used as the input for constructing the hollow notch kernel in the next processing window;

[0136] Because excessively large write-back increments can cause drastic jumps between processing windows in the side protection zone, while excessively small increments lose their feedback significance, the preset increment range for the expansion or contraction of the side protection zone boundary is 0.05 to 0.3, with a typical value of 0.1; the preset increment range for adjusting the protection filter coefficient is 0.05 to 0.2, with a typical value of 0.1; and the number of processing windows for determining continuous labeling or a continuous decrease in confidence is 3 to 8, with a typical value of 4.

[0137] The write-back correction only applies to the side protection zone and its protection filter coefficients, and does not include the carrier core suppression zone in the side protection zone.

[0138] It should be noted that the execution process of the write-back boundary constraint is as follows: before each extension branch is executed, it is checked whether the boundary of the extended side protection zone is still outside the carrier core suppression zone. If the extended boundary enters the range of the carrier core suppression zone, the boundary is truncated at the outer edge of the carrier core suppression zone. The order range of the carrier core suppression zone and the value of the suppression filter coefficient remain unchanged during the write-back correction process. Only the side protection zone and its protection filter coefficient participate in the write-back adjustment.

[0139] The output of the write-back correction includes: candidate carrier number, corrected side-protection zone order range, corrected protection filter coefficient sequence, correction branch type, the number of consecutive processing windows used for this correction, and the trend of average confidence change. A specific output example is: candidate carrier number = C03, corrected side-protection zone order range. The corrected protection filter coefficient sequence is [0.62, 0.71, 0.78, 0.81], the corrected branch type is extended, the number of consecutive processing windows is 4, and the average confidence trend is increasing.

[0140] Through the above technical solution, this embodiment, by collecting the effective fault candidate components and their multi-window homology confidence in order of order neighborhood, writes them back to the side protection zone and protection filter coefficient of the next processing window, and uses the carrier core suppression zone as an inviolable boundary to constrain the write-back range. This upgrades the entire notch filter-Kalman cascade from a single serial process to a closed-loop structure with mutual feedback between the front and rear stages, providing a feedback channel for the accumulation and retention of early fault modulation information during long-term operation of the transmission device.

[0141] See Figure 2 As shown, this scheme proposes an acoustic signal noise reduction system based on notch filtering and improved Kalman filtering to implement the aforementioned acoustic signal noise reduction method based on notch filtering and improved Kalman filtering, including:

[0142] The order spectrum generation module is used to acquire the operating sound signal of the transmission device, divide it into multiple processing windows according to the sliding time window, and perform order characterization on each processing window to obtain the order spectrum.

[0143] The closure degree forming module is used to determine candidate order carriers in the order spectrum, extract order residuals in the preset order neighborhoods on both sides of the carriers, and form residual closure degree based on the relative order distance, envelope change and phase continuation relationship of the order residuals.

[0144] The notch core generation module is used to divide the carrier core suppression region and the side protection region within a preset order neighborhood using the residual closure degree. For the order sampling points of the carrier core suppression region, the suppression filter coefficient is determined by the carrier energy concentration degree and the relative order distance. For the order sampling points of the side protection region, the protection filter coefficient is determined by the residual closure degree and the relative order distance. The suppression filter coefficient and the protection filter coefficient are arranged in order of order to form a hollow notch core.

[0145] The notch residual determination module is used to use a hollow notch core to weight the order acoustic spectrum and reconstruct the notch, obtain the notch-before acoustic signal, determine the notch differential residual based on the difference between the running acoustic signal and the notch-before acoustic signal, and determine the side protection residual based on the retained acoustic signal of the side protection zone.

[0146] The Kalman prediction module is used to perform Kalman prediction based on the acoustic signal after notch filtering as the observation sequence, combined with the state estimation and noise statistics parameters of the initial or previous processing window, to obtain innovative residuals.

[0147] The same-origin discrimination module is used to map the notch differential residual, side protection residual and innovation residual to the same order of representation, and to perform same-origin discrimination according to the relative order distance, envelope direction and phase continuity relationship. The same-origin innovation residual is defined as structural innovation quantity and formed as fault candidate component, and the non-same-origin innovation residual is defined as random innovation quantity.

[0148] The noise reduction and extraction module is used to update the noise statistics parameters using random innovation quantities and suppress random noise in the notch-filtered acoustic signal. It also uses fault candidate components to correct the state estimation and obtain the noise-reduced fault acoustic signal.

[0149] In another embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above embodiments.

[0150] In one embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps described above.

[0151] In one embodiment, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and executes the computer instructions, causing the computer device to perform the steps described above.

[0152] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for acoustic signal noise reduction based on notch filtering and improved Kalman filtering, characterized in that, The method includes: The operating sound signal of the transmission device is acquired, divided into multiple processing windows according to the sliding time window, and the order of each processing window is characterized to obtain the order sound spectrum. Candidate order carriers are identified in the order spectrum, and order residuals are extracted in the preset order neighborhoods on both sides of them. The residual closure degree is formed based on the relative order distance, envelope change and phase continuation relationship of the order residuals. The carrier core suppression region and the side protection region are divided in the preset order neighborhood using the residual closure degree. For the order sampling points of the carrier core suppression region, the suppression filter coefficient is determined by the carrier energy concentration degree and the relative order distance. For the order sampling points of the side protection region, the protection filter coefficient is determined by the residual closure degree and the relative order distance. The suppression filter coefficient and the protection filter coefficient are arranged in order of order to form a hollow notch core. The hollow notch core is used to weight the order acoustic spectrum and reconstruct the notch signal to obtain the notch signal. The notch differential residual is determined based on the difference between the running acoustic signal and the notch signal. The side protection residual is determined based on the retained acoustic signal of the side protection zone. Using the acoustic signal after notch filtering as the observation sequence, Kalman prediction is performed by combining the state estimation of the initial or previous processing window and noise statistical parameters to obtain innovative residuals. The differential residual of the notch filter, the side protection residual, and the innovation residual are mapped to the same order of representation. Homology is determined according to the relative order distance, envelope direction, and phase continuity relationship. Homology innovation residuals are defined as structural innovation quantities and form fault candidate components, while non-homology innovation residuals are defined as random innovation quantities. The noise statistics parameters are updated using the random innovation quantity and the random noise in the notch-filtered acoustic signal is suppressed. The state estimation is corrected using the fault candidate component, and the noise-reduced fault acoustic signal is obtained.

2. The method according to claim 1, characterized in that, The residual closure is determined based on the relative order distance, envelope change, and phase continuation relationship of the order residuals, specifically including: For the order residuals in the preset order neighborhood on both sides of the candidate order carrier, calculate the relative order distance from the order sampling point corresponding to each order residual to the candidate order carrier, and convert it into distance proximity according to the distance attenuation law. Extract the envelope amplitude trajectory of the order residuals on both sides of the candidate order carrier, compare the same direction of the envelope amplitude trajectories on both sides and judge the corresponding change trend to obtain the envelope correspondence degree. The phase change trajectory of the order residuals on both sides of the candidate order carrier is extracted, and the continuity deviation of the phase change trajectory on both sides is determined along the order to obtain the phase continuity. When the phase continuity is lower than the preset continuity condition, the distance proximity and envelope correspondence are reduced by a preset attenuation ratio. When the envelope correspondence does not meet the preset correspondence condition, the phase continuity is reduced by a preset attenuation ratio to obtain the distance proximity, envelope correspondence and phase continuity after conditional coupling. The distance proximity, envelope correspondence, and phase continuity after conditional coupling are combined according to a preset fusion method to synthesize the residual closure of the neighborhood corresponding to the candidate carrier.

3. The method according to claim 2, characterized in that, The suppression filter coefficients and the guard filter coefficients are arranged in order of order to form a hollow notch kernel, specifically including: Within the carrier core suppression region, the suppression reference strength is determined based on the carrier energy concentration corresponding to each order sampling point, and the suppression reference strength is adjusted by distance attenuation in combination with the relative order distance from the sampling point to the candidate order carrier to obtain the suppression filter coefficient. Within the adjacent protection zone, the protection reference strength is determined based on the residual closure of the neighborhood of each order sampling point, and the protection reference strength is monotonically constrained and adjusted in conjunction with the relative order distance from the sampling point to the candidate order carrier to obtain the protection filter coefficient. A transition order zone is defined at the junction of the carrier core suppression region and the side protection zone. For the order sampling points in the transition order zone, the residual closure of the corresponding order neighborhood is used as the fusion weight to weight the adjacent suppression filter coefficients and protection filter coefficients to obtain the transition filter coefficients. The suppression filter coefficient, transition filter coefficient, and protection filter coefficient are connected sequentially according to their order to form a hollow notch core.

4. The method according to claim 1, characterized in that, The notch differential residual, side protection residual, and innovative residual are mapped to the same order of representation, and homology is determined according to relative order distance, envelope direction, and phase continuity relationship. Specifically, this includes: The differential residual, side protection residual, and innovative residual are uniformly mapped to the same order representation according to the order coordinates of the current processing window, and the amplitude and phase of the three are normalized to obtain the differential residual component, protection residual component, and innovative residual component. For each order sampling point in the innovative residual component, its order position is matched with the order neighborhood of the differential residual component and the protection residual component to obtain the neighborhood matching discrimination result. For the innovative residual component order sampling points that meet the neighbor matching discrimination results, the direction of their envelope amplitude change is compared with the direction of their envelope amplitude change in the same order neighboring protected residual component to obtain the envelope direction matching discrimination results. For the innovative residual component order sampling points that meet the neighbor matching discrimination results, the phase direction along the order direction is continuously compared with the phase direction of the differential residual component or the guard residual component in the same order neighborhood to obtain the phase continuity matching discrimination results. When the neighbor matching discrimination result, the envelope direction matching discrimination result, and the phase continuation matching discrimination result all satisfy the preset homogeneity condition, the corresponding innovative residual component order sampling point is designated as a structural innovation quantity and forms a fault candidate component; otherwise, the corresponding innovative residual component order sampling point is designated as a random innovation quantity.

5. The method according to claim 1, characterized in that, Before dividing the processing window into multiple processing windows according to the sliding time window, a phase reference holding window mechanism based on operating condition fluctuations is also included, specifically including: The operating sound signal is scanned for a short time according to the preset initial window. The main order drift, carrier phase change, and operating condition indication obtained from the synchronously acquired speed indication or load indication are extracted within the initial window. The operating condition fluctuation index is formed based on the main order drift, carrier phase change, and operating condition indication. When the operating condition fluctuation index is not higher than the preset stability threshold, the window length of the sliding time window is extended by the preset upper limit and the basic overlap rate is adopted. When the operating condition fluctuation index is higher than the preset stability threshold, the window length of the sliding time window is shortened by the preset lower limit and the overlap rate is increased, so as to obtain the window length and overlap rate configuration of the current processing window. The acoustic signal is segmented according to the window length and overlap rate of the current processing window, and the order representation of adjacent processing windows is phase-shifted and corrected using the carrier phase of adjacent processing windows in the overlapping interval as the alignment reference, so as to obtain the processing window sequence with phase reference alignment.

6. The method according to claim 5, characterized in that, After defining the residuals of homogeneous innovations as structural innovation quantities and forming candidate fault components, a cross-window confirmation mechanism based on multi-window homogeneity confidence is also included, specifically: The fault candidate components in the current processing window are matched one by one with the fault candidate components recorded in the previous processing windows according to their order position, envelope change trajectory and phase continuation direction, under the processing window sequence aligned with the phase reference, to obtain the multi-window homology confidence of the current fault candidate components. When the confidence level of multi-window homology meets the preset confirmation conditions, the corresponding fault candidate component is marked as a valid fault candidate component. When the confidence level of multi-window homology does not meet the preset confirmation condition but is not lower than the preset mismatch condition, the corresponding fault candidate component is placed into the observation queue and its order position and envelope change trajectory are retained. When the confidence level of multi-window homology is consistently lower than the preset mismatch condition, the corresponding fault candidate component is converted into a random innovation quantity and included in the random innovation quantity set.

7. The method according to claim 6, characterized in that, After updating the noise statistics parameters using random innovation and suppressing random noise in the notch-filtered acoustic signal, and correcting the state estimate using fault candidate components to obtain the denoised fault acoustic signal, a write-back correction mechanism for the side protection boundary is also included, specifically including: Extract the valid fault candidate components and their multi-window homology confidence scores from the current processing window output. Then, group the valid fault candidate components into the side protection zone of the corresponding candidate order carrier according to their order positions to obtain the order neighborhood label to be written back. For the order neighborhood that is marked as a valid fault candidate component in several consecutive processing windows, during the construction of the hollow notch core in the next processing window, the order range of the corresponding side protection zone is expanded along the direction of increasing protection strength and the corresponding protection filter coefficient is adjusted. For the order neighborhood where the confidence of multi-window homology continues to decrease in several consecutive processing windows, during the construction of the hollow notch kernel in the next processing window, the order range of the corresponding side protection zone is shrunk along the direction of decreasing protection strength and the corresponding protection filter coefficient is adjusted. The write-back correction only applies to the side protection zone and its protection filter coefficients, and does not include the carrier core suppression zone in the side protection zone.

8. A sound signal noise reduction system based on notch filtering and improved Kalman filtering, characterized in that, The method for implementing acoustic signal noise reduction based on notch filtering and improved Kalman filtering as described in any one of claims 1-7 includes: The order spectrum generation module is used to acquire the operating sound signal of the transmission device, divide it into multiple processing windows according to the sliding time window, and perform order characterization on each processing window to obtain the order spectrum. The closure degree forming module is used to determine candidate order carriers in the order spectrum, extract order residuals in the preset order neighborhoods on both sides of the carriers, and form residual closure degree based on the relative order distance, envelope change and phase continuation relationship of the order residuals. The notch core generation module is used to divide the carrier core suppression region and the side protection region within a preset order neighborhood using the residual closure degree. For the order sampling points of the carrier core suppression region, the suppression filter coefficient is determined by the carrier energy concentration degree and the relative order distance. For the order sampling points of the side protection region, the protection filter coefficient is determined by the residual closure degree and the relative order distance. The suppression filter coefficient and the protection filter coefficient are arranged in order of order to form a hollow notch core. The notch residual determination module is used to use a hollow notch core to weight the order acoustic spectrum and reconstruct the notch, obtain the notch-before acoustic signal, determine the notch differential residual based on the difference between the running acoustic signal and the notch-before acoustic signal, and determine the side protection residual based on the retained acoustic signal of the side protection zone. The Kalman prediction module is used to perform Kalman prediction based on the acoustic signal after notch filtering as the observation sequence, combined with the state estimation and noise statistics parameters of the initial or previous processing window, to obtain innovative residuals. The same-origin discrimination module is used to map the notch differential residual, side protection residual and innovation residual to the same order of representation, and to perform same-origin discrimination according to the relative order distance, envelope direction and phase continuity relationship. The same-origin innovation residual is defined as structural innovation quantity and formed as fault candidate component, and the non-same-origin innovation residual is defined as random innovation quantity. The noise reduction and extraction module is used to update the noise statistics parameters using random innovation quantities and suppress random noise in the notch-filtered acoustic signal. It also uses fault candidate components to correct the state estimation and obtain the noise-reduced fault acoustic signal.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images