Grain depot belt conveyor fault sound and vibration feature extraction method based on time domain synchronous averaging

Abstract

The invention relates to the field of grain storage equipment, and discloses a grain depot belt conveyor fault sound and vibration feature extraction method based on time domain synchronous averaging, and the method comprises the following steps: S1, carrying out the multi-channel synchronous collection of sound and vibration signals in the operation process of a grain depot belt conveyor, and obtaining an original sound and vibration data set; s2, the original sound vibration data set is preprocessed, noise interference is removed, and data standardization is completed; and S3, constructing a time domain synchronous average model based on the belt conveyor operation period signal, performing synchronous average processing on the preprocessed sound vibration data, and extracting a periodic sound vibration component. By constructing a time domain synchronous average model based on an operation periodic signal, model parameters can be dynamically adjusted in combination with periodic characteristics of actual operation of a belt conveyor, and periodic fault sound and vibration components are specifically separated and extracted from sound and vibration signals of mixed aperiodic noise. Aperiodic interference factors such as grain flow noise and mechanical interference in the grain depot environment are effectively filtered out.

Description

Technical Field

[0001] This invention relates to the field of grain storage equipment technology, specifically to a method for extracting the acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging. Background Technology

[0002] As a key piece of equipment in grain storage and transportation, the operating status of belt conveyors directly affects the efficiency and safety of storage operations. During long-term operation, typical faults such as drum eccentricity and bearing wear generate periodic acoustic and vibration signals. However, due to non-periodic noises such as grain flow noise and mechanical interference in the grain storage environment, and the fact that the operating cycle of the belt conveyor is easily affected by load fluctuations, the periodic fault characteristics are severely masked. Existing fault diagnosis methods based on acoustic and vibration signals struggle to accurately separate periodic fault components from non-periodic interference when processing such noisy and periodically unstable signals, limiting the effectiveness of fault feature extraction and consequently affecting the accuracy of fault identification. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging, thus solving the problems mentioned in the background section.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a method for extracting the acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging, comprising the following steps:

[0005] S1. Simultaneously collect the acoustic and vibration signals of the grain depot belt conveyor during operation through multiple channels to obtain the original acoustic and vibration dataset;

[0006] S2. Preprocess the original acoustic and vibration dataset to remove noise interference and standardize the data.

[0007] S3. Construct a time-domain synchronous averaging model based on the belt conveyor operating cycle signal, perform synchronous averaging on the preprocessed acoustic vibration data, and extract the periodic acoustic vibration components.

[0008] S4. Extract features from the periodic acoustic vibration components to obtain a set of feature parameters containing fault information;

[0009] S5. Input the feature parameter set into the preset fault classification model to complete the fault type identification and generate the sound and vibration feature extraction results of the grain depot belt conveyor fault.

[0010] Preferably, S1 includes:

[0011] A multi-channel acquisition array composed of acoustic and vibration sensors is deployed at the drive end, driven end, and intermediate section of the belt conveyor.

[0012] The sampling frequency was set to 20-50kHz, and the acquisition time was no less than three operating cycles of the belt conveyor to obtain the original acoustic and vibration dataset composed of acoustic signals and vibration acceleration signals.

[0013] Preferably, S2 includes:

[0014] Wavelet thresholding was used to denoise the original acoustic and vibration dataset, and a soft thresholding function was set to suppress high-frequency noise components.

[0015] The denoised acoustic vibration data is mapped to the [0,1] interval using the Z-score normalization method to obtain a normalized acoustic vibration dataset.

[0016] Preferably, the construction of the time-domain synchronous averaging model based on the belt conveyor operating cycle signal includes:

[0017] Obtain the periodic reference value T0 and the periodic fluctuation coefficient δ from the belt conveyor's operating cycle signal;

[0018] Using the periodic baseline value T0 as the time axis unit, the length of the sliding time window is set to k×T0, where k is a positive integer and 1≤k≤5;

[0019] A dynamic weight allocation rule is constructed based on the periodic fluctuation coefficient δ, and differential weights are assigned to acoustic vibration data at different periodic positions within the window.

[0020] The model input is defined as preprocessed acoustic vibration data and time series labels, and the output is the predicted value of the periodic acoustic vibration component. The loss function is constructed by minimizing the deviation between the predicted value and the actual periodic component, and the parameter initialization and structure determination of the time-domain synchronous averaging model are completed.

[0021] Preferably, S3 includes:

[0022] The belt conveyor motor speed signal is acquired by an encoder, and the belt conveyor operating cycle T is calculated.

[0023] Based on the operating period T, the preprocessed acoustic and vibration data are divided into several periodic subsequences;

[0024] The time-domain alignment of each periodic subsequence is performed, and the time-domain synchronous average signal is obtained by periodically superimposing and averaging. The periodic acoustic vibration component is then extracted from it.

[0025] in:

[0026] When the operating cycle of the belt conveyor fluctuates, the length of the periodic subsequence is normalized by interpolation.

[0027] Set an average number threshold. When the signal-to-noise ratio improvement rate after iterative averaging is lower than the threshold, stop the time-domain synchronous averaging operation.

[0028] Preferably, S4 includes:

[0029] Temporal features of periodic acoustic vibration components are extracted, and peak value, kurtosis, variance, and impulse factor are calculated.

[0030] Frequency domain features of periodic acoustic vibration components are extracted, and power spectral density is obtained through Fourier transform. Feature frequencies and frequency band energy ratios are then extracted.

[0031] By fusing time-domain and frequency-domain features, a feature parameter set is constructed.

[0032] Principal component analysis is used to reduce the dimensionality of the feature parameter set, retaining principal components whose cumulative contribution rate exceeds a preset threshold;

[0033] The principal components after dimensionality reduction are scaled by minimax normalization to obtain a standardized feature matrix.

[0034] Preferably, the temporal feature extraction of the periodic acoustic vibration components includes:

[0035] The periodic acoustic vibration components are converted into a discrete time-domain signal sequence x(i), where i = 1, 2, ..., n, and n is the number of sampling points;

[0036] Calculate the peak value: Take the maximum value in the discrete time domain signal sequence x(i), that is, peak value = max|x(i)|;

[0037] Calculate kurtosis using K = (1 / n)∑[x(i)-μ] ∧ 4 / σ ∧ 4. Calculate, where μ is the sequence mean and σ is the sequence standard deviation;

[0038] Calculate variance: using the formula σ 2 = (1 / (n-1))Σ[x(i)-μ] ∧ 2. Calculate, where μ is the sequence mean;

[0039] Calculate the impulse factor: take the ratio of the peak value to the average absolute value of the sequence, that is, impulse factor = peak value / [(1 / n)∑|x(i)|].

[0040] Preferably, S5 includes:

[0041] The fault classification model is a support vector machine-based classification model, and its training process includes:

[0042] Obtain a historical fault acoustic and vibration feature sample set, and divide it into a training set and a test set;

[0043] Set the kernel function type and penalty parameters, train the model using the training set, and verify the model's classification accuracy using the test set.

[0044] Preferably, the historical fault acoustic and vibration feature sample set contains at least 5 types of standardized feature matrices corresponding to typical belt conveyor faults, with no less than 300 samples for each type of fault.

[0045] The training set and the test set were randomly divided in a 7:3 ratio, and stratified sampling was used to maintain the consistency of the distribution of fault types in the two sets.

[0046] The kernel function types include linear kernel functions, polynomial kernel functions, and radial basis function kernel functions, wherein the parameter γ of the radial basis function takes values ​​in the range of [0.01, 10], and the order of the polynomial kernel function takes values ​​in the range of [2, 5].

[0047] The penalty parameter C has a value range of [0.1, 100], and parameter combination training is performed by traversing the value range;

[0048] The maximum number of iterations during training is set to 1000. Training is stopped when the change in training error is less than 1e-5 over 50 consecutive iterations.

[0049] The model's classification accuracy, precision, and recall are calculated using the test set, and the weighted average of the three metrics is used as the basis for evaluating the model's performance.

[0050] Preferably, the interpolation method includes linear interpolation or cubic spline interpolation:

[0051] When the periodic fluctuation coefficient is ≤5%, linear interpolation is used to supplement the missing data by linear fitting of adjacent sampling points;

[0052] When the periodic fluctuation coefficient is greater than 5%, cubic spline interpolation is used to achieve a smooth transition of the periodic subsequence by constructing a cubic polynomial function.

[0053] This invention provides a method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging.

[0054] It has the following beneficial effects:

[0055] 1. This invention constructs a time-domain synchronous averaging model based on the operating cycle signal, which can dynamically adjust the model parameters according to the actual operating cycle characteristics of the belt conveyor. It can specifically separate and extract the periodic fault acoustic vibration component from the acoustic vibration signal mixed with non-periodic noise, effectively filtering non-periodic interference factors such as grain flow noise and mechanical interference in the grain depot environment, significantly improving the signal-to-noise ratio of the fault characteristic signal, thereby solving the problem that the periodic fault characteristics are masked by environmental noise and mechanical interference.

[0056] 2. This invention addresses the periodic fluctuations caused by load changes and other factors during belt conveyor operation. It normalizes the length of the periodic subsequence through interpolation and adaptively selects a suitable interpolation method based on the amplitude of the periodic fluctuations, ensuring the continuity and integrity of the periodic signal under different fluctuation levels. This reduces the interference of periodic fluctuations on feature extraction, making the extraction process of periodic fault features more stable and laying the foundation for accurate acquisition of subsequent feature parameters.

[0057] 3. This invention uses multi-channel synchronous acquisition of acoustic and vibration signals from different locations on the belt conveyor, which can comprehensively capture the fault characteristics of key parts of the belt conveyor. At the same time, by fusing time-domain and frequency-domain features for extraction and combining dimensionality reduction processing to remove redundant information, not only are the key fault-related features preserved, but also the interference of irrelevant information is reduced. This provides more identifiable feature parameters for fault classification and helps to improve the accuracy of feature extraction. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the method steps of the present invention. Detailed Implementation

[0059] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Example:

[0061] Please see the appendix Figure 1 This invention provides a method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging, comprising the following steps:

[0062] S1. Simultaneously collect the acoustic and vibration signals of the grain depot belt conveyor during operation through multiple channels to obtain the original acoustic and vibration dataset;

[0063] S2. Preprocess the original acoustic and vibration dataset to remove noise interference and standardize the data.

[0064] S3. Construct a time-domain synchronous averaging model based on the belt conveyor operating cycle signal, perform synchronous averaging on the preprocessed acoustic vibration data, and extract the periodic acoustic vibration components.

[0065] S4. Extract features from the periodic acoustic vibration components to obtain a set of feature parameters containing fault information;

[0066] S5. Input the feature parameter set into the preset fault classification model to complete the fault type identification and generate the sound and vibration feature extraction results of the grain depot belt conveyor fault.

[0067] Specifically, the process begins with multi-channel synchronous acquisition of acoustic and vibration signals during operation to ensure that the raw data covers the status information of all key parts of the conveyor belt. Then, preprocessing removes environmental noise and interference, laying the foundation for subsequent analysis. Next, a time-domain synchronous averaging model is constructed based on the conveyor belt's operating cycle to specifically extract periodic fault acoustic and vibration components masked by noise, overcoming the interference of non-periodic noise by focusing on periodic features. Then, by fusing time-domain and frequency-domain features, a feature parameter set containing fault information is constructed. Finally, a pre-defined fault classification model is used to identify the fault type. The entire process forms a closed loop of "acquisition-purification-focusing on periodic features-feature quantification-identification," solving the problem of periodic fault features being masked in the complex environment of grain depots and ensuring the effectiveness of the entire chain from raw signals to fault identification.

[0068] S1 includes:

[0069] A multi-channel acquisition array composed of acoustic and vibration sensors is deployed at the drive end, driven end, and intermediate section of the belt conveyor.

[0070] The sampling frequency was set to 20-50kHz, and the acquisition time was no less than three operating cycles of the belt conveyor to obtain the original acoustic and vibration dataset composed of acoustic signals and vibration acceleration signals.

[0071] Specifically, a combination of acoustic and vibration sensors is used to form a data acquisition array, which is deployed at the drive end (near the power source, where faults are prone to occur), the driven end (load-bearing point), and the middle section (the main area of ​​belt operation) of the conveyor belt to comprehensively capture acoustic and vibration signals from key parts of the conveyor belt. The sampling frequency is set to 20-50kHz, which can cover the mid-to-high frequency characteristics generated by mechanical faults (such as high-frequency vibrations from bearing wear) while avoiding redundant data; the acquisition time is no less than three operating cycles to ensure that the raw data contains complete periodic signal characteristics. Multi-location acquisition ensures the spatial integrity of the data, and reasonable sampling parameters ensure the capture of periodic features, providing high-quality raw data support for subsequent extraction of periodic components.

[0072] S2 includes:

[0073] Wavelet thresholding was used to denoise the original acoustic and vibration dataset, and a soft thresholding function was set to suppress high-frequency noise components.

[0074] The denoised acoustic vibration data is mapped to the [0,1] interval using the Z-score normalization method to obtain a normalized acoustic vibration dataset.

[0075] Specifically, the original acoustic and vibration dataset is denoised using wavelet thresholding, and high-frequency noise components are suppressed using a soft thresholding function. The multi-resolution characteristics of wavelet transform can accurately distinguish the high-frequency components of noise and signal, and soft thresholding can preserve signal edge features while denoising. The denoised data is mapped to the [0,1] interval using the Z-score normalization method to eliminate the dimensional differences between acoustic and vibration signals of different magnitudes (such as sound pressure and vibration acceleration). This reduces the interference of environmental noise (such as grain flow noise) on the effective signal, and the normalization also gives the data a unified analytical scale, providing a consistent input for the stable operation of the subsequent time-domain synchronous averaging model.

[0076] The time-domain synchronous averaging model constructed based on the belt conveyor operating cycle signal includes:

[0077] Obtain the periodic reference value T0 and the periodic fluctuation coefficient δ from the belt conveyor's operating cycle signal;

[0078] Using the periodic baseline value T0 as the time axis unit, the length of the sliding time window is set to k×T0, where k is a positive integer and 1≤k≤5;

[0079] A dynamic weight allocation rule is constructed based on the periodic fluctuation coefficient δ, and differential weights are assigned to acoustic vibration data at different periodic positions within the window.

[0080] The model input is defined as preprocessed acoustic vibration data and time series labels, and the output is the predicted value of the periodic acoustic vibration component. The loss function is constructed by minimizing the deviation between the predicted value and the actual periodic component, and the parameter initialization and structure determination of the time-domain synchronous averaging model are completed.

[0081] Specifically, the model first obtains the periodic reference value and periodic fluctuation coefficient from the conveyor belt's operating cycle signal, serving as the basic parameters for adapting the model to the actual operating state. Using the periodic reference value as the time axis unit, a sliding time window is set to achieve segmented focusing of the continuous signal. Based on the periodic fluctuation coefficient, a dynamic weight allocation rule is constructed, assigning differentiated weights to the acoustic and vibration data at different cycle positions within the window. The weight of cycle segments with larger fluctuations is reduced, while the weight of stable segments is increased, minimizing the interference of fluctuations on the average result. Finally, by defining the input (preprocessed data and time series labels), output (predicted values ​​of periodic components), and loss function (minimizing the deviation between the predicted and actual cycle components), the model parameters are initialized and the structure is determined. This allows the analysis strategy to be adjusted according to the real-time cycle characteristics of the conveyor belt, thereby accurately identifying and amplifying periodic fault characteristics under non-periodic noise interference.

[0082] S3 includes:

[0083] The belt conveyor motor speed signal is acquired by an encoder, and the belt conveyor operating cycle T is calculated.

[0084] Based on the operating period T, the preprocessed acoustic and vibration data are divided into several periodic subsequences;

[0085] The time-domain alignment of each periodic subsequence is performed, and the time-domain synchronous average signal is obtained by periodically superimposing and averaging. The periodic acoustic vibration component is then extracted from it.

[0086] in:

[0087] When the operating cycle of the belt conveyor fluctuates, the length of the periodic subsequence is normalized by interpolation.

[0088] Set an average number threshold. When the signal-to-noise ratio improvement rate after iterative averaging is lower than the threshold, stop the time-domain synchronous averaging operation.

[0089] Specifically, the encoder acquires motor speed signals and calculates the operating cycle T to provide a time reference for data segmentation. The preprocessed data is divided into several periodic subsequences in units of cycle T. Time-domain alignment (correcting the starting phase of each subsequence) and superposition averaging are used to enhance the energy of the periodic signal (aperiodic noise is weakened due to random distribution). When there are fluctuations in the operating cycle, interpolation is used to normalize the length of the periodic subsequences to ensure the consistency of each subsequence in the time dimension. At the same time, an averaging threshold is set. When the signal-to-noise ratio improvement rate after iterative averaging is lower than the threshold, the operation is stopped to improve efficiency while ensuring the effect.

[0090] S4 includes:

[0091] Temporal features of periodic acoustic vibration components are extracted, and peak value, kurtosis, variance, and impulse factor are calculated.

[0092] Frequency domain features of periodic acoustic vibration components are extracted, and power spectral density is obtained through Fourier transform. Feature frequencies and frequency band energy ratios are then extracted.

[0093] By fusing time-domain and frequency-domain features, a feature parameter set is constructed.

[0094] Principal component analysis is used to reduce the dimensionality of the feature parameter set, retaining principal components whose cumulative contribution rate exceeds a preset threshold;

[0095] The principal components after dimensionality reduction are scaled by minimax normalization to obtain a standardized feature matrix.

[0096] Specifically, in the time domain, peak value (reflecting instantaneous impact intensity), kurtosis (reflecting the steepness of the signal distribution), variance (reflecting signal dispersion), and impulse factor (reflecting the relative intensity of the impact) are calculated to capture the temporal dynamic characteristics of the fault. In the frequency domain, the power spectral density is obtained through Fourier transform, and characteristic frequencies (such as the characteristic frequencies of bearing faults) and frequency band energy ratios (reflecting the energy distribution of different frequency bands) are extracted to reflect the frequency characteristics of the fault. After fusing the time-domain and frequency-domain features to construct a parameter set, principal component analysis is used for dimensionality reduction (retaining principal components with cumulative contribution rates exceeding a threshold) to remove feature redundancy. Finally, the scale of the principal components after dimensionality reduction is unified through max-min normalization. The advantage of this design is that multi-domain features comprehensively cover the temporal dynamics and frequency-domain fingerprint of the fault, while dimensionality reduction and standardization optimize the feature quality, providing highly discriminative input parameters for subsequent classification models.

[0097] The extraction of time-domain features from the periodic acoustic vibration components includes:

[0098] The periodic acoustic vibration components are converted into a discrete time-domain signal sequence x(i), where i = 1, 2, ..., n, and n is the number of sampling points;

[0099] Calculate the peak value: Take the maximum value in the discrete time domain signal sequence x(i), that is, peak value = max|x(i)|;

[0100] Calculate kurtosis using K = (1 / n)∑[x(i)-μ] ∧ 4 / σ ∧ 4. Calculate, where μ is the sequence mean and σ is the sequence standard deviation;

[0101] Calculate variance: using the formula σ 2 = (1 / (n-1))Σ[x(i)-μ] ∧ 2. Calculate, where μ is the sequence mean;

[0102] Calculate the impulse factor: take the ratio of the peak value to the average absolute value of the sequence, that is, impulse factor = peak value / [(1 / n)∑|x(i)|].

[0103] Specifically, after converting the periodic acoustic vibration components into a discrete time-domain signal sequence, the peak value is obtained by taking the maximum absolute value of the sequence, directly reflecting the strongest impact in the signal (such as the instantaneous vibration caused by roller eccentricity); the degree of kurtosis reflects the proportion of impact components in the signal (the more severe the fault, the larger the kurtosis value); the variance is obtained by calculating the squared mean of the deviations of the sequence from the mean, reflecting the overall dispersion of the signal (faults increase signal fluctuations); the impulse factor is calculated by the ratio of the peak value to the average absolute value of the sequence, amplifying the proportion of impact features in a stable signal (such as the intermittent impact caused by bearing wear). These quantification methods accurately characterize the time-domain features of the periodic acoustic vibration components from different dimensions, transforming abstract signal features into numerical parameters that can be directly used for classification, thus improving the discriminative power of the features.

[0104] S5 includes:

[0105] The fault classification model is a support vector machine-based classification model, and its training process includes:

[0106] Obtain a historical fault acoustic and vibration feature sample set, and divide it into a training set and a test set;

[0107] Set the kernel function type and penalty parameters, train the model using the training set, and verify the model's classification accuracy using the test set.

[0108] Specifically, the process begins by acquiring a historical set of acoustic and vibration feature samples from faults, containing feature parameters for different fault types. Effective model training is achieved by dividing the dataset into a training set (for model learning) and a test set (for verifying generalization ability). A kernel function type (mapping features to a high-dimensional space to handle nonlinear relationships) and a penalty parameter (balancing the model's fit to training samples and its generalization ability) are then set. The model parameters are iteratively optimized using the training set, and the classification accuracy is verified using the test set. SVM's advantage lies in its strong classification ability with small sample data, making it suitable for scenarios with limited accumulated fault samples in grain depot conveyor belts. A reasonable training process ensures that the model can accurately identify the fault type corresponding to newly extracted feature parameters.

[0109] The historical fault acoustic and vibration feature sample set contains standardized feature matrices corresponding to at least 5 types of typical belt conveyor faults, with no less than 300 samples for each type of fault.

[0110] The training set and the test set were randomly divided in a 7:3 ratio, and stratified sampling was used to maintain the consistency of the distribution of fault types in the two sets.

[0111] The kernel function types include linear kernel functions, polynomial kernel functions, and radial basis function kernel functions, wherein the parameter γ of the radial basis function takes values ​​in the range of [0.01, 10], and the order of the polynomial kernel function takes values ​​in the range of [2, 5].

[0112] The penalty parameter C has a value range of [0.1, 100], and parameter combination training is performed by traversing the value range;

[0113] The maximum number of iterations during training is set to 1000. Training is stopped when the change in training error is less than 1e-5 over 50 consecutive iterations.

[0114] The model's classification accuracy, precision, and recall are calculated using the test set, and the weighted average of the three metrics is used as the basis for evaluating the model's performance.

[0115] Specifically, the historical sample set covers at least five typical fault types (such as roller eccentricity and bearing wear), with no fewer than 300 samples for each type, ensuring sample diversity and representativeness. The training and test sets are randomly divided in a 7:3 ratio, and stratified sampling is used to maintain a consistent distribution of fault types in both sets, avoiding sample bias from affecting model generalization. The kernel functions include linear kernels (handling linearly separable features), multinomial kernels (capturing nonlinear interactions between features), and radial basis kernels (adapting to complex nonlinear relationships), with parameter ranges limited to narrow the optimization space. Penalty parameters are trained by combining values ​​across their ranges to balance the risk of overfitting and underfitting. The training process sets a maximum number of iterations and a convergence condition (error change less than a threshold after 50 consecutive iterations) to ensure stable model convergence. Finally, performance is evaluated using a weighted average of classification accuracy, precision, and recall, comprehensively verifying the model's reliability. These detailed designs enable the SVM model to more accurately learn the mapping relationship between fault features and types, improving classification stability and accuracy.

[0116] The interpolation method includes linear interpolation or cubic spline interpolation:

[0117] When the periodic fluctuation coefficient is ≤5%, linear interpolation is used to supplement the missing data by linear fitting of adjacent sampling points;

[0118] When the periodic fluctuation coefficient is greater than 5%, cubic spline interpolation is used to achieve a smooth transition of the periodic subsequence by constructing a cubic polynomial function.

[0119] Specifically, when the periodic fluctuation coefficient is ≤5% (small fluctuations), linear interpolation is used to supplement missing data through linear fitting of adjacent sampling points. This method is computationally simple and can preserve signal trends, making it suitable for scenarios with small fluctuations. When the fluctuation coefficient is >5% (large fluctuations), cubic spline interpolation is used to achieve a smooth transition between adjacent points by constructing a cubic polynomial function. This avoids signal distortion caused by linear interpolation during large fluctuations and better preserves the continuity of the periodic signal. This design, which dynamically selects the interpolation method based on the fluctuation amplitude, ensures that the normalization processing of the periodic subsequence can adapt to fluctuations while maintaining the physical characteristics of the signal under different periodic stability conditions. This provides high-quality subsequence data for subsequent superposition and averaging to extract periodic components, further improving the stability of feature extraction.

[0120] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for extracting acoustic and vibration features of grain depot belt conveyor faults based on time-domain synchronous averaging, characterized in that, Includes the following steps: S1. Simultaneously collect the acoustic and vibration signals of the grain depot belt conveyor during operation through multiple channels to obtain the original acoustic and vibration dataset; S2. Preprocess the original acoustic and vibration dataset to remove noise interference and standardize the data. S3. Construct a time-domain synchronous averaging model based on the belt conveyor operating cycle signal, perform synchronous averaging on the preprocessed acoustic vibration data, and extract the periodic acoustic vibration components. S4. Extract features from the periodic acoustic vibration components to obtain a set of feature parameters containing fault information; S5. Input the feature parameter set into the preset fault classification model to complete the fault type identification and generate the sound and vibration feature extraction results of the grain depot belt conveyor fault.

2. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 1, characterized in that, S1 includes: A multi-channel acquisition array composed of acoustic and vibration sensors is deployed at the drive end, driven end, and intermediate section of the belt conveyor. The sampling frequency was set to 20-50kHz, and the acquisition time was no less than three operating cycles of the belt conveyor to obtain the original acoustic and vibration dataset composed of acoustic signals and vibration acceleration signals.

3. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 1, characterized in that, S2 includes: Wavelet thresholding was used to denoise the original acoustic and vibration dataset, and a soft thresholding function was set to suppress high-frequency noise components. The denoised acoustic vibration data is mapped to the [0,1] interval using the Z-score normalization method to obtain a normalized acoustic vibration dataset.

4. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 1, characterized in that, The time-domain synchronous averaging model constructed based on the belt conveyor operating cycle signal includes: Obtain the periodic reference value T0 and the periodic fluctuation coefficient δ from the belt conveyor's operating cycle signal; Using the periodic baseline value T0 as the time axis unit, the length of the sliding time window is set to k×T0, where k is a positive integer and 1≤k≤5; A dynamic weight allocation rule is constructed based on the periodic fluctuation coefficient δ, and differential weights are assigned to acoustic vibration data at different periodic positions within the window. The model input is defined as preprocessed acoustic vibration data and time series labels, and the output is the predicted value of the periodic acoustic vibration component. The loss function is constructed by minimizing the deviation between the predicted value and the actual periodic component, and the parameter initialization and structure determination of the time-domain synchronous averaging model are completed.

5. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 1, characterized in that, S3 includes: The belt conveyor motor speed signal is acquired by an encoder, and the belt conveyor operating cycle T is calculated. Based on the operating period T, the preprocessed acoustic and vibration data are divided into several periodic subsequences; The time-domain alignment of each periodic subsequence is performed, and the time-domain synchronous average signal is obtained by periodically superimposing and averaging. The periodic acoustic vibration component is then extracted from it. in: When the operating cycle of the belt conveyor fluctuates, the length of the periodic subsequence is normalized by interpolation. Set an average number threshold. When the signal-to-noise ratio improvement rate after iterative averaging is lower than the threshold, stop the time-domain synchronous averaging operation.

6. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 1, characterized in that, S4 includes: Temporal features of periodic acoustic vibration components are extracted, and peak value, kurtosis, variance, and impulse factor are calculated. Frequency domain features of periodic acoustic vibration components are extracted, and power spectral density is obtained through Fourier transform. Feature frequencies and frequency band energy ratios are then extracted. By fusing time-domain and frequency-domain features, a feature parameter set is constructed. Principal component analysis is used to reduce the dimensionality of the feature parameter set, retaining principal components whose cumulative contribution rate exceeds a preset threshold; The principal components after dimensionality reduction are scaled by minimax normalization to obtain a standardized feature matrix.

7. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 6, characterized in that, The extraction of time-domain features from the periodic acoustic vibration components includes: The periodic acoustic vibration components are converted into a discrete time-domain signal sequence x(i), where i = 1, 2, ..., n, and n is the number of sampling points; Calculate the peak value: Take the maximum value in the discrete time domain signal sequence x(i), that is, peak value = max|x(i)|; Calculate kurtosis: by K = (1 / n)∑[x(i)-μ]^4 / σ^4, where μ is the mean of the sequence and σ is the standard deviation of the sequence; Calculate variance: using the formula σ 2 = (1 / (n-1))∑[x(i)-μ] ∧ 2. Calculate, where μ is the sequence mean; Calculate the impulse factor: take the ratio of the peak value to the average absolute value of the sequence, that is, impulse factor = peak value / [(1 / n)∑|x(i)|].

8. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 1, characterized in that, S5 includes: The fault classification model is a support vector machine-based classification model, and its training process includes: Obtain a historical fault acoustic and vibration feature sample set, and divide it into a training set and a test set; Set the kernel function type and penalty parameters, train the model using the training set, and verify the model's classification accuracy using the test set.

9. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 8, characterized in that, The historical fault acoustic and vibration feature sample set contains standardized feature matrices corresponding to at least 5 types of typical belt conveyor faults, with no less than 300 samples for each type of fault. The training set and the test set were randomly divided in a 7:3 ratio, and stratified sampling was used to maintain the consistency of the distribution of fault types in the two sets. The kernel function types include linear kernel functions, polynomial kernel functions, and radial basis function kernel functions, wherein the parameter γ of the radial basis function takes values ​​in the range of [0.01, 10], and the order of the polynomial kernel function takes values ​​in the range of [2, 5]. The penalty parameter C has a value range of [0.1, 100], and parameter combination training is performed by traversing the value range; The maximum number of iterations during training is set to 1000. Training is stopped when the change in training error is less than 1e-5 over 50 consecutive iterations. The model's classification accuracy, precision, and recall are calculated using the test set, and the weighted average of the three metrics is used as the basis for evaluating the model's performance.

10. The method for extracting acoustic and vibration features of grain depot belt conveyors based on time-domain synchronous averaging according to claim 5, characterized in that, The interpolation method includes linear interpolation or cubic spline interpolation: When the periodic fluctuation coefficient is ≤5%, linear interpolation is used to supplement the missing data by linear fitting of adjacent sampling points; When the periodic fluctuation coefficient is greater than 5%, cubic spline interpolation is used to achieve a smooth transition of the periodic subsequence by constructing a cubic polynomial function.

Images