Vibration signal feature extraction method and system based on signal processing

Abstract

The invention discloses a vibration signal feature extraction method and system based on signal processing, and relates to the technical field of signal processing, and the method comprises the steps: obtaining a vibration signal of target mechanical equipment, and carrying out the preprocessing of the vibration signal; decomposing the preprocessed vibration signal by adopting multi-scale wavelet transform and frequency domain analysis to obtain time domain, frequency domain and time-frequency domain features; carrying out dimension reduction processing on the extracted feature data by using a deep learning model, and carrying out sequence feature processing on the data after dimension reduction to obtain dimension-reduced feature data; and training a deep learning mode, analyzing the dimension reduction feature data by using the trained deep learning model, identifying abnormality in the vibration signal, and outputting an abnormality monitoring result and fault early warning information. The method can provide reliable fault early warning under different working conditions, and is suitable for on-line monitoring and intelligent operation and maintenance of various mechanical devices.

Description

Technical Field

[0001] The present invention relates to the technical field of vibration signal processing, and particularly to a vibration signal feature extraction method and system based on signal processing. Background Art

[0002] The vibration signals of mechanical equipment carry rich operation information. Using this signal for feature extraction, fault diagnosis, and predictive maintenance has become an important means to improve the operation stability of equipment. However, vibration signals often have complex time-varying and non-linear characteristics. Traditional feature extraction methods are difficult to handle the extraction of multi-scale features and time-frequency domain analysis, resulting in the loss of feature information or inaccurate analysis results. In recent years, the application of artificial intelligence technology in vibration signal feature extraction has gradually emerged, and its deep learning model can effectively improve the accuracy of feature extraction and the accuracy of fault detection. Summary of the Invention

[0003] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title. However, such simplifications or omissions shall not be used to limit the scope of the present invention.

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a vibration signal feature extraction method and system based on signal processing, which can solve the problems mentioned in the background art.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a vibration signal feature extraction method based on signal processing, which includes obtaining the vibration signal of a target mechanical equipment and preprocessing the vibration signal;

[0008] Decomposing the preprocessed vibration signal by using multi-scale wavelet transform and frequency domain analysis to obtain time domain, frequency domain, and time-frequency domain features;

[0009] Using a deep learning model to perform dimensionality reduction processing on the extracted feature data, and performing sequence feature processing on the data after dimensionality reduction to obtain dimensionality reduction feature data;

[0010] Training the deep learning model, and using the trained deep learning model to analyze the dimensionality reduction feature data, identify the anomalies in the vibration signal, and output anomaly monitoring results and fault warning information.

[0011] As a preferred embodiment of the vibration signal feature extraction method based on signal processing according to the present invention, wherein: obtaining the vibration signal of the target mechanical equipment includes obtaining the vibration signal of the target mechanical equipment through an acceleration sensor and a velocity sensor;

[0012] Preprocessing the vibration signal includes conditioning the signal output by the sensor using an amplification process and a filtering process, amplifying the signal to a level suitable for analog-to-digital converter conversion, and using a low-pass filter to filter out unwanted high-frequency noise;

[0013] Design an anti-aliasing low-pass filter so that its cut-off frequency f c Satisfy the following conditions, and the calculation formula is as follows:

[0014]

[0015] f s ≥2f max ,

[0016] Wherein, f s Is the sampling frequency, and f max Is the highest frequency of the signal;

[0017] Perform analog-to-digital conversion to quantize the continuous analog signal into a discrete digital signal, and set the quantization level Q. The quantization level Q is set as:

[0018] Q = 2 N

[0019] Wherein, N is the number of bits of the analog-to-digital conversion;

[0020] The existing quantization error e q The calculation formula is as follows:

[0021]

[0022] Wherein, x is the original signal value of the input, which is the amplitude of the analog signal at a specific sampling moment, Is the quantized signal value, that is, the nearest quantization level;

[0023] During the process, it is also necessary to estimate the theoretical signal-to-noise ratio of the analog-to-digital conversion. The calculation formula is as follows:

[0024] SNR(DB) = 6.02N + 1.76

[0025] The calculation formula of the sampling signal during the sampling process is as follows:

[0026] x[n] = x(nT s )

[0027]

[0028] Among them, T s is the sampling frequency, x[n] is the discrete-time signal after sampling, representing the signal amplitude at the nth sampling moment, and n is an integer index representing the sampling point number.

[0029] As a preferred solution of the vibration signal feature extraction method based on signal processing according to the present invention, wherein: the preprocessing of the vibration signal includes detrending, denoising, and normalization;

[0030] The detrending is to eliminate the trend term in the signal and use a polynomial for fitting. The calculation formula is as follows:

[0031] y = a 0 + a 1 x + a 2 x 2 +... + a n x n

[0032] Among them, y represents the signal value after fitting, that is, the output variable, x represents the independent variable, corresponding to the time or space dimension, a 0 , a 1 , a 2 ,..., a n are the coefficients of the polynomial, determined through the fitting process, and n is the order of the polynomial;

[0033] The denoising reduces noise interference by using a filter or wavelet denoising method;

[0034] The normalization normalizes the signal amplitude. The calculation formula is as follows:

[0035]

[0036] Among them, x represents the original signal value, x ′ is the signal value after normalization, x max and x min are respectively the minimum and maximum values in the signal data. After normalization processing, the original signal is mapped into the interval [0, 1].

[0037] As a preferred solution of the vibration signal feature extraction method based on signal processing according to the present invention, wherein: the decomposition of the preprocessed vibration signal by using multi-scale wavelet transform and frequency domain analysis includes,

[0038] Performing multi-scale wavelet decomposition on the vibration signal to decompose the vibration signal into components in different frequency bands;

[0039] Obtaining time-domain features includes calculating the mean value, root mean square value, variance, peak value, kurtosis, skewness, peak factor, margin factor, waveform factor, and impulse factor;

[0040] Obtaining frequency-domain features includes performing a Fourier transform on the signal and calculating the spectrum of the signal;

[0041] Obtaining time-frequency domain features includes using the short-time Fourier transform (STFT) and wavelet transform to analyze the vibration signal in different time windows or by stretching and translating wavelet basis functions, and obtaining the spectrum information at different moments and the characteristics of the signal at different scales and translation positions.

[0042] As a preferred embodiment of the vibration signal feature extraction method based on signal processing according to the present invention, wherein: the dimensionality reduction processing of the extracted feature data using a deep learning model includes,

[0043] Performing a convolution operation on the feature data using a convolutional neural network, through a convolutional layer, an activation function, a pooling layer, and a fully connected layer;

[0044] Applying a rectified linear unit as the activation function;

[0045] Adopting a max pooling or average pooling method to pool the activation values in the pooling layer of the convolutional neural network;

[0046] Applying a long short-term memory network (LSTM) to process the feature data after dimensionality reduction by the convolutional neural network.

[0047] As a preferred embodiment of the vibration signal feature extraction method based on signal processing according to the present invention, wherein: the convolutional neural network consists of a convolutional layer, an activation function, a pooling layer, and a fully connected layer;

[0048] For a vibration signal, the convolution calculation formula is as follows:

[0049]

[0050] where z i is the output vibration signal, representing the response of the system to the input at time t, x(τ) is the input vibration signal, representing the excitation or external force applied to the system at time τ, h(t - τ) is the impulse response function or unit impulse response of the system, representing the response at time t to an instantaneous unit impulse input applied at time τ, and the activation function adopts a rectified linear unit, and the calculation formula is as follows:

[0051] a i = ReLU(z i ) = max(0, z i )

[0052] where the ReLU function outputs 0 when x ≤ 0 and outputs x itself when x > 0;

[0053] The pooling layer performs pooling on the activation values. The calculation formula for max pooling is as follows:

[0054] p i = max a iS+j

[0055] The calculation formula for average pooling is as follows:

[0056]

[0057] Among them, S is the pooling stride, i and j represent the position (i, j) after pooling, and M is the number of pixel points in the pooling window area.

[0058] As a preferred scheme of the vibration signal feature extraction method based on signal processing described in the present invention, wherein: the output of the abnormal monitoring result and the fault warning information includes

[0059] Receiving the dimensionality-reduced feature data of the vibration signal;

[0060] Selecting the corresponding deep learning model according to the data type. For the original vibration signal, a one-dimensional convolutional neural network is used. For the time-frequency image, a two-dimensional convolutional neural network is used. For the time series data, a long short-term memory network is used;

[0061] Designing the structure of the selected deep learning model, which is generally divided into an input layer, a hidden layer, and an output layer. The hidden layer includes a convolutional layer, a recurrent layer, or a fully connected layer. The calculation formula for the fully connected layer is as follows:

[0062] z (L) = W (L) x (L-1) + b (L)

[0063] Among them, W (L) is the weight matrix of the fully connected layer, b (L) is the bias vector of the fully connected layer, and x (L-1) is the output feature of the previous layer;

[0064] The calculation formula for the output layer is as follows:

[0065]

[0066] Among them, y k is the probability of predicting the k-th type of fault, exp() is the exponential function, and z k is the linear combination result of the corresponding category;

[0067] The loss function is calculated and defined in the form of cross-entropy loss. The calculation formula is as follows:

[0068]

[0069] Among them, K is the number of groups of empirical data, and y k is the output of the output layer;

[0070] For backpropagation and parameter update, first calculate the gradient and then update the parameters. The calculation formula is as follows:

[0071]

[0072] Among them, α is the learning rate, and θ t is the parameter at the current non-updated moment, and θ t+1 is the parameter after update at the next moment, and L is the loss function;

[0073] Execute the forward propagation process, calculate the outputs of each layer in the model until the final prediction result is obtained;

[0074] Calculate the cross-entropy loss function to evaluate the difference between the prediction result and the actual label;

[0075] Apply the backpropagation algorithm to train the model, update the model parameters by the gradient descent method, and minimize the loss function;

[0076] Evaluate the performance of the trained deep learning model on the test set;

[0077] Output the anomaly monitoring result and the fault warning information. When the deep learning model detects an abnormal signal, provide a device operation status report.

[0078] In a second aspect, the present invention provides a vibration signal feature extraction system based on signal processing, which includes: a signal acquisition module, a feature extraction module, a feature dimensionality reduction module, and an anomaly output module;

[0079] The signal acquisition module is used to acquire the vibration signal of the target mechanical equipment and preprocess the vibration signal;

[0080] The feature extraction module is used to decompose the preprocessed vibration signal by using multi-scale wavelet transform and frequency domain analysis to obtain time-domain, frequency-domain, and time-frequency domain features;

[0081] The feature dimensionality reduction module is used to perform dimensionality reduction processing on the extracted feature data by using a deep learning model and perform sequence feature processing on the dimensionality-reduced data to obtain dimensionality-reduced feature data;

[0082] The anomaly output module is used to train the deep learning mode and analyze the dimensionality-reduced feature data by using the trained deep learning model to identify the anomalies in the vibration signal and output the anomaly monitoring result and the fault warning information.

[0083] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the processor executes the computer program, the steps of the vibration signal feature extraction method based on signal processing are implemented.

[0084] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by a processor, the steps of the vibration signal feature extraction method based on signal processing are implemented.

[0085] Compared with the prior art, the beneficial effects of the present invention are that the accuracy of vibration signal feature extraction and the real-time performance of fault detection are significantly improved, reliable fault warnings can be provided under different working conditions, and it is applicable to the online monitoring and intelligent operation and maintenance of various mechanical equipment. Description of the Drawings

[0086] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0087] Figure 1 It is a method flowchart of a vibration signal feature extraction method and system based on signal processing provided by an embodiment of the present invention;

[0088] Figure 2 It is an internal structure diagram of a computer device of a vibration signal feature extraction method and system based on signal processing provided by an embodiment of the present invention;

[0089] Figure 3 It is a feature extraction flowchart of a vibration signal feature extraction method and system based on signal processing provided by an embodiment of the present invention. Detailed Embodiments

[0090] To make the above objects, features, and advantages of the present invention more understandable, the following will give a detailed description of the specific embodiments of the present invention in conjunction with the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0091] In the following description, numerous specific details are set forth to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Persons skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0092] Secondly, as used herein, an "embodiment" or "embodiments" refers to specific features, structures, or characteristics that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it an isolated or alternative embodiment mutually exclusive of other embodiments.

[0093] Embodiment 1, referring to Figures 1 - 3 , which is the first embodiment of the present invention. This embodiment provides a method for extracting vibration signal features based on signal processing, including:

[0094] This application can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to elaborate in detail how to implement the method for extracting vibration signal features based on signal processing;

[0095] Figure 1 The flowchart of a method for extracting vibration signal features based on signal processing and a system thereof is shown, including:

[0096] S1: Obtain the vibration signal of the target mechanical equipment and preprocess the vibration signal;

[0097] Furthermore, obtaining the vibration signal of the target mechanical equipment includes,

[0098] Obtain the vibration signal of the target mechanical equipment through devices such as acceleration sensors and velocity sensors;

[0099] During the preprocessing process, use the amplification process and the filtering process to condition the signal output by the sensor, amplify the signal to a level suitable for conversion by an analog-to-digital converter (ADC), and use a low-pass filter to filter out unwanted high-frequency noise;

[0100] Design an anti-aliasing low-pass filter to ensure that its cut-off frequency is lower than half of the sampling frequency. According to the Nyquist sampling theorem, prevent signal components higher than the Nyquist frequency from aliasing into the sampled signal;

[0101] Design a low-pass filter such that its cut-off frequency f c Satisfies the following conditions, that is:

[0102]

[0103] where, f sis the sampling frequency, which needs to satisfy the Nyquist sampling theorem, that is:

[0104] f s ≥2f max

[0105] where f max is the highest frequency of the signal. The sampling frequency must be at least twice the highest frequency of the signal. This filter filters out the signal components higher than the Nyquist frequency before sampling to prevent aliasing.

[0106] Perform analog-to-digital conversion to quantize the continuous analog signal into a discrete digital signal. Set the quantization level and estimate the theoretical signal-to-noise ratio of the analog-to-digital conversion. The quantization level Q is set to:

[0107] Q = 2 N

[0108] where N is the number of bits of the analog-to-digital conversion;

[0109] The existing quantization error e q The calculation formula is as follows:

[0110]

[0111] where x is the original signal value of the input, which can be the amplitude of the analog signal at a specific sampling moment, is the quantized signal value, that is, the nearest quantization level;

[0112] During the process, it is also necessary to estimate the theoretical signal-to-noise ratio of the analog-to-digital conversion. The calculation formula is as follows:

[0113] SNR(DB) = 6.02N + 1.76

[0114] The calculation formula of the sampling signal during the sampling process is as follows:

[0115] x[n] = x(nT s )

[0116]

[0117] where T s is the sampling frequency, x[n] is the discrete-time signal after sampling, representing the signal amplitude at the nth sampling moment, and n is an integer index representing the sampling point number;

[0118] Digital signal sampling represents the signal amplitude of the discrete-time signal after sampling at each sampling moment, where the sampling frequency meets the requirements of the Nyquist sampling theorem.

[0119] Further, the preprocessing of the vibration signal includes three processes: detrending, denoising, and normalization. Preprocessing the sampling signal can effectively improve the accuracy of extraction;

[0120] Detrending is to eliminate the trend term in the signal, usually using polynomial fitting, and the calculation formula is as follows:

[0121] y = a 0 + a 1 x + a 2 x 2 +... + a n x n

[0122] Among them, y represents the signal value after fitting, that is, the output variable, x represents the independent variable, usually corresponding to dimensions such as time or space, a 0 , a 1 , a 2 ,..., a n are the coefficients of the polynomial, which can be determined through the fitting process, and n is the order of the polynomial, which determines the complexity of the polynomial;

[0123] The denoising process reduces noise interference by using filters or wavelet denoising methods;

[0124] The normalization process normalizes the signal amplitude to facilitate the comparison of different signals, and the calculation formula is as follows:

[0125]

[0126] Among them, x represents the original signal value, x ′ is the signal value after normalization, x max and x min are the minimum and maximum values in the signal data respectively. After normalization processing, the original signal is mapped into the interval [0, 1].

[0127] S2: Decompose the preprocessed vibration signal by using multi-scale wavelet transform and frequency domain analysis to obtain time-domain features, frequency-domain features, and time-frequency domain features;

[0128] Further, perform multi-scale wavelet decomposition on the vibration signal to decompose the vibration signal into components of different frequency bands.

[0129] Further, obtaining time-domain features includes,

[0130] Calculating mathematical quantities such as mean, root mean square value, variance, peak value, kurtosis, skewness, peak factor, margin factor, waveform factor, and impulse factor to quantitatively describe the statistical characteristics and dynamic behavior of the vibration signal,

[0131] The mean value μ of the DC component or average value of the reaction signal is calculated as follows:

[0132]

[0133] The root mean square value RMS of the reaction signal energy magnitude and the signal effective value is calculated as follows:

[0134]

[0135] The variance σ that measures the dispersion degree of the signal amplitude 2 , is calculated as follows:

[0136]

[0137] The peak value is the maximum absolute value x of the signal peak , is calculated as follows:

[0138] x peak =max(|x(n)|)

[0139] The kurtosis K used to detect impulsive or shock components and characterize the sharpness of the signal distribution is calculated as follows:

[0140]

[0141] The skewness S that measures the symmetry of the signal amplitude distribution is calculated as follows:

[0142]

[0143] The crest factor CF that reflects the ratio of the peak value to the effective value in the signal is calculated as follows:

[0144]

[0145] The margin factor MF that measures the ratio of the peak value to the average amplitude is calculated as follows:

[0146]

[0147] The form factor FF that measures the ratio of the effective value to the average amplitude is calculated as follows:

[0148]

[0149] The impulse factor IF that reflects the ratio of the peak value to the average absolute value is calculated as follows:

[0150]

[0151] By extracting the above mathematical quantities, it can provide a strong basis for signal classification, pattern recognition, fault detection and diagnosis, etc., and can also improve the accuracy and effectiveness of analysis.

[0152] Furthermore, obtaining time-domain features includes

[0153] By performing spectral analysis on the signal, extracting frequency-domain features, performing Fourier transform on the signal, and calculating the spectrum X of the signal f , the calculation formula is as follows:

[0154]

[0155] For discrete signals, use discrete Fourier transform or fast Fourier transform, and the calculation formula is as follows:

[0156]

[0157] where k is the frequency index, and its value range is k = 0, 1, 2,..., N - 1, corresponding to different frequency components. X[k] is the complex value of the frequency signal at the k-th frequency point, which contains the amplitude and phase information of this frequency component. N is the total number of sampling points of the signal, and x[n] is the value of the time-domain signal at the n-th sampling point;

[0158] Calculate the amplitude of the spectrum, and the calculation formula is as follows:

[0159]

[0160] The spectrum power P[k] is the square of the spectrum amplitude, which reflects the energy of each frequency component, and the calculation formula is as follows:

[0161] P[k] = |X[k]|^2 2

[0162] The spectrum bandwidth BW reflects the width of the spectrum, and the calculation formula is as follows:

[0163]

[0164] The spectrum kurtosis K s (f) is used to detect non-Gaussian components or transient features, and the calculation formula is as follows:

[0165]

[0166] where E{·} represents the expected value.

[0167] Furthermore, obtaining time-frequency domain features includes

[0168] Using the Short-Time Fourier Transform (STFT) and wavelet transform, analyze the vibration signal in different time windows or by stretching and translating the wavelet basis function to obtain the spectral information at different times and the characteristics of the signal at different scales and translation positions, thereby extracting time-frequency domain features;

[0169] Apply the Short-Time Fourier Transform to calculate the spectrum X(t,f) of the signal in different time windows. The calculation formula is as follows:

[0170]

[0171] where ω(τ) is the window function;

[0172] It should be noted that the window function is a method for localizing signals in both the time and frequency domains. By multiplying the window function with the original signal, a segment of the signal at a certain moment or time period can be selected for spectral analysis; the window function usually has a finite length and a specific shape, and its selection directly affects the time resolution and frequency resolution of the analysis. Commonly used window functions include rectangular windows, Hanning windows, Hamming windows, Gaussian windows, etc.; they limit the range of the signal in the time domain and affect the main lobe width and side lobe attenuation in the frequency domain; the application of the window function can effectively analyze and extract the instantaneous frequency characteristics of non-stationary signals, thereby obtaining the spectral information of the signal at different times;

[0173] Then, analyze the time-frequency characteristics of the signal by stretching and translating the wavelet basis function. The calculation formula is as follows:

[0174]

[0175] where W(a,b) is the wavelet coefficient, representing the video characteristics of the signal at scale a and time position b, x(t) is the original time-domain signal, ψ(t) is the mother wavelet, a is the scale parameter, and b is the translation parameter.

[0176] S3: Use a deep learning model to perform dimensionality reduction on the extracted feature data and perform sequence feature processing on the data after dimensionality reduction to obtain dimensionality-reduced feature data;

[0177] Furthermore, the dimensionality reduction process includes,

[0178] Use a Convolutional Neural Network (CNN) to perform convolution operations on the feature data, reducing data redundancy and retaining key information through convolutional layers, activation functions, pooling layers, and fully connected layers;

[0179] In the convolutional neural network, apply the Rectified Linear Unit (ReLU) as the activation function to truncate negative values to zero and keep positive values linearly passing through to simplify the calculation;

[0180] The maximum pooling or average pooling method is used to pool the activation values in the pooling layer of the convolutional neural network, further reducing the feature dimension;

[0181] The long short-term memory network (LSTM) is applied to capture the long-term and short-term dependencies in the sequence data and process the feature data after dimensionality reduction by the convolutional neural network.

[0182] Specifically, the convolutional neural network mainly consists of a convolutional layer, an activation function, a pooling layer, and a fully connected layer. The convolutional neural network mainly consists of a convolutional layer, an activation function, a pooling layer, and a fully connected layer. For vibration signals, the convolution calculation formula is as follows:

[0183]

[0184] where z i is the output vibration signal, representing the response of the system to the input at time t, x(τ) is the input vibration signal, representing the excitation or external force applied to the system at time τ, h(t - τ) is the impulse response function or unit impulse response of the system, representing the response at time t to the instantaneous unit impulse input applied at τ. The activation function uses the rectified linear unit, and the calculation formula is as follows:

[0185] a i = ReLU(z i ) = max(0, z i )

[0186] where the ReLU function outputs 0 when x ≤ 0 and outputs x itself when x > 0. The ReLU has the characteristic of truncating negative values to zero and keeping positive values linearly passing through, with simple calculation, avoiding complex function evaluation and reducing the calculation cost;

[0187] The pooling layer pools the activation values, and the calculation formula for maximum pooling is as follows:

[0188] p i = max a iS+j

[0189] The calculation formula for average pooling is as follows:

[0190]

[0191] where S is the pooling stride, i and j represent the position (i, j) after pooling, and M is the number of pixel points in the pooling window area.

[0192] Furthermore, the sequence feature processing includes,

[0193] In the long short-term memory network, the information flow is controlled by an input gate, a forget gate, and an output gate to update the state of the memory cell to adapt to the time series characteristics; the long short-term memory network is a special recurrent neural network that can effectively capture the long-term and short-term dependencies in the sequence data;

[0194] Calculate the candidate memory cell state, use the hyperbolic tangent (tanh) activation function to limit its value range to [-1, 1], and provide new alternative memory content;

[0195] Update the state of the memory cell of the long short-term memory network, combine the memory state at the previous moment and the input at the current moment, and integrate the information through element-wise multiplication operation;

[0196] Generate the hidden state vector of the long short-term memory network as the output at the current moment for subsequent processing or prediction.

[0197] Specifically, the long short-term memory network unit introduces a memory cell and a gating mechanism, including an input gate, a forget gate, and an output gate. The calculation formula of the input gate is as follows:

[0198] i t =σ(W i x t +U i h t-1 +b i )

[0199] Where b i is the bias vector of the input gate, h t-1 is the hidden state vector at time step t-1, σ is the Sigmoid activation function that can map the output to the interval [0, 1], x t is the input feature vector at time step t, U i is the weight matrix from the previous hidden state h t-1 to the input gate, and W i is the weight matrix from the input to the input gate;

[0200] The calculation formula of the forget gate is as follows:

[0201] f t =σ(W f x t +U f h t-1 +b f )

[0202] Where W f is the weight matrix from the input x t to the forget gate, U f is the weight matrix from the previous hidden moment h t-1 to the forget gate, bf is the bias vector of the forget gate;

[0203] When updating the memory cell, first calculate the candidate memory cell state, and the calculation formula is as follows:

[0204]

[0205] Where, is the candidate memory cell state vector at time step t, and its value range is [-1, 1], providing new alternative memory contents; tanh is the hyperbolic tangent activation function, W c is the weight matrix input to the candidate memory cell; U c is the weight matrix from the previous hidden state h t-1 to the memory candidate cell; b c is the bias vector of the candidate memory cell;

[0206] Subsequently, update the memory cell state, and the calculation formula is as follows:

[0207]

[0208] Where, C t is the memory cell state vector at time step t, which can store the long-term memory information of the network, C t-1 is the memory cell state vector at time step t-1, and ⊙ is element-wise multiplication, that is, the elements at the corresponding positions of the vectors are multiplied;

[0209] The calculation formula of the output gate is as follows:

[0210] o t = σ(W o x t + U o h t-1 + b o )

[0211] h t = o t ⊙ tanh(C t )

[0212] Where, o t is the output gate vector at time step t, and its value range is [0, 1], which determines the influence degree of the memory cell state on the hidden state, W o is the weight matrix input to the output gate, U o is the weight matrix from the previous h t-1 hidden time to the output gate, b o is the bias vector of the output gate, h t is the hidden state vector at time step t, as the output at the current moment, which can be used for subsequent processing or prediction.

[0213] S4: Train a deep learning model, and analyze the dimensionality-reduced feature data using the trained deep learning model to identify anomalies in the vibration signal, and output an anomaly monitoring result and a fault warning message.

[0214] Further, outputting an anomaly monitoring result and a fault warning message includes

[0215] Receiving the dimensionality-reduced feature data of the vibration signal;

[0216] Selecting a corresponding deep learning model according to the data type. For the original vibration signal, a one-dimensional convolutional neural network (1D CNN) is used. For the time-frequency image, a two-dimensional convolutional neural network (2D CNN) is used. For the time series data, a long short-term memory network (LSTM) is used;

[0217] Design the structure of the selected deep learning model, which is generally divided into an input layer, a hidden layer, and an output layer. The hidden layer includes a convolutional layer, a recurrent layer, or a fully connected layer; The calculation formula of the fully connected layer is as follows:

[0218] z (L) =W (L) x (L-1) +b (L)

[0219] where W (L) is the weight matrix of the fully connected layer, b (L) is the bias vector of the fully connected layer, and x (L-1) is the output feature of the previous layer;

[0220] The calculation formula of the output layer is as follows:

[0221]

[0222] where y k is the probability of predicting the kth type of fault, exp() is the exponential function, and z k is the linear combination result of the corresponding category;

[0223] Loss function calculation, defined in the way of cross-entropy loss, and the calculation formula is as follows:

[0224]

[0225] where K is the number of groups of empirical data, and y k is the output of the output layer;

[0226] Backpropagation and parameter update, first calculate the gradient, and then update the parameters. The calculation formula is as follows:

[0227]

[0228] where α is the learning rate, and θ t is the parameter at the current unupdated moment, and θ t+1 is the parameter updated at the next moment, and L is the loss function;

[0229] Execute the forward propagation process, and calculate the outputs of each layer in the model until the final prediction result is obtained;

[0230] Calculate the cross-entropy loss function to evaluate the difference between the prediction result and the actual label;

[0231] Apply the backpropagation algorithm to train the model, and update the model parameters by the gradient descent method to minimize the loss function;

[0232] Evaluate the performance of the trained deep learning model on the test set to verify its accuracy and generalization ability;

[0233] Output the anomaly monitoring result and the fault warning information, and provide the device operation status report when the deep learning model detects an abnormal signal.

[0234] Furthermore, this embodiment also provides a vibration signal feature extraction system based on signal processing, including: a signal acquisition module, a feature extraction module, a feature dimensionality reduction module, and an anomaly output module;

[0235] The signal acquisition module is used to acquire the vibration signal of the target mechanical equipment and preprocess the vibration signal;

[0236] The feature extraction module is used to decompose the preprocessed vibration signal by using multi-scale wavelet transform and frequency domain analysis to obtain time domain, frequency domain, and time-frequency domain features;

[0237] The feature dimensionality reduction module is used to perform dimensionality reduction processing on the extracted feature data by using a deep learning model, and perform sequence feature processing on the data after dimensionality reduction to obtain dimensionality reduction feature data;

[0238] The anomaly output module is used to train the deep learning model, and analyze the dimensionality reduction feature data by using the trained deep learning model to identify the anomalies in the vibration signal and output the anomaly monitoring result and the fault warning information.

[0239] This embodiment also provides a computer device, which can be a terminal, and its internal structure diagram can be as Figure 2As shown in the figure. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, carrier network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it realizes a vibration signal feature extraction method based on signal processing. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0240] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by the processor, the following steps are implemented: obtaining the vibration signal of the target mechanical equipment and preprocessing the vibration signal;

[0241] Decomposing the preprocessed vibration signal by using multi-scale wavelet transform and frequency domain analysis to obtain time domain, frequency domain, and time-frequency domain features;

[0242] Using a deep learning model to perform dimensionality reduction processing on the extracted feature data, and performing sequence feature processing on the data after dimensionality reduction to obtain dimensionality reduction feature data;

[0243] Training the deep learning model, and using the trained deep learning model to analyze the dimensionality reduction feature data, identify the anomalies in the vibration signal, and output the anomaly monitoring result and the fault warning information.

[0244] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not restrictive. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

[0245] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript, etc.

[0246] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0247] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0248] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0249] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0250] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to cover these changes and modifications.

Claims

1. A vibration signal feature extraction method based on signal processing, characterized in that: The method comprises obtaining a vibration signal of a target mechanical device and preprocessing the vibration signal; Multi-scale wavelet transform and frequency domain analysis are used to decompose the preprocessed vibration signal to obtain the time domain, frequency domain and time-frequency domain features; The extracted feature data is subjected to dimensionality reduction processing using a deep learning model, and sequence feature processing is performed on the reduced-dimensional data to obtain reduced-dimensional feature data; A deep learning model is trained, and the reduced-dimensional feature data is analyzed using the trained deep learning model to identify anomalies in the vibration signal, and output anomaly monitoring results and fault warning information.

2. The vibration signal feature extraction method based on signal processing as claimed in claim 1, characterized in that: The obtaining of the vibration signal of the target mechanical equipment includes obtaining the vibration signal of the target mechanical equipment through an acceleration sensor and a velocity sensor; Preprocessing the vibration signal includes conditioning the signal output by the sensor using an amplification process and a filtering process, amplifying the signal to a level suitable for conversion by an analog-to-digital converter, and filtering out unwanted high-frequency noise using a low-pass filter; Design an anti-aliasing low-pass filter with a cutoff frequency f c If the following conditions are met, the calculation formula is as follows: f s ≥2f max , Among them, f s is the sampling frequency, f max is the highest frequency of the signal; Perform analog-to-digital conversion, quantize the continuous analog signal into a discrete digital signal, set the quantization level Q, and the quantization level Q is set to: Q＝2 N Where N is the number of bits for analog-to-digital conversion; The quantization error e q The calculation formula is as follows: Among them, x is the original input signal value, which is the amplitude of the analog signal at a specific sampling time. is the quantized signal value, i.e., the most recent quantization level; The theoretical signal-to-noise ratio of analog-to-digital conversion also needs to be estimated in the process. The calculation formula is as follows: SNR(DB)=6.02N+1.76 The calculation formula of the sampling signal during the sampling process is as follows: x[n]=x(nT s ) Among them, T s is the sampling frequency, x[n] is the discrete time signal after sampling, which represents the signal amplitude at the nth sampling moment, and n is an integer index representing the sampling point number.

3. The vibration signal feature extraction method based on signal processing as claimed in claim 2, characterized in that: The preprocessing of the vibration signal includes detrending, denoising and normalizing; The detrending is to eliminate the trend term in the signal, and a polynomial is used for fitting. The calculation formula is as follows: y=a0+a1x+a2x 2 +...+a n x n Among them, y represents the signal value after fitting, that is, the output variable, and x represents the independent variable, corresponding to the time or space dimension, a0, a1, a2, ..., a n are the coefficients of the polynomial, determined by the fitting process, and n is the order of the polynomial; The denoising can reduce noise interference by using a filter or a wavelet denoising method; The normalization normalizes the signal amplitude, and the calculation formula is as follows: Among them, x represents the original signal value, x ′ is the normalized signal value, x max and x min are the minimum and maximum values ​​in the signal data respectively. After normalization, the original signal is mapped to the [0,1] interval.

4. The vibration signal feature extraction method based on signal processing as claimed in claim 3, characterized in that: Decomposing the pre-processed vibration signal by using multi-scale wavelet transform and frequency domain analysis includes performing multi-scale wavelet decomposition on the vibration signal to decompose the vibration signal into components of different frequency bands; Obtaining time domain features includes calculating mean, RMS value, variance, peak, kurtosis, skewness, crest factor, margin factor, shape factor and pulse factor; Obtaining frequency domain features includes performing Fourier transform on the signal and calculating the spectrum of the signal; Obtaining time-frequency domain features includes using short-time Fourier transform and wavelet transform to analyze the vibration signal in different time windows or by scaling and translating wavelet basis functions to obtain spectrum information at different times and characteristics of the signal at different scales and translation positions.

5. The vibration signal feature extraction method based on signal processing as claimed in claim 4, characterized in that: The dimensionality reduction processing of the extracted feature data using the deep learning model includes: Using a convolutional neural network to perform a convolution operation on the feature data, through a convolutional layer, an activation function, a pooling layer, and a fully connected layer; Apply rectified linear unit as activation function; Pooling the activation values ​​in the pooling layer of the convolutional neural network using a maximum pooling or average pooling method; The long short-term memory network is used to process the feature data after dimensionality reduction by the convolutional neural network.

6. The vibration signal feature extraction method based on signal processing as claimed in claim 5, characterized in that: The convolutional neural network is composed of a convolutional layer, an activation function, a pooling layer and a fully connected layer; For vibration signals, the convolution calculation formula is as follows: Among them, z i is the output vibration signal, which indicates the response of the system to the input at time t, x(τ) is the input vibration signal, which indicates the excitation or external force applied to the system at time τ, h(t-τ) is the impulse response function or unit impulse response of the system, which indicates the response to the instantaneous unit pulse input applied at time τ at time t, and the activation function adopts a linear rectifier unit, which is calculated as follows: and i =ReLU(from i )=max(0,z i ) Among them, the ReLU function outputs 0 when x≤0, and outputs x itself when x>0; The pooling layer pools the activation values, and the calculation formula for the maximum pooling is as follows: p i =maxa iS+j The calculation formula for average pooling is as follows: Among them, S is the pooling step size, i and j represent the position (i, j) after pooling, and M is the number of pixels in the pooling window area.

7. The vibration signal feature extraction method based on signal processing according to claim 6, characterized in that: The output abnormality monitoring results and fault warning information include: receiving dimension-reduced feature data of a vibration signal; Select the corresponding deep learning model according to the data type. For the original vibration signal, a one-dimensional convolutional neural network is used, for the time-frequency image, a two-dimensional convolutional neural network is used, and for the time series data, a long short-term memory network is used. Design the structure of the selected deep learning model, which is generally divided into input layer, hidden layer and output layer. The hidden layer contains convolutional layer, recurrent layer or fully connected layer. The calculation formula of the fully connected layer is as follows: z (L) =W (L) x (L-1) +b (L) Among them, W (L) is the weight matrix of the fully connected layer, b (L) is the bias vector of the fully connected layer, x (L-1) is the output feature of the previous layer; The calculation formula of the output layer is as follows: Among them, y k is the probability of predicting the kth type of fault, exp() is the exponential function, z k is the linear combination result of the corresponding category; The loss function is calculated in the form of cross entropy loss, and the calculation formula is as follows: Among them, K is the number of groups of empirical data, y k is the output of the output layer; Back propagation and parameter update, first calculate the gradient, then update the parameters, the calculation formula is as follows: Among them, α is the learning rate, θ t is the parameter at the current time when it is not updated, θ t+1 is the updated parameter at the next moment, and L is the loss function; Perform a forward propagation process to calculate the output of each layer in the model until the final prediction result is obtained; Calculate the cross entropy loss function to evaluate the difference between the predicted result and the actual label; Applying a back-propagation algorithm to train the model, updating the model parameters by a gradient descent method, and minimizing the loss function; Evaluate the performance of the trained deep learning model on the test set; Output abnormal monitoring results and fault warning information, and provide equipment operation status report when the deep learning model detects abnormal signals.

8. A vibration signal feature extraction system based on signal processing, based on the vibration signal feature extraction method based on signal processing according to any one of claims 1 to 7, characterized in that: Including signal acquisition module, feature extraction module, feature dimension reduction module and abnormal output module; The signal acquisition module is used to obtain the vibration signal of the target mechanical equipment and pre-process the vibration signal; The feature extraction module is used to decompose the pre-processed vibration signal by using multi-scale wavelet transform and frequency domain analysis to obtain time domain, frequency domain and time-frequency domain features; The feature dimension reduction module is used to perform dimension reduction processing on the extracted feature data using a deep learning model, and perform sequence feature processing on the dimension-reduced data to obtain dimension-reduced feature data; The abnormality output module is used to train a deep learning model, and uses the trained deep learning model to analyze the reduced-dimensionality feature data, identify abnormalities in the vibration signal, and output abnormality monitoring results and fault warning information.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the vibration signal feature extraction method based on signal processing according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the vibration signal feature extraction method based on signal processing according to any one of claims 1 to 7 are implemented.

Images