Powder particle size measurement method, device, equipment and storage medium

Abstract

This invention provides a method, apparatus, device, and storage medium for measuring powder particle size, belonging to the field of acoustic emission technology. The method for measuring powder particle size includes: acquiring an acoustic emission signal; performing wavelet packet transform on the acoustic emission signal to obtain multiple sub-component signals; reconstructing the phase space of the sub-component signals to obtain the phase point matrix of the acoustic emission signal; inputting the phase point matrix into a convolutional neural network, causing the convolutional neural network to output target feature data; inputting the target feature data into a long short-term memory network to obtain a particle size distribution prediction result corresponding to the acoustic emission signal; and obtaining powder particle size data based on the particle size distribution prediction result. This method meets the need for real-time online measurement of powder particle size and improves the accuracy and precision of particle size measurement.

Description

Technical Field

[0001] This invention relates to the field of acoustic emission technology, and more specifically to a method for measuring powder particle size, a device for measuring powder particle size, an electronic device, and a readable storage medium. Background Technology

[0002] In thermal power plants, pulverized coal particle size is an important factor affecting the economic and safe operation of coal-fired boilers, especially boilers that use lean coal or anthracite as their main fuel, which are more susceptible to the influence of pulverized coal particle size.

[0003] Existing technology uses lasers to irradiate a dispersed group of particles. The laser irradiation produces reflected / scattered light, which is then detected. The particle size distribution in the group is measured based on the intensity signal of the detected reflected / scattered light. However, when the light intensity detector is placed on the optical axis of the light source, the intensity of the light directly incident on the light intensity detector is very high because no particles are dispersed, thus the particle size cannot be measured accurately. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, device, and storage medium for measuring powder particle size, in order to solve the problem that powder particle size cannot be measured in real time and accurately.

[0005] To achieve the above objectives, embodiments of the present invention provide a method for measuring powder particle size, comprising:

[0006] Acquire acoustic emission signals;

[0007] The acoustic emission signal is subjected to wavelet packet transform to obtain multiple sub-component signals;

[0008] The phase space of the sub-component signals is reconstructed to obtain the phase point matrix of the acoustic emission signal;

[0009] The phase point matrix is ​​input into a convolutional neural network, so that the convolutional neural network outputs target feature data;

[0010] The target feature data is input into a long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal;

[0011] Based on the particle size distribution prediction results, the powder particle size data is obtained.

[0012] Optionally, performing wavelet packet transform on the acoustic emission signal to obtain multiple sub-component signals includes:

[0013] According to Formula 1, wavelet packet decomposition is performed on the acoustic emission signal to obtain wavelet packet coefficients of multiple components; wherein, Formula 1 includes:

[0014]

[0015] in, and Here are the wavelet packet coefficients; j is the decomposition level, l and k are the translation parameters, n is the frequency parameter, and h is the frequency parameter. k-2l As a low-pass filter component, g k-2l For high-pass filtering components, j∈{i,i-1,...1};

[0016] Based on the wavelet packet coefficients, the sub-component signal corresponding to the wavelet packet coefficients is obtained; wherein, the sub-component signal includes the time-domain information and frequency-domain information of the sub-component.

[0017] Optionally, before performing wavelet packet decomposition on the acoustic emission signal according to formula one to obtain the wavelet packet coefficients of multiple components, the method further includes:

[0018] Obtain the wavelet center frequency and wavelet sampling frequency;

[0019] Based on the wavelet center frequency, the wavelet sampling frequency, the preset decomposition level, and Formula 2, the pseudo-frequency corresponding to the preset decomposition level is calculated; wherein, Formula 2 includes:

[0020]

[0021] Where k is the preset decomposition level, f k f is the pseudo-frequency corresponding to the preset number of decomposition layers k. c f is the center frequency of the wavelet. s The wavelet sampling frequency;

[0022] Obtain the frequency of the useful signal in the acoustic emission signal;

[0023] Based on the frequency of the useful signal, the pseudo-frequency, and Formula 3, the optimal number of wavelet packet decomposition layers is calculated; wherein, Formula 3 includes:

[0024]

[0025] Where L is the optimal wavelet packet decomposition layer, f k＝L f is the frequency corresponding to the optimal wavelet packet decomposition level when the decomposition level is equal to the optimal decomposition level. sig The frequency of the useful signal is defined as min, where min represents the minimum value and max represents the maximum value.

[0026] The wavelet packet decomposition of the acoustic emission signal according to formula A yields wavelet packet coefficients for multiple components, including:

[0027] The acoustic emission signal is decomposed into wavelet packets according to Formula 1 and the optimal number of wavelet packet decomposition layers to obtain the wavelet packet coefficients of the multiple components.

[0028] Optionally, the step of reconstructing the phase space of the sub-component signal to obtain the phase point matrix of the acoustic emission signal includes: obtaining the time series of the sub-component signal based on the time domain information of the sub-component signal;

[0029] The phase space of the time series of the sub-component signals is reconstructed according to Formula 4 to obtain the phase point matrix of the acoustic emission signal; wherein, Formula 4 includes:

[0030] X={x i ,x i+τ ,x i+2τ ,...,x i+(m-1)τ},

[0031] Where M is the time series length of the reconstructed acoustic emission signal, X is the M×m matrix of the phase point, x is the time series of the acoustic emission signal, m is the embedding dimension, τ is the delay time, and i∈[1,M].

[0032] Optionally, the step of performing phase space reconstruction on the time series of the sub-component signals according to Formula 4 to obtain the phase point matrix of the acoustic emission signal includes:

[0033] Based on the time sequence of the acoustic emission signal and Formula 5, the optimal delay time is calculated; wherein, Formula 5 includes:

[0034]

[0035] Where I(τ) is the mutual information function, P(x(t)) is the probability of x(t) appearing in the time series of the acoustic emission signal, P(x(t+τ)) is the probability of x(t+τ) appearing in the time series of the acoustic emission signal, P(x(t),x(t+τ)) is the probability of x(t) and x(t+τ) appearing together in the time series of the acoustic emission signal, and P(x(1)) is the probability of the first signal appearing in the time series of the acoustic emission signal.

[0036] Obtain the vector space of the acoustic emission signal after spatial reconstruction based on the optimal delay time;

[0037] Based on the vector space and Formula Six, the distances between nearest neighbor vectors of different dimensions are calculated; wherein, Formula Six includes:

[0038]

[0039] in, The maximum norm number in the (m+1)-dimensional space. Y is the maximum norm number in m-dimensional space. n For the vector space with Y μ(n) The nearest neighbor vector, a(i,m) is the distance between the nearest neighbor vectors under different dimensions, i∈[1,M];

[0040] The optimal embedding dimension is calculated based on the distance between the nearest neighbor vectors at different dimensions, the optimal delay time, and Formula 7; wherein Formula 7 includes:

[0041]

[0042] Where E(m) is the mean of a(i,m), E(m+1) is the mean of a(i,m+1), E1(m) is the transformation rate between E(m) and E(m+1), and m is the optimal embedding dimension.

[0043] Substituting the optimal embedding dimension and the optimal delay time into Equation 4 yields the phase point matrix of the acoustic emission signal. Optionally, inputting the target feature data into a Long Short-Term Memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal includes:

[0044] The reconstructed phase point matrix of the acoustic emission signal is input into the convolutional neural network, which performs convolution calculation on the phase point matrix according to Formula 8 to obtain convolutional layer data. The convolutional layer data is then downsampled to obtain pooling layer data, and the pooling layer data is flattened to obtain the target feature data. Formula 8 includes:

[0045]

[0046] in, For the convolutional layer data, X i+m,j+m f is the value in the M-th row and m-th column of the phase point matrix. cov () represents the activation function to be selected, ω M,m Let b be the weight of the convolution kernel in rows M and columns m. M,m denoted as kernel bias, and k is the sliding window size.

[0047] Optionally, the step of inputting the target feature data into a long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal includes:

[0048] The target feature data is input into the long short-term memory network, which then predicts the target feature data based on the target prediction model to obtain predicted feature data. The predicted feature data is weighted using an attention mechanism to obtain a particle size distribution prediction result corresponding to the emitted signal. The target prediction model is obtained by training the long short-term memory network based on historical target feature data.

[0049] In a second aspect of the present invention, a powder particle size measuring device is provided, comprising:

[0050] The signal acquisition module is used to acquire acoustic emission signals;

[0051] The signal transformation module is used to perform wavelet packet transformation on the acoustic emission signal to obtain multiple sub-component signals;

[0052] The signal reconstruction module is used to reconstruct the phase space of the sub-component signals to obtain the phase point matrix of the acoustic emission signal;

[0053] The convolution module is used to input the phase point matrix into the convolutional neural network, so that the convolutional neural network outputs target feature data;

[0054] The prediction module is used to input the target feature data into a long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal;

[0055] The data acquisition module is used to obtain powder particle size data based on the particle size distribution prediction results.

[0056] In a third aspect of the present invention, an electronic device is provided, comprising: a processor and a memory, the memory storing machine-readable instructions executable by the processor, wherein the machine-readable instructions, when executed by the processor, perform the steps described in any possible implementation of the first aspect.

[0057] In a fourth aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing instructions that cause a machine to perform the steps described in any possible implementation of the first aspect.

[0058] In this embodiment of the invention, an acoustic emission signal is acquired, and wavelet packet transform is performed on the acoustic emission signal to obtain sub-component signals at multiple levels. The phase space of the sub-component signals is reconstructed to obtain energy feature data of the sub-component signals. The energy feature data is then sequentially input into a convolutional neural network and a long short-term memory network, so that the long short-term memory network outputs a particle size distribution prediction result corresponding to the emission signal. Based on the prediction result, the powder particle size data is further obtained.

[0059] This invention utilizes wavelet packet transform to perform multi-level segmentation of the acoustic emission signal, dividing it into sub-component signals of multiple frequency bands. This improves the localization function of the signal spectrum and enhances the time-frequency resolution of the acoustic emission signal. Furthermore, this invention employs phase space reconstruction technology to reconstruct the sub-component signals into a dynamic system with the same topological properties as the original acoustic emission signal system. This yields multi-dimensional features of the acoustic emission signal, which are then input into a CNN (Convolutional Neural Network)-ALSTM (Attention Long Short Term Memory) network. The CNN performs dimensionality reduction and noise reduction on the multi-dimensional feature data, sending the feature data with higher weights to the ALSTM. The ALSTM combines the prediction results from the prediction model with the attention mechanism to extract the prediction results with higher weights as the particle size prediction results, improving the accuracy of particle size prediction. This solves the problem of existing technologies being unable to measure powder particle size accurately and in real time, meeting the need for real-time online measurement of powder particle size and improving the precision and accuracy of particle size measurement.

[0060] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0061] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0062] Figure 1 This is a schematic flowchart of an embodiment of the powder particle size measurement method of the present invention;

[0063] Figure 2 This is a schematic diagram of the acoustic emission signal formation process;

[0064] Figure 3 This is a schematic diagram of three-layer wavelet packet decomposition;

[0065] Figure 4 It is a CNN-ALSTM combined model structure;

[0066] Figure 5 This is a schematic diagram of the functional modules of the powder particle size measuring device involved in this application. Detailed Implementation

[0067] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this application.

[0069] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0070] Example 1

[0071] Please refer to Figure 1 , Figure 1 This is a schematic diagram of a powder particle size measurement method provided in this embodiment.

[0072] Step S100: Acquire acoustic emission signal.

[0073] The acoustic emission signal is an elastic wave generated by the collision of powder particles with the waveguide rod. It should be understood that this embodiment uses acoustic emission technology, such as... Figure 2 As shown, the collision between powder particles and the waveguide rod causes vibration on the surface of the waveguide rod. The vibration displacement generated at the collision point propagates along the waveguide rod in the form of elastic waves. The acoustic emission sensor detects the superposition of elastic waves in various directions. The elastic waves are converted into electrical signals by the acoustic emission sensor, amplified, filtered, and then collected. Under ideal conditions, the powder particle flow impacts the waveguide rod in the form of dispersed single particles, generating separable acoustic emission signals. The obtained acoustic emission signals can be expressed by Formula 9, which is:

[0074] V(t) = s(t) * g(t) * i(t),

[0075] Where s(t) is the acoustic emission source function, g(t) is the elastic wave transfer function, i(t) is the acoustic emission instrument response function, * denotes convolution, and V(t) is the acoustic emission signal.

[0076] Step S200: Perform wavelet packet transform on the acoustic emission signal to obtain multiple sub-component signals.

[0077] like Figure 3As shown, wavelet packet transform, based on wavelet transform, further decomposes both low-frequency and high-frequency subbands at each level of signal decomposition. Finally, by minimizing a cost function, the optimal signal decomposition path is calculated, and this path is used to decompose the original signal. By selecting different resolutions for various frequencies at different levels, a series of time-domain waveforms across the entire frequency band are provided at various scales, thus improving the signal frequency resolution.

[0078] Sub-component signals are sub-signals of the acoustic emission signal in different frequency bands.

[0079] In one specific embodiment, step S200 includes:

[0080] Step S210: Perform wavelet packet decomposition on the acoustic emission signal according to Formula 1 to obtain wavelet packet coefficients of multiple components; wherein, Formula 1 includes:

[0081]

[0082] in, and Here are the wavelet packet coefficients; j is the decomposition level, l and k are the translation parameters, n is the frequency parameter, and h is the frequency parameter. k-2l As a low-pass filter component, g k-2l For high-pass filtering components, j∈{i,i-1,...1}.

[0083] The number of decomposition levels refers to the number of different frequency band sub-components that are divided when decomposing an acoustic emission signal. It can be understood that before performing wavelet packet decomposition, the number of decomposition levels needs to be determined. Specifically, determining the number of decomposition levels involves: calculating the pseudo-frequency corresponding to the assumed decomposition level based on the wavelet center frequency, the wavelet usage frequency, the assumed number of decomposition levels, and Formula 2. Formula 2 includes:

[0084]

[0085] Where k is the assumed number of decomposition levels, f k f is the pseudo-frequency corresponding to the assumed number of decomposition levels k. c f is the wavelet center frequency. s The wavelet sampling frequency;

[0086] Assuming the optimal decomposition level is L, to ensure denoising accuracy, the useful signal frequencies of the acoustic emission signal should all be included within the entire range of the aforementioned pseudo-frequency range. Therefore, as shown in Formula 3, the maximum decomposition level of the wavelet packet can be determined based on the minimum frequency of the useful signal in the acoustic emission signal. This maximum decomposition level is the optimal wavelet packet decomposition level. Formula 3 specifically includes:

[0087]

[0088] Where L is the optimal wavelet packet decomposition layer, f k＝L f is the frequency corresponding to the optimal number of decomposition layers. sig The frequency of the useful signal is denoted by min, which represents the minimum value, and max represents the maximum value.

[0089] It is important to understand that before performing wavelet transform, appropriate wavelet basis functions need to be selected. The selection criteria for wavelet basis functions are quantitatively characterized by analyzing the similarity between the reconstructed acoustic emission signal and the original acoustic emission signal. Based on the wavelet packet decomposition formula expressed in Equation 1 above, the wavelet packet reconstruction formula is obtained. Equation 10 represents the wavelet packet reconstruction. Specifically, Equation 10 includes:

[0090]

[0091] in, These are the reconstructed wavelet packet coefficients. and S1 represents the low-pass and high-pass filters in wavelet packet decomposition. The reconstructed signal corresponding to the reconstructed wavelet packet coefficients can be found based on the reconstructed wavelet packet coefficients.

[0092] The noise-free acoustic emission signal and the noisy acoustic emission signal are obtained, denoted as S and S', respectively. N According to S1, S and S N Using Formula 11, the relative error, denoted as e, is calculated. Specifically, Formula 11 includes:

[0093]

[0094] Where S is a noiseless acoustic emission signal, S N S1 is the reconstructed acoustic emission signal containing noise, and γ is the acoustic emission signal containing noise. d1 (γ d1 ≥0) is the overall deviation factor, γ d2 (γ d2 ≥0) is the extreme value deviation factor, and γ d1 +γ d2 =1, where N is the length of the signal. Let S be the Euclidean distance, representing the local deviation between the denoised and reconstructed acoustic emission signal and the noise-free emission signal. maxS-S1 represents the local deviation between the denoised and reconstructed acoustic emission signal and the noise-free emission signal. The local deviation between the reconstructed acoustic emission signal after noise reduction and the acoustic emission signal containing noise is represented by a value of 0.5. In order to balance the overall deviation and the local deviation, the overall deviation and the local deviation are both set to 0.5.

[0095] The reconstruction factor is further defined as the reciprocal of the relative error, as shown in Formula Twelve:

[0096]

[0097] The larger the reconstruction factor, the better the wavelet packet processing effect. Therefore, the wavelet packet basis function that satisfies the largest reconstruction factor is selected.

[0098] Step S220: Obtain the sub-component signal corresponding to the wavelet packet coefficients based on the wavelet packet coefficients; wherein, the S1 sub-component signal includes the time domain information and frequency domain information of the sub-component.

[0099] Based on the wavelet packet coefficients, the sub-component signals corresponding to the wavelet packet coefficients can be retrieved from the wavelet packet sub-band tree.

[0100] It is understandable that since the sub-component signals are acoustic emission signals at different frequency bands, the time domain information of the sub-components is consistent with the time domain information of the original acoustic emission signal. That is, the time domain information of the original acoustic emission signal can be obtained from the time domain information of the sub-components.

[0101] This embodiment determines the optimal wavelet packet basis function based on the decomposed and reconstructed acoustic emission signal before wavelet packet decomposition. Then, it determines the maximum decomposition level of the wavelet packet, i.e., the optimal decomposition level, based on the minimum frequency of the useful signal in the acoustic emission signal. The acoustic emission signal is then decomposed based on the maximum decomposition level and the optimal wavelet packet basis function to obtain the wavelet packet decomposition coefficients. The corresponding sub-component signals are obtained based on the wavelet packet decomposition coefficients, which ensures the noise reduction effect of the acoustic emission signal and improves the time-frequency resolution of the acoustic emission signal.

[0102] Step S300: Perform phase space reconstruction on the sub-component signals to obtain the phase point matrix of the acoustic emission signals.

[0103] Phase space reconstruction technology proposes that the evolution of any component in a system is determined by other components that interact with it, and the information of related components is implicit in the development process of any component. That is, the original dynamic system model can be reconstructed using an observation of the system. In this embodiment, phase space reconstruction technology is used to recover and characterize the original dynamic system based on the existing time series.

[0104] The feature data is data that includes the time-frequency domain features of the reconstructed sub-component signals.

[0105] In one specific embodiment, step S300, the method further includes:

[0106] Step S310: Obtain the time series of the sub-component signal based on the time domain information of the sub-component signal.

[0107] The time-domain information is the change of the sub-component signal over time.

[0108] A time series is a sequence formed by arranging the values ​​of sub-component signals at different times in chronological order.

[0109] Step S320: Reconstruct the phase space of the time series of the sub-component signal according to Formula 4 to obtain the phase point matrix of the acoustic emission signal; wherein, Formula 4 includes:

[0110] X={x i ,x i+τ ,x i+2τ ,...,x i+(m-1)τ},

[0111] Where M is the time series length of the reconstructed acoustic emission signal, X is a matrix with a phase point dimension of M×m, x is the time series of the acoustic emission signal, m is the embedding dimension, τ is the delay time, and i∈[1,M].

[0112] In this embodiment, phase space reconstruction is performed using the time series of sub-component signals. Since the time domain information of the sub-component signals is consistent with the time domain information of the acoustic emission signals, the time series of the sub-component signals is also the same as the time series of the acoustic emission signals. The phase point matrix of the reconstructed acoustic emission signals can be obtained by performing phase space reconstruction on the time series of the sub-component signals.

[0113] It is important to understand that before obtaining the phase point matrix in Formula 4 above, it is necessary to determine the phase space reconstruction delay time and the embedding dimension, as follows:

[0114] First, the optimal delay time is calculated based on the time series of the acoustic emission signal, specifically including:

[0115] Based on the time series of the acoustic emission signal and Formula 5, the optimal delay time is calculated; where Formula 5 includes:

[0116]

[0117] Where I(τ) is the mutual information function, P(x(t)) is the probability of x(t) appearing in the time series of the acoustic emission signal, P(x(t+τ)) is the probability of x(t+τ) appearing in the time series of the acoustic emission signal, P(x(t),x(t+τ)) is the probability of x(t) and x(t+τ) appearing together in the time series of the acoustic emission signal, and P(x(1)) is the probability of the first signal appearing in the time series of the acoustic emission signal.

[0118] The optimal delay time τ mentioned above is the delay time corresponding to the first local minimum value reached by the mutual information function.

[0119] Then, using the Cao method, based on the aforementioned optimal delay time and the vector space of the spatially reconstructed acoustic emission signal, the embedding dimension is calculated, specifically including:

[0120] Obtain the vector space of the acoustic emission signal after spatial reconstruction based on the optimal delay time. Calculate the distance between nearest neighbor vectors of different dimensions according to the vector space and Formula 6. Formula 6 includes:

[0121]

[0122] in, The maximum norm number in the (m+1)-dimensional space. Y is the maximum norm number in m-dimensional space. n For vector space with Y μ(n) The nearest neighbor vector, a(i,m) is the distance between the nearest neighbor vectors in different dimensions, i∈[1,M];

[0123] The optimal embedding dimension is calculated based on the distance between nearest neighbor vectors at different dimensions, the optimal delay time, and Formula 7; where Formula 7 includes:

[0124]

[0125] Where E(m) is the mean of a(i,m), E(m+1) is the mean of a(i,m+1), E1(m) is the transformation rate between E(m) and E(m+1), and m is the optimal embedding dimension.

[0126] When the change of E1(m) gradually reaches a stable state, the m at the stable point is the minimum embedding dimension, that is, the optimal just-embedded dimension.

[0127] Substituting the optimal embedding dimension and the optimal delay time into Formula 4, the phase point matrix of the acoustic emission signal is obtained.

[0128] This embodiment calculates the optimal delay time and optimal embedding dimension for phase space reconstruction of the acoustic emission signal before performing phase space reconstruction. Then, it performs phase space reconstruction on the time series of the acoustic emission signal based on the optimal delay time and optimal embedding dimension to obtain the phase point matrix of the acoustic emission signal. This realizes the recovery and characterization of the driving force system of the acoustic emission signal, providing a data foundation for subsequent steps.

[0129] Step S400: Input the phase point matrix into the convolutional neural network so that the convolutional neural network outputs the target feature data.

[0130] Convolutional Neural Networks (CNNs) are a type of feedforward neural network that includes convolutional computation and has a deep structure. The basic structure includes an input layer, convolutional layers, pooling layers, fully connected layers, and an output layer. Features are extracted through convolutional and pooling layers, gradually transforming low-level features into high-level features. The high-level features are then classified through fully connected layers and the output layer to produce a one-dimensional vector.

[0131] The target feature data is a one-dimensional vector containing features with weights.

[0132] This step extracts feature data related to powder particle size from the acoustic emission signal using convolutional and pooling layers. This data can be energy data, but is not limited to energy data.

[0133] In one specific embodiment, step S400 includes:

[0134] Step S410: Input the phase point matrix of the reconstructed acoustic emission signal into the convolutional neural network, so that the convolutional neural network performs convolution calculation on the phase point matrix according to Formula 8 to obtain convolutional layer data, then performs dimensionality reduction and downsampling on the convolutional layer data to obtain pooling layer data, and flattens the pooling layer data to obtain the target feature data; wherein, Formula 8 includes:

[0135]

[0136] in, For convolutional layer data, X i+m,j+m f is the value in the M-th row and m-th column of the phase point matrix. cov () represents the activation function to be selected, ω M,m Let b be the weight of the convolution kernel in rows M and columns m. M,m denoted as kernel bias, and k is the sliding window size.

[0137] It is understandable that, such as Figure 4 As shown, this embodiment uses convolutional layers to extract features from the input phase matrix, pooling layers to perform dimensionality reduction and downsampling through filters and sliding window strides to remove interference and noise information, maximizing the activation of features with high weights, and flattening layers to flatten the features with high weights, resulting in a one-dimensional array that can be recognized by the Long Short-Term Memory (LSTM) network.

[0138]

[0139] In this embodiment, feature data is input into a convolutional neural network. The convolutional layer in the convolutional neural network extracts features from the point matrix, the pooling layer reduces the dimensionality of the feature data, activates features with large weights, removes interference and noise from the features, and the flattening layer flattens the feature data to obtain a vector. There is no need to manually select features, and the convolutional and pooling layers extract features multiple times, which improves the feature extraction effect.

[0140] Step S500: Input the target feature data into the long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal.

[0141] Long Short-Term Memory (LSTM) networks are a type of recurrent neural network that can learn the long-term and short-term dependencies between time-series data. They can autonomously learn the contribution of historical data to the predicted data based on the input data. In particular, this type of neural network overcomes the shortcoming of other neural networks that cannot learn for long periods of time.

[0142] The particle size distribution prediction results are obtained by using a long short-term memory network to predict the particle size of the target features.

[0143] like Figure 4 As shown, this step combines a long short-term memory network with a convolutional neural network. The convolutional neural network extracts features from the data, and the long short-term memory network trains a model based on historical feature data to obtain a prediction model. The prediction model is then used to predict the target feature data to obtain the particle size distribution prediction result of the acoustic emission signal. Finally, the particle size distribution prediction result is output through the output layer of the long short-term memory network.

[0144] It is understandable that the acoustic emission signal contains frequency domain information. The instantaneous power of the acoustic emission signal can be obtained from the instantaneous amplitude of the acoustic emission signal. From the instantaneous power of the acoustic emission signal, the energy data of the acoustic emission signal can be further obtained. Since the energy data can reflect the particle size of the powder under certain operating conditions, the convolutional neural network extracts features from the phase matrix to extract the energy features of the acoustic emission signal. The long short-term memory network then predicts the particle size of the powder based on the energy features to obtain the particle size distribution prediction result corresponding to the acoustic emission signal.

[0145] In one specific embodiment, step S500 includes:

[0146] Step S510: Input the target feature data into the Long Short-Term Memory network, so that the Long Short-Term Memory network predicts the target feature data according to the target prediction model, obtains the predicted feature data, and uses the attention mechanism to assign weights to the predicted feature data to obtain the particle size distribution prediction result corresponding to the transmitted signal; wherein, the target prediction model is obtained by training the Long Short-Term Memory network based on historical target feature data.

[0147] Attention mechanisms reallocate resources, or weights, based on the importance of objects. The core idea is to find the correlations between existing data and then highlight certain important features.

[0148] Historical target feature data refers to target feature data that existed prior to the current measurement period.

[0149] It is understandable that by adding an attention mechanism at the end of the hidden layer of the Long Short-Term Memory network, this embodiment can assign weights to each vector in the hidden layer, allocate more computing power to data with larger weights, and thus associate the output of the hidden layer with the final prediction result of the model, thereby improving the accuracy of the prediction.

[0150] This embodiment inputs the target feature data into a long short-term memory network, which then uses a target prediction model to predict the target feature data, obtains predicted feature data, and then uses an attention mechanism to assign weights to the predicted feature data. The predicted feature data with larger weights is output as the particle size distribution prediction result, thereby improving the accuracy of the prediction.

[0151] Step S600: Obtain powder particle size data based on the particle size distribution prediction results.

[0152] This embodiment acquires the acoustic emission signal, performs wavelet packet transform on the acoustic emission signal to obtain multiple sub-component signals, and then reconstructs the phase space of the sub-component signals to obtain the phase point matrix of the acoustic emission signal. The phase point matrix is ​​input into a convolutional neural network, which extracts features from the phase point matrix to obtain target feature data. The target feature data is then input into a long short-term memory network, which predicts the target feature data to obtain the particle size distribution prediction result corresponding to the acoustic emission signal. Based on the particle size distribution prediction result, the corresponding powder particle size data is obtained. This solves the problem that existing technologies cannot measure powder particle size in real time and accurately, meets the need for real-time online measurement of powder particle size, and improves the accuracy and precision of particle size measurement.

[0153] Example 2

[0154] Please refer to Figure 5 , Figure 5 This is a schematic diagram of the structure of a powder particle size measuring device 200 provided in an embodiment of this application.

[0155] Signal acquisition module 210 is used to acquire acoustic emission signals;

[0156] The signal transformation module 220 is used to perform wavelet packet transformation on the acoustic emission signal to obtain multiple sub-component signals;

[0157] The signal reconstruction module 230 is used to reconstruct the phase space of the sub-component signals to obtain the phase point matrix of the acoustic emission signal;

[0158] The convolution module 240 is used to input the phase point matrix into the convolutional neural network, so that the convolutional neural network outputs the target feature data;

[0159] The prediction module 250 is used to input the target feature data into the long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal;

[0160] The data acquisition module 260 is used to obtain powder particle size data based on the particle size distribution prediction results.

[0161] It should be understood that this device corresponds to the above-described method embodiment for measuring powder particle size and is capable of performing the various steps involved in the above method embodiment. The specific functions of this device can be found in the description above, and detailed descriptions are omitted here to avoid repetition. The device includes at least one software function module that can be stored in memory or embedded in the device's operating system (OS) in the form of software or firmware.

[0162] Example 3

[0163] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0164] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0165] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0166] Example 4

[0167] This invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, are adapted to perform the various steps described in any possible implementation of the powder particle size measurement method.

[0168] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0169] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0170] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0171] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0172] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not describe the various possible combinations separately.

[0173] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0174] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0175] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for measuring the particle size of powder, characterized in that, include: Acquire acoustic emission signals; The acoustic emission signal is subjected to wavelet packet transform to obtain multiple sub-component signals; The phase space of the sub-component signals is reconstructed to obtain the phase point matrix of the acoustic emission signal; The phase point matrix is ​​input into a convolutional neural network, so that the convolutional neural network outputs target feature data; The target feature data is input into a long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal; Based on the particle size distribution prediction results, the powder particle size data is obtained; The step of reconstructing the phase space of the sub-component signals to obtain the phase point matrix of the acoustic emission signal includes: Based on the time-domain information of the sub-component signals, the time series of the sub-component signals is obtained; The phase space of the time series of the sub-component signals is reconstructed according to Formula 4 to obtain the phase point matrix of the acoustic emission signal; wherein, Formula 4 includes: ， Where M is the time series length of the reconstructed acoustic emission signal, and X is the dimension of the phase point. The matrix is ​​given by x, where x is the time series of the acoustic emission signal, and m is the embedding dimension. To delay time, ; The phase space reconstruction of the time series of the sub-component signals according to Formula 4, to obtain the phase point matrix of the acoustic emission signal, includes: Based on the time sequence of the acoustic emission signal and Formula 5, the optimal delay time is calculated; wherein, Formula 5 includes: ， in, For mutual information functions, for The probability of occurrence of the acoustic emission signal in the time series. for The probability of occurrence of the acoustic emission signal in the time series. for and The probability of them occurring together in the time series of the acoustic emission signals The probability that the first signal in the time series of the acoustic emission signal appears in the time series of the acoustic emission signal; Obtain the vector space of the acoustic emission signal after spatial reconstruction based on the optimal delay time; Based on the vector space and Formula Six, the distances between nearest neighbor vectors of different dimensions are calculated; wherein, Formula Six includes: ， in, The maximum norm number in the (m+1)-dimensional space. The maximum norm number in m-dimensional space. For the vector space and The nearest neighbor vector, The distance between the nearest neighbor vectors in the different dimensions. ; The optimal embedding dimension is calculated based on the distance between the nearest neighbor vectors at different dimensions, the optimal delay time, and Formula 7; wherein Formula 7 includes: ， in, for The mean, for The mean, for and The transformation rate, where m is the optimal embedding dimension; Substituting the optimal embedding dimension and the optimal delay time into Formula 4, the phase point matrix of the acoustic emission signal is obtained.

2. The method for measuring powder particle size according to claim 1, characterized in that, The wavelet packet transform of the acoustic emission signal yields multiple sub-component signals, including: According to Formula 1, wavelet packet decomposition is performed on the acoustic emission signal to obtain wavelet packet coefficients of multiple components; wherein, Formula 1 includes: ， in, and Here are the wavelet packet coefficients; j is the decomposition level; l and k are the translation parameters; and n is the frequency parameter. It is a low-pass filter component. It is a high-pass filter component. ; Based on the wavelet packet coefficients, the sub-component signal corresponding to the wavelet packet coefficients is obtained; wherein, the sub-component signal includes the time-domain information and frequency-domain information of the sub-component.

3. The method for measuring powder particle size according to claim 2, characterized in that, Before performing wavelet packet decomposition on the acoustic emission signal according to formula one to obtain the wavelet packet coefficients of multiple components, the method further includes: Obtain the wavelet center frequency and wavelet sampling frequency; Based on the wavelet center frequency, the wavelet sampling frequency, the preset decomposition level, and Formula 2, the pseudo-frequency corresponding to the preset decomposition level is calculated; wherein, Formula 2 includes: ， Where k is the preset decomposition level. The pseudo-frequency corresponding to the preset number of decomposition layers k, The wavelet center frequency is, The wavelet sampling frequency; Obtain the frequency of the useful signal in the acoustic emission signal; Based on the frequency of the useful signal, the pseudo-frequency, and Formula 3, the optimal number of wavelet packet decomposition layers is calculated; wherein, Formula 3 includes: ， Where L is the optimal wavelet packet decomposition layer. This refers to the frequency corresponding to the optimal wavelet packet decomposition level when the decomposition level is equal to the number of decomposition levels. The frequency of the useful signal is defined as min, where min represents the minimum value and max represents the maximum value. The wavelet packet decomposition of the acoustic emission signal according to formula A yields wavelet packet coefficients for multiple components, including: The acoustic emission signal is decomposed into wavelet packets according to Formula 1 and the optimal number of wavelet packet decomposition layers to obtain the wavelet packet coefficients of the multiple components.

4. The method for measuring powder particle size according to claim 1, characterized in that, The step of inputting the target feature data into a long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal includes: The reconstructed phase point matrix of the acoustic emission signal is input into the convolutional neural network, which performs convolution calculation on the phase point matrix according to Formula 8 to obtain convolutional layer data. The convolutional layer data is then downsampled to obtain pooling layer data, and the pooling layer data is flattened to obtain the target feature data. Formula 8 includes: ， in, For the convolutional layer data, The value in the M-th row and m-th column of the phase point matrix. To select an activation function, The weights are M rows and m columns of the convolution kernel. denoted as kernel bias, and k is the sliding window size.

5. The method for measuring powder particle size according to claim 4, characterized in that, The step of inputting the target feature data into a long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal includes: The target feature data is input into the long short-term memory network, which then predicts the target feature data based on the target prediction model to obtain predicted feature data. The predicted feature data is weighted using an attention mechanism to obtain a particle size distribution prediction result corresponding to the emitted signal. The target prediction model is obtained by training the long short-term memory network based on historical target feature data.

6. A device for measuring the particle size of powder, characterized in that, include: The signal acquisition module is used to acquire acoustic emission signals; The signal transformation module is used to perform wavelet packet transformation on the acoustic emission signal to obtain multiple sub-component signals; The signal reconstruction module is used to reconstruct the phase space of the sub-component signals to obtain the phase point matrix of the acoustic emission signal; The convolution module is used to input the phase point matrix into the convolutional neural network, so that the convolutional neural network outputs target feature data; The prediction module is used to input the target feature data into a long short-term memory network to obtain the particle size distribution prediction result corresponding to the acoustic emission signal; The data acquisition module is used to obtain powder particle size data based on the particle size distribution prediction results; The step of reconstructing the phase space of the sub-component signals to obtain the phase point matrix of the acoustic emission signal includes: Based on the time-domain information of the sub-component signals, the time series of the sub-component signals is obtained; The phase space of the time series of the sub-component signals is reconstructed according to Formula 4 to obtain the phase point matrix of the acoustic emission signal; wherein, Formula 4 includes: ， Where M is the time series length of the reconstructed acoustic emission signal, and X is the dimension of the phase point. The matrix is ​​given by x, where x is the time series of the acoustic emission signal, and m is the embedding dimension. To delay time, ; The phase space reconstruction of the time series of the sub-component signals according to Formula 4, to obtain the phase point matrix of the acoustic emission signal, includes: Based on the time sequence of the acoustic emission signal and Formula 5, the optimal delay time is calculated; wherein, Formula 5 includes: ， in, For mutual information functions, for The probability of occurrence of the acoustic emission signal in the time series. for The probability of occurrence of the acoustic emission signal in the time series. for and The probability of them occurring together in the time series of the acoustic emission signals The probability that the first signal in the time series of the acoustic emission signal appears in the time series of the acoustic emission signal; Obtain the vector space of the acoustic emission signal after spatial reconstruction based on the optimal delay time; Based on the vector space and Formula Six, the distances between nearest neighbor vectors of different dimensions are calculated; wherein, Formula Six includes: ， in, The maximum norm number in the (m+1)-dimensional space. The maximum norm number in m-dimensional space. For the vector space and The nearest neighbor vector, The distance between the nearest neighbor vectors in the different dimensions. ; The optimal embedding dimension is calculated based on the distance between the nearest neighbor vectors at different dimensions, the optimal delay time, and Formula 7; wherein Formula 7 includes: ， in, for The mean, for The mean, for and The transformation rate, where m is the optimal embedding dimension; Substituting the optimal embedding dimension and the optimal delay time into Formula 4, the phase point matrix of the acoustic emission signal is obtained.

7. An electronic device, characterized in that, include: A processor and a memory, the memory storing machine-readable instructions executable by the processor, which, when executed by the processor, perform the method for measuring powder particle size as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions for causing the machine to perform the method for measuring the particle size of powder as described in any one of claims 1-5.

Images