Particle fluid flow rate measurement system and method based on passive acoustics and ai compensation

Abstract

This invention relates to the field of particulate fluid velocity detection technology, and in particular to a particulate fluid velocity measurement system and method based on passive acoustics and AI compensation. The system includes a passive acoustic acquisition module, a signal processing unit, an AI compensation module, a fusion processing module, a communication module, a visualization terminal, a dynamic weight adjustment module, a low-power monitoring module, and an environmental monitoring module. The signal processing unit includes a filtering and denoising module, a time-domain feature extraction module, and a frequency-domain feature extraction module. The filtering and denoising module is connected to the passive acoustic acquisition module, and the time-domain feature extraction module is connected to the filtering and denoising module. This approach solves the technical problems of existing particulate fluid velocity measurement methods, such as poor response to low flow rates and discontinuous flow fields, susceptibility to particle disturbances and pipe wall interference leading to measurement distortion, sensor wear and blockage requiring frequent maintenance, and difficulty in integrating intelligent compensation of multiple physical features.

Description

Technical Field

[0001] This invention relates to the field of particulate fluid velocity detection technology, and in particular to a particulate fluid velocity measurement system and method based on passive acoustics and AI compensation. Background Technology

[0002] With the widespread application of particulate fluids in chemical, pharmaceutical, energy, and food processing industries, flow velocity measurement, as a key parameter, is of great significance for process optimization and safety control. Currently, the mainstream methods for measuring particulate fluid flow velocity mainly include:

[0003] Electromagnetic induction flow meter:

[0004] This type of sensor utilizes the Faraday principle of electromagnetic induction and is suitable for conductive liquids, but it is ineffective for measuring non-conductive fluids (such as some oils or particulate gas-solid two-phase flows), and the sensor is easily interfered with in high-temperature and high-dust environments.

[0005] Differential pressure flow meters (such as orifice plates and venturi flow meters):

[0006] The flow velocity is estimated by measuring the pressure difference across the fluid. This method is simple in structure, but it is susceptible to clogging by particle deposition, and the relationship between pressure difference and flow velocity is no longer linear in the low Reynolds number range, making it unsuitable for dynamically changing particulate flows.

[0007] Mechanical impeller flow meter:

[0008] The flow rate is sensed by rotating components, but in high-wear, high-dust or strong-impact flow fields, the lifespan is short, the response is sluggish, and maintenance is frequent, so it is gradually being phased out.

[0009] In summary, existing methods for measuring particulate fluid velocity have several drawbacks, including poor response to low flow rates and discontinuous flow fields, susceptibility to measurement distortion due to particle disturbances and pipe wall interference, sensor wear and blockage requiring frequent maintenance, and difficulty in integrating intelligent compensation based on multiple physical features. Summary of the Invention

[0010] The purpose of this invention is to provide a particulate fluid velocity measurement system and method based on passive acoustics and AI compensation, aiming to solve the technical problems of existing particulate fluid velocity measurement methods, such as poor response to low flow rates and discontinuous flow fields, susceptibility to particle disturbances and pipe wall interference leading to measurement distortion, sensor wear and blockage requiring frequent maintenance, and difficulty in integrating intelligent compensation of multiple physical features.

[0011] To achieve the above objectives, this invention employs a particulate fluid velocity measurement system based on passive acoustics and AI compensation.

[0012] The system includes a passive acoustic acquisition module, a signal processing unit, an AI compensation module, a fusion processing module, a communication module, a visualization terminal, a dynamic weight adjustment module, a low-power monitoring module, and an environmental monitoring module. The signal processing unit includes a filtering and denoising module, a time-domain feature extraction module, and a frequency-domain feature extraction module. The filtering and denoising module is connected to the passive acoustic acquisition module, the time-domain feature extraction module is connected to the filtering and denoising module, the frequency-domain feature extraction module is connected to the time-domain feature extraction module, the AI ​​compensation module is connected to the frequency-domain feature extraction module, and the fusion processing module is connected to the AI ​​compensation module. The communication module is located between the AI ​​compensation module and the visualization terminal. The dynamic weight adjustment module is connected to the fusion processing module, the low-power monitoring module is connected to the passive acoustic acquisition module, and the environmental monitoring module is connected to the fusion processing module.

[0013] The passive acoustic acquisition module is used to acquire data and transmit it to the signal processing unit;

[0014] The filtering and denoising module in the signal processing unit is used to filter and denoise the data. The filtered and denoised data is then transmitted to the time-domain feature extraction module for time-domain feature extraction. The frequency-domain feature extraction module is used to extract frequency-domain features from the data.

[0015] The AI ​​compensation module performs high-precision correction on the initial estimated flow rate value obtained by passive acoustics based on the Transformer structure or LSTM neural network, and the fusion processing module performs fusion processing on the data collected by the passive acoustic acquisition module and the data predicted by the AI ​​compensation module based on preset weights to obtain the final output flow rate.

[0016] The communication module transmits the final output flow rate data to the visualization terminal.

[0017] The dynamic weight adjustment module is used to adjust the fusion weight of the fusion processing module, and the dynamic weight adjustment logic is set by using a lookup table method or a rule engine.

[0018] The low-power monitoring module is used to acquire and process signals triggered by events.

[0019] The environmental monitoring module is used to collect environmental characteristics, including temperature, humidity, and particle concentration. The particulate fluid velocity measurement system based on passive acoustics and AI compensation also includes a management module connected to the weighted dynamic adjustment module.

[0020] This invention also provides a method for measuring particulate fluid velocity based on passive acoustics and AI compensation, applicable to the particulate fluid velocity measurement system based on passive acoustics and AI compensation as described above.

[0021] Includes the following steps:

[0022] After constructing and training the AI ​​compensation module, it is deployed in the measurement system.

[0023] The passive acoustic acquisition module non-invasively acquires the impact and friction acoustic emission signals between particles and the pipe wall or between particles inside the pipe, and the signal processing unit performs filtering, noise reduction, and time-domain and frequency-domain feature extraction on the acoustic signals.

[0024] The deployed AI compensation module is used to perform high-precision correction on the flow velocity value by combining the extracted time-domain and frequency-domain features. The fusion processing module then performs fusion processing on the data collected by the passive acoustic acquisition module and the data predicted by the AI ​​compensation module based on preset weights to obtain the final output flow velocity.

[0025] The communication module transmits the final output flow rate data to the visualization terminal.

[0026] The specific method for constructing and training the AI ​​compensation module and then deploying it in the measurement system is as follows:

[0027] First, feature construction and dataset preparation are carried out.

[0028] Then, the AI ​​model structure is designed using the LSTM sequence model structure or the Transformer structure, and the model is trained and optimized using the dataset;

[0029] The optimized AI model structure is deployed in the measurement system.

[0030] The data collection steps during dataset preparation are as follows:

[0031] Standard flow meters were deployed under different flow rates, particle concentrations, and material conditions.

[0032] Simultaneously acquire acoustic emission signals and reference flow velocity values, and record timestamps and sensor numbers;

[0033] The raw data is processed in a sliding window (e.g., 50ms) to extract the corresponding feature vectors and the true flow velocity values ​​as labels;

[0034] Data preprocessing is then performed:

[0035] All input features are normalized.

[0036] Downsampling is performed on curve-type features such as PSD to unify dimensions;

[0037] The flow rate labels are divided into low / medium / high speed bins to aid in data balancing.

[0038] This invention discloses a particulate fluid velocity measurement system and method based on passive acoustics and AI compensation. In practical use, the passive acoustic acquisition module first non-invasively acquires acoustic emission signals from the impact and friction between particles and the pipe wall or between particles within the pipe. The signal processing unit then filters, denoises, and extracts time-domain and frequency-domain features from the acoustic signals. The AI ​​compensation module, combined with the extracted time-domain and frequency-domain features, performs high-precision correction on the velocity value. The fusion processing module, based on preset weights, fuses the data acquired by the passive acoustic acquisition module with the data predicted by the AI ​​compensation module to obtain the final output velocity. The communication module transmits the final output velocity data to the visualization terminal. This approach solves the technical problems of existing particulate fluid velocity measurement methods, such as poor response to low-velocity and discontinuous flow fields, susceptibility to particle disturbances and pipe wall interference leading to measurement distortion, sensor wear and blockage requiring frequent maintenance, and difficulty in integrating intelligent compensation of multiple physical features.

[0039] This invention features a completely non-invasive passive acoustic structure, which enhances the system's durability and adaptability in highly abrasive and high-particle-concentration media.

[0040] The AI ​​compensation module is introduced to dynamically compensate for flow velocity errors caused by flow field, particle distribution, and noise interference, significantly improving measurement accuracy under low flow velocity or special working conditions.

[0041] Differentiated feature design, integrating time delay features and multi-dimensional spectral features, breaks through the accuracy bottleneck of traditional signal processing algorithms and achieves measurement accuracy within ±2% FS;

[0042] It supports universal flow velocity monitoring for solid-liquid two-phase or multi-phase fluids such as mud, coal slurry, and mineral slurry.

[0043] This invention provides a flow velocity measurement technology that requires no structural contact and can stably sense the behavior of particulate fluids in complex noisy environments. In particular, it can perform dynamic compensation and feature fusion analysis for unstable flow fields and particulate disturbance scenarios, thereby improving the overall system's stability, versatility, and accuracy at low flow velocities. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the first embodiment of the present invention.

[0046] Figure 2 This is a schematic diagram of the second embodiment of the present invention.

[0047] Figure 3 This is a flowchart of the particle fluid velocity measurement based on passive acoustics and AI compensation according to the present invention.

[0048] 101-Passive acoustic acquisition module, 102-Signal processing unit, 103-AI compensation module, 104-Fusion processing module, 105-Communication module, 106-Visualization terminal, 107-Weight dynamic adjustment module, 108-Management module, 109-Low power monitoring module, 110-Environmental monitoring module, 111-Filtering and noise reduction module, 112-Time domain feature extraction module, 113-Frequency domain feature extraction module, 201-Login module, 202-Authentication module, 203-Access control module, 204-Anomaly detection module, 205-Alarm module. Detailed Implementation

[0049] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, but should not be construed as limiting the present invention.

[0050] The first embodiment of this application is as follows:

[0051] Please see Figure 1 , Figure 1 This is a schematic diagram of the first embodiment of the present invention.

[0052] This invention provides a particulate fluid velocity measurement system based on passive acoustics and AI compensation, including a passive acoustic acquisition module 101, a signal processing unit 102, an AI compensation module 103, a fusion processing module 104, a communication module 105, a visualization terminal 106, a weight dynamic adjustment module 107, a management module 108, a low-power monitoring module 109, and an environmental monitoring module 110. The signal processing unit 102 includes a filtering and noise reduction module 111, a time-domain feature extraction module 112, and a frequency-domain feature extraction module 113. The aforementioned solution solves the technical problems of existing particulate fluid velocity measurement methods, such as poor response to low flow rates and discontinuous flow fields, susceptibility to particle disturbances and pipe wall interference leading to measurement distortion, sensor wear and blockage requiring frequent maintenance, and difficulty in integrating intelligent compensation of multiple physical features.

[0053] In this specific embodiment, the passive acoustic acquisition module 101 is used to acquire data and transmit it to the signal processing unit 102;

[0054] The filtering and denoising module 111 in the signal processing unit 102 is used to filter and denoise the data. The filtered and denoised data is then transmitted to the time-domain feature extraction module 112 for time-domain feature extraction. The frequency-domain feature extraction module 113 is used to extract frequency-domain features from the data.

[0055] The AI ​​compensation module 103 performs high-precision correction on the initial estimated flow rate value measured by passive acoustics based on the Transformer structure or LSTM neural network, and the fusion processing module 104 performs fusion processing on the data collected by the passive acoustic acquisition module 101 and the data predicted by the AI ​​compensation module 103 based on preset weights to obtain the final output flow rate.

[0056] The AI ​​compensation module 103 selects the model structure according to the characteristics of the working condition data: LSTM is suitable for continuous time signal processing, and Transformer is suitable for multi-feature fusion and complex pattern recognition.

[0057] The communication module 105 transmits the final output flow rate data to the visualization terminal 106.

[0058] The filtering and denoising module 111 is connected to the passive acoustic acquisition module 101, the time-domain feature extraction module 112 is connected to the filtering and denoising module 111, the frequency-domain feature extraction module 113 is connected to the time-domain feature extraction module 112, the AI ​​compensation module 103 is connected to the frequency-domain feature extraction module 113, the fusion processing module 104 is connected to the AI ​​compensation module 103, and the communication module 105 is disposed between the AI ​​compensation module 103 and the visualization terminal 106. In specific use, the passive acoustic acquisition module 101 first non-invasively collects the impact and friction acoustic emission signals between particles and the pipe wall or between particles inside the pipe, and then the signal processing unit 102 processes the acoustic signals. The process involves filtering, denoising, and extracting time and frequency domain features. The AI ​​compensation module 103, combined with the extracted time and frequency domain features, performs high-precision correction on the flow velocity value. The fusion processing module 104, based on preset weights, fuses the data acquired by the passive acoustic acquisition module 101 with the data predicted by the AI ​​compensation module 103 to obtain the final output flow velocity. The communication module 105 transmits the final output flow velocity data to the visualization terminal 106. This approach solves the technical problems of existing particulate fluid velocity measurement methods, such as poor response to low flow velocities and discontinuous flow fields, susceptibility to particle disturbances and pipe wall interference leading to measurement distortion, sensor wear and blockage requiring frequent maintenance, and difficulty in integrating intelligent compensation of multiple physical features.

[0059] Secondly, the weight dynamic adjustment module 107 is connected to the fusion processing module 104;

[0060] The weight dynamic adjustment module 107 is used to adjust the fusion weight of the fusion processing module 104, and uses a lookup table method or a rule engine to set the dynamic weight adjustment logic.

[0061] Meanwhile, the management module 108 is connected to the weight dynamic adjustment module 107. The management module 108 is used to manage the weight dynamic adjustment module 107, so that the administrator can adjust the logic, etc.

[0062] Furthermore, the low-power monitoring module 109 is connected to the passive acoustic acquisition module 101, and the low-power monitoring module 109 continuously monitors the acoustic signal energy. In the "silent" state, it only maintains basic signal monitoring, and the AI ​​compensation module 103 remains dormant. When a sudden acoustic emission event is detected (such as a signal above a certain amplitude or a spectral change), the comparator circuit triggers the wake-up module, activating the passive acoustic acquisition module 101 and the AI ​​compensation module 103 to perform signal acquisition and prediction. After acquisition is complete, it re-enters the low-power monitoring state according to set conditions, realizing an event-driven working mode.

[0063] Furthermore, the environmental monitoring module 110 is connected to the fusion processing module 104. The environmental monitoring module 110 monitors relevant environmental parameters in real time. Environmental parameters may affect the propagation of acoustic signals and measurement results. The environmental monitoring module 110 transmits the monitored data to the fusion processing module 104 so that the influence of environmental factors can be considered during the data fusion process, thereby further improving the measurement accuracy.

[0064] Using the particulate fluid velocity measurement system based on passive acoustics and AI compensation in this embodiment, in specific use, the passive acoustic acquisition module 101 first non-invasively collects the acoustic emission signals of impact friction between particles and the pipe wall or between particles inside the pipe. The signal processing unit 102 then filters, denoises, and extracts time-domain and frequency-domain features from the acoustic signals. The AI ​​compensation module 103 uses the extracted time-domain and frequency-domain features to perform high-precision correction on the velocity value and transmits it to the fusion processing module 104. The fusion processing module 104 fuses the data collected by the passive acoustic acquisition module 101 and the data predicted by the AI ​​compensation module 103 based on preset weights to obtain the final output velocity. The communication module 105 transmits the final output velocity data to the visualization terminal 106. This method solves the technical problems of existing particulate fluid velocity measurement methods, such as poor response to low velocity and discontinuous flow fields, susceptibility to particle disturbance and pipe wall interference leading to measurement distortion, sensor wear and blockage requiring frequent maintenance, and difficulty in integrating intelligent compensation of multiple physical features.

[0065] This invention features a completely non-invasive passive acoustic structure, which enhances the system's durability and adaptability in highly abrasive and high-particle-concentration media.

[0066] The AI ​​compensation module 103 is introduced to dynamically compensate for flow velocity errors caused by flow field, particle distribution and noise interference, and significantly improve the measurement accuracy under low flow velocity or special working conditions.

[0067] Differentiated feature design, integrating time delay features and multi-dimensional spectral features, breaks through the accuracy bottleneck of traditional signal processing algorithms and achieves measurement accuracy within ±2% FS;

[0068] It supports universal flow velocity monitoring for solid-liquid two-phase or multi-phase fluids such as mud, coal slurry, and mineral slurry.

[0069] This invention provides a flow velocity measurement technology that requires no structural contact and can stably sense the behavior of particulate fluids in complex noisy environments. In particular, it can perform dynamic compensation and feature fusion analysis for unstable flow fields and particulate disturbance scenarios, thereby improving the overall system's stability, versatility, and accuracy at low flow velocities.

[0070] The second embodiment of this application is as follows:

[0071] Based on the first embodiment, please refer to Figure 2 , Figure 2 This is a schematic diagram of the second embodiment of the present invention.

[0072] The present invention provides a particulate fluid velocity measurement system based on passive acoustics and AI compensation, and further includes a login module 201, an identity verification module 202, an access control module 203, an anomaly detection module 204, and an alarm module 205.

[0073] In this specific embodiment, the login module 201 is used to log in to the visualization terminal 106, and the identity verification module 202 is used to verify the identity of the administrator logging into the visualization terminal 106. When the administrator enters a username and password in the login module 201, the front end first performs hash encryption on the password. The encrypted password is sent to the server along with the username. After receiving the data, the server queries the database for the stored password corresponding to the username (the password stored in the database is also hash-encrypted). The server compares the received encrypted password with the encrypted password stored in the database. If they match, the login is successful; otherwise, the login fails.

[0074] The permission control module 203 is connected to the identity verification module 202. The permission control module 203 controls the operation permissions based on the role access control (RBAC) permission allocation algorithm and the identity information transmitted from the identity verification module 202.

[0075] Secondly, the anomaly detection module 204 is connected to the passive acoustic acquisition module 101, and the alarm module 205 is connected to the anomaly detection module 204. The anomaly detection module 204 monitors the signals acquired by the measurement system and the calculated flow velocity data in real time to determine whether there are any abnormal situations, such as signal sudden changes or flow velocity exceeding the normal range.

[0076] When an anomaly is detected, an alarm signal is promptly issued through the alarm module 205 to notify relevant personnel for handling.

[0077] Using the particulate fluid velocity measurement system based on passive acoustics and AI compensation in this embodiment, when the administrator enters a username and password in the login module 201, the front end first performs hash encryption on the password. The encrypted password is then sent to the server along with the username. After receiving the data, the server queries the database for the stored password corresponding to the username (the password stored in the database is also hash-encrypted). The server compares the received encrypted password with the encrypted password stored in the database. If they match, the login is successful; otherwise, the login fails.

[0078] The permission control module 203 is connected to the authentication module 202. The permission control module 203 controls the operation permissions based on the role access control (RBAC) permission allocation algorithm and the identity information transmitted from the authentication module 202.

[0079] Please see Figure 3 , Figure 3This is a flowchart of the particulate fluid velocity measurement based on passive acoustics and AI compensation according to the present invention. The present invention also provides a particulate fluid velocity measurement method based on passive acoustics and AI compensation, applicable to the particulate fluid velocity measurement system based on passive acoustics and AI compensation as described above.

[0080] Includes the following steps:

[0081] S1. After constructing and training the AI ​​compensation module 103, deploy it in the measurement system;

[0082] In this specific implementation, the AI ​​compensation module 103 is constructed, trained, and then deployed in the measurement system in the following manner:

[0083] First, feature construction and dataset preparation are carried out.

[0084] Feature input dimensions include:

[0085] Cross-correlation delay (Δt);

[0086] Power spectral density (PSD);

[0087] Main frequency, spectral centroid, spectral drift;

[0088] High-frequency / low-frequency energy ratio (e.g., E_high / E_low);

[0089] Optional two-dimensional spectrogram (such as Mel spectrogram);

[0090] Environmental parameters such as temperature, pressure, and humidity;

[0091] Pipe material code (e.g., carbon steel, PVC);

[0092] Physical rough flow velocity (converted from Δt and upstream / downstream distance).

[0093] Then, the AI ​​model structure is designed using the LSTM sequence model structure or the Transformer structure, and the model is trained and optimized using the dataset;

[0094] Model training methods

[0095] Loss function: Mean Squared Error (MSE)

[0096] ;

[0097] Optimizer and hyperparameter settings:

[0098] Adam optimizer (initial learning rate 0.001, weight decay 0.01).

[0099] Batch Size: 64~128;

[0100] Epoch count: 100~200 (Early Stopping monitoring validation set RMSE);

[0101] Model evaluation metrics:

[0102] RMSE (Root Mean Square Error), MAE (Mean Absolute Error), R² (Goodness of Fit);

[0103] Meanwhile, a key verification section was established in the low-speed range of 0.1 to 1.0 m / s.

[0104] The optimized AI model structure is deployed in the measurement system.

[0105] After training with PyTorch or TensorFlow, export the model as a TFLite / ONNX / C++ model;

[0106] After quantization and compression, the chips are deployed to embedded MCUs (such as STM32H7) or edge AI chips (such as K210, ARM Cortex-A55).

[0107] The data collection steps during dataset preparation are as follows:

[0108] Standard flow meters were deployed under different flow rates, particle concentrations, and material conditions.

[0109] Simultaneously acquire acoustic emission signals and reference flow velocity values, and record timestamps and sensor numbers;

[0110] The raw data is processed in a sliding window (e.g., 50ms) to extract the corresponding feature vectors and the true flow velocity values ​​as labels;

[0111] Data preprocessing is then performed:

[0112] All input features are normalized.

[0113] Downsampling is performed on curve-type features such as PSD to unify dimensions;

[0114] The flow rate labels are divided into low / medium / high speed bins to aid in data balancing.

[0115] S2. The passive acoustic acquisition module 101 non-invasively acquires the impact and friction acoustic emission signal between particles in the tube and the tube wall or between particles, and the signal processing unit 102 performs filtering, noise reduction, and time-domain and frequency-domain feature extraction on the acoustic signal.

[0116] For this specific implementation method, the data acquisition process is as follows:

[0117] 1. Sensor arrangement: Select sensor groups with different sensitivities according to the pipe diameter. Generally, a symmetrical arrangement is adopted (the upstream and downstream spacing is set according to the pipe diameter, and it is recommended to be more than 30cm).

[0118] Based on the pipe diameter and particulate matter concentration, select a piezoelectric acoustic emission sensor with appropriate sensitivity and frequency response range (such as the R15α or WD series). The sensor uses a magnetic base or pipe clamp + industrial adhesive fixing structure, installed symmetrically upstream and downstream on the outer wall of the pipe, with adjustable spacing (recommended greater than 30 cm). During installation, ensure the sensor is tightly fitted to the pipe wall to avoid loosening that could affect signal coupling. Use shielded signal cables for wiring, running along the pipe to the acquisition unit, and protect them with cable ties or metal conduit.

[0119] 2. Sampling mechanism: Multi-channel synchronous acquisition is performed through a high-speed analog-to-digital converter (ADC, sampling rate not less than 1MSPS), and a sliding window (typically 50ms) is used to extract effective signal segments;

[0120] The sensor signal is first amplified by a low-noise preamplifier and then input to a multi-channel high-speed analog-to-digital converter module (Analog Devices AD9230 is recommended, with a sampling rate of 20 MSPS and a resolution of 12 bits).

[0121] Synchronous acquisition card (such as NI PXI-5122 or STM32+FPGA board);

[0122] Multi-channel isochronous sampling is performed, and a 50ms sliding window is used to extract feature segments during acquisition. It has pre-trigger and gating acquisition functions so that target segment data can be latched and stored after specific acoustic emission characteristics (sudden energy jump) are identified.

[0123] 3. Calibration environment: Perform on-site flow velocity calibration using a standard flow meter (such as an electromagnetic flow meter) to build training and test datasets;

[0124] A high-precision electromagnetic flowmeter (such as ABB FSM4000 or Micromotion ELITE) is installed in parallel on the test pipeline and its data is collected synchronously with that of the acoustic sensor. The calibration process includes the following steps:

[0125] Start the flow meter and acoustic emission system to record synchronously;

[0126] Adjust the pipe flow rate (0.1~3 m / s) to cover the target measurement range;

[0127] Maintain a stable flow rate for 2 minutes at each setting, and collect acoustic emission signals and corresponding flow data;

[0128] Environmental parameters such as temperature and pressure are recorded. After cleaning and preprocessing, the dataset is used for AI model training and testing.

[0129] When processing data:

[0130] Filtering and noise reduction:

[0131] Use a bandpass filter (20kHz~200kHz) to eliminate low-frequency vibrations and high-frequency electromagnetic interference;

[0132] Unstructured noise is separated using wavelet thresholding or EMD methods;

[0133] Principal component analysis (PCA) is used to decorrelate multiple signals.

[0134] To ensure signal quality and suppress low-frequency structural vibration, high-frequency electromagnetic interference and fluid background noise under complex operating conditions.

[0135] Temporal feature extraction:

[0136] Extract indicators such as peak amplitude, root mean square (RMS) signal, envelope energy, and zero crossover rate;

[0137] The cross-correlation function algorithm is used to extract the cross-correlation time delay Δt between upstream and downstream acoustic signals.

[0138] In the filtered and denoised signal, time-domain features are extracted for initial flow velocity estimation and AI model auxiliary input:

[0139] Data Acquisition and Control Steps:

[0140] The system data acquisition and control uses a 50ms sliding window (10ms step size), and the data is continuously collected from multiple windows before being sent to the AI ​​prediction module.

[0141] The control terminal supports RS485 or CAN bus access to PLC / gateway systems, enabling remote control of window triggering and data uploading.

[0142] Feature extraction items:

[0143] RMS (Root Mean Square): Measures the amount of acoustic energy and reflects the activity level of particles in a fluid;

[0144] Peak amplitude and envelope energy: for detecting strong transient events, such as high-speed particle impacts;

[0145] Zero crossover rate (ZCR): Evaluates the high-frequency components of a signal;

[0146] Cross-correlation delay Δt: The cross-correlation function algorithm is used to calculate the maximum correlation peak shift between upstream and downstream channels to estimate particle velocity.

[0147] ;

[0148] The obtained Δt, combined with the sensor spacing, can be used to calculate the "physical initial estimated flow velocity," which serves as one of the inputs to the AI ​​compensation model.

[0149] Frequency domain feature extraction:

[0150] Use Fast Fourier Transform (FFT) to extract the main frequency, spectral centroid, spectral width, etc.

[0151] Calculate the high / low frequency energy ratio (e.g., E_high / E_low) and extract spectral drift features;

[0152] Optional Mel spectrogram extraction can be used for AI model reinforcement learning.

[0153] To model the nonlinear relationship between acoustic emission characteristics and flow velocity, the system performs frequency domain analysis on each sampling window:

[0154] Processing module: The STM32 core or FPGA module uses the FFT core to perform a 1024-point Fast Fourier Transform (FFT) with a resolution better than 1.95 kHz.

[0155] Frequency domain characteristic terms:

[0156] Peak Frequency: Identify the frequency point with the highest energy in the frequency spectrum;

[0157] Spectral centroid and spectral width: reflect the location and dispersion of energy distribution;

[0158] Energy ratio E_high / E_low: The energy ratio of high frequency (>100kHz) to low frequency (20kHz~60kHz), used to identify the changes in the impact frequency distribution of particles with different flow velocities;

[0159] Frequency drift: Compares the change in the main frequency before and after the window, used for dynamic modeling of flow velocity fluctuations;

[0160] Optional Mel spectrogram: Converts the spectrum of each window into a Mel scale plot, which is then input into the AI ​​model as a two-dimensional feature, improving the model's ability to express complex spectral features.

[0161] S3. Using the deployed AI compensation module 103, the flow velocity value is corrected with high precision by combining the extracted time domain and frequency domain features. The fusion processing module 104 performs fusion processing on the data collected by the passive acoustic acquisition module 101 and the data predicted by the AI ​​compensation module 103 based on preset weights to obtain the final output flow velocity.

[0162] S4. The communication module 105 transmits the final output flow rate data to the visualization terminal 106.

[0163] The above description discloses only one preferred embodiment of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art will understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A particulate fluid velocity measurement system based on passive acoustics and AI compensation, characterized in that, The system includes a passive acoustic acquisition module, a signal processing unit, an AI compensation module, a fusion processing module, a communication module, a visualization terminal, a dynamic weight adjustment module, a low-power monitoring module, and an environmental monitoring module. The signal processing unit includes a filtering and denoising module, a time-domain feature extraction module, and a frequency-domain feature extraction module. The filtering and denoising module is connected to the passive acoustic acquisition module, the time-domain feature extraction module is connected to the filtering and denoising module, the frequency-domain feature extraction module is connected to the time-domain feature extraction module, the AI ​​compensation module is connected to the frequency-domain feature extraction module, and the fusion processing module is connected to the AI ​​compensation module. The communication module is located between the AI ​​compensation module and the visualization terminal. The dynamic weight adjustment module is connected to the fusion processing module, the low-power monitoring module is connected to the passive acoustic acquisition module, and the environmental monitoring module is connected to the fusion processing module. The passive acoustic acquisition module is used to acquire data and transmit it to the signal processing unit; The filtering and denoising module in the signal processing unit is used to filter and denoise the data. The filtered and denoised data is then transmitted to the time-domain feature extraction module for time-domain feature extraction. The frequency-domain feature extraction module is used to extract frequency-domain features from the data. The AI ​​compensation module performs high-precision correction on the initial estimated flow rate value obtained by passive acoustics based on the Transformer structure or LSTM neural network, and the fusion processing module performs fusion processing on the data collected by the passive acoustic acquisition module and the data predicted by the AI ​​compensation module based on preset weights to obtain the final output flow rate. The communication module transmits the final output flow rate data to the visualization terminal. The dynamic weight adjustment module is used to adjust the fusion weight of the fusion processing module, and the dynamic weight adjustment logic is set by using a lookup table method or a rule engine. The low-power monitoring module is used to acquire and process signals triggered by events. The environmental monitoring module is used to collect environmental characteristics, including temperature, humidity, and particle concentration.

2. The particulate fluid velocity measurement system based on passive acoustics and AI compensation as described in claim 1, characterized in that, The particulate fluid velocity measurement system based on passive acoustics and AI compensation also includes a management module, which is connected to the weight dynamic adjustment module.

3. A method for measuring particulate fluid velocity based on passive acoustics and AI compensation, applied to the particulate fluid velocity measurement system based on passive acoustics and AI compensation as described in claim 2, characterized in that, Includes the following steps: After constructing and training the AI ​​compensation module, it is deployed in the measurement system. The passive acoustic acquisition module non-invasively acquires the impact and friction acoustic emission signals between particles and the pipe wall or between particles inside the pipe, and the signal processing unit performs filtering, noise reduction, and time-domain and frequency-domain feature extraction on the acoustic signals. The deployed AI compensation module is used to perform high-precision correction on the flow velocity value by combining the extracted time-domain and frequency-domain features. The fusion processing module then performs fusion processing on the data collected by the passive acoustic acquisition module and the data predicted by the AI ​​compensation module based on preset weights to obtain the final output flow velocity. The communication module transmits the final output flow rate data to the visualization terminal.

4. The particulate fluid velocity measurement method based on passive acoustics and AI compensation as described in claim 3, characterized in that, The specific method for constructing and training the AI ​​compensation module and then deploying it in the measurement system is as follows: First, feature construction and dataset preparation are carried out. Then, the AI ​​model structure is designed using the LSTM sequence model structure or the Transformer structure, and the model is trained and optimized using the dataset; The optimized AI model structure is deployed in the measurement system.

5. The particulate fluid velocity measurement method based on passive acoustics and AI compensation as described in claim 4, characterized in that, When preparing the dataset, the data collection steps are as follows: Standard flow meters were deployed under different flow rates, particle concentrations, and material conditions. Simultaneously acquire acoustic emission signals and reference flow velocity values, and record timestamps and sensor numbers; The raw data is processed by sliding a window, and the corresponding feature vectors and real flow velocity values ​​are extracted as labels. Data preprocessing is then performed: All input features are normalized. Downsampling is performed on curve-type features such as PSD to unify dimensions; The flow rate labels are divided into low / medium / high speed bins to aid in data balancing.

Images