Device operation state monitoring method, device and system

Abstract

This invention provides a method, apparatus, and system for monitoring the operating status of equipment, belonging to the field of acoustic monitoring technology. The equipment includes a device wall and internal components. The method includes: acquiring acoustic signals from the equipment during a certain operating time; extracting modal features of the acoustic signals and, based on the modal features, determining the component generating the current acoustic signal, wherein the generating component is the device wall or an internal component; and, based on the generating component, calling a corresponding acoustic analysis algorithm to analyze the acoustic signal and determine the state of the generating component. This invention extracts modal features from the acquired acoustic signals, determines the generating component, and sets corresponding acoustic analysis algorithms for different components to perform comprehensive and accurate fault detection. The detection method is simple, efficient, and accurate.

Description

Technical Field

[0001] This invention relates to the field of acoustic monitoring technology, specifically to a method for monitoring equipment operating status, a device for monitoring equipment operating status, a system for monitoring equipment operating status, and a machine-readable storage medium. Background Technology

[0002] The operational status of equipment directly affects the operational safety and economic benefits of an enterprise. Equipment includes both dynamic and static equipment. For static equipment, fault diagnosis primarily involves process flow simulation and online measurement of process parameters. Process flow simulation enables optimized control and operation of chemical processes, but its successful application presupposes that the equipment structure is "normal." When equipment experiences unpredictable mechanical failures due to scaling, corrosion, vibration, overload, or other reasons, or when operating conditions exceed the model's prediction range, it becomes difficult to accurately determine whether the equipment's operational status is normal. Online testing technologies for process parameters such as temperature, pressure, flow rate, and composition during equipment operation can only provide superficial information about the equipment's operational status. Once mechanical or operational faults occur, these conventional detection methods are insufficient to pinpoint the root cause.

[0003] Therefore, existing technologies have proposed using acoustic signals for equipment fault detection. However, since the equipment walls and internal components are interconnected or close to each other, when any component of the wall or internal component generates an acoustic signal, the acoustic sensors installed on the wall or internal component may collect the corresponding acoustic signal. Therefore, existing technologies do not identify and distinguish the components that generate the acoustic signal, but only use the same acoustic analysis method to monitor all acoustic signals. This results in poor applicability, poor monitoring effect, and low identification accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a method, apparatus, and system for monitoring the operating status of equipment, so as to at least solve the problems of low applicability, poor monitoring effect, and low identification accuracy in the prior art, which does not identify and distinguish the components that generate sound signals and only uses the same acoustic analysis method to cover and monitor all sound signals.

[0005] To achieve the above objectives, a first aspect of the present invention provides a method for monitoring the operating status of equipment, the equipment comprising a device wall and internal components, the method comprising:

[0006] Acquire acoustic signals from the device over a period of time.

[0007] Extract the modal features of the acoustic signal, and based on the modal features, determine the generating component of the current acoustic signal, wherein the generating component is the device wall or an internal component;

[0008] Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component.

[0009] Optionally, modal features of the acoustic signal are extracted, and based on the modal features, the generating component of the acoustic signal is determined, including:

[0010] The acoustic signal is processed using synchronous compressed wavelets to obtain its modal characteristics;

[0011] The first preset mode is reconstructed from the modal features to obtain the time-domain signal that first arrives at the preset mode;

[0012] Based on the time-domain signal that arrives first in the preset mode, the arrival time of the acoustic signal within that operating time period is determined.

[0013] The component that generates the acoustic signal is determined based on the arrival time of the acoustic signal.

[0014] Optionally, based on the time-domain signal that first arrives at the preset mode, the arrival time of the acoustic signal within this operating time period is determined, including:

[0015] The signal envelope is obtained by processing the time-domain signal that first arrives at the preset mode using the Hilbert transform.

[0016] Based on the signal envelope, determine the global maximum point;

[0017] The time window is determined based on the global maximum value and the preset time delay;

[0018] Within the time window, the first global minimum point is calculated using the AIC function;

[0019] Based on the first global minimum point, the range of the neighborhood is determined, the midpoint of the neighborhood is the first global minimum point, and the length of the neighborhood is a preset time delay.

[0020] Within the specified range, the second global minimum point is calculated using the AIC function and used as the arrival time of the acoustic signal.

[0021] Optionally, the acoustic signal generating component is determined based on the acoustic signal arrival time, including:

[0022] At least one time difference value is obtained by subtracting the arrival times of the acoustic signals from each acoustic sensor.

[0023] Based on the time difference values, a signal time difference matrix is ​​constructed;

[0024] Based on the aforementioned signal time difference matrix, multiple spatial distances are obtained;

[0025] From the spatial distance, a predetermined number of minimum spatial distances that can form a predetermined shape are determined as the acoustic signal generation area, so as to determine the acoustic signal generation component.

[0026] Optionally, the equipment is petrochemical equipment.

[0027] Optionally, based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including:

[0028] If the component that generates the acoustic signal is determined to be the device wall:

[0029] Based on the acoustic signal, the entropy parameter and acoustic reconstruction data of the acoustic signal are obtained;

[0030] Based on the time domain, frequency domain, and entropy parameter of the acoustic signal, the first theoretical damage state of the device wall is determined;

[0031] Based on the acoustic signal and the acoustic reconstruction data, the second theoretical damage state of the vessel wall is determined;

[0032] Based on the first theoretical damage state and the second theoretical damage state, the damage state of the vessel wall is determined, including the presence of damage and the absence of damage.

[0033] Optionally, based on the time domain, frequency domain, and entropy parameter of the acoustic signal, a first theoretical damage state of the vessel wall is determined, including:

[0034] The time-domain, frequency-domain, and entropy parameters of the acoustic signal are used as inputs to the hypersphere model to obtain the generalized distance between the acoustic signal and the center of the hypersphere model. The hypersphere model is obtained by training SVDD with a training dataset, which includes the time-domain, frequency-domain, and entropy parameters of historical acoustic signals when the vessel wall generates anomalies.

[0035] If the generalized distance is less than or equal to the generalized radius of the hypersphere of the hypersphere model, then the first theoretical damage state of the device wall is determined to be no damage.

[0036] If the generalized distance is greater than the generalized radius of the hypersphere in the hypersphere model, then the first theoretical damage state of the device wall is determined to be that damage exists;

[0037] Based on the acoustic signal and the acoustic reconstruction data, the second theoretical damage state of the vessel wall is determined, including:

[0038] Calculate the mean square error between the acoustic signal and the acoustic reconstruction data;

[0039] If the mean square error is less than or equal to the reconstruction threshold, then the second theoretical damage state of the device wall is determined to be no damage.

[0040] If the mean square error is greater than the reconstruction threshold, then the second theoretical damage state of the device wall is determined to be that damage exists.

[0041] Optionally, based on the first theoretical damage state and the second theoretical damage state, the damage state of the vessel wall is determined, including:

[0042] If both the first theoretical damage state and the second theoretical damage state are present, then the damage state of the vessel wall is determined to be present.

[0043] If both the first theoretical damage state and the second theoretical damage state are non-damaged, then the damage state of the vessel wall is determined to be non-damaged.

[0044] If the first theoretical damage state and the second theoretical damage state are respectively the presence of damage and the absence of damage, then the judgment coefficient is calculated based on the generalized distance and the mean square error.

[0045] If the judgment coefficient is less than the preset threshold, the damage state of the device wall is determined to be no damage.

[0046] If the judgment coefficient is greater than or equal to the preset threshold, then the damage state of the device wall is determined to be present.

[0047] Optionally, the method further includes:

[0048] When the damage condition of the equipment wall is determined to be present:

[0049] The damage stress is determined based on the total acoustic emission count of the acoustic signal;

[0050] If the damage stress is less than or equal to the first preset stress, then the degree of damage to the equipment wall during this operating period is determined to be initial damage.

[0051] If the damage stress is greater than the first preset stress and less than or equal to the second preset stress, then the degree of damage to the equipment wall during this operating period is determined to be low damage.

[0052] If the damage stress is greater than the second preset stress and less than or equal to the third preset stress, then the degree of damage to the equipment wall during this operating period is determined to be medium damage.

[0053] If the damage stress is greater than the third preset stress, then the damage level of the equipment wall during this operating period is determined to be high damage.

[0054] Optionally, the internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle;

[0055] Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including:

[0056] If the component that generates the acoustic signal is determined to be a wing valve:

[0057] The acoustic signal is preprocessed to obtain the processed acoustic signal;

[0058] Based on the processed acoustic signal, the flow rate percentage of each wing valve during the operating time is determined, and the flow rate percentage is determined by the energy value in the processed acoustic signal.

[0059] Based on the flow rate ratio of each wing valve, the operating status of the wing valve during this operating period is determined.

[0060] Optionally, the acoustic signal includes energy parameters;

[0061] The acoustic signal is preprocessed to obtain a processed acoustic signal, including:

[0062] For each wing valve:

[0063] Acoustic signals with energy values ​​greater than a first energy threshold are defined as the first valid signals;

[0064] The first effective signal was corrected using clustering algorithms, the Laida criterion, the quartile method, and cross-validation to obtain the processed acoustic feature parameters.

[0065] Optionally, the internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle;

[0066] Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including:

[0067] If the component that generates the acoustic signal is determined to be a wing valve:

[0068] Based on the acoustic signals, the catalyst leakage risk factor of the wing valve is determined;

[0069] Based on the catalyst runoff risk coefficient, the operating status of the wing valve during this operating period is determined.

[0070] Optionally, based on the acoustic signal, the catalyst leakage risk factor of the wing valve is determined, including:

[0071] The acoustic signal of each wing valve is input into the opening and closing cycle recognition model to obtain the actual opening and closing cycle corresponding to each wing valve.

[0072] Based on the energy parameters and ringing count parameters corresponding to each actual opening and closing cycle of each wing valve, the least squares method is used to perform linear fitting to calculate the comprehensive slope of the energy parameters and ringing count parameters of each wing valve as a function of time in each actual opening and closing cycle.

[0073] Based on the combined slope and the average value of the combined slope of each wing valve, the deflection coefficient corresponding to each wing valve is obtained;

[0074] Based on the flow deflection coefficient corresponding to each wing valve, the catalyst runoff risk coefficient of the wing valve is calculated.

[0075] Optionally, the overall slope can be calculated using the following formula:

[0076] k ij =p·k′ ij +(1-p)·k″ ij ;

[0077] Where, k ij The slope is the composite slope corresponding to the j-th actual opening and closing cycle of the i-th wing valve; p is the weighting coefficient; k′ ij Let k″ be the slope of the energy parameter change over time during the j-th actual opening and closing cycle of the i-th wing valve; ij The slope of the ringing count parameter as a function of time during the j-th opening and closing cycle of the i-th wing valve;

[0078] The deflection coefficient is calculated using the following formula:

[0079]

[0080] Where, δ i k is the deflection coefficient of the i-th wing valve; ij This is the comprehensive slope corresponding to the j-th actual opening and closing cycle of the i-th wing valve; Let be the average of the combined slopes of the i-th wing valve.

[0081] The catalyst runaway risk factor is calculated using the following formula:

[0082]

[0083] Where η is the catalyst runoff risk coefficient of the internal components; δ i The deflection coefficient of the wing valve; The average deflection coefficient for all wing valves,

[0084] Optionally, the internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle;

[0085] Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including:

[0086] If the component that generates the acoustic signal is determined to be a wing valve:

[0087] The acoustic signal of each wing valve is input into the opening and closing cycle recognition model to obtain the actual opening and closing cycle of each wing valve.

[0088] Multiply the catalyst flow rate of the wing valve by the actual opening and closing cycle to obtain the actual load-bearing weight of the wing valve;

[0089] The operating status of the wing valve during this operating period is determined based on the difference between the actual load-bearing weight and the preset load-bearing weight of the wing valve.

[0090] Optionally, the method further includes:

[0091] Acquire the structural parameters of the wing valve, the historical acoustic signals of the wing valve during a certain operating period, and the operating parameters of the wing valve;

[0092] Based on the structural parameters and operating parameters of the wing valve, the theoretical opening and closing cycle of the wing valve is determined;

[0093] Data samples were constructed based on the historical acoustic signals and theoretical opening and closing cycles of the wing valve. The constructed data samples were then used to train the support vector machine algorithm to obtain the opening and closing cycle recognition model.

[0094] Optionally, based on the structural parameters and the operating parameters, the theoretical opening and closing cycle of the wing valve is determined, including:

[0095] The theoretical opening and closing cycle of the wing valve is calculated using the following formula:

[0096]

[0097] Among them, T c ΔP represents the theoretical opening and closing period of the wing valve. c For the pressure drop of the wing valve; R c θ is the radius of the feed leg of the wing valve; c The inclination angle of the slanted pipe section of the wing valve; m cp The equilibrium catalyst amount for the wing valve; A c S is the valve plate mounting angle for the wing valve. cf v is the sealing area of ​​the valve plate of the wing valve; c S is the inlet linear velocity of the wing valve; c ρ is the inlet area of ​​the wing valve; cThe inlet concentration of the wing valve.

[0098] Optionally, if the wing valve is a two-stage wing valve arranged in series, and the inlet linear velocity of the wing valve of the later stage cannot be directly obtained, determining the theoretical opening and closing cycle of the wing valve based on the structural parameters and the operating parameters further includes:

[0099] The theoretical opening and closing cycle of the wing valve is calculated using the following formula:

[0100]

[0101] Among them, T Z1 This represents the theoretical opening and closing cycle of the subsequent stage wing valve; ΔP c For the pressure drop of the subsequent stage wing valve; R c θ is the radius of the feed leg of the subsequent stage wing valve; c The inclination angle of the inclined pipe section of the subsequent stage wing valve; m cp The equilibrium catalyst amount for the subsequent stage wing valve; A c The valve plate mounting angle for the subsequent stage wing valve; S cf v is the valve plate sealing area of ​​the subsequent stage wing valve; c S is the inlet linear velocity of the preceding stage wing valve; c ρ is the inlet area of ​​the subsequent stage wing valve; c The inlet concentration of the next-stage wing valve; η z denoted as cyclone separation efficiency of internal components; Z is the efficiency correction coefficient; d is the average particle size of regenerated catalyst; a1, a2, a3, a4 and a5 are constants; δ1 and δ2 are constants, and δ1 is less than δ2.

[0102] Optionally, the internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle;

[0103] Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including:

[0104] If the acoustic signal generating component is determined to be a double-acting slide valve:

[0105] The acoustic signal is preprocessed to obtain the processed acoustic signal;

[0106] Based on the processed acoustic signal, the crack length of the double-acting slide valve during this operating period is determined;

[0107] Based on the crack length, determine the degree of damage to the double-acting slide valve during this operating period;

[0108] Based on the degree of damage to the double-acting slide valve, the operating status of the double-acting slide valve during this operating period is determined.

[0109] Optionally, the acoustic signal includes: ringing count parameters, energy parameters, and frequency parameters;

[0110] The acoustic signal is preprocessed to obtain a processed acoustic signal, including:

[0111] Acoustic signals with energy values ​​greater than a first energy threshold and peak frequencies greater than a first frequency threshold are identified as the second valid signals.

[0112] The second effective signal is corrected using clustering algorithms, quartile method, Laida criterion and cross-validation to obtain the processed acoustic signal.

[0113] Optionally, the degree of damage to the double-acting spool valve can be determined based on crack length, including:

[0114] If the crack length is less than or equal to the first preset length, then it is determined that the double-acting slide valve is undamaged during this operating period.

[0115] If the crack length is greater than the first preset length and less than or equal to the second preset length, the damage level of the double-acting slide valve during this operating period is determined to be low damage.

[0116] If the crack length is greater than the second preset length and less than or equal to the third preset length, the damage level of the double-acting slide valve during this operating period is determined to be medium damage.

[0117] If the crack length is greater than the third preset length, the damage level of the double-acting slide valve during this operating period is determined to be high damage.

[0118] Optionally, the internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle;

[0119] Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including:

[0120] If the acoustic signal generating component is determined to be a tray or a float valve:

[0121] Based on the acoustic signal, acoustic waveform data and contour plots of the tray or float valve during the operating time of that period are obtained;

[0122] Based on the acoustic waveform data or the contour map, determine whether there are any abnormal points in the tray or float valve during the operating time period.

[0123] If an anomaly is found in the tray or float valve during the operating period, the operating status of the tray or float valve during the operating period is determined based on the contour map. The operating status includes normal operation and abnormal operation.

[0124] Optionally, based on the acoustic signal, a contour map is obtained for the duration of the operation, including:

[0125] The total duration of a single signal is obtained based on the sampling rate and sampling length of the acoustic signal;

[0126] The total duration of a single signal is divided into frames and transformed by time and frequency to obtain the frame frequency, the time point for spectrum analysis, and the energy spectral density.

[0127] Based on the frame frequency, spectral analysis time point, and energy spectral density, the distribution of contour maps of the corresponding areas of the time-frequency intensity cloud map is extracted to obtain the contour maps of the tray or floating valve during the operating time of that period.

[0128] Optionally, based on the acoustic waveform data or the contour map, determine whether there are any abnormal points in the tray or float valve during this operating period, including:

[0129] The acoustic waveform data is used as input to the first anomaly detection model to determine the anomaly detection result, or

[0130] The contour map is used as input to the second anomaly detection model to determine the anomaly detection result;

[0131] The first anomaly detection model is obtained by training the autoencoder using the first training dataset, which includes historical acoustic waveform data of the tray or float valve.

[0132] The second anomaly detection model is obtained by training the autoencoder using the second training dataset, which includes historical contour maps of trays or float valves.

[0133] Optionally, if it is determined that there are abnormal points in the tray or float valve during this operating period, the operating status of the tray or float valve during this operating period is determined based on the contour map, including:

[0134] If the contour map after the anomaly point is not in a steady state, then the operating state of the tray or float valve is determined to be abnormal.

[0135] If the contour map after the outlier is in a steady state, then determine the degree of deviation between the contour map before the outlier and the contour map after the outlier.

[0136] If the deviation is less than or equal to the preset threshold, the operating status of the tray or float valve is determined to be normal operation;

[0137] If the deviation exceeds the preset threshold, the operating status of the tray or float valve is determined to be abnormal.

[0138] Optionally, the internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle;

[0139] Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including:

[0140] If the acoustic signal generating component is determined to be a tube bundle:

[0141] Based on the acoustic signal, the RMS spread entropy and acoustic signature characteristics of the tube bundle during the operating time of that segment are determined.

[0142] Based on the RMS dispersion entropy and the acoustic signature, the operating status of the tube bundle during this operating period is determined.

[0143] Optionally, based on the acoustic signal, the RMS spread entropy of the acoustic signal of the tube bundle during this operating time is determined, including:

[0144] The acoustic signal is processed by scaling to obtain a multi-scale signal;

[0145] Determine the RMS values ​​of multi-scale signals and form a data sequence;

[0146] The RMS scattering entropy value of the data sequence is calculated and used as the RMS scattering entropy of the acoustic signal.

[0147] Optionally, based on the acoustic signal, the acoustic signature characteristics of the tube bundle during this operating time are determined, including:

[0148] Convert the acoustic emission signal into a time-domain signal;

[0149] Based on the time-domain signal, a linear spectrum is obtained using Fourier transform;

[0150] The linear spectrum is converted into a Mel spectrum using a Mel frequency filter bank.

[0151] The logarithmic energy and logarithmic spectrum of the Mel spectrum are taken, and the first derivative is performed to obtain the Mel frequency cepstral coefficients.

[0152] Based on the Mel frequency cepstral coefficients, the voiceprint features are obtained.

[0153] Optionally, based on the RMS scattering entropy and the acoustic signature features, the operating state of the tube bundle during this operating time period is determined, including:

[0154] The RMS scattering entropy and the acoustic signature feature are used as inputs to the leakage identification model to obtain the leakage identification result. The leakage identification model is obtained by training a convolutional neural network using a training dataset. The first training dataset includes the corresponding RMS scattering entropy and acoustic signature feature obtained based on the historical acoustic signal of the tube bundle.

[0155] Based on the leak identification results, the operating status of the tube bundle during that operating period is determined.

[0156] A second aspect of the present invention provides a device for monitoring the operating status of equipment, the equipment comprising a wall and internal components, the device comprising:

[0157] The parameter acquisition module is used to acquire the acoustic signals of the device during a certain operating time.

[0158] The component generation determination module is used to extract the modal features of the acoustic signal and determine the component generating the current acoustic signal based on the modal features. The component generating the signal is a device wall or an internal component.

[0159] The component state determination module is used to call the corresponding acoustic analysis algorithm based on the acoustic signal generating component, and to analyze the acoustic signal based on the called acoustic analysis algorithm to determine the state of the acoustic signal generating component.

[0160] A third aspect of the present invention provides a device operation status monitoring system, the device comprising a device wall and internal components, the system comprising:

[0161] Multiple acoustic sensors are mounted on the device to collect acoustic signals;

[0162] The aforementioned equipment operation status monitoring device is connected to the acoustic sensor.

[0163] Optionally, the acoustic sensor includes:

[0164] The internally hollow cylindrical shell has a signal output terminal and a signal detection terminal at its upper and lower ends, respectively.

[0165] The backing pad is installed inside the housing by contacting and fitting with the inner wall of the housing through the side wall, and the lower end of the backing pad is provided with an installation space;

[0166] A piezoelectric wafer and a mass block are disposed in the installation space, with the mass block located on top of the piezoelectric wafer, and the negative electrode of the piezoelectric wafer is connected to the inner wall of the housing;

[0167] A diaphragm is disposed at the signal detection end of the housing and does not contact the piezoelectric crystal.

[0168] A terminal block is fixed to the signal output end of the housing; the terminal block is connected to the positive electrode of the piezoelectric crystal through a wire passing through the backing pad, and is used to transport the charge generated by the vibration of the piezoelectric crystal outward.

[0169] Optionally, the installation space includes:

[0170] A first installation space and a second installation space, wherein a first step is formed between the first installation space and the second installation space;

[0171] The piezoelectric wafer is disposed in the first mounting space, and the mass block is disposed in the second mounting space, with the mass block in contact with the step surface of the first step portion.

[0172] A fourth aspect of the present invention provides a machine-readable storage medium storing instructions that cause a machine to perform the above-described device operation status monitoring method.

[0173] This technical solution targets acoustic signals collected during equipment operation. First, it extracts the modal features of the acoustic signals. Then, based on the modal features, it identifies the generating components (such as equipment walls or internal components). After identifying the generating components of the acoustic signals, it calls the corresponding acoustic analysis algorithm to perform comprehensive and accurate fault detection and analysis. The detection and analysis method is simple, and the detection efficiency and accuracy are high.

[0174] 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

[0175] 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:

[0176] Figure 1 This is a flowchart of the equipment operation status monitoring method provided by the present invention;

[0177] Figure 2 This is a flowchart of the component for determining the generation of acoustic signals provided by the present invention;

[0178] Figure 3 This is a comparative diagram of the arrival times of acoustic signals calculated using different methods, provided by the present invention.

[0179] Figure 4 This is a flowchart for determining the damage state of the device wall provided by the present invention;

[0180] Figure 5 This is a flowchart of the first method for determining the operating state of a wing valve provided by the present invention;

[0181] Figure 6 This is a flowchart of the second method for determining the operating state of a wing valve provided by the present invention;

[0182] Figure 7 This is a flowchart of the third method for determining the operating state of a wing valve provided by the present invention;

[0183] Figure 8 This is a flowchart for determining the operating state of a double-acting slide valve provided by the present invention;

[0184] Figure 9 This is a flowchart for determining the operating status of a tray or float valve provided by the present invention;

[0185] Figure 10 This is a flowchart for determining the operating state of the tube bundle provided by the present invention;

[0186] Figure 11 This is a schematic diagram of the equipment operation status monitoring device provided by the present invention;

[0187] Figure 12 This is a schematic diagram of the equipment operation status monitoring system provided by the present invention;

[0188] Figure 13 This is a schematic diagram of the acoustic sensor provided by the present invention;

[0189] Figure 14 This is a cross-sectional schematic diagram of the acoustic sensor provided by the present invention;

[0190] Figure 15 This is a cross-sectional schematic diagram of the housing and backing pad provided by the present invention.

[0191] Explanation of reference numerals in the attached figures

[0192] 10 - Parameter Acquisition Module; 20 - Component Determination Module; 30 - Component Status Determination Module;

[0193] 1-Housing; 2-Backing pad; 3-Piezoelectric crystal;

[0194] 4-Mass block; 5-Diaphragm; 6-Terminal;

[0195] 7-Wire; 8-Connecting piece; 11-Second step;

[0196] 12 - Third step; 13 - External thread; 21 - Installation space;

[0197] 81 - Mounting hole; 101 - Signal detection terminal; 102 - Signal output terminal;

[0198] 211 - First installation space; 212 - Second installation space; 213 - First step. Detailed Implementation

[0199] 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 present invention.

[0200] Example 1

[0201] This embodiment provides a method for monitoring the operating status of equipment. The equipment includes equipment walls and internal components, such as... Figure 1 As shown, the method includes:

[0202] S1. Acquire acoustic signals from the device during a certain operating period;

[0203] S2. Extract the modal features of the acoustic signal, and based on the modal features, determine the acoustic signal generating component, wherein the generating component includes internal components and the device wall corresponding to each internal component;

[0204] S3. Based on the acoustic signal generating component, call the corresponding acoustic analysis algorithm to analyze the acoustic signal and determine the state of the acoustic signal generating component.

[0205] Specifically, after determining the state of the acoustic signal generating component, the state of the device wall has a smaller impact on the device's operating state compared to the operating state of the internal components. Therefore, in another embodiment, the device's operating state is determined based on the operating state of the internal components and / or the state of the device wall. When the number of device wall states that are abnormal is less than a first preset threshold, the device's operating state is determined to be normal; when the number of device wall states that are abnormal is greater than or equal to a second preset threshold, the device's operating state is determined to be abnormal, where the second preset threshold is greater than the first preset threshold; when the operating state of any internal component is abnormal, the device's operating state is determined to be abnormal.

[0206] More specifically, in one embodiment, the equipment applicable to this solution is petrochemical equipment, that is, it can be used in the chemical industry. Petrochemical equipment may include components with the same damage and failure mechanism, such as at least one of reaction equipment, concentration and crystallization equipment, filtration and separation equipment, drying equipment, molding equipment, mixing, homogenizing and emulsifying equipment, pulverizing and grinding equipment, feeding and conveying equipment, hydrogen production equipment, flue gas energy recovery equipment, fluid mechanics experimental equipment, heat exchange equipment and refrigeration equipment. Each of the above-mentioned equipment includes at least one internal component; the internal component may include at least one of the following: a wing valve, a double-acting slide valve, a tray, a float valve, a tube bundle, and a component with the same working principle as the wing valve, double-acting slide valve, tray, float valve or tube bundle. For internal components where acoustic sensors can be installed, multiple acoustic sensors are installed on both the equipment wall and the internal component to collect acoustic signals. For internal components where acoustic sensors cannot be installed, multiple acoustic sensors are installed only on the equipment wall, with at least one of them located within a predetermined distance (e.g., within 30cm of the connection point) at the junction of the internal component and the equipment wall. This sensor arrangement ensures that acoustic signals can be collected even for internal components where acoustic sensors cannot be installed, thus accurately identifying the component that generates the acoustic signal.

[0207] More specifically, modal characteristics include at least one of the following: frequency characteristics, amplitude characteristics, time-domain characteristics, and phase characteristics. Frequency characteristics include: center frequency, frequency bandwidth, and frequency resolution. Center frequency is an important parameter describing the frequency distribution of an acoustic signal. It roughly represents the frequency location where the energy of the acoustic signal is concentrated. For example, for a narrowband noise signal, its center frequency can well reflect the main frequency component of the noise. In speech signals, the center frequency of vowels can help distinguish different vowels; for example, the center frequencies of / a / and / i / are significantly different, with / a / having a relatively lower center frequency and / i / having a higher center frequency. Frequency bandwidth refers to the frequency range contained in an acoustic signal. Wideband signals contain richer frequency components and can convey more detailed information. For example, symphonic signals in music have a wide frequency bandwidth, covering a variety of frequency components from low-frequency drumbeats to high-frequency string sounds. Narrowband signals, on the other hand, have a narrow frequency range; for example, simple monotone signals, such as pure-tone signals in pure-tone audiometry, have a narrow frequency bandwidth and typically only have one main frequency component. This refers to the fineness with which an acoustic signal can be distinguished along the frequency axis. High frequency resolution allows for more accurate analysis of the characteristics of different frequency components in an acoustic signal. In spectrum analysis, frequency resolution can be improved by increasing the data length or employing more advanced algorithms (such as high-resolution FFT algorithms). For example, when analyzing complex machine vibration acoustic signals to detect faults, high frequency resolution helps distinguish subtle frequency differences between normal operation and fault conditions, as different fault modes may produce subtle frequency variations within specific frequency ranges.

[0208] Amplitude characteristics include peak amplitude, root mean square (RMS) amplitude, and amplitude dynamic range. Peak amplitude is the maximum amplitude value reached by an audio signal over a period of time. It is crucial for assessing the upper limit of an audio signal's intensity. For example, peak amplitude is a key indicator when evaluating the performance of audio equipment. If the peak amplitude of an audio signal exceeds the processing range of the equipment, it will lead to clipping distortion. In acoustic measurements, such as measuring the intensity of an explosion, peak amplitude can be used to measure the maximum sound pressure level generated at the moment of the explosion. RMS amplitude is a statistical description of the amplitude of an audio signal, reflecting its average power. In audio processing, it reflects the actual loudness of the sound better than peak amplitude. For example, for a periodic audio signal, RMS amplitude can be obtained by averaging the square of the signal amplitude over one period and then taking the square root. In audio broadcasting, RMS amplitude is often used to control the broadcast level of the audio signal to ensure that the loudness heard by the listener is relatively stable. Amplitude dynamic range refers to the difference between the maximum and minimum amplitude of an audio signal. It reflects the degree of variation in the amplitude of the audio signal. In musical performances, symphonic music typically has a wider dynamic range, ranging from very soft (ppp) to very loud (fff), while some electronic music may have a relatively smaller dynamic range. A good audio system should be able to reproduce the dynamic range of the sound signal well, thus providing listeners with a more realistic auditory experience.

[0209] Temporal characteristics include waveform shape, rise time and fall time, and duration. The waveform shape of an acoustic signal can intuitively reflect the characteristics of the sound. For example, a sine wave is the simplest waveform, representing a pure tone of a single frequency. Complex speech or music signals, on the other hand, have waveforms composed of multiple frequency components. Different waveforms, such as square waves and triangle waves, also have their own unique acoustic characteristics. Square waves contain abundant odd harmonics, and their spectral energy distribution differs significantly from that of sine waves; this difference in waveform shape leads to different timbre effects. Rise time refers to the time it takes for an acoustic signal to rise from its initial amplitude to its peak amplitude, while fall time is the time it takes for the peak amplitude to fall to a lower amplitude. In pulsed acoustic signals, rise time and fall time are crucial for describing the characteristics of the pulse. For example, in acoustic imaging technology, the rise time and fall time of short pulsed acoustic signals (such as ultrasonic pulses) affect the imaging resolution. Short rise and fall times can make the pulse sharper, thereby improving imaging accuracy. The duration of an acoustic signal is also important for identifying the type of sound. For example, short-duration sounds might be transient sounds like key presses or clicks, while long-duration sounds might be continuous background music or spoken words. Audio editing software allows you to create different audio effects by trimming the duration of sound signals, such as editing a long piece of background music into a short clip suitable as a ringtone.

[0210] Phase characteristics include initial phase and phase difference. Initial phase is the phase value of the acoustic signal at the start of the signal. When multiple acoustic signals of the same frequency are superimposed, different initial phases will cause changes in the amplitude and waveform shape of the synthesized signal. For example, in audio spatial positioning technology, by adjusting the initial phase of sounds of the same frequency emitted by different speakers, the perceived spatial location of the sound can be altered. Phase difference refers to the phase difference between two or more acoustic signals. In a stereo system, the phase difference between the left and right channel sounds is crucial for creating a sense of spatiality. When the phase difference between the left and right channel sounds is appropriate, the listener can perceive that the sound is coming from a specific direction. Furthermore, in acoustic signal processing, such as beamforming technology, directional reception of sound sources can be achieved by controlling the phase difference between the acoustic signals received by different array elements.

[0211] Example 2

[0212] This embodiment also provides a method for determining the acoustic signal generating component based on modal characteristics, such as... Figure 2 As shown, it specifically includes:

[0213] S21. First, the acoustic signal is processed using synchronous compressed wavelets to obtain the modal characteristics of the acoustic signal, specifically including:

[0214] Synchronous compressed wavelet transform (wavelet decomposition, frequency redistribution based on wall / internal component modal features, and synchronous compression) is used to extract different modal features of the signals received by each sensor. A suitable wavelet basis is selected based on the wall / internal component modal features; instantaneous frequency is calculated, and differentiation and phase analysis are performed: the time derivative of each scale component in the wavelet coefficient matrix is ​​calculated to determine its phase change rate. The instantaneous frequency is calculated using the phase change rate, reflecting the change of the acoustic signal frequency over time. Time-frequency rearrangement (synchronous compression): a time-frequency plane is constructed by converting the time-scale plane to a time-frequency plane for more intuitive observation of the time-frequency characteristics of the acoustic signal. The time-frequency spectrum is rearranged based on the calculated instantaneous frequency, compressing the wavelet coefficients near the instantaneous frequency to concentrate the energy more in that area. This step improves the time-frequency resolution, making the time-frequency representation of the acoustic signal clearer and more accurate. The synchronous compressed wavelet transform result is obtained: the time-frequency representation after synchronous compression is the result of the synchronous compressed wavelet transform. This result has higher resolution in the time-frequency domain and can more accurately reflect the time-frequency characteristics of the acoustic signal.

[0215] S22. Reconstruct the first preset mode from the modal features to obtain the time domain signal that first arrives at the preset mode;

[0216] Data reconstruction involves transforming preset modal components from one geometric form to another, and from one format to another. This includes structural conversion, format conversion, and type replacement (data splicing, data trimming, data compression, etc.) to achieve uniformity in the structure, format, and type of preset modal component data, and to connect and fuse multi-source and heterogeneous data. Specifically, in this embodiment, based on the time-frequency ridge trend of different modal signals, the mode of the acoustic signal is determined from the time-frequency ridge of the high-resolution time-frequency analysis results. Acoustic components of different frequencies are identified, and the variation law of acoustic modes over time is extracted to determine the fastest main characteristic mode (which may be S0, A0, S1, A1, etc., with different preset modes corresponding to the container wall and different internal components), which is the first preset mode to arrive. Furthermore, inverse wavelet transform is used to reconstruct the first preset mode from the modal features to obtain the time-domain signal that arrives first.

[0217] S23. Based on the time-domain signal that arrives first in the preset mode, determine the arrival time of the acoustic signal within the operating time segment;

[0218] Specifically, the arrival time of the acoustic signal for each acoustic sensor is determined using the following method:

[0219] Based on the acoustic signal, the signal envelope is calculated using the Hilbert transform.

[0220] Specifically, in this embodiment, the signal envelope is calculated using the following formulas (3) and (4):

[0221]

[0222] Based on the signal envelope, determine the global maximum point;

[0223] The time window is determined based on the global maximum value and the preset time delay.

[0224] More specifically, the global maximum point is the time point corresponding to the maximum value on the ordinate, and the time window N is set between the starting point t0 and t1. MAX +t AM Between, where t MAX The global maximum point, t AM The set time delay is defined as one cycle of the fastest mode of the signal in the structure, thereby ensuring that at least one preset mode arrives first within the entire time window.

[0225] Within the time window, the first global minimum point is calculated using the AIC function;

[0226] The function definition of AIC is as follows:

[0227] AIC(tw )=t w ·log(var(R w (t w ,1))) +(T w -t w -1)·log(var(R w (1+t w ,T w )))(1)

[0228] The domain range is determined based on the first global minimum point;

[0229] Wherein, the midpoint value of the domain range is the first global minimum point, and the length of the domain range is a preset time delay;

[0230] Within the specified range, the second global minimum point is calculated using the AIC function and used as the arrival time of the acoustic signal.

[0231] Determining the start time of transient fault signals is crucial for improving source localization accuracy, and accurately acquiring the arrival time of acoustic emission signals at the sensor is key. Traditionally, a first-crossing-threshold method is used, and the choice of threshold is very important. A low threshold will cause premature triggering, while a high threshold will reduce localization accuracy. For signals with very small amplitude or high noise, this method will lead to localization errors. The AIC algorithm can accurately determine the arrival time of acoustic emission signals, and experimental verification shows that the AIC function outperforms the Hinkley standard in a range of signal-to-noise ratios.

[0232] Based on the traditional AIC algorithm, a two-step AIC algorithm based on dispersion curves is proposed, which can extract arrival times more accurately. The AIC information criterion is a standard for measuring the goodness of fit of a statistical model. It is based on the concept of entropy and can balance the complexity of the estimated model with its goodness of fit to the data. A time series can be divided into two locally stable periods, corresponding to the non-informative part (noise) and the informative part (signal) of the time series containing the first arrival of the acoustic emission event. Each period can be fitted by an autoregressive model. Assuming that the experimental data time series contains the start time of the acoustic emission event, the sequence is divided into two periods at the moment the acoustic emission wave begins: the pre-arrival time and the post-arrival time. The AIC function returns a minimum value, which occurs at the moment the signal begins. The starting position of the signal can be easily determined through image analysis.

[0233] The time series R is divided into two parts, with w as the dividing point, w∈[1,N], N is the length of the signal, var is the variance function, and T W For the last sample in the time series, t WFor any sample in the time series, R W (t W ,1) indicates that the variance function is calculated from the starting point to the current value t. W R W (1+t W ,T W The variance function is calculated over a period of 1 + t. W To T W All sample values. The variance function var is defined as:

[0234]

[0235] However, when using AIC to select the arrival time, the time of the absolute minimum may differ significantly from the signal arrival time. This can happen, for example, when multiple modes arrive at different times, and a later-arriving mode is stronger than the first-arriving mode. This can severely impact positioning accuracy. Furthermore, the performance of AIC largely depends on the choice of the time window N. At low signal-to-noise ratios, the effect of AIC is also not significant. In addition, different modes propagate at different speeds and exhibit dispersion characteristics. Moreover, different modes have different amplitudes, and large differences in amplitude can cause serious problems in conventional positioning, thus affecting positioning accuracy.

[0236] To accurately identify arrival times, the time window N is configured such that it begins within the noise, starts at the beginning of the original signal, and ends after the signal amplitude reaches its maximum value. After this point, only the final portion of the acoustic emission event, its reflection, and the noise remain, affecting the AIC function result. Arrival time can be indicated by variations in either frequency or amplitude within the time series; the characteristic function should enhance this variation, improving the resolution between noise and acoustic emission signals. The original shape of the signal is characterized by the easily computed and widely used absolute value function CF(i) = |R(i)|, and the signal envelope calculated using the Hilbert transform.

[0237] To verify the applicability of the optimized AIC algorithm and accurately extract the arrival time of the acoustic emission wave, a numerical simulation of ultrasonic guided wave propagation was performed using multiphysics simulation software. A sinusoidal excitation was applied to simulate a single-frequency acoustic emission excitation signal, with its center frequency fc set to 150 kHz. In the numerical simulation, the finite element boundary conditions were free to ensure comparability between the simulation and subsequent experiments. Tetrahedral mesh elements were generated in all finite element models. To ensure computational accuracy and convergence, at least 20 meshes were required for the wavelength of the highest frequency wave. The time step should not exceed 1 / 20 of the highest frequency; since the studied frequency was around 150 kHz, the time step and the maximum element mesh size could be 0.3 μs and 0.8 mm, respectively. Receiver points were designed at distances of 100 mm, 100 mm, and 141 mm from the excitation point in three different directions to comprehensively verify the time extraction capability of the optimized AIC.

[0238] like Figure 3 As shown in the diagram, this invention provides a comparative structural diagram of the arrival time extracted by the optimized AIC (The second AIC) and the traditional AIC (The first AIC) using this technical solution. In the diagram, the horizontal axis represents time, and the vertical axis represents the normalized amplitude. The arrival time is extracted using both the optimized AIC and the traditional AIC, and then compared with the theoretically calculated time. The comparison between the arrival time extracted by the two algorithms and the theoretical arrival time shows that the optimized AIC is superior to the traditional AIC, and the determined arrival time is very close to the theoretical arrival time.

[0239] The mean absolute error of arrival time extraction by the two algorithms can be calculated using formula (6), and the accuracy of arrival time extraction by both algorithms can be evaluated.

[0240]

[0241] Among them, t calc,i t represents the arrival time calculated and extracted by the algorithm. method,i The arrival time is the theoretically calculated value, and N is the number of sensors. Detailed results are shown in Table 1. Compared to the traditional AIC, the mean absolute error of the arrival time using the optimized AIC decreased from 6.88 μs to 2.68 μs, demonstrating that the optimized AIC can improve the accuracy of arrival time extraction and reduce errors. The standard deviation of the arrival time extracted by the two algorithms can be calculated using formula (7) to evaluate their accuracy.

[0242] As shown in Table 1, the standard deviation decreased from 6.91 μs to 2.92 μs, which proves that optimizing AIC can reduce dispersion, make the positioning system more stable, and improve the accuracy of time of arrival extraction.

[0243] Table 1. MAE and Standard Deviation of First Arrival Time

[0244]

[0245] In this scheme, the first step of the algorithm is applicable to most acoustic emission signals, but the accuracy of the first step decreases when the amplitude of the first incident mode is very small compared to that of subsequent incident modes. In such signals, changing the time window setting eliminates this limitation, and the second step improves the estimation of the first step to enhance accuracy and precision.

[0246] S24. Based on the arrival time of the acoustic signals from each acoustic sensor, determine the component that generates the acoustic signals.

[0247] Specifically, the acoustic signal generating components are determined in the following ways:

[0248] At least one time difference value is obtained by subtracting the arrival times of the acoustic signals from each acoustic sensor.

[0249] Based on the time difference value, a signal time difference matrix is ​​constructed.

[0250] Based on the aforementioned signal time difference matrix, multiple spatial distances are obtained;

[0251] Specifically, based on the signal time difference matrix, multiple spatial distances are obtained, including:

[0252] The spatial distance is calculated using the following formula:

[0253]

[0254] Among them, D k For spatial distance; TR(i,j) is the signal time difference matrix; R k (i,j) is the preset time difference matrix.

[0255] From the spatial distance, a predetermined number of minimum spatial distances that can form a predetermined shape are determined as the acoustic signal generation area, so as to determine the acoustic signal generation component.

[0256] Specifically, the preset shape and preset quantity correspond to each other and are both related to the arrangement of the acoustic sensors. If the acoustic sensors are arranged in a rectangular structure on the device wall, the regions corresponding to the four minimum spatial distances that can form a rectangle are determined from the spatial distances and used as the acoustic signal generation regions.

[0257] Specifically, in this embodiment, to locate the acoustic signal generation area, a novel positioning method based on a time difference matrix is ​​proposed. This method is an improvement on the traditional Delta T technique. The traditional Delta T technique uses a threshold crossover method to determine the wave arrival time. However, simple threshold crossover can lead to incorrect positioning because the signal initially falls below the threshold level. The preprocessing of the provided positioning method is the same as that of the traditional Delta T. First, an artificial source needs to be created in advance. By constructing signals of wall cracks and internal component damage through the artificial source, a preset time difference matrix can be obtained.

[0258] A rectangular node grid is constructed on the device wall, with four nodes forming one grid area. Signals are acquired, and after optimization using the AIC algorithm, the arrival time of each node relative to each sensor is obtained to construct the time difference value. When there are N sensors, for each node, there exists... Time difference. This Group time difference constructs a {1×C}N 2 A one-dimensional matrix.

[0259] The one-dimensional time difference matrix is ​​entirely contained in the time difference matrix library. The time difference of each sensor pair is calculated based on the actual acoustic emission event of the fault, resulting in a one-dimensional matrix composed of a set of time difference values. This matrix is ​​compared with k matrices in the time difference matrix library to calculate the spatial distance, obtaining k distinct spatial distances. By iterating through these k values, the nodes corresponding to the four smallest spatial distances that can form a complete rectangle are found, thus determining the fault source region and ultimately identifying the acoustic signal generating component. The smaller the calculated spatial distance, the better the two time difference matrices match; otherwise, the greater the difference between them.

[0260] To verify the effectiveness of the acoustic signal generating component identification method, 50 acoustic signals were triggered on both the vessel wall and the internal components to locate the acoustic signal generating component. These signals were then collected using acoustic sensors on both the vessel wall and the internal components. The actual number of collected signals was 86 on the vessel wall and 93 on the internal components, indicating some signal duplication. However, analysis using the above method confirmed that 50 signals were emitted from both the vessel wall and the internal components. Therefore, the proposed method for locating the acoustic signal generating component can pinpoint the specific component that generates the acoustic signal, thereby enabling fault diagnosis of that specific component.

[0261] Example 3

[0262] In this embodiment, as Figure 4 As shown, after determining that the component generating the acoustic signal is the equipment wall, the equipment wall state recognition algorithm is used as the analysis algorithm for situations such as cracking and impact on the equipment wall, including:

[0263] S31. Based on the acoustic signal, obtain the entropy parameter and acoustic reconstruction data of the acoustic signal; S32. Based on the time domain, frequency domain and entropy parameter of the acoustic signal, determine the first theoretical damage state of the device wall; S33. Based on the acoustic signal and acoustic reconstruction data, determine the second theoretical damage state of the device wall; S34. Based on the first theoretical damage state and the second theoretical damage state, determine the damage state of the device wall, including the presence of damage and the absence of damage.

[0264] Specifically, the entropy parameter includes types such as information entropy, conditional entropy, cross entropy, and relative entropy. Preferably, in this embodiment, the entropy parameter is information entropy, which is first calculated using the following formula:

[0265]

[0266] Among them, H S Information entropy; p i Let be the probability of an acoustic signal occurring with respect to the i-th voltage value of an acoustic emission signal.

[0267] Furthermore, the acoustic signal is used as the input to the data reconstruction model to obtain the acoustic reconstruction data of the acoustic signal. The data reconstruction model is obtained by training the autoencoder based on the training dataset, which includes the time domain, frequency domain, and entropy parameters of the historical acoustic signal when the vessel wall generates anomalies. Specifically, it includes the time domain, frequency domain, and entropy parameters of the historical acoustic signal generated by the vessel wall under abnormal states such as material tension, three-point bending, stress corrosion, flow erosion, hydrogen-induced cracking, and fatigue fracture.

[0268] If the generalized distance is less than or equal to the generalized radius of the hypersphere of the hypersphere model, the first theoretical damage state of the container wall is determined to be no damage; if the generalized distance is greater than the generalized radius of the hypersphere of the hypersphere model, the first theoretical damage state of the container wall is determined to be damage.

[0269] The expression for the hypersphere model is as follows:

[0270]

[0271] Where d is the generalized distance between the acoustic signal and the center of the hypersphere of the hypersphere model; α i and α j σ is the Lagrange multiplier operator; K is the kernel function; σ is a constant; z is the eigenvalue of the test data, x i and x j These are the eigenvalues ​​of the time-domain, frequency-domain, and entropy parameters of the acoustic signal.

[0272] The generalized radius *r* of the hypersphere can be calculated from the distance from any support vector on the hypersphere's descriptive boundary to the hypersphere's center *a*. Therefore, the generalized radius of the hypersphere in the hypersphere model is:

[0273]

[0274] Where, d r α is the generalized radius of the acoustic signal relative to the center of the hypersphere in the hypersphere model; i and α j σ is the Lagrange multiplier operator; K is the kernel function; σ is a constant; x sv For support vectors; x i and x j These are the eigenvalues ​​of the time-domain, frequency-domain, and entropy parameters of the acoustic signal.

[0275] In this embodiment, the first damage state of the equipment wall is determined based on the time domain, frequency domain, and entropy parameters of the acoustic signal. This enables effective identification of the defect propagation signal of the equipment wall under complex environments and strong background noise, realizes multi-dimensional acoustic identification of damage state, improves the accuracy of acoustic signal identification under strong background noise environment, further improves the accuracy of acoustic monitoring, and provides a basis for equipment operation and maintenance planning.

[0276] In addition, to further improve the accuracy of identification and reduce the error caused by using only one method to determine whether there is damage to the vessel wall, a second theoretical damage state of the vessel wall is determined based on the acoustic signal and the acoustic reconstruction data, including:

[0277] Calculate the mean square error between the acoustic signal and the acoustic reconstruction data;

[0278] If the mean square error is less than or equal to the reconstruction threshold, then the second theoretical damage state of the device wall is determined to be no damage.

[0279] If the mean square error is greater than the reconstruction threshold, then the second theoretical damage state of the device wall is determined to be that damage exists.

[0280] Among them, after obtaining two independent damage assessment results using the above two methods, the damage state of the vessel wall is determined based on the first theoretical damage state and the second theoretical damage state, including:

[0281] If both the first theoretical damage state and the second theoretical damage state indicate that damage exists, then the two judgment results are the same, and the damage state of the vessel wall is directly determined to be that damage exists.

[0282] If both the first theoretical damage state and the second theoretical damage state are no damage, it means that the two judgment results are the same, and the damage state of the vessel wall is directly determined to be no damage.

[0283] If the first theoretical damage state and the second theoretical damage state are respectively the presence of damage and the absence of damage, it means that the two judgment results are not consistent and one of them has a judgment error. Then, the judgment coefficient is calculated based on the generalized distance and the mean square error.

[0284] If the judgment coefficient is less than the preset threshold, the damage state of the device wall is determined to be no damage.

[0285] If the judgment coefficient is greater than or equal to the preset threshold, then the damage state of the device wall is determined to be present.

[0286] The judgment coefficient is calculated using the following formula:

[0287]

[0288] Where y is the judgment coefficient; d is the generalized distance; e is the mean squared error; and t is the reconstruction threshold.

[0289] In another embodiment, the method further includes: when the damage state of the equipment wall is determined to be that damage exists:

[0290] The damage stress is determined based on the total acoustic emission count of the acoustic signal; the degree of damage to the device wall is determined based on the damage stress.

[0291] If the damage stress is less than or equal to the first preset stress, then the degree of damage to the equipment wall during this operating period is determined to be initial damage.

[0292] If the damage stress is greater than the first preset stress and less than or equal to the second preset stress, then the degree of damage to the equipment wall during this operating period is determined to be low damage.

[0293] If the damage stress is greater than the second preset stress and less than or equal to the third preset stress, then the degree of damage to the equipment wall during this operating period is determined to be medium damage.

[0294] If the damage stress is greater than the third preset stress, then the damage level of the equipment wall during this operating period is determined to be high damage.

[0295] The damage stress is calculated using the following formula:

[0296]

[0297] Where σ is the damage stress; N is the total acoustic emission count; D and C are constants; η is the fitting coefficient; σ max β is the material strength of the equipment wall; 温度 β is the temperature effect coefficient. 压力 β is the pressure influence coefficient. 流量 This represents the flow rate impact coefficient.

[0298] In this embodiment, when the damage to the equipment wall is classified as initial damage, it indicates the least severe damage, generally considered to be in the crack nucleation stage, at which point the risk is low. When the damage is classified as low damage, it is generally considered to be in the early stage of steady-state propagation, posing a certain degree of risk and requiring continuous monitoring. When the damage is classified as moderate damage, it is generally considered to be in the late stage of steady-state propagation, a stage with higher risk, requiring depressurization and retesting to reduce operational risks. When the damage is classified as medium damage, it is generally considered to be in unstable propagation, a stage with the highest risk, requiring immediate shutdown and maintenance to reduce safety risks. Therefore, based on the degree of equipment wall damage, four alarm levels are established, covering different damage stages and enabling accurate early warning.

[0299] In this embodiment, the acquisition of acoustic signals determines whether damage has occurred to the equipment wall. Once damage is confirmed, the degree of damage is determined based on the acquired data, enabling tiered early warning. This approach further reduces data processing volume and saves computing power, while maintaining high accuracy in damage warnings, thus ensuring safe production. Using the same data for verification, the accuracy rate of directly analyzing the acquired acoustic signals using a conventional neural network model is 83%, while the accuracy rate using the proposed method reaches 98%, demonstrating a significant improvement.

[0300] Example 4

[0301] In this embodiment, if the acoustic signal generating component is a wing valve, such as Figure 5 As shown, it includes:

[0302] First, obtain the acoustic characteristic parameters of each wing valve in each stage of the wing valve during a certain operating time;

[0303] S41. Perform data preprocessing on the acoustic signal to obtain the processed acoustic signal;

[0304] S42. Based on the processed acoustic signal, determine the flow rate percentage of each wing valve during the operating time segment, wherein the flow rate percentage is determined by the energy value in the acoustic signal;

[0305] S43. Based on the flow rate ratio of each wing valve, determine the operating status of the internal components during this operating period.

[0306] Specifically, both the settler and the regenerator are equipped with multi-stage wing valve assemblies for material discharge. Each stage of the wing valve assembly includes multiple wing valves. During the production process of the catalytic cracking unit, the amount of material discharged from the unit is closely related to the opening and closing of the wing valves and the amount of material discharged in each opening and closing cycle. Since the wing valves are located inside the unit, there is no real-time monitoring method for the working status of the wing valves in the existing technology. Therefore, it is impossible to provide accurate guidance for subsequent processes such as material discharge. Therefore, in this embodiment, acoustic signals are acquired for each wing valve during a certain operating period. Sensors for acquiring these acoustic signals can be placed near the wing valve plate on the feed leg of the catalytic cracking unit, and sensors can also be placed at the lifting bolts of the catalytic cracking unit to jointly monitor the catalyst's operating status in the feed leg. The acoustic signals are preprocessed to obtain processed acoustic signals. Preprocessing removes abnormal and useless signals from the acoustic signal acquisition process, ensuring a more accurate acoustic signal. This allows for a more precise determination of the flow ratio of each wing valve in each stage of the catalytic cracking unit during that operating period. By calculating the flow ratio of each wing valve in each stage, the operating status of the catalytic cracking unit during that period is determined. If the catalyst flow rate of a certain wing valve is high, contributing significantly to production, the catalyst's scouring effect on the feed leg and valve plate of that wing valve is excessive, increasing the risk of catalyst runoff. The calculation of the flow ratio between each stage of the wing valves includes the flow ratio between each stage of the settling tank and the flow ratio between each stage of the regenerator. Operating status includes normal operation and abnormal operation. Specifically, for the same stage of wing valves, if the calculated flow ratios between the same stage of wing valves are relatively close, and the difference between the flow ratios of the same stage of wing valves is less than or equal to a preset difference value, then the catalytic cracking unit (settling tank or regenerator) is determined to be operating normally; if the calculated difference between the flow ratios of the same stage of wing valves is greater than or equal to a preset difference value, then the catalytic cracking unit (settling tank or regenerator) is determined to be operating abnormally.

[0307] In this embodiment, the catalyst flows slowly into the feed leg. Due to the obstruction of the wing valve plate, it accumulates inside the feed leg. When the wing valve plate opens, a large amount of catalyst flows out. During the outflow process, friction is generated on the feed leg. In this embodiment, this friction signal is used as the acoustic signal during the operation of each wing valve, rather than the acoustic signal of the valve plate hitting the feed leg during the opening and closing of the wing valve. The acoustic signal of the valve plate hitting the feed leg will be rejected as an abnormal signal to ensure the accuracy of the calculation results.

[0308] Furthermore, the acoustic signal is the energy parameter of the wing valve during a certain operating time.

[0309] Acoustic signals include energy parameters. During the opening and closing of the wing valve, the energy parameters have obvious characteristics and are easy to distinguish, thus enabling accurate monitoring of the catalyst flow process and determining whether the wing valve is open or closed.

[0310] For each wing valve, its energy parameters over a certain operating period can be stored as a separate energy parameter set for easy differentiation and subsequent data processing. The above data preprocessing is performed on the energy parameter set corresponding to each wing valve to ensure data accuracy.

[0311] Because a single reactor in a catalytic cracking unit contains multiple wing valves (e.g., a settling tank with six wing valves in one stage), an excessively short opening and closing cycle (high frequency) of any wing valve indicates excessive catalyst flow at that location. This results in excessive scouring of the feed leg and valve plate by the catalyst, potentially causing catalyst runoff and unplanned shutdowns if the feed leg or valve plate wears through. Therefore, ideally, the flow rates of the six wing valves should be equal or nearly equal. This necessitates obtaining the energy parameters for each wing valve within the reactor and using these parameters to determine the flow rate ratios among the wing valves in each stage, thus establishing a comparative relationship between wing valves within the same stage.

[0312] Further, the acoustic signal is preprocessed to obtain the processed acoustic signal, including: for each wing valve:

[0313] Acoustic signals with energy values ​​greater than a first energy threshold are defined as the first valid signals;

[0314] The first effective signal is corrected using clustering algorithms, the Laida criterion, the quartile method, and cross-validation to obtain the processed acoustic signal.

[0315] By performing preliminary statistical analysis on the collected data, a suitable threshold can be obtained as the first energy threshold, which can filter out the carpet background noise signal in the energy distribution. Because the settling tank generates strong background noise during production and is in a fully enclosed state, coupled with the influence of the waveguide rod on sound wave propagation, it is impossible to obtain the background noise signal and the valve cover opening / closing signal reflecting the flow state of the wing valve. If wavelet denoising is used, the wavelet node coefficients can only be set to the same value, which is equivalent to treating the noise as white noise with a certain energy and performing full-band noise reduction processing. Therefore, by performing statistical analysis on the collected signals, a suitable threshold is defined, which can filter out the carpet background noise signal in the energy distribution, achieving a filtering effect.

[0316] Furthermore, the first effective signal is corrected using clustering algorithms, the Laida criterion, the quartile method, and cross-validation to obtain the processed acoustic signal, including:

[0317] For each wing valve:

[0318] Using the K-means clustering algorithm, the first effective signal of the wing valve is classified to obtain the first type of energy parameters and the second type of energy parameters of the wing valve, wherein the energy value of each energy parameter in the first type of energy parameters is greater than the energy parameter with the largest energy value in the second type of energy parameters;

[0319] All energy parameters in the first category of energy parameters of the wing valve are sorted in ascending order of energy value to form a box plot. The upper limit value is determined in the box plot as the second energy threshold of the wing valve.

[0320] Based on the first valid signal of the wing valve, the third energy threshold of the wing valve is determined using the Laida criterion;

[0321] The energy parameters with energy values ​​greater than the second and third energy thresholds in the first category of energy parameters of the wing valve are removed, and the remaining energy parameters are used as the first acoustic characteristic parameters.

[0322] The original acoustic signal corresponding to the second type of energy parameter of the wing valve is divided into K subsets. Based on K-fold cross-validation, an autoencoder is used for training and testing. The subsets with reconstruction errors greater than a set threshold after testing are deleted to obtain the remaining energy feature parameters, which are used as the second acoustic feature parameters.

[0323] The first acoustic characteristic parameter and the second acoustic characteristic parameter are used as the processed acoustic characteristic parameters of the wing valve.

[0324] Specifically, for the original acoustic signal, K-fold cross-validation is used to sequentially train and test the autoencoder, and mean square error is used to evaluate the reconstruction error. During each test, signals with a reconstruction error e greater than the reconstruction threshold t are removed. Here, the reconstruction threshold t is the maximum reconstruction error of the autoencoder on the training set at each test. After K training and testing cycles, the filtered acoustic signal is obtained, and the corresponding energy feature parameter is recorded as the second acoustic feature parameter.

[0325] For each wing valve, due to the slight differences between wing valves, even wing valves of the same level will have different second and third energy thresholds. Therefore, it is necessary to calculate the corresponding second and third energy thresholds for each wing valve separately.

[0326] Taking a specific wing valve as an example: First, for the 50 first valid signals of this wing valve, the K-means clustering algorithm is used, setting K=2 groups, to automatically cluster the signals into two parts: high energy and low energy. The high energy part corresponds to the first type of energy parameter, and the low energy part corresponds to the second type of energy parameter. Since abnormal signals are generated by sudden events such as impacts and crack propagation, and the energy values ​​are relatively large, abnormal signal detection is only performed on the high-energy clusters formed by k-means clustering. For the high-energy clusters formed by clustering, the quantiles of the high-energy cluster data are used to identify the outliers. All energy values ​​are arranged from smallest to largest and divided into four equal parts (rounded up if less than the minimum), and a box plot is drawn. The upper limit value is determined in the box plot as the second energy threshold. When the energy value in the cluster is greater than the box limit value, the outlier is determined. When the line graph is displayed, such data points can be considered as suspected outliers. At the same time, in order to avoid errors caused by the randomness of the K-means clustering algorithm, the Raida criterion in statistics will be used to check the outlier signals. Based on the first valid signal, the third energy threshold is determined using the Raida criterion. Energy parameters in the first type of energy parameters whose energy values ​​are all greater than the second and third energy thresholds are identified as outliers and removed. The remaining energy parameters in the first type of energy parameters and all energy parameters in the second type of energy parameters are used as the processed acoustic signal as the final valid signal. In order to ensure that the number of valid signals in each channel is the same, the number of signals removed, m, needs to be recorded. The m signals with the largest energy values ​​in the background noise are selected to supplement the valid signal.

[0327] Furthermore, all energy parameters in the first category of energy parameters of the wing valve are sorted in ascending order of energy value to form a box plot. The upper threshold value in the box plot is determined as the second energy threshold of the wing valve, including:

[0328] The energy values ​​of the energy parameters located at the 25th percentile of the box plot are taken as the upper quartile, and the energy values ​​of the energy parameters located at the 75th percentile of the box plot are taken as the lower quartile.

[0329] The upper quartile value is calculated using the following formula as the second energy threshold of the wing valve: E2 = Q3 + 1.5(Q3 - Q1), where Q3 is the upper quartile and Q1 is the lower quartile.

[0330] Furthermore, based on the first effective signal of the wing valve, the third energy threshold of the wing valve is determined using the Raida criterion, including:

[0331] Obtain the mean energy and standard deviation of all energy parameters in the first valid signal of the wing valve;

[0332] The third energy threshold of the wing valve is calculated using the following formula: E3 = μ + 3σ, where E3 is the third energy threshold; μ is the mean of all energy parameters in the first valid signal; and σ is the standard deviation of all energy parameters in the first valid signal.

[0333] Furthermore, based on the processed acoustic signal, the flow rate percentage of each wing valve during this operating period is determined, including:

[0334] The flow rate percentage of each wing valve in each stage of the wing valve can be calculated using the following formula: Among them, Q ij E represents the flow rate percentage of the j-th wing valve in the i-th stage. ij is the sum of the processed acoustic signal energy values ​​of the j-th wing valve in the i-th stage, and n is the total number of wing valves in the i-th stage.

[0335] Because the operating status of the wing valves cannot be observed during production, and it is impossible to correlate effective signals with the opening and closing status of the wing valves in real time, and considering that the flow rate of the wing valves is related to the activity of the wing valve cover, energy parameters are used to describe the activity level of each wing valve's opening and closing over a period of time. However, there is no necessary connection between wing valves at different stages. Therefore, in this embodiment, based on the processed acoustic signal, the flow rate percentage of each wing valve in the same stage is determined during the operating time. By adding the energy values ​​of the effective signals of each wing valve in the same stage, the ratio of the energy value of each wing valve in that stage to the sum of the energy values ​​of all wing valves in that stage is obtained. This ratio is used as the flow rate percentage of each wing valve in the same stage, reflecting the activity level of each wing valve in the same stage under the current state, thereby obtaining the operating status during that operating time.

[0336] In another implementation, if the acoustic signal generating component is the same as the working principle of the wing valve, the above-mentioned scheme can also be used for fault analysis, identification and diagnosis, so as to obtain the operating status of the component with the same working principle as the wing valve.

[0337] Example 5

[0338] In this embodiment, if the acoustic signal generating component is a wing valve, such as Figure 6 As shown, it includes:

[0339] Acquire acoustic signals during the operation of each wing valve;

[0340] S51. Based on the acoustic signal, determine the catalyst runoff risk coefficient of the wing valve;

[0341] S52. Based on the catalyst runoff risk coefficient, determine the operating status of the wing valve during this operating period.

[0342] Specifically, firstly, acoustic signals are acquired during the operation of each wing valve in the settling tank or regenerator of the catalytic cracking unit. Sensors for acquiring these acoustic signals can be placed near the wing valve plate on the feed leg of the catalytic cracking unit, and sensors are also placed at the lifting lug bolts of the catalytic cracking unit to jointly monitor and identify the catalyst's operating status in the feed leg. Analyzing the acquired acoustic signals yields the catalyst runoff risk coefficients for the catalytic cracking unit, including the catalyst runoff risk coefficients for the settling tank and the regenerator. After obtaining the catalyst runoff risk coefficients (for both the settling tank and the regenerator), the current operating status of the catalytic cracking unit (settling tank or regenerator) can be accurately determined. The operating status includes normal operation and abnormal operation. Specifically, if the acquired catalyst runoff risk coefficient is less than or equal to a preset risk threshold, the catalytic cracking unit (settling tank or regenerator) is determined to be operating normally; if the acquired catalyst runoff risk coefficient is greater than the preset risk threshold, the catalytic cracking unit (settling tank or regenerator) is determined to be operating abnormally.

[0343] In this embodiment, the catalyst flows slowly into the feed leg and accumulates inside the feed leg due to the obstruction of the wing valve. When the valve plate of the wing valve opens, a large amount of catalyst flows out and generates friction on the feed leg. In this embodiment, this friction signal is used as the acoustic signal during the operation of each wing valve, rather than the acoustic signal of the valve plate hitting the feed leg during the opening and closing of the wing valve.

[0344] The method further includes: when the catalytic cracking unit (sedimenter or regenerator) is operating normally, there is no need to adjust the process parameters of the unit; when the catalytic cracking unit (sedimenter or regenerator) is not operating normally, the operating parameters of the catalytic cracking unit (sedimenter or regenerator), including working pressure and catalyst flow rate, are adjusted to adjust the opening and closing cycle of the corresponding wing valve, so as to ensure the stable operation of the catalytic cracking unit.

[0345] Specifically, based on the acoustic signals, determining the catalyst leakage risk factor of the wing valve includes:

[0346] The acoustic signal of each wing valve is input into the opening and closing cycle identification model to obtain the actual opening and closing cycle corresponding to each wing valve. The acoustic signal includes: frequency domain characteristic parameters, energy parameters and ringing count parameters.

[0347] Based on the energy parameters and ringing count parameters corresponding to each actual opening and closing cycle of each wing valve, the least squares method is used to perform linear fitting to calculate the comprehensive slope of the energy parameters and ringing count parameters of each wing valve as a function of time in each actual opening and closing cycle.

[0348] Based on the combined slope and the average value of the combined slope of each wing valve, the deflection coefficient corresponding to each wing valve is obtained.

[0349] The flow deviation coefficient can be understood as the catalyst flow deviation during the operation of each wing valve. The opening and closing cycle corresponding to each wing valve is the time stamp of two wing valve opening states.

[0350] Acoustic signals include frequency domain characteristic parameters, energy parameters, and ringing count parameters. During the opening and closing of the wing valve, the energy parameters and ringing count parameters have obvious characteristics and are easy to distinguish, thus enabling accurate identification of the wing valve's operating state, i.e., open and closed. In order to distinguish environmental noise, the frequency domain characteristic parameter - average frequency - is added as the input to the opening and closing cycle identification model, which also has obvious characteristics for the opening and closing of the wing valve.

[0351] Because a single reactor in a catalytic cracking unit contains multiple wing valves (e.g., a settling tank with six wing valves), if the opening and closing cycle of any wing valve in the settling tank is too short (i.e., the opening and closing frequency is too high), it indicates that the catalyst flow rate at that location is too high. This results in excessive scouring of the feed leg and valve plate by the catalyst. If the feed leg or valve plate wears through, it will lead to catalyst runoff and unplanned shutdowns. Therefore, the ideal state is for all six wing valves to have the same opening and closing cycle. Thus, it is necessary to acquire the acoustic signal corresponding to each wing valve in the reactor and input it into the opening and closing cycle recognition model to obtain the opening and closing cycle for each wing valve. Based on the opening and closing cycle of each wing valve, the catalyst runoff risk coefficient of the catalytic cracking unit is assessed. When the overall risk coefficient exceeds a preset risk threshold, an early warning is issued, and the opening and closing cycle of the wing valves is adjusted in a timely manner by adjusting process parameters such as pressure and catalyst flow rate to ensure long-term stable operation of the equipment.

[0352] Furthermore, the method also includes:

[0353] Obtain a training dataset, which includes: frequency domain feature parameters, energy parameters during the operation of the wing valve under different pressures and different catalyst flow rates, and ringing count parameters;

[0354] The training dataset is input into the support vector machine algorithm to train the opening and closing periodicity recognition model.

[0355] Specifically, during the production process of the catalytic cracking unit, acoustic signals during the opening and closing of the wing valve under different operating conditions are first collected, including acoustic signals during the opening and closing of the wing valve under different pressures and catalyst flow rates, and a training dataset is formed. The training dataset specifically includes: frequency domain feature parameters, energy parameters during the operation of the wing valve under different pressures and catalyst flow rates, and ringing count parameters. After obtaining the training dataset, the training dataset is input into a support vector machine algorithm to train the opening and closing cycle recognition model. More specifically, the training dataset can also be input into a neural network model for training to obtain the opening and closing cycle recognition model.

[0356] In another implementation, while acquiring the training dataset, the acquired data is divided into a training dataset, a correction dataset, and a validation dataset in a ratio of 8:1:1. The opening and closing periodicity recognition model is trained using the training dataset. Then, the model parameters are corrected using the correction dataset to improve the model's accuracy. Finally, the model is validated using the validation dataset to obtain the final opening and closing periodicity recognition model.

[0357] In another implementation, the method for training the opening and closing period recognition model further includes:

[0358] Acquire the structural parameters of the wing valve, the historical acoustic signals of the wing valve during a certain operating period, and the operating parameters of the wing valve;

[0359] Based on the structural parameters and operating parameters of the wing valve, the theoretical opening and closing cycle of the wing valve is determined;

[0360] Data samples were constructed based on the historical acoustic signals and theoretical opening and closing cycles of the wing valve. The constructed data samples were then used to train the support vector machine algorithm to obtain the opening and closing cycle recognition model.

[0361] The theoretical opening and closing cycle of the wing valve is calculated using the following formula:

[0362]

[0363] Among them, T c ΔP represents the theoretical opening and closing period of the wing valve. c For the pressure drop of the wing valve; R c θ is the radius of the feed leg of the wing valve; c The inclination angle of the slanted pipe section of the wing valve; m cp The equilibrium catalyst amount for the wing valve; A c S is the valve plate mounting angle for the wing valve. cf v is the sealing area of ​​the valve plate of the wing valve; c S is the inlet linear velocity of the wing valve; c ρ is the inlet area of ​​the wing valve; c The inlet concentration of the wing valve.

[0364] Furthermore, based on the opening and closing cycle of each wing valve, the deflection coefficient corresponding to each wing valve is determined, including:

[0365] Based on the energy parameters and ringing count parameters corresponding to each opening and closing cycle of each wing valve, the least squares method is used to perform linear fitting to calculate the comprehensive slope of the energy parameters and ringing count parameters of each wing valve as a function of time in each opening and closing cycle.

[0366] Based on the combined slope and the average value of the combined slope of each wing valve, the deflection coefficient corresponding to each wing valve is obtained.

[0367] Furthermore, the overall slope is calculated using the following formula:

[0368] k ij =p·k′ ij +(1-p)·k″ ij ;

[0369] Where, k ij The slope is the overall slope corresponding to the j-th opening and closing cycle of the i-th wing valve; p is the weighting coefficient; k′ ij Let k″ be the slope of the energy parameter change with time during the j-th opening and closing cycle of the i-th wing valve; ij Let be the slope of the ringing count parameter as a function of time during the j-th opening and closing cycle of the i-th wing valve.

[0370] Furthermore, after calculating the combined slope for each opening and closing cycle of each wing valve, the deflection coefficient of each wing valve is calculated using the following formula:

[0371]

[0372] Where, δ i k is the deflection coefficient of the i-th wing valve; ij The comprehensive slope corresponding to the j-th opening and closing cycle of the i-th wing valve; Let be the average of the combined slopes of the i-th wing valve.

[0373] Furthermore, the catalyst runoff risk coefficient is calculated using the following formula: Where η is the catalyst runoff risk coefficient of the catalytic cracking unit; δ i Let be the deflection coefficient of the i-th wing valve; The average deflection coefficient for all wing valves,

[0374] In this scheme, the actual opening and closing cycle of the wing valve is first obtained through the model. Then, the slope of the energy parameter and ringing count parameter of the wing valve changing with time in each actual opening and closing cycle is calculated. Finally, the two slopes are fused to calculate the deflection coefficient. Using this method, the calculation results are more accurate and can intuitively reflect the state of the entire cycle.

[0375] In another implementation, if the acoustic signal generating component is the same as the working principle of the wing valve, the above-mentioned scheme can also be used for fault analysis, identification and diagnosis, so as to obtain the operating status of the component with the same working principle as the wing valve.

[0376] Example 6

[0377] In this embodiment, after determining that the acoustic signal generating component is a wing valve, as follows: Figure 7 As shown, the acoustic analysis algorithm includes:

[0378] S61. Input the acoustic signal of each wing valve into the opening and closing cycle recognition model to obtain the actual opening and closing cycle of each wing valve.

[0379] S62. Multiply the catalyst flow rate of the wing valve by the actual opening and closing cycle to obtain the actual load-bearing weight of the wing valve.

[0380] S63. Based on the difference between the actual load-bearing weight and the preset load-bearing weight of the wing valve, determine the operating status of the wing valve during this operating period.

[0381] Among them, structural parameters, acoustic signals and operating parameters of the wing valve during a certain operating time are obtained;

[0382] Specifically, in this embodiment, the structural parameters include: the material leg radius of the wing valve, the inclination angle of the inclined pipe section of the wing valve, the valve plate installation angle of the wing valve, the valve plate seal of the wing valve, and the inlet area of ​​the wing valve.

[0383] Specifically, the acoustic signals include: frequency domain characteristic parameters, energy parameters, and ringing count parameters; the operating parameters during the operation of the internal components include: pressure drop of the wing valve, equilibrium catalyst quantity of the wing valve, inlet linear velocity of the wing valve, and inlet concentration of the wing valve.

[0384] In addition, due to significant noise interference at the acoustic signal acquisition site, preprocessing of the acquired acoustic signals is necessary to improve their accuracy. This preprocessing includes setting a noise removal threshold to filter out carpet background noise signals from the energy distribution of the acoustic signals. More specifically, all signals are sorted in descending order of energy value, and the top 50 signals are selected as valid signals based on the noise removal threshold, while background noise signals with energy below this threshold are filtered out.

[0385] The theoretical opening and closing cycle of the wing valve of the internal component is determined based on the structural parameters and the operating parameters.

[0386] In this embodiment, if the internal component is provided with only one-stage wing valve or with multiple stages of wing valves, each stage of wing valve is independent of each other, wherein each stage of wing valve contains multiple independently configured wing valves, then the theoretical opening and closing cycle of the wing valves of the internal component is calculated using the following formula:

[0387]

[0388] Among them, T c ΔP represents the theoretical opening and closing period of the wing valve. c For the pressure drop of the wing valve; R c θ is the radius of the feed leg of the wing valve; c The inclination angle of the slanted pipe section of the wing valve; m cp The equilibrium catalyst amount for the wing valve; A c S is the valve plate mounting angle for the wing valve. cf v is the sealing area of ​​the valve plate of the wing valve; c S is the inlet linear velocity of the wing valve; c ρ is the inlet area of ​​the wing valve; c The inlet concentration of the wing valve.

[0389] If the internal component is equipped with only two-stage wing valves, and the two-stage wing valves are arranged in series, and the inlet linear velocity of the wing valve of the subsequent wing valve cannot be directly obtained due to the obstruction of the preceding wing valve, wherein each stage of the wing valve contains multiple independently arranged wing valves, then the opening and closing cycle of the wing valve of the internal component is calculated using the following formula:

[0390]

[0391] Among them, T Z1 This represents the theoretical opening and closing cycle of the subsequent stage wing valve; ΔP c For the pressure drop of the subsequent stage wing valve; R c θ is the radius of the feed leg of the subsequent stage wing valve; c The inclination angle of the inclined pipe section of the subsequent stage wing valve; m cp The equilibrium catalyst amount for the subsequent stage wing valve; A c The valve plate mounting angle for the subsequent stage wing valve; S cf v is the valve plate sealing area of ​​the subsequent stage wing valve; c S is the inlet linear velocity of the preceding stage wing valve; c ρ is the inlet area of ​​the subsequent stage wing valve; c The inlet concentration of the next-stage wing valve; η zZ is the cyclone separation efficiency of the internal components; d is the efficiency correction coefficient; a1, a2, a3, a4 and a5 are constants, and their order from smallest to largest is: a1, a3, a2, a4 and a5; δ1 and δ2 are constants, and δ1 is less than δ2.

[0392] The theoretical opening and closing cycle of the wing valve calculated using the above formula is more accurate, which can further improve the data accuracy in the subsequent model training process, thereby improving the output accuracy of the model.

[0393] Data samples are constructed based on the acoustic signals and the opening and closing cycle of the wing valve. An opening and closing cycle recognition model is trained using the data samples. The opening and closing cycle recognition model is used to obtain the actual opening and closing cycle of the corresponding wing valve based on the acoustic signals of the input internal components during a certain operating time.

[0394] Specifically, data samples constructed using acoustic signals and the opening and closing cycle of the wing valve are used as input to a support vector machine algorithm for model training, resulting in the opening and closing cycle recognition model. After training, the acoustic signals of the internal component during a specific operating period can be directly obtained as input to the model, yielding an accurate real-time opening and closing cycle of the wing valve within that operating period. The opening and closing cycle recognition model in Example 4 can also be obtained using the training method described in this embodiment.

[0395] In another implementation, in addition to using the support vector machine algorithm for model training, convolutional neural networks or similar methods can also be used for model training.

[0396] Based on the catalyst flow rate of the wing valve and the wing valve opening and closing cycle of the internal component obtained through the opening and closing cycle identification model, the operating status of the internal component during this operating period is determined.

[0397] Specifically, in this embodiment, the operating status of the internal component during this operating time is determined through the following steps:

[0398] The acoustic signal of each wing valve is input into the opening and closing cycle recognition model to obtain the opening and closing cycle corresponding to each wing valve.

[0399] Based on the catalyst flow rate of the wing valve and the wing valve opening and closing cycle of the internal components obtained through the opening and closing cycle identification model, the actual load-bearing weight of the wing valve is determined.

[0400] Specifically, the actual load-bearing weight of the wing valve is calculated using the following formula: m 实际 =q m ×T, where m 实际 q represents the actual load-bearing weight of the wing valve. mdenoted as , where is the catalyst flow rate of the wing valve; T is the wing valve opening and closing cycle of the internal component obtained through the opening and closing cycle identification model.

[0401] Based on the difference between the actual load-bearing weight and the preset load-bearing weight of the wing valve, the operating status of the internal components during this operating period is determined.

[0402] Specifically, the difference in the preset load-bearing weight is the initial weight of the material accumulating in the wing valve, which is the design value corresponding to the internal components. Under this setting, the opening and closing cycle of the wing valve is a certain value. However, as the wing valve works, due to the wear between the wing valve and the material, the actual load-bearing weight of the wing valve eventually changes, and the opening and closing cycle of the wing valve also changes accordingly.

[0403] Therefore, in this embodiment, if the difference between the actual load-bearing weight of the wing valve and the preset load-bearing weight is less than the preset value, it indicates that the wing valve is in a stable operating state with minimal wear; if the difference between the actual load-bearing weight of the wing valve and the preset load-bearing weight is greater than or equal to the preset value, it indicates that the wing valve is in an unstable operating state with significant wear. Furthermore, if the number of wing valves in an unstable operating state within the same stage of the internal component is less than a preset number, it is determined that the internal component (settler or regenerator) is in a normal operating state; if the number of wing valves in an unstable operating state within the same stage of the internal component is greater than or equal to the preset number, it is determined that the internal component (settler or regenerator) is in an abnormal operating state.

[0404] In another implementation, if the acoustic signal generating component is the same as the working principle of the wing valve, the above-mentioned scheme can also be used for fault analysis, identification and diagnosis, so as to obtain the operating status of the component with the same working principle as the wing valve.

[0405] Example 7

[0406] In this embodiment, if the acoustic signal generating component is a double-acting slide valve, such as Figure 8 As shown, the method includes:

[0407] Acquire the acoustic signal of the double-acting slide valve during a certain operating time;

[0408] S71. Perform data preprocessing on the acoustic signal to obtain the processed acoustic signal;

[0409] S72. Based on the processed acoustic signal, determine the crack length of the double-acting slide valve during the operating time of this segment;

[0410] S73. Based on the crack length, determine the degree of damage to the double-acting slide valve during this operating period;

[0411] S74. Based on the degree of damage to the double-acting slide valve, determine the operating status of the double-acting slide valve during this operating period.

[0412] Specifically, a double-acting slide valve is installed in the regenerator of the catalytic cracking unit. This valve, through an actuator, adjusts the valve opening to regulate the flue gas outlet flow, thereby controlling the regeneration pressure. The valve contains two valve plates that move in opposite directions, controlling the regenerator pressure by adjusting the opening degree, maintaining a stable pressure difference between the regenerator and the reactor. Its normal operation directly affects the safety and stability of the entire unit. During operation, the double-acting slide valve is prone to guide rail bolt breakage, leading to guide rail detachment or tilting, valve plate disengagement, and valve stem bending deformation, causing unplanned shutdowns and seriously affecting safe production. Since the regenerator is located inside, current technology lacks real-time monitoring methods for its operating status, thus failing to provide accurate guidance for subsequent processes such as feed rate.

[0413] Furthermore, the acoustic signals are the ringing count parameters, energy parameters, and frequency parameters of the double-acting slide valve during a certain operating time.

[0414] Specifically, in this embodiment, the damage and cracking signal of bolt fracture is found to be a high-frequency, high-energy burst signal with a peak frequency of 150-200kHz; the guide rail friction signal is a low-frequency, low-energy continuous signal with a peak frequency of ≤75kHz; the background noise signal is a low-frequency, low-energy signal with a certain regularity, which is submerged in the acoustic signal monitored during the normal operation of the slide valve; there may also be abnormal signals mixed in with the damage and cracking signal. Therefore, by including the acoustic signal into energy parameters and frequency parameters, and considering the obvious characteristics of energy parameters and frequency parameters during the operation of the double-acting slide valve, it is easy to distinguish the damage and cracking of the double-acting slide valve.

[0415] Further, the acoustic signal is preprocessed to obtain a processed acoustic signal, including:

[0416] Acoustic signals with energy values ​​greater than a first energy threshold and peak frequencies greater than a first frequency threshold are identified as the second valid signals.

[0417] The second effective signal is corrected using clustering algorithms, quartile method, Laida criterion and cross-validation to obtain the processed acoustic signal.

[0418] Specifically, by performing preliminary statistical analysis on the collected data, a suitable energy threshold can be obtained as the first energy threshold, and a suitable frequency threshold can be obtained as the first frequency threshold. The combination of the first energy threshold and the first frequency threshold can filter out the carpet background noise signal in the energy distribution. Because the settling tank generates strong background noise during production and is in a fully enclosed state, coupled with the influence of the waveguide rod on sound wave propagation, it is impossible to obtain the background noise signal and the valve cover opening / closing signal reflecting the flow state of the double-acting slide valve. If wavelet denoising is used, the wavelet node coefficients can only be set to the same value, which is equivalent to treating the noise as white noise with a certain energy and performing full-band noise reduction processing. Therefore, by performing statistical analysis on the collected signals, a suitable threshold is defined, which can filter out the carpet background noise signal in the energy distribution, achieving a filtering effect.

[0419] In this embodiment, for a certain dual-acting slide valve, the energy values ​​of all acquired acoustic signals are sorted in descending order, and the peak frequencies of all signals are sorted in descending order. There are 50 acoustic signals that satisfy both the energy value and the peak frequency being greater than a first energy threshold and a first frequency threshold. This allows background noise signals with energy values ​​lower than the first energy threshold to be filtered out, and the 50 signals are determined as the first valid signals. The first energy threshold and the first frequency threshold can adaptively separate the valid signals from the carpet-like background noise at the bottom under different environments, different acquisition parameters, and different sampling times.

[0420] Simultaneously, anomalous acoustic emission signals were removed using statistical and clustering methods. The energy value was 3.4838 × 10⁻⁶. -4 The signal is an abnormal state signal, which can be regarded as being generated due to some unexpected impact during the production process. As a singular value, it is judged as an abnormal valid signal. In order to remove the influence of the singular value, the C-means clustering algorithm combined with the Laida criterion is used to remove it. The second valid signal is then corrected to obtain the processed acoustic signal.

[0421] Furthermore, using clustering algorithms, quartile methods, the Laida criterion, and cross-validation, the first effective parameter is corrected to obtain the processed acoustic feature parameters, including:

[0422] The energy parameters of the first effective parameter are corrected using C-means clustering, quartile method, Laida criterion and K-fold cross-validation to obtain the second effective parameter;

[0423] The frequency parameters of the first effective parameter were corrected using the C-means clustering algorithm, quartile method, Laida criterion and K-fold cross-validation to obtain the third effective parameter;

[0424] The second effective parameter and the third effective parameter are used as the processed acoustic signal.

[0425] Specifically, since the acoustic signal in this embodiment is the energy parameter and frequency parameter of the double-acting slide valve during a certain operating time, the data of the energy parameter and frequency parameter are corrected respectively to improve the accuracy of the data.

[0426] Furthermore, the energy parameters of the first effective parameter are corrected using C-means clustering, quartile method, Laida criterion, and K-fold cross-validation to obtain the second effective parameter, including:

[0427] Using the C-means clustering algorithm, the energy parameters of the first effective parameter are classified to obtain a first class of energy parameters and a second class of energy parameters, wherein the energy value of each energy parameter in the first class of energy parameters is greater than the energy parameter with the largest energy value in the second class of energy parameters;

[0428] Based on the first type of energy parameters, the second energy threshold is determined using the quartile method;

[0429] Based on the energy parameters of the first effective parameter, the third energy threshold is determined using the Laida criterion;

[0430] Energy parameters with energy values ​​greater than the second and third energy thresholds in the first category of energy parameters are removed, and the remaining energy parameters are used as the first acoustic feature parameters.

[0431] The original acoustic signal corresponding to the second type of energy parameter is divided into K subsets. Based on K-fold cross-validation, an autoencoder is used for training and testing. The subsets with reconstruction errors greater than a set threshold after testing are deleted to obtain the remaining energy parameters, which are used as the second acoustic feature parameters.

[0432] The first acoustic feature parameter and the second acoustic feature parameter are used as the second effective parameter;

[0433] Specifically, taking a double-acting spool valve as an example: First, for the 50 first valid signals of the double-acting spool valve, the C-means clustering algorithm is used, set into two groups, automatically clustering the signals into high-energy and low-energy parts. The high-energy part corresponds to the first type of energy parameter, and the low-energy part corresponds to the second type of energy parameter. Since abnormal signals are generated by sudden events such as impact and crack propagation, and have large energy values, abnormal signal detection is only performed on the high-energy clusters formed by C-means clustering. For the high-energy clusters formed by clustering, the quantiles of the high-energy cluster data are used to identify the outliers. All energy values ​​are arranged from smallest to largest and divided into four equal parts (rounded up if less than the minimum), and a box plot is drawn. The upper threshold value is determined in the box plot as the second energy threshold. When the energy value in the cluster is greater than the upper threshold value of the box plot, such data points can be considered as suspected outliers. At the same time, for To avoid errors caused by the randomness of the C-means clustering algorithm, the Laida criterion from statistics will be used to verify abnormal signals. Based on the first valid signal, a third energy threshold will be determined using the Laida criterion. Energy parameters in the first class of energy parameters whose energy values ​​are all greater than the second and third energy thresholds will be identified as outliers and removed. The remaining energy parameters in the first class of energy parameters will be used as the first acoustic feature parameters. In addition, to reduce the error caused by the C-means clustering algorithm, the original acoustic signals corresponding to the second class of energy parameters will be divided into K subsets. Based on K-fold cross-validation, an autoencoder will be used for training and testing. Subsets with reconstruction errors greater than a set threshold after testing will be deleted, and the remaining energy parameters will be used as the second acoustic feature parameters. Finally, the first and second acoustic feature parameters will be used as the second valid parameters. During the processing, to ensure that the number of valid signals in each channel is the same, the number of signals m removed needs to be recorded, and the m signals with the largest energy values ​​in the background noise will be selected to supplement the valid signals.

[0434] Furthermore, all energy parameters in the first category of energy parameters are sorted in ascending order of energy value to form a box plot. The energy values ​​of the energy parameters at the 25th percentile of the box plot are taken as the upper quartiles, and the energy values ​​of the energy parameters at the 75th percentile of the box plot are taken as the lower quartiles.

[0435] The upper quartile value is calculated using the following formula as the second energy threshold: E2 = Q′3 + 1.5(Q′3 - Q′1), where Q′3 is the upper quartile and Q′1 is the lower quartile.

[0436] Based on the first valid signal, the third energy threshold is determined using the Laida criterion, including:

[0437] Obtain the mean energy and standard deviation of all energy parameters in the first valid signal;

[0438] The third energy threshold is calculated using the following formula: E3 = μ1 + 3σ1, where E3 is the third energy threshold; μ1 is the mean energy of all energy parameters in the first valid signal; and σ1 is the standard deviation of the energy of all energy parameters in the first valid signal.

[0439] Furthermore, the first effective signal is classified using the C-means clustering algorithm to obtain a first class of frequency parameters and a second class of frequency parameters, wherein the peak frequency of each frequency parameter in the first class of frequency parameters is greater than the energy parameter with the largest peak frequency in the second class of frequency parameters.

[0440] Based on the first type of frequency parameters, the second frequency threshold is determined using the quartile method;

[0441] Based on the first valid signal, the third frequency threshold is determined using the Raida criterion;

[0442] The frequency parameters in the first type of frequency parameters that are greater than the second frequency threshold and greater than the third frequency threshold are taken as the second valid signal.

[0443] Furthermore, based on the first type of frequency parameters, the second frequency threshold is determined using the quartile method, including:

[0444] All frequency parameters in the first category of frequency parameters are sorted in ascending order of peak frequency value to form a box plot. The peak frequency value of the frequency parameter located at 25% of the box plot is taken as the upper quartile, and the peak frequency value of the frequency parameter located at 75% of the box plot is taken as the lower quartile.

[0445] The upper quartile value is calculated using the following formula as the second frequency threshold: f2 = Q″3 + 1.5(Q″3 - Q″1), where Q″3 is the upper quartile and Q″1 is the lower quartile.

[0446] Based on the first valid signal, the third energy threshold is determined using the Laida criterion, including:

[0447] Obtain the mean peak frequency and standard deviation of peak frequency for all frequency parameters in the first valid signal;

[0448] The third frequency threshold is calculated using the following formula: f3 = μ2 + 3σ2, where f3 is the third energy threshold; μ2 is the mean peak frequency of all frequency parameters in the first effective signal; and σ2 is the standard deviation of the peak frequency of all frequency parameters in the first effective signal.

[0449] Specifically, the processing steps for the frequency parameter are the same as those for the energy parameter, ultimately yielding the third effective parameter, which will not be elaborated here.

[0450] In addition, after obtaining the processed acoustic signal through the above scheme, the crack length of the double-acting slide valve during the operating time is determined based on the sum of the ringing count parameters in the processed acoustic signal.

[0451] The crack length of the double-acting spool valve is calculated using the following formula:

[0452]

[0453] Where a is the crack length; p is the fitting constant; C is the proportional constant; and N is the sum of the ringing count parameters in the processed acoustic characteristic parameters.

[0454] Furthermore, based on the crack length, the degree of damage to the double-acting spool valve is determined, including:

[0455] If the crack length is less than or equal to the first preset length, then it is determined that the double-acting slide valve is undamaged during this operating period.

[0456] If the crack length is greater than the first preset length and less than or equal to the second preset length, the damage level of the double-acting slide valve during this operating period is determined to be low damage.

[0457] If the crack length is greater than the second preset length and less than or equal to the third preset length, the damage level of the double-acting slide valve during this operating period is determined to be medium damage.

[0458] If the crack length is greater than the third preset length, the damage level of the double-acting slide valve during this operating period is determined to be high damage.

[0459] The first preset length is less than the second preset length, and the second preset length is less than the third preset length.

[0460] Specifically, if the damage level of the double-acting slide valve is determined to be low, observation operation is adopted. If the number of processed acoustic signals increases during the observation operation, the operation speed needs to be reduced; if an increasing trend is observed, the operation should be stopped immediately for maintenance. If the damage level of the double-acting slide valve is determined to be medium, the operation speed needs to be reduced, and maintenance should be carried out at an opportune time. If the number of processed acoustic signals increases during this process, the operation should be stopped immediately for maintenance if an increasing trend is observed. If the damage level of the double-acting slide valve is determined to be high, the operation should be stopped immediately for maintenance, and a comprehensive overhaul of the regenerator should be carried out. To verify the effectiveness of the double-acting slide valve damage monitoring, 1200 sets of acoustic signals were collected. This scheme and conventional recognition models were used to identify bolt fracture (400 sets), guide rail friction (400 sets), and noise signals (400 sets). Compared with conventional recognition models, such as the CNN model with an accuracy of 86.50% and the empirical database matching model with an accuracy of 74.58%, the accuracy of the model in this scheme is 97.08%.

[0461] In another implementation, if the acoustic signal generating component is the same as the double-acting slide valve in operation, the above-mentioned scheme can also be used for fault analysis, identification and diagnosis, so as to obtain the operating status of the component with the same operating principle as the double-acting slide valve.

[0462] Example 8

[0463] In this embodiment, the acoustic signal generating component is a tray or a float valve, such as... Figure 9 As shown, it includes:

[0464] Taking a tower as an example: Acoustic signals of the tower during a certain operating time are acquired. In this embodiment, the acoustic signals are collected by an acoustic sensor.

[0465] S81. Based on the acoustic signal, obtain the acoustic waveform data and contour map of the tower during the operating time of this segment;

[0466] Specifically, acoustic waveform data is a graphical representation of the amplitude of sound. It is obtained by superimposing simple sine waves of different amplitudes and phases at various frequencies. The horizontal axis of the waveform graph is time, and the vertical axis is amplitude, which represents the change of the total amplitude of the superimposed sine waves of all frequencies over time.

[0467] Furthermore, based on the acoustic signal, a contour map is obtained for this operating time period, including:

[0468] The total duration of a single signal is obtained based on the sampling rate and sampling length of the acoustic signal;

[0469] The total duration of a single signal is divided into frames and transformed by time and frequency to obtain the frame frequency, the time point for spectrum analysis, and the energy spectral density.

[0470] Based on the frame frequency, the time point of the spectrum analysis, and the energy spectral density, the distribution of contour maps of the corresponding areas of the time-frequency intensity cloud map is extracted to obtain the contour map of the tower during the operating time of that period.

[0471] Specifically, the acoustic signature distribution feature extraction is performed by dividing and calculating the total duration of a single signal based on the acoustic signal sampling rate and sampling length. Statistical analysis of the signal is used to determine the sampling locations of stationary signals, followed by frame segmentation. Through time-frequency transformation, the frame frequency, spectral analysis time points, and energy spectral density are calculated. Then, the calculation results are used to extract the distribution of contour maps of the corresponding areas in the time-frequency intensity cloud map, ultimately obtaining the contour map of the tower during that operating period.

[0472] S82. Based on the acoustic waveform data or the contour map, determine whether there are any abnormal points in the tower during the operating period, including: the start time of the abnormal point; and determine the cause of the abnormal point. The cause of the abnormal point in the tower during the operating period may be the tower tray flipping and the float valve falling off, thereby achieving more detailed monitoring of other operating states.

[0473] Furthermore, based on the acoustic waveform data or the contour map, determine whether there are any anomalies during this operating time period, including:

[0474] The acoustic waveform data is used as input to the first anomaly detection model to determine the anomaly detection result, or

[0475] The contour map is used as input to the second anomaly detection model to determine the anomaly detection result.

[0476] Specifically, by using a pre-trained first anomaly detection model to determine the anomaly detection result, or by using a pre-trained second anomaly detection model to determine the anomaly detection result, the accuracy of the detection result can be guaranteed.

[0477] Furthermore, the method also includes:

[0478] Obtain a first training dataset, which includes historical acoustic waveform data;

[0479] Based on the first training dataset and the autoencoder, a first anomaly detection model is trained.

[0480] Obtain a second training dataset, which includes historical contour maps;

[0481] Based on the second training dataset and the autoencoder, a second anomaly detection model is trained.

[0482] Specifically, the first training dataset contains historical acoustic waveform data obtained based on historical acoustic signals. These historical acoustic signals include acoustic signals generated by airflow friction before and after tray flipping and float valve detachment within the tower. Using this first training dataset and an autoencoder, a first anomaly detection model is trained. This model can not only identify whether anomalies exist during tower operation but also determine the cause of the anomalies, i.e., anomalies caused by tray flipping or float valve detachment. The second training dataset contains historical contour maps obtained based on historical acoustic signals. These historical acoustic signals include acoustic signals generated by airflow friction before and after tray flipping and float valve detachment within the tower. Using this second training dataset and an autoencoder, a second anomaly detection model is trained. This model can not only identify whether anomalies exist during tower operation but also determine the cause of the anomalies, i.e., anomalies caused by tray flipping or float valve detachment. The steps for obtaining historical contour maps from historical acoustic signals are the same as described above and will not be repeated here. Sliding window sampling is performed on the data obtained from the first and second training datasets. The window size is S and the overlap between windows is A. The dataset is divided into training and test sets, with the proportion of the training set being 50%-80%.

[0483] This autoencoder learns the encoding format of "normal data," so when a dataset is provided to it, it encodes and decodes it according to this format. If the error between the decoded dataset and the input dataset is within a certain range, the input dataset is considered "normal"; otherwise, it is considered "abnormal." After determining a threshold, it can be assumed that if the error after decoding a new dataset exceeds that threshold, it is considered abnormal data.

[0484] S83. If it is determined that there are abnormal points in the tower during the operating period, the operating status of the tower during the operating period is determined based on the contour map. The operating status includes normal operation and abnormal operation.

[0485] In this embodiment, after the sudden abnormal event ends, the airflow state inside the tower will change, causing a change in the airflow acoustic signal. That is, the sudden abnormal event leads to a change in the operating state (abnormal response) inside the tower, but the change in the airflow acoustic signal is weak, and the subtle abnormal change is difficult to identify. To address this problem, a soundprint distribution feature extraction method is used to enhance the signal features and improve the accuracy.

[0486] Furthermore, if it is determined that there are abnormal points in the tower crane during this operating period, the operating status of the tower crane during this operating period is determined based on the contour map, including:

[0487] If the contour map after the anomaly point is not in a steady state, then the tower's operating status is determined to be abnormal.

[0488] Specifically, in this embodiment, after the sudden abnormal event ends, the airflow state inside the tower will change, causing a change in the airflow acoustic signal. That is, the sudden abnormal event leads to a change in the tower's operating state (abnormal response), but the change in the airflow acoustic signal is weak, and the subtle abnormal change is difficult to identify. To address this problem, a sound signature distribution feature extraction method is used to enhance the signal features and improve accuracy. Specifically, the steady state after the abnormal point is determined as follows: the differences between several consecutive signals located at the abnormal point are judged. If the difference between several consecutive signals is less than a preset difference, it proves that the tower has reached a new steady state after the abnormal point. If there is a difference between signals that is not less than the preset difference, it proves that the tower has not reached a steady state after the abnormal point, thus confirming that the tower is abnormal.

[0489] Furthermore, if it is determined that there are abnormal points in the tower crane during this operating period, the operating status of the tower crane during this operating period is determined based on the contour map, including:

[0490] If the contour map after the anomaly point is in a steady state, the tower's operating status is determined based on the degree of deviation between the contour map before the anomaly point and the contour map after the anomaly point.

[0491] Furthermore, based on the degree of deviation between the contour maps before and after the anomaly points, the operating status of the tower is determined, including:

[0492] If the deviation is less than or equal to the preset threshold, the tower's operating status is determined to be normal operation;

[0493] If the deviation exceeds the preset threshold, the tower's operating status is determined to be abnormal.

[0494] Specifically, in this embodiment, when an abnormal signal is detected, the current time t1 is recorded, where t1 is the point where the abnormal event occurs. From the current time t1, after a preset time, when the change in the MSE value of the signal within N consecutive windows is less than the threshold D, it is considered to have reached the second steady state. At this time, the time t2 of the first window signal within the N windows is recorded, where t2 is the end point of the abnormal event. The value of N is determined based on the sampling rate, signal processing capability, and processing effect. After obtaining time t2, the signal difference before t1 and after t2 is compared. The comparison method is as follows: if the signal after t2 is an abnormal signal, K consecutive samples are collected from both before t1 and after t2. Following the chronological order, the samples before t1 and after t2 are paired, and the difference between the corresponding two sets of signal acoustic signatures is calculated. If the signal after t2 is a normal signal, it indicates that the signals before t1 and after t2 are the same, and the tower's operating state is determined to be normal. If the signals before t1 and after t2 are different, the tower's operating state is determined to be abnormal.

[0495] Furthermore, the degree of deviation is calculated using the following formula:

[0496] Where MSE represents the degree of deviation; Y i A contour map showing the area in front of the outlier. This is a contour map following the outlier points.

[0497] To verify the monitoring effectiveness of trays or float valves, 1000 sets of acoustic signals were collected. This scheme and conventional recognition models were used to identify tray tipping (250 sets), float valve detachment (250 sets), airflow changes (250 sets), and noise interference (250 sets). Compared to conventional recognition models, such as the CNN model (73.20% accuracy) and the empirical database matching model (82.50% accuracy), the model in this scheme achieved an accuracy of 98.1%.

[0498] In another implementation, if the acoustic signal generating component is the same as the working principle of the tray or float valve, the above-mentioned scheme can also be used for fault analysis, identification and diagnosis, so as to obtain the operating status of the component with the same working principle as the tray or float valve.

[0499] Example 9

[0500] In this embodiment, the acoustic signal generating component is a tube bundle, such as... Figure 10 As shown, it includes:

[0501] Acquire acoustic signals from the heat exchanger over a period of time.

[0502] Specifically, in this embodiment, the acoustic signal is acquired by an acoustic sensor. During normal operation of the heat exchanger, the acoustic signal monitored is a stable fluid flow acoustic signal A. When a tube bundle leak occurs, the internal and external pressure difference causes the medium to flow through the leak hole and friction leak hole, which superimposes with the flow of the heat exchange fluid to form an acoustic signal B. For the identification of a leak event, it is necessary to accurately identify this subtle change.

[0503] S91. Based on the acoustic signal, determine the RMS spread entropy and acoustic signature characteristics of the acoustic signal during the operating time segment.

[0504] Specifically, in this embodiment, the RMS spread entropy of the acoustic signal is obtained in the following manner, such as... Figure 9 As shown, it specifically includes:

[0505] The acoustic signal is processed by scaling to obtain a multi-scale signal;

[0506] In the process of scaling, the scale factor is usually set to 20 or 30.

[0507] Determine the RMS values ​​of multi-scale signals and form a data sequence;

[0508] Specifically, the root mean square (RMS) of a multi-scale signal x(t) that varies continuously with time over the time interval (0, t) can be expressed as:

[0509]

[0510] Where rms is the root mean square RMS; x(t) is the acoustic signal; and τ is the scaling factor.

[0511] Meanwhile, the following formula data sequence is used:

[0512]

[0513] The RMS scattering entropy value of the data sequence is calculated and used as the RMS scattering entropy of the acoustic signal.

[0514] The RMS scatter entropy value of the data sequence is calculated using the following formula:

[0515]

[0516] Where MDE is the RMS scattering entropy value; The data sequence is represented by τ, which is a scale factor and takes a positive integer value. If τ = 1, it represents the original time-domain signal. When τ > 1, the original time-domain signal is divided into τ scaled sequences of length N / τ. By coarsening the original signal, τ multi-scale signals are obtained. However, during the coarsening process, when τ is too small, it is difficult to completely extract the state feature information from the picked-up switch machine vibration signal. As the scale factor τ increases, if τ is too large, the number of discrete points in the resulting new sequence will be greatly reduced, which may lead to... The feature information represented by the obtained entropy value is lost. Therefore, a coarse-grained method of continuous translation and averaging can be used to improve the accuracy and computational stability of the entropy value. m, c, and d are constants. When extracting multi-scale RMS scatter entropy, four parameters need to be selected: number of categories c, embedding dimension m, time delay d, and scale factor τ. Usually, the embedding dimension m and the number of categories c should not be too small or too large. m is usually 2 or 3, c is an integer between 4 and 8, and the time delay d is generally 1. The length of the data to be processed should be greater than 2000.

[0517] In this embodiment, to address the complex changes in acoustic signals after leakage, RMS scatter entropy is calculated quickly, is less affected by abrupt signal changes, and considers the magnitude relationship between amplitudes. To improve the recognizability of leakage signals, RMS scatter entropy information of acoustic detection signals at different scales is extracted. Combined with the low-frequency continuity and other characteristics of leakage acoustic signals, the signal data is coarsely processed, and a multi-scale RMS scatter entropy feature extraction method based on RMS is established, which can ensure the accuracy of the data.

[0518] Specifically, in this embodiment, the acoustic signature features of the acoustic signal are obtained in the following manner, including:

[0519] The acoustic signal is subjected to time-domain transformation, Fourier transform, filtering, and differentiation to obtain the voiceprint features.

[0520] The voiceprint features are obtained using the following methods:

[0521] Convert the acoustic emission signal into a time-domain signal;

[0522] Specifically, time-domain signals can describe the relationship between mathematical functions or physical signals and time. For example, the time-domain waveform of an acoustic signal can express how the signal changes over time.

[0523] Based on the time-domain signal, a linear spectrum is obtained using Fourier transform;

[0524] The Fourier transform can be performed using either the Discrete Fourier Transform or the Fast Fourier Transform. Alternatively, the Laplace transform or the Z-transform can be used.

[0525] The linear spectrum is converted into a Mel spectrum using a Mel frequency filter bank.

[0526] Specifically, because frequency domain signals have a lot of redundancy, filter banks can simplify the amplitude in the frequency domain, using a single value to represent each frequency band. This is achieved by multiplying the amplitude spectrum obtained from the FFT with the frequency of each filter and accumulating the results; the resulting value represents the energy of the frame data in the corresponding frequency band of that filter.

[0527] The logarithmic energy and logarithmic spectrum of the Mel spectrum are taken, and the first derivative is performed to obtain the Mel frequency cepstral coefficients.

[0528] Specifically, the cepstral spectrum means performing a Fourier transform on the time-domain signal, taking the logarithm, and then performing an inverse Fourier transform. It can be divided into complex cepstral spectrum, real cepstral spectrum, and power cepstral spectrum; we are using the power cepstral spectrum. Cepstral analysis can be used to decompose signals, transforming the convolution of two signals into their sum. Specifically, by taking the logarithmic energy and logarithmic spectrum of the Mel spectrum and performing first-order differentiation, the Mel frequency cepstral coefficients are obtained, which can further amplify the difference between the anomalous signal and the generated signal, facilitating the detection of anomalous signals.

[0529] Based on the Mel frequency cepstral coefficients, the voiceprint features are obtained.

[0530] S92. Based on the RMS dispersion entropy and the acoustic signature, determine the operating status of the heat exchanger during this operating period.

[0531] Specifically, in this embodiment, the RMS dispersion entropy and the voiceprint features are used as inputs to the leakage identification model to obtain the leakage identification result;

[0532] Based on the leak identification results, the operating status of the heat exchanger during this operating period is determined.

[0533] More specifically, the input to the leak identification model is the RMS scattering entropy and the acoustic signature features, and the output of the leak identification model is 0 or 1. 0 represents that the signal is not a normal acoustic signal of the heat exchanger tube bundle operation state, i.e., an abnormal signal; 1 represents that the signal is a normal acoustic signal of the heat exchanger tube bundle operation state.

[0534] In this embodiment, the method further includes:

[0535] Obtain a training dataset, which includes the corresponding RMS scattering entropy and voiceprint features obtained based on historical acoustic signals;

[0536] Based on the training dataset and convolutional neural network, a leak detection model is trained.

[0537] Specifically, before real-time monitoring, based on historical acoustic signals collected by acoustic sensors (including acoustic signals from normal and abnormal operation of the heat exchanger tube bundles), the corresponding RMS scattering entropy and acoustic signature features are obtained using the method described above. These RMS scattering entropy and acoustic signature features are then used as a training set to train a convolutional neural network, resulting in a leak detection model. During real-time monitoring, the corresponding RMS scattering entropy and acoustic signature features are first obtained based on the actual acoustic signals collected. These features are then used as input to the leak detection model, with the final output being either 0 or 1. 0 represents that the signal is not a normal acoustic signal indicating normal operation of the heat exchanger tube bundles, i.e., an abnormal signal; 1 represents that the signal is a normal acoustic signal indicating normal operation of the heat exchanger tube bundles.

[0538] More specifically, convolutional neural networks possess translation invariance, enabling them to better capture anomalous signal features. This allows for more accurate and effective identification of abnormal events. A leak detection model is built using a one-dimensional convolutional neural network, consisting of several convolutional layers and fully connected layers, including:

[0539] The received signal is divided according to the window size W, and the windows overlap by Z.

[0540] During the training phase, assuming the input is x, some noise is added to x to generate... . After encoding and decoding, the signal is reconstructed into x′. x and x′ are then input to a discriminator, which determines whether the signal is a generated signal or a real signal. Ideally, after training, a generator capable of reconstructing x~ into a distribution similar to x can be obtained.

[0541] During the testing phase, x is input into the generator, and the difference between x and the generator's output x′ is judged. When the difference is greater than a certain value, x is an abnormal signal and the output is 0; when the difference is less than a certain value, x is a normal signal and the output is 1.

[0542] To verify the monitoring effect of the heat exchanger, four sensors were installed on the outer heat exchanger head of Qilu Petrochemical, collecting 800 sets of acoustic signals with different leakage orifice diameters. 200 sets of noise signals were mixed in for identification. The accuracy of conventional models such as the CNN model was 79.50%, the accuracy of the empirical database matching model was 74.60%, and the identification accuracy of the online acoustic monitoring and diagnosis model for heat exchanger tube bundle leakage was 97.3%.

[0543] In another implementation, if the acoustic signal generating component is the same as the tube bundle in working principle, the above-mentioned scheme can also be used for fault analysis, identification and diagnosis, so as to obtain the operating status of the component with the same working principle as the tube bundle.

[0544] Example 10

[0545] In this embodiment, as Figure 11 As shown, this embodiment provides a device for monitoring the operating status of equipment. The device includes a device wall and internal components. The device includes:

[0546] The parameter acquisition module 10 is used to acquire the acoustic signals of the device during a certain operating time.

[0547] The component determination module 20 is used to extract the modal features of the acoustic signal and determine the component that generates the current acoustic signal based on the modal features. The component that generates the signal is a device wall or an internal component.

[0548] The component state determination module 30 is used to call the corresponding acoustic analysis algorithm based on the acoustic signal generating component, and to analyze the acoustic signal based on the called acoustic analysis algorithm to determine the state of the acoustic signal generating component.

[0549] Example 11

[0550] In this embodiment, as Figure 12 As shown, this embodiment provides a device operation status monitoring system. The device includes a device wall and internal components. The system includes:

[0551] Multiple acoustic sensors are mounted on the device to collect acoustic signals;

[0552] The aforementioned equipment operation status monitoring device is connected to the acoustic sensor.

[0553] Furthermore, to ensure the accuracy of the data collected by the acoustic sensor, in this embodiment, as follows: Figure 13-15 As shown, an acoustic sensor is provided, such as Figure 13-15 As shown, it includes:

[0554] The shell 1 is hollow inside, and the upper and lower ends of the shell 1 are the signal output terminal 102 and the signal detection terminal 101, respectively.

[0555] The backing pad 2 is disposed inside the housing 1 through contact and cooperation with the inner wall of the housing 1 via its side wall. The end of the backing pad 2 near the signal detection end 101 is provided with an installation space 21.

[0556] A piezoelectric wafer 3 and a mass block 4 are disposed in the installation space 21, with the mass block 4 located on top of the piezoelectric wafer 3, and the negative electrode of the piezoelectric wafer 3 is connected to the housing 1;

[0557] The diaphragm 5 is disposed at the signal detection end 101 of the housing 1 and does not contact the piezoelectric crystal 3;

[0558] Terminal 6 is fixed to the signal output terminal 102 of the housing 1; terminal 6 is connected to the positive electrode of the piezoelectric crystal 3 through a wire 7 passing through the backing pad 2, and is used to transmit the charge generated by the vibration of the piezoelectric crystal 3 to the outside.

[0559] In this embodiment, the housing 1 can be configured as a cylindrical structure; the backing pad 2 is connected and fixed to the housing 1 by adhesive.

[0560] The piezoelectric crystal 3 achieves signal conversion through the piezoelectric effect. The mechanism is as follows: piezoelectric crystals have low symmetry. When deformed under external force, the relative displacement of positive and negative ions in the unit cell causes the centers of positive and negative charges to no longer coincide, leading to macroscopic polarization of the crystal. The surface charge density of the crystal is equal to the projection of the polarization intensity onto the surface normal. Therefore, when a piezoelectric material deforms under pressure, opposite charges appear on its two ends. When receiving an acoustic emission signal, the crystal vibrates, generating opposite charges. When the piezoelectric material is polarized in an electric field, the displacement of the charge centers causes material deformation. In addition to self-supporting capabilities, it also possesses self-diagnostic, self-adaptive, and self-repairing functions.

[0561] The positive and negative electrodes used in the piezoelectric crystal 3 are typically silver electrodes, which are easy to solder. The negative electrode of the piezoelectric crystal 3 is connected to the inner wall of the housing 1. Specifically, this can be done by using conductive adhesive to connect the negative electrode of the piezoelectric crystal 3 to the inner wall of the housing 1, thereby fixing the piezoelectric crystal 3 to the housing 1. Alternatively, a through hole can be provided on the backing pad 2, through which the negative electrode of the piezoelectric crystal 3 is connected to the inner wall of the housing 1.

[0562] The mass and position of mass block 4 significantly affect the sensor's measurement range and sensitivity. Generally, the heavier the mass block 4, the larger the sensor's measurement range, but the lower the sensitivity; conversely, the lighter the mass block 4, the smaller the sensor's measurement range, but the higher the sensitivity. Furthermore, the position of mass block 4 is also crucial, as it determines the corresponding output voltage value. Ideally, mass block 4 should be positioned at the sensor's center of gravity to minimize mechanical noise and ensure the sensor's sensitivity and accuracy.

[0563] More specifically, a through hole is provided in the middle of the terminal 6, and the wire is located in the through hole. The wire can be sealed and fixed in the housing 1 with insulating glue. When the positive electrode of the piezoelectric crystal 3 is located in the middle, the wire passes through the middle of the backing pad 2 and the corresponding mass block 4, and connects to the positive electrode of the piezoelectric crystal 3.

[0564] Furthermore, the installation space 21 includes: a first installation space 211 and a second installation space 212, wherein a first step portion 213 is formed between the first installation space 211 and the second installation space 212;

[0565] The piezoelectric wafer 3 is disposed in the first mounting space 211, and the mass block 4 is disposed in the second mounting space 212. The mass block 4 is in contact with the step surface of the first step portion 213.

[0566] In this embodiment, in order to install and fix the piezoelectric chip 3 and the mass block 4, an installation space 21 is provided at the end of the backing pad 2 near the signal detection end 101, and includes a first installation space 211 and a second installation space 212 with different accommodating spaces. A first step portion 213 is formed between the first installation space 211 and the second installation space 212, and the mass block 4 is locked and fixed through the first step portion 213.

[0567] Furthermore, the piezoelectric wafer 3 is a multilayer lead zirconate titanate piezoelectric thin film.

[0568] In this embodiment, the piezoelectric crystal 3 made of lead zirconate titanate is fabricated using a series-parallel stacked structure to achieve a narrow bandwidth effect. The stacked structure is bonded with conductive adhesive, resulting in a wide bandwidth, good electroacoustic conversion efficiency, and easier matching with the excitation circuit.

[0569] Furthermore, the acoustic sensor has a response frequency of 20-400 kHz.

[0570] Furthermore, the signal detection end 101 of the housing 1 is provided with a second step portion 11, and the diaphragm 5 is fixed on the second step portion 11.

[0571] In this embodiment, to ensure the structural strength and secure installation of the diaphragm 5, a second step 11 is provided at the signal detection end 101 of the housing 1. The diaphragm 5 is fixed by interference fit between it and the inner wall of the housing 1, or by adhesive bonding. The housing 1 protects the piezoelectric crystal 3, ensuring its service life. The diaphragm 5 can be made of a thin iron sheet, and through holes can be provided on the diaphragm 5 to transmit acoustic emission signals.

[0572] Furthermore, the signal output terminal 102 of the housing 1 is provided with a third step portion 12, and the wiring terminal 6 is fixed on the third step portion 12.

[0573] In this embodiment, in order to install and fix the terminal block 6 and ensure structural strength, a third step portion 12 is provided at the signal output end 102 of the housing 1. The terminal block 6 is fixed by interference fit with the inner wall of the housing 1 or by adhesive bonding.

[0574] Furthermore, the housing 1 is made of metal; the backing pad 2 is made of damping material.

[0575] In this embodiment, the housing 1 is made of metal and serves as the transmission medium to ground the housing. The backing pad 2 is a damping material used for vibration and noise to ensure the accuracy of the acquired signal. Specifically, it can be made of rubber, plastic damping plates, or a combination of rubber and foam plastic.

[0576] In another embodiment, a connecting piece 8 is provided on the outer surface of the housing 1, and the connecting piece 8 has a mounting hole 81. The acoustic sensor with this structure is suitable for monitoring devices with through holes, such as those in the housing 1. During installation, the signal detection end 101 is inserted into the through hole, and the connecting piece 8 is fixed to the monitoring device using screws passing through the mounting hole 81, thus securing the acoustic sensor. The connecting piece 8 can be replaced with a flange. To reduce vibration, a rubber damping ring is fitted onto the housing 1, allowing for vibration damping after the acoustic sensor is fixed, ensuring more accurate data acquisition.

[0577] In another embodiment, an external thread 13 is provided on the outer surface between the signal detection end 101 of the housing 1 and the connecting piece 8. The acoustic sensor with this structure is suitable for monitoring devices with through holes, where internal threads are provided. During installation, the signal detection end 101 is inserted into the through hole, and a threaded connection is achieved between the external thread 13 on the outer surface of the housing and the internal thread of the through hole. Furthermore, the depth of the signal detection end 101 can be controlled by rotating the acoustic sensor to ensure more accurate data acquisition.

[0578] In addition, embodiments of the present invention also provide a machine-readable storage medium storing instructions that cause a machine to execute the above-described device operation status monitoring method.

[0579] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for monitoring the operating status of equipment, characterized in that, The device includes a device wall and internal components, and the method includes: Acquire acoustic signals from the device over a period of time. Extract the modal features of the acoustic signal, and based on the modal features, determine the generating component of the current acoustic signal, wherein the generating component is the device wall or an internal component; Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component.

2. The method according to claim 1, characterized in that, Extract modal features of the acoustic signal, and based on the modal features, determine the components that generate the acoustic signal, including: The acoustic signal is processed using synchronous compressed wavelets to obtain its modal characteristics; The first preset mode is reconstructed from the modal features to obtain the time-domain signal that first arrives at the preset mode; Based on the time-domain signal that arrives first in the preset mode, the arrival time of the acoustic signal within that operating time period is determined. The component that generates the acoustic signal is determined based on the arrival time of the acoustic signal.

3. The method according to claim 2, characterized in that, Based on the time-domain signal that first arrives at the preset mode, the arrival time of the acoustic signal within this operating time is determined, including: The signal envelope is obtained by processing the time-domain signal that first arrives at the preset mode using the Hilbert transform. Based on the signal envelope, determine the global maximum point; The time window is determined based on the global maximum value and the preset time delay; Within the time window, the first global minimum point is calculated using the AIC function; Based on the first global minimum point, the range of the neighborhood is determined, the midpoint of the neighborhood is the first global minimum point, and the length of the neighborhood is a preset time delay. Within the specified range, the second global minimum point is calculated using the AIC function and used as the arrival time of the acoustic signal.

4. The method according to claim 2, characterized in that, Based on the arrival time of the acoustic signal, the components that generate the acoustic signal are determined, including: At least one time difference value is obtained by subtracting the arrival times of the acoustic signals from each acoustic sensor. Based on the time difference values, a signal time difference matrix is ​​constructed; Based on the aforementioned signal time difference matrix, multiple spatial distances are obtained; From the spatial distance, a predetermined number of minimum spatial distances that can form a predetermined shape are determined as the acoustic signal generation area, so as to determine the acoustic signal generation component.

5. The method according to claim 1, characterized in that, The equipment in question is a petrochemical equipment.

6. The method according to claim 1, characterized in that, Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including: If the component that generates the acoustic signal is determined to be the device wall: Based on the acoustic signal, the entropy parameter and acoustic reconstruction data of the acoustic signal are obtained; Based on the time domain, frequency domain, and entropy parameter of the acoustic signal, the first theoretical damage state of the device wall is determined; Based on the acoustic signal and the acoustic reconstruction data, the second theoretical damage state of the vessel wall is determined; Based on the first theoretical damage state and the second theoretical damage state, the damage state of the vessel wall is determined, including the presence of damage and the absence of damage.

7. The method according to claim 6, characterized in that, Based on the time domain, frequency domain, and entropy parameter of the acoustic signal, the first theoretical damage state of the vessel wall is determined, including: The time-domain, frequency-domain, and entropy parameters of the acoustic signal are used as inputs to the hypersphere model to obtain the generalized distance between the acoustic signal and the center of the hypersphere model. The hypersphere model is obtained by training SVDD with a training dataset, which includes the time-domain, frequency-domain, and entropy parameters of historical acoustic signals when the vessel wall generates anomalies. If the generalized distance is less than or equal to the generalized radius of the hypersphere of the hypersphere model, then the first theoretical damage state of the device wall is determined to be no damage. If the generalized distance is greater than the generalized radius of the hypersphere in the hypersphere model, then the first theoretical damage state of the device wall is determined to be that damage exists; Based on the acoustic signal and the acoustic reconstruction data, the second theoretical damage state of the vessel wall is determined, including: Calculate the mean square error between the acoustic signal and the acoustic reconstruction data; If the mean square error is less than or equal to the reconstruction threshold, then the second theoretical damage state of the device wall is determined to be no damage. If the mean square error is greater than the reconstruction threshold, then the second theoretical damage state of the device wall is determined to be that damage exists.

8. The method according to claim 6, characterized in that, Based on the first theoretical damage state and the second theoretical damage state, the damage state of the vessel wall is determined, including: If both the first theoretical damage state and the second theoretical damage state are present, then the damage state of the vessel wall is determined to be present. If both the first theoretical damage state and the second theoretical damage state are non-damaged, then the damage state of the vessel wall is determined to be non-damaged. If the first theoretical damage state and the second theoretical damage state are respectively the presence of damage and the absence of damage, then the judgment coefficient is calculated based on the generalized distance and the mean square error. If the judgment coefficient is less than the preset threshold, the damage state of the device wall is determined to be no damage. If the judgment coefficient is greater than or equal to the preset threshold, then the damage state of the device wall is determined to be present.

9. The method according to claim 6, characterized in that, The method further includes: When the damage condition of the equipment wall is determined to be present: The damage stress is determined based on the total acoustic emission count of the acoustic signal; If the damage stress is less than or equal to the first preset stress, then the degree of damage to the equipment wall during this operating period is determined to be initial damage. If the damage stress is greater than the first preset stress and less than or equal to the second preset stress, then the degree of damage to the equipment wall during this operating period is determined to be low damage. If the damage stress is greater than the second preset stress and less than or equal to the third preset stress, then the degree of damage to the equipment wall during this operating period is determined to be medium damage. If the damage stress is greater than the third preset stress, then the damage level of the equipment wall during this operating period is determined to be high damage.

10. The method according to claim 1, characterized in that, The internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle; Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including: If the component that generates the acoustic signal is determined to be a wing valve: The acoustic signal is preprocessed to obtain the processed acoustic signal; Based on the processed acoustic signal, the flow rate percentage of each wing valve during the operating time is determined, and the flow rate percentage is determined by the energy value in the processed acoustic signal. Based on the flow rate ratio of each wing valve, the operating status of the wing valve during this operating period is determined.

11. The method according to claim 10, characterized in that, The acoustic signal includes energy parameters; The acoustic signal is preprocessed to obtain a processed acoustic signal, including: For each wing valve: Acoustic signals with energy values ​​greater than a first energy threshold are defined as the first valid signals; The first effective signal was corrected using clustering algorithms, the Laida criterion, the quartile method, and cross-validation to obtain the processed acoustic feature parameters.

12. The method according to claim 1, characterized in that, The internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle; Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including: If the component that generates the acoustic signal is determined to be a wing valve: Based on the acoustic signals, the catalyst leakage risk factor of the wing valve is determined; Based on the catalyst runoff risk coefficient, the operating status of the wing valve during this operating period is determined.

13. The method according to claim 12, characterized in that, Based on the acoustic signals, the catalyst leakage risk factor of the wing valve is determined, including: The acoustic signal of each wing valve is input into the opening and closing cycle recognition model to obtain the actual opening and closing cycle corresponding to each wing valve. Based on the energy parameters and ringing count parameters corresponding to each actual opening and closing cycle of each wing valve, the least squares method is used to perform linear fitting to calculate the comprehensive slope of the energy parameters and ringing count parameters of each wing valve as a function of time in each actual opening and closing cycle. Based on the combined slope and the average value of the combined slope of each wing valve, the deflection coefficient corresponding to each wing valve is obtained; Based on the flow deflection coefficient corresponding to each wing valve, the catalyst runoff risk coefficient of the wing valve is calculated.

14. The method according to claim 13, characterized in that, The overall slope is calculated using the following formula: k ij =p·k i ′ j +(1-p)·k i ′ j ′; Where, k ij The slope is the composite slope corresponding to the j-th actual opening and closing cycle of the i-th wing valve; p is the weighting coefficient; k i ′ j Let k be the slope of the energy parameter change with time during the j-th actual opening and closing cycle of the i-th wing valve; i ′ j ′ represents the slope of the ringing count parameter as a function of time during the j-th opening and closing cycle of the i-th wing valve; The deflection coefficient is calculated using the following formula: Where, δ i k is the deflection coefficient of the i-th wing valve; ij This is the comprehensive slope corresponding to the j-th actual opening and closing cycle of the i-th wing valve; Let be the average of the combined slopes of the i-th wing valve. The catalyst runaway risk factor is calculated using the following formula: Where η is the catalyst runoff risk coefficient of the internal components; δ i The deflection coefficient of the wing valve; The average deflection coefficient for all wing valves, 15. The method according to claim 1, characterized in that, The internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle; Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including: If the component that generates the acoustic signal is determined to be a wing valve: The acoustic signal of each wing valve is input into the opening and closing cycle recognition model to obtain the actual opening and closing cycle of each wing valve. Multiply the catalyst flow rate of the wing valve by the actual opening and closing cycle to obtain the actual load-bearing weight of the wing valve; The operating status of the wing valve during this operating period is determined based on the difference between the actual load-bearing weight and the preset load-bearing weight of the wing valve.

16. The method according to claim 13 or 15, characterized in that, The method further includes: Acquire the structural parameters of the wing valve, the historical acoustic signals of the wing valve during a certain operating period, and the operating parameters of the wing valve; Based on the structural parameters and operating parameters of the wing valve, the theoretical opening and closing cycle of the wing valve is determined; Data samples were constructed based on the historical acoustic signals and theoretical opening and closing cycles of the wing valve. The constructed data samples were then used to train the support vector machine algorithm to obtain the opening and closing cycle recognition model.

17. The method according to claim 16, characterized in that, Based on the structural parameters and the operating parameters, the theoretical opening and closing cycle of the wing valve is determined, including: The theoretical opening and closing cycle of the wing valve is calculated using the following formula: Among them, T c ΔP represents the theoretical opening and closing period of the wing valve. c For the pressure drop of the wing valve; R c θ is the radius of the feed leg of the wing valve; c The inclination angle of the slanted pipe section of the wing valve; m cp The equilibrium catalyst amount for the wing valve; A c S is the valve plate mounting angle for the wing valve. cf v is the sealing area of ​​the valve plate of the wing valve; c S is the inlet linear velocity of the wing valve; c ρ is the inlet area of ​​the wing valve; c The inlet concentration of the wing valve.

18. The method according to claim 17, characterized in that, If the wing valve is a two-stage wing valve arranged in series, and the inlet linear velocity of the subsequent wing valve cannot be directly obtained, determining the theoretical opening and closing cycle of the wing valve based on the structural parameters and the operating parameters further includes: The theoretical opening and closing cycle of the wing valve is calculated using the following formula: Among them, T Z1 This represents the theoretical opening and closing cycle of the subsequent stage wing valve; ΔP c For the pressure drop of the subsequent stage wing valve; R c θ is the radius of the feed leg of the subsequent stage wing valve; c The inclination angle of the inclined pipe section of the subsequent stage wing valve; m cp The equilibrium catalyst amount for the subsequent stage wing valve; A c The valve plate mounting angle for the subsequent stage wing valve; S cf v is the valve plate sealing area of ​​the subsequent stage wing valve; c S is the inlet linear velocity of the preceding stage wing valve; c ρ is the inlet area of ​​the subsequent stage wing valve; c The inlet concentration of the next-stage wing valve; η z denoted as cyclone separation efficiency of internal components; Z is the efficiency correction coefficient; d is the average particle size of regenerated catalyst; a1, a2, a3, a4 and a5 are constants; δ1 and δ2 are constants, and δ1 is less than δ2.

19. The method according to claim 1, characterized in that, The internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle; Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including: If the acoustic signal generating component is determined to be a double-acting slide valve: The acoustic signal is preprocessed to obtain the processed acoustic signal; Based on the processed acoustic signal, the crack length of the double-acting slide valve during this operating period is determined; Based on the crack length, determine the degree of damage to the double-acting slide valve during this operating period; Based on the degree of damage to the double-acting slide valve, the operating status of the double-acting slide valve during this operating period is determined.

20. The method according to claim 19, characterized in that, The acoustic signals include: ringing count parameters, energy parameters, and frequency parameters; The acoustic signal is preprocessed to obtain a processed acoustic signal, including: Acoustic signals with energy values ​​greater than a first energy threshold and peak frequencies greater than a first frequency threshold are identified as the second valid signals. The second effective signal is corrected using clustering algorithms, quartile method, Laida criterion and cross-validation to obtain the processed acoustic signal.

21. The method according to claim 20, characterized in that, The degree of damage to the double-acting spool valve is determined based on crack length, including: If the crack length is less than or equal to the first preset length, then it is determined that the double-acting slide valve is undamaged during this operating period. If the crack length is greater than the first preset length and less than or equal to the second preset length, the damage level of the double-acting slide valve during this operating period is determined to be low damage. If the crack length is greater than the second preset length and less than or equal to the third preset length, the damage level of the double-acting slide valve during this operating period is determined to be medium damage. If the crack length is greater than the third preset length, the damage level of the double-acting slide valve during this operating period is determined to be high damage.

22. The method according to claim 1, characterized in that, The internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle; Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including: If the acoustic signal generating component is determined to be a tray or a float valve: Based on the acoustic signal, acoustic waveform data and contour plots of the tray or float valve during the operating time of that period are obtained; Based on the acoustic waveform data or the contour map, determine whether there are any abnormal points in the tray or float valve during the operating time period. If an anomaly is found in the tray or float valve during the operating period, the operating status of the tray or float valve during the operating period is determined based on the contour map. The operating status includes normal operation and abnormal operation.

23. The method according to claim 22, characterized in that, Based on the acoustic signal, a contour map of the operating time segment is obtained, including: The total duration of a single signal is obtained based on the sampling rate and sampling length of the acoustic signal; The total duration of a single signal is divided into frames and transformed by time and frequency to obtain the frame frequency, the time point for spectrum analysis, and the energy spectral density. Based on the frame frequency, the time point of the spectrum analysis, and the energy spectral density, the distribution of contour maps of the corresponding areas of the time-frequency intensity cloud map is extracted to obtain the contour maps of the tray or floating valve during the operating time of that period.

24. The method according to claim 22, characterized in that, Based on the acoustic waveform data or the contour map, determine whether there are any abnormal points in the tray or float valve during this operating period, including: The acoustic waveform data is used as input to the first anomaly detection model to determine the anomaly detection result, or The contour map is used as input to the second anomaly detection model to determine the anomaly detection result; The first anomaly detection model is obtained by training the autoencoder using the first training dataset, which includes historical acoustic waveform data of the tray or float valve. The second anomaly detection model is obtained by training the autoencoder using the second training dataset, which includes historical contour maps of trays or float valves.

25. The method according to claim 22, characterized in that, If an anomaly is identified in the tray or float valve during this operating period, the operating status of the tray or float valve during this period is determined based on the contour map, including: If the contour map after the anomaly point is not in a steady state, then the operating state of the tray or float valve is determined to be abnormal. If the contour map after the outlier is in a steady state, then determine the degree of deviation between the contour map before the outlier and the contour map after the outlier. If the deviation is less than or equal to the preset threshold, the operating status of the tray or float valve is determined to be normal operation; If the deviation exceeds the preset threshold, the operating status of the tray or float valve is determined to be abnormal.

26. The method according to claim 1, characterized in that, The internal components include at least one of a wing valve, a double-acting slide valve, a tray, a float valve, and a tube bundle; Based on the acoustic signal generating component, a corresponding acoustic analysis algorithm is invoked to analyze the acoustic signal and determine the state of the acoustic signal generating component, including: If the acoustic signal generating component is determined to be a tube bundle: Based on the acoustic signal, the RMS spread entropy and acoustic signature characteristics of the tube bundle during the operating time of that segment are determined. Based on the RMS dispersion entropy and the acoustic signature, the operating status of the tube bundle during this operating period is determined.

27. The method according to claim 26, characterized in that, Based on the acoustic signal, the RMS spread entropy of the acoustic signal of the tube bundle during this operating period is determined, including: The acoustic signal is processed by scaling to obtain a multi-scale signal; Determine the RMS values ​​of multi-scale signals and form a data sequence; The RMS scattering entropy value of the data sequence is calculated and used as the RMS scattering entropy of the acoustic signal.

28. The method according to claim 26, characterized in that, Based on the acoustic signal, the acoustic signature characteristics of the tube bundle during this operating period are determined, including: Convert the acoustic emission signal into a time-domain signal; Based on the time-domain signal, a linear spectrum is obtained using Fourier transform; The linear spectrum is converted into a Mel spectrum using a Mel frequency filter bank. The logarithmic energy and logarithmic spectrum of the Mel spectrum are taken, and the first derivative is performed to obtain the Mel frequency cepstral coefficients. Based on the Mel frequency cepstral coefficients, the voiceprint features are obtained.

29. The method according to claim 26, characterized in that, Based on the RMS dispersion entropy and the acoustic signature characteristics, the operating status of the tube bundle during this operating time period is determined, including: The RMS scattering entropy and the acoustic signature feature are used as inputs to the leakage identification model to obtain the leakage identification result. The leakage identification model is obtained by training a convolutional neural network using a training dataset. The first training dataset includes the corresponding RMS scattering entropy and acoustic signature feature obtained based on the historical acoustic signal of the tube bundle. Based on the leak identification results, the operating status of the tube bundle during that operating period is determined.

30. A device for monitoring the operating status of equipment, characterized in that, The equipment includes equipment walls and internal components, and the device includes: The parameter acquisition module is used to acquire the acoustic signals of the device during a certain operating time. The component generation determination module is used to extract the modal features of the acoustic signal and determine the component generating the current acoustic signal based on the modal features. The component generating the signal is a device wall or an internal component. The component state determination module is used to call the corresponding acoustic analysis algorithm based on the acoustic signal generating component, and to analyze the acoustic signal based on the called acoustic analysis algorithm to determine the state of the acoustic signal generating component.

31. A device operation status monitoring system, characterized in that, The device includes a device wall and internal components, and the system includes: Multiple acoustic sensors are mounted on the device to collect acoustic signals; The equipment operation status monitoring device according to claim 30 is connected to the acoustic sensor.

32. The equipment operation status monitoring system according to claim 31, characterized in that, The acoustic sensor includes: The internally hollow cylindrical shell (1) has a signal output terminal (102) and a signal detection terminal (101) at its upper and lower ends, respectively. The backing pad (2) is disposed inside the housing (1) through contact and cooperation with the inner wall of the housing (1) via the side wall, and the lower end of the backing pad (2) is provided with an installation space (21); A piezoelectric wafer (3) and a mass block (4) are arranged in the installation space (21), with the mass block (4) located above the piezoelectric wafer (3), and the negative electrode of the piezoelectric wafer (3) is connected to the inner wall of the housing (1); A diaphragm (5) is disposed at the signal detection end (101) of the housing (1) and does not contact the piezoelectric crystal (3); A terminal block (6) is fixed to the signal output terminal (102) of the housing (1); the terminal block (6) is connected to the positive electrode of the piezoelectric crystal (3) through a wire (7) passing through the backing pad (2), and is used to transmit the charge generated by the vibration of the piezoelectric crystal (3) to the outside.

33. The equipment operation status monitoring system according to claim 32, characterized in that, The installation space (21) includes: A first mounting space (211) and a second mounting space (212), the first mounting space (211) and the... A first step (213) is formed between the second installation spaces (212); The piezoelectric wafer (3) is disposed in the first mounting space (211), the mass block (4) is disposed in the second mounting space (212), and the mass block (4) is in contact with the step surface of the first step portion (213).

34. A machine-readable storage medium storing instructions for causing a machine to perform the device operation status monitoring method according to any one of claims 1-29.

Images