Two-parameter intelligent closed-loop diaphragm pump aging test system and method

Abstract

The application relates to the technical field of diaphragm pump aging test, and discloses a double-parameter intelligent closed-loop diaphragm pump aging test system and method. The method obtains a temperature frequency spectrum and a vibration frequency spectrum through fast Fourier transform, calculates a frequency domain coherence coefficient to identify temperature-vibration coupling characteristics, adopts a spectrum peak tracking algorithm to realize real-time identification of a resonance frequency band and a drift trend thereof, constructs a multi-objective optimization function considering aging acceleration efficiency and resonance risk, solves an optimal temperature stress amplitude and a vibration stress frequency by using a particle swarm algorithm, generates a stress loading instruction avoiding a resonance zone through an adaptive PID controller, and dynamically updates model parameters based on a closed-loop feedback mechanism. The application solves the problem of resonance effect amplification caused by traditional fixed-frequency stress loading, and realizes fine adaptive stress control.

Description

Technical Field

[0001] This invention relates to the field of diaphragm pump testing technology, and more specifically, to a dual-parameter intelligent closed-loop diaphragm pump aging test system and method. Background Technology

[0002] During the aging test of diaphragm pumps, temperature and vibration stress are applied to accelerate the aging process in order to assess their long-term reliability. As a critical fluid transfer device, diaphragm pumps are widely used in chemical, pharmaceutical, and semiconductor manufacturing industries, and their reliability directly affects production safety and product quality.

[0003] Traditional aging testing methods typically employ fixed-frequency stress loading, accelerating the degradation process of diaphragm pumps by setting constant temperature and vibration stress frequencies. This method is based on time-domain signal analysis, primarily monitoring changes in temperature and vibration amplitude, and assessing the degradation status of the diaphragm pump through simple threshold judgments.

[0004] However, traditional methods have the following significant drawbacks: First, when the applied stress frequency is close to the diaphragm pump's natural resonant frequency, even with moderate stress intensity, the resonance effect can cause abnormally large stress concentrations in local areas, leading to non-uniform degradation and accelerated local fatigue. This distorts the aging test results and fails to accurately reflect the degradation pattern of the diaphragm pump under normal operating conditions. Second, during the aging process, the material properties and structural characteristics of the diaphragm pump evolve, causing the natural frequency to drift. Traditional methods rely solely on time-domain signal analysis and cannot capture the changes in the frequency-domain coupling relationship between temperature and vibration. Therefore, they cannot adjust the stress loading strategy in a timely manner to adapt to these dynamic changes, resulting in insufficiently precise stress loading control. These drawbacks lead to the technical problems of low efficiency and insufficient accuracy in aging tests. Summary of the Invention

[0005] This invention provides a dual-parameter intelligent closed-loop diaphragm pump aging test method, which solves the technical problems of local fatigue acceleration caused by resonance effect and insufficient precision in stress loading control in related technologies.

[0006] This invention discloses a dual-parameter intelligent closed-loop diaphragm pump aging test method, comprising: acquiring real-time temperature data and vibration acceleration data of the diaphragm pump, performing a fast Fourier transform to generate a temperature spectrum and a vibration spectrum; calculating the cross-spectral density function of the temperature-vibration signal based on the temperature spectrum and the vibration spectrum, calculating the frequency domain coherence coefficient by the ratio of the cross-spectral density function to the temperature signal auto-spectral density function and the vibration signal auto-spectral density function, extracting frequency points that satisfy the coherence threshold to form a coherence peak frequency set; for the frequency points in the coherence peak frequency set, extracting the spectral peaks in the neighborhood of the frequency points in the vibration spectrum, marking the spectral peaks as potential resonance frequencies when they satisfy the peak discrimination condition, performing linear fitting on the potential resonance frequencies identified in multiple consecutive time windows to obtain the resonance frequency drift trend, and determining the resonance frequency. The center frequency and bandwidth of the band are determined. A multi-objective optimization function is constructed, considering both aging acceleration efficiency and resonance risk. The aging acceleration efficiency function is based on the coupling effect of temperature stress and vibration stress frequency, while the resonance risk function characterizes the proximity of the stress frequency to the resonance center frequency. Constraints are set requiring that the distance between the stress frequency and the resonance center frequency is greater than the safe frequency interval. The optimal temperature stress amplitude and the optimal vibration stress frequency are obtained by solving the problem. The optimal temperature stress amplitude and the optimal vibration stress frequency are input into an adaptive PID controller to generate a stress loading command sequence that avoids the resonance zone. Stress adjustment is performed and new temperature-vibration feedback data is acquired. The frequency domain coherence coefficient is recalculated based on the new spectral data. When the coherence change rate exceeds the threshold, the coherence threshold and peak discrimination coefficient are updated. The stress loading strategy for the next cycle is iteratively optimized.

[0007] This invention discloses a system for performing the aforementioned dual-parameter intelligent closed-loop diaphragm pump aging test method, comprising: a temperature sensor for acquiring the time-domain temperature sequence of key parts of the diaphragm pump; a vibration acceleration sensor for acquiring the time-domain vibration acceleration sequence; a stress loading device for performing stress adjustment according to the stress loading command sequence; and a data processing unit for performing fast Fourier transform, calculating the frequency domain coherence coefficient, identifying the resonance frequency band, solving the multi-objective optimization function, generating the stress loading command sequence, and updating the model parameters.

[0008] Further, the step of calculating the frequency domain coherence coefficient includes: calculating the temperature-vibration cross-spectral density function based on the temperature spectrum and vibration spectrum, wherein the temperature-vibration cross-spectral density function characterizes the correlation between the temperature signal and the vibration signal in the frequency domain; calculating the auto-spectral density function of the temperature signal, wherein the auto-spectral density function of the temperature signal is obtained by multiplying the temperature spectrum by its conjugate; calculating the auto-spectral density function of the vibration signal, wherein the auto-spectral density function of the vibration signal is obtained by multiplying the vibration spectrum by its conjugate; and calculating the frequency domain coherence coefficient, wherein the frequency domain coherence coefficient is the ratio of the square of the magnitude of the temperature-vibration cross-spectral density function to the product of the temperature signal auto-spectral density function and the vibration signal auto-spectral density function.

[0009] Further, the steps of determining the center frequency and frequency bandwidth of the resonance frequency band include: for each frequency point in the coherent peak frequency set, extracting the spectral amplitude in the neighborhood of that frequency point in the vibration spectrum, calculating the spectral peak value and the precise frequency corresponding to the peak value in the neighborhood; determining whether the spectral peak value is greater than the product of the peak value discrimination coefficient and the average amplitude of the vibration spectrum, and if the condition is met, marking the corresponding frequency as a potential resonance frequency; arranging the potential resonance frequencies identified in multiple consecutive time windows in chronological order to form a resonance frequency time series, and performing linear fitting on the resonance frequency time series to obtain the resonance frequency drift trend; determining the center frequency of the resonance frequency band as the current resonance frequency based on the current resonance frequency and the resonance frequency drift trend; finding frequency points in the vibration spectrum where the amplitude on both sides of the resonance peak value drops to half of the resonance peak value, and the frequency bandwidth is the frequency difference between the two frequency points.

[0010] Further, the multi-objective optimization function includes: defining stress adjustment parameters including temperature stress amplitude and vibration stress frequency; constructing an optimization objective function by weighted summing of the negative value of the aging acceleration efficiency function and the resonance risk function, wherein the aging acceleration efficiency function is calculated based on the Arrhenius model and the vibration fatigue cumulative damage theory, characterizing the multiple relationship between the aging rate and the reference state under a given stress parameter, wherein the Arrhenius model part calculates the exponential function of the product of the reciprocal difference between the temperature stress and the reference temperature and the activation energy, and the vibration fatigue cumulative damage part calculates the power of the vibration stress frequency; constructing a resonance risk function by calculating the exponential function of the square of the difference between the stress frequency and the resonance center frequency divided by the negative value of the frequency risk bandwidth parameter, characterizing that the risk value is greater when the stress frequency is closer to the resonance center frequency; setting optimization constraints, including upper and lower limit constraints on the temperature stress amplitude, upper and lower limit constraints on the vibration stress frequency, and a constraint that the distance between the stress frequency and the resonance center frequency is greater than the safe frequency interval; and solving the optimization objective function using the particle swarm optimization algorithm to obtain the optimal temperature stress amplitude and the optimal vibration stress frequency.

[0011] Furthermore, the execution process of the particle swarm optimization algorithm includes: initializing the particle swarm, where the position vector of each particle represents a set of candidate stress parameters including temperature stress amplitude and vibration stress frequency, and the velocity vector represents the direction and magnitude of parameter adjustment; randomly initializing the particle position and velocity within the constraint range; recording the individual optimal position and global optimal position of each particle; during the iteration process, the particle velocity update is based on the product of the inertia weight coefficient and the current velocity, plus the product of the first acceleration coefficient, the first random number, and the difference between the individual optimal position and the current position, plus the product of the second acceleration coefficient, the second random number, and the difference between the global optimal position and the current position; the particle position is updated to the current position plus the updated velocity; for the updated particle position, if it exceeds the constraint range, it is projected back to the constraint boundary; if it violates the distance constraint between the stress frequency and the resonance center frequency, a penalty term is added to the objective function using the penalty function method; calculating the fitness value of each particle, updating the individual optimal position and the global optimal position; the iteration process continues until the objective function converges or the maximum number of iterations is reached, and the global optimal solution is output.

[0012] Further, after determining the center frequency and bandwidth of the resonance frequency band, the method further includes: extracting time-domain degradation features, including the mean, standard deviation, and trend coefficient of the temperature time-domain sequence, and the mean, standard deviation, and peak factor of the vibration acceleration time-domain sequence, to form a time-domain feature vector; extracting frequency-domain features, including the resonance frequency, resonance peak, peak frequency-domain coherence coefficient, and bandwidth, to form a frequency-domain feature vector; inputting the time-domain feature vector and the frequency-domain feature vector into the time-domain feature processing branch and the frequency-domain feature processing branch of a two-stream neural network, respectively, each branch containing two fully connected layers and an activation function layer, outputting a time-domain hidden representation and a frequency-domain hidden representation, respectively; concatenating the time-domain hidden representation and the frequency-domain hidden representation to obtain a fused feature vector, and performing a nonlinear transformation through the feature fusion layer to generate a comprehensive degradation state vector; using the comprehensive degradation state vector as an additional input, calculating a weighted score through a weight adjustment parameter vector, and adaptively adjusting the weight coefficients of aging acceleration efficiency and resonance risk in the multi-objective optimization function.

[0013] Furthermore, the data transmission process of the temporal feature processing branch is as follows: the temporal feature vector undergoes a linear transformation through the weight matrix and bias vector of the first fully connected layer, and then the first layer output is obtained through a modified linear unit activation function; the first layer output undergoes a linear transformation through the weight matrix and bias vector of the second fully connected layer, and then the temporal hidden representation is obtained through a modified linear unit activation function; the data transmission process of the frequency domain feature processing branch is as follows: the frequency domain feature vector undergoes a linear transformation through the weight matrix and bias vector of the first fully connected layer, and then the first layer output is obtained through a modified linear unit activation function; the first layer output undergoes a linear transformation through the weight matrix and bias vector of the second fully connected layer, and then the frequency domain hidden representation is obtained through a modified linear unit activation function.

[0014] Furthermore, the step of iteratively optimizing the stress loading strategy for the next cycle includes: after the stress loading device performs stress adjustment according to the stress loading command sequence, it re-collects temperature data and vibration acceleration data after a preset time interval to obtain new temperature spectrum and vibration spectrum; based on the new spectrum data, it recalculates the frequency domain coherence coefficient and compares it with the historical coherence coefficient to calculate the coherence change rate; if the absolute value of the coherence change rate is greater than the coherence change threshold, it is determined that the frequency domain coupling relationship has changed significantly, and the coherence threshold and peak discrimination coefficient are updated, wherein the update of the coherence threshold is the old threshold plus the product of the first update learning rate and the coherence change rate, and the update of the peak discrimination coefficient is the old coefficient plus the product of the second update learning rate and the resonant peak change rate; the performance boundary parameters of the diaphragm pump are monitored, including the maximum allowable temperature and the maximum allowable vibration acceleration. If the measured temperature or vibration acceleration exceeds the preset proportion of the performance boundary, the upper limit value in the optimization constraint is adjusted.

[0015] Furthermore, it also includes: statistically analyzing the resonance frequency evolution data over multiple consecutive periods, using a long short-term memory time series prediction model to predict the resonance frequencies for multiple future periods, and obtaining a predicted resonance frequency sequence; in multi-objective optimization, incorporating the predicted resonance frequency sequence into the constraints, planning the stress frequency trajectory for multiple future periods in advance, avoiding passing through the predicted resonance frequency region during the stress frequency adjustment process, and calculating the stress frequency sequence for multi-period joint optimization.

[0016] This invention introduces a temperature-vibration frequency domain coherence analysis mechanism, using the cross-spectral density function to quantify the coupling relationship between temperature and vibration signals in the frequency domain, enabling the identification of frequency domain coupling characteristics that traditional time-domain analysis cannot capture. It employs a peak tracking algorithm to identify and track the resonance frequency and its drift trend in real time, dynamically capturing the evolution of the diaphragm pump's inherent frequency during aging. Based on resonance frequency constraints, a multi-objective optimization function is constructed to balance aging acceleration efficiency and resonance risk. By forcing the stress frequency away from the resonance frequency band through constraints, it effectively solves the problem of resonance effect amplification caused by the stress application frequency approaching the resonance frequency. An adaptive PID controller is used to generate... This invention employs a stress loading command to achieve refined stress control that avoids the resonance region, solving the problems of localized fatigue acceleration and non-uniform degradation caused by resonance, and realizing adaptive stress loading control optimized across the entire frequency band. By introducing a dual-stream neural network to fuse time-domain and frequency-domain features, it can extract multi-dimensional representations of the diaphragm pump's degradation state and adaptively adjust the weight coefficients of the optimization target, enabling the stress loading strategy to dynamically adjust according to the degree of degradation, thus improving the intelligence and adaptability of stress control. Through a closed-loop feedback mechanism to update the frequency-domain coherence model parameters and performance boundary parameters, and through multi-period forward-looking optimization based on the evolution law of the resonance frequency, it achieves continuous optimization and long-term stability of the stress loading strategy. In summary, this invention solves the technical problems of resonance effect amplification and insufficient precision in stress loading control caused by traditional fixed-frequency stress loading, achieving the technical effects of improving the accuracy and efficiency of aging tests, avoiding non-uniform degradation caused by resonance, and realizing refined adaptive stress control. Attached Figure Description

[0017] Figure 1 This is the main flowchart of the dual-parameter intelligent closed-loop diaphragm pump aging test method of the present invention. Detailed Implementation

[0018] In the aging test of diaphragm pumps, it is necessary to accelerate the aging process by applying temperature and vibration stress to assess their long-term reliability. Traditional aging test methods typically employ a fixed-frequency stress loading approach. However, this method has significant drawbacks: when the stress application frequency approaches the diaphragm pump's natural resonant frequency, even with moderate stress intensity, the resonance effect can lead to abnormally large stress concentrations in localized areas, causing non-uniform degradation and accelerated local fatigue. This distorts the aging test results and fails to accurately reflect the degradation pattern of the diaphragm pump under normal operating conditions. Furthermore, during the aging process, the material properties and structural characteristics of the diaphragm pump evolve, causing the natural frequency to drift. Traditional methods rely solely on time-domain signal analysis and cannot capture the changes in the frequency-domain coupling relationship between temperature and vibration. Therefore, they cannot adjust the stress loading strategy in a timely manner to adapt to these dynamic changes, resulting in insufficiently precise stress loading control and affecting the accuracy and efficiency of the aging test.

[0019] This embodiment provides a dual-parameter intelligent closed-loop diaphragm pump aging test method, applied to a diaphragm pump aging test platform. The platform includes a temperature sensor, a vibration acceleration sensor, a stress loading device, and a data processing unit. According to an embodiment of this method, the method includes the following steps: Step 100: Acquire real-time temperature and vibration acceleration data of the diaphragm pump, perform Fast Fourier Transform, and generate temperature and vibration spectra. Specifically, the temperature sensor uses a sampling frequency Collect temperature time-domain series of key components of diaphragm pump Vibration acceleration sensor at sampling frequency Acquisition of vibration acceleration time-domain sequences The sampling time window length is For temperature time-domain sequences Perform a fast Fourier transform to obtain the temperature spectrum. ; for vibration acceleration time-domain sequences Perform a fast Fourier transform to obtain the vibration spectrum. ,in Indicates frequency.

[0020] Step 200: Calculate the frequency domain coherence coefficient of the temperature-vibration signal using the cross-spectral density function, and extract the set of coherence peak frequencies. Specifically, based on the temperature spectrum and vibration spectrum Calculate the cross-spectral density function of the temperature-vibration signal. The cross-spectral density function characterizes the correlation between the temperature signal and the vibration signal in the frequency domain. The frequency domain coherence coefficient is calculated. The calculation formula is:

[0021] in, It is a temperature-vibrational cross-spectral density function. Let be the autospectral density function of the temperature signal. Let be the autospectral density function of the vibration signal. This represents the modulo operation. It should be noted that the autospectral density function... Through temperature spectrum The product of its conjugate is obtained, i.e. ,in for The conjugate of ; self-spectral density function Vibration spectrum The product of its conjugate is obtained, i.e. .

[0022] Traversing frequency range Identify and satisfy , where As a coherence threshold, these frequency points form a set of coherence peak frequencies. .

[0023] Step 300: Based on the coherence peak frequency set, use the spectral peak tracking algorithm to identify potential resonance frequency bands and their drift trends. Specifically, for the set of coherent peak frequencies Each frequency point in In the vibration spectrum Extract the neighborhood of this frequency point The spectral amplitude within, of which This is the width of the frequency window. Calculate the spectral peak value within this neighborhood. and the precise frequency corresponding to the peak value .

[0024] Determine the peak value of the spectrum Does it meet the requirements? ,in The peak discrimination coefficient, For vibration Spectrum In frequency range The average amplitude within the range. If the condition is met, then the frequency... Marked as Potential resonant frequency.

[0025] Continuous The potential resonant frequencies identified within each time window are arranged in chronological order to form a resonant frequency time series. By performing linear fitting on the time series, the resonant frequency drift trend can be obtained. Based on the current resonant frequency and drift trend Determine the center frequency of the resonance frequency band. and frequency bandwidth The frequency bandwidth Determined based on the full width at half maximum (FWHM) of the resonance peak.

[0026] It should be noted that the center frequency of the above-mentioned resonant frequency band Take the current resonant frequency Frequency bandwidth Determined in the vibration spectrum as follows: Find the resonance peak value in the middle. The amplitude on both sides of the corresponding frequency decreases to frequency points and Then the frequency bandwidth .

[0027] Step 400: Based on the resonant frequency band and temperature-vibration data, calculate the optimal stress adjustment parameters using a multi-objective optimization algorithm. Specifically, the stress adjustment parameters are defined including the temperature stress amplitude. and vibration stress frequency A multi-objective optimization function is constructed that simultaneously considers the need to accelerate aging efficiency and avoid resonant frequencies:

[0028] in, The normalized aging acceleration efficiency function characterizes the aging rate under a given stress parameter. The resonance risk function characterizes the degree of proximity between the stress frequency and the resonance frequency; and For the weighting coefficients, satisfying .

[0029] Aging Acceleration Efficiency Function Calculations based on the Arrhenius model and the cumulative damage theory of vibration fatigue:

[0030] in, To activate energy, Boltzmann's constant, For reference temperature, This is the vibration fatigue index. Because... The calculation results include the dimension of frequency (Hz). ), and resonance risk function Since the aging acceleration efficiency function is dimensionless, the two cannot be directly weighted and summed. Therefore, it is necessary to normalize the aging acceleration efficiency function, and a range-based normalization method is used to... Mapped to The normalized aging acceleration efficiency function is obtained from the interval. This eliminates the influence of dimensions on multi-objective optimization and ensures the consistency of dimensions in each term of the optimization function.

[0031] Resonance Risk Function The calculation formula is:

[0032] in, This is the frequency risk bandwidth parameter, with a value of [value to be filled in]. , This indicates the absolute value operation. This function characterizes the stress frequency... The closer to the resonant center frequency The higher the risk level, the greater the risk.

[0033] The optimization constraints include:

[0034] in, and These are the lower and upper limits of temperature stress, respectively. and These are the lower and upper limits of the vibration stress frequency, respectively. To ensure a safe frequency interval, the distance between the stress frequency and the resonant center frequency must be greater than this safe interval.

[0035] The above multi-objective optimization problem is solved using the particle swarm optimization algorithm to obtain the optimal temperature stress amplitude. and optimal vibration stress frequency The particle swarm optimization algorithm takes a multi-objective optimization function, constraints, and particle swarm parameters as input, and outputs the global optimal solution that satisfies the constraints. .

[0036] The execution process of the aforementioned particle swarm optimization algorithm is as follows: Initialize the particle swarm size to... Each particle position vector Represents a set of candidate stress parameters, velocity vector This indicates the direction and magnitude of parameter adjustment. Particle positions are randomly initialized within the constraints. and speed Record the individual optimal position of each particle. and the global optimal position .

[0037] In the In the next iteration, the particles The speed update formula is:

[0038] particle The position update formula is:

[0039] in, This is the inertia weighting coefficient, with a value range of [value range missing]. , and The acceleration coefficient is usually taken as a value of , and for Random numbers within the interval.

[0040] For the updated particle position If it exceeds the range of the constraints, then project it back to the constraint boundary; for the third constraint... If a particle's position violates this constraint, a penalty term is added to the objective function using a penalty function method. The fitness value (i.e., the objective function value) of each particle is calculated, and the individual's optimal position is updated. and the global optimal position The iterative process continues until the objective function converges (the change in the global optimum during successive iterations is less than a threshold). (or reaching the maximum number of iterations) Output the global optimal solution .

[0041] Step 500: Input the stress adjustment parameters into the adaptive PID controller to generate a stress loading command sequence that avoids the resonance region. Specifically, the adaptive PID controller is based on the optimal temperature stress amplitude. Generate temperature control commands According to the optimal vibration stress frequency Generate vibration control commands .

[0042] Temperature control command The calculation formula is:

[0043] in, For temperature error, For real-time measured temperature values, , , These are the proportional, integral, and derivative coefficients for temperature control, respectively.

[0044] Vibration control commands Includes frequency and amplitude commands; the frequency command is directly set to... The amplitude command is adjusted using PID control based on the vibration acceleration error.

[0045] in, For vibration acceleration error, The target vibration acceleration amplitude, The vibration acceleration amplitude is measured in real time. , , These are the proportional, integral, and derivative coefficients for vibration control, respectively.

[0046] Temperature control command and vibration control commands The output is sent to the stress loading device, which performs the stress loading operation according to the instruction sequence.

[0047] In this embodiment of the application, in order to improve the accuracy of resonant frequency identification and the robustness of stress adjustment, the following steps are further included between step 300 and step 400: Step 310: Input the resonant frequency features and temporal degradation features into a two-stream neural network and fuse them to generate a comprehensive degradation state vector. Specifically, time-domain degradation features are extracted, including temperature time-domain sequences. mean Standard deviation and trend coefficient and vibration acceleration time-domain sequence mean Standard deviation and peak factor , forming time-domain feature vectors .

[0048] Extract frequency domain features, including resonant frequencies. Resonance peak Peak frequency domain coherence coefficient and frequency bandwidth , forming frequency domain feature vectors .

[0049] Due to the time-domain feature vector and frequency domain eigenvectors The dimensions and numerical ranges of the various feature components differ significantly (e.g., temperature is in °C, vibration acceleration is in m / s², and frequency is in Hz). To eliminate the influence of dimensions on neural network training, data preprocessing of the feature vectors is necessary. Specifically, for time-domain feature vectors... and frequency domain eigenvectors Perform Z-score standardization on each part to obtain the standardized time-domain feature vectors. and frequency domain eigenvectors This ensures that the mean of each feature component is 0 and the standard deviation is 1, thereby guaranteeing the comparability of features with different dimensions on a numerical scale.

[0050] A two-stream neural network consists of a temporal feature processing branch, a frequency domain feature processing branch, and a feature fusion layer. The input to the temporal feature processing branch is the standardized temporal feature vector. It includes a first fully connected layer, a second fully connected layer, and corresponding activation function layers. The output of the first fully connected layer is... The output of the second fully connected layer is a temporal hidden representation. The input to the frequency domain feature processing branch is the standardized frequency domain feature vector. It includes a first fully connected layer, a second fully connected layer, and corresponding activation function layers. The output of the first fully connected layer is... The output of the second fully connected layer is a frequency domain hidden representation. ,in For the hidden layer dimension.

[0051] The data transfer process of the aforementioned time-domain feature processing branch is represented as follows:

[0052]

[0053] in, This is the weight matrix of the first fully connected layer. This is the bias vector for the first fully connected layer. This is the weight matrix of the second fully connected layer. This is the bias vector for the second fully connected layer. To modify the activation function of the linear unit.

[0054] The data transfer process of the aforementioned frequency domain feature processing branch is represented as follows:

[0055]

[0056] in, This is the weight matrix of the first fully connected layer. This is the bias vector for the first fully connected layer. This is the weight matrix of the second fully connected layer. This is the bias vector for the second fully connected layer.

[0057] Temporal hidden representation Frequency domain hidden representation The features are concatenated to obtain the fused feature vector. Then, a nonlinear transformation is performed through a feature fusion layer to generate a comprehensive degenerate state vector. :

[0058] in, The weight matrix of the feature fusion layer. This is the bias vector for the feature fusion layer. The dimension of the comprehensive degenerate state vector.

[0059] The aforementioned two-stream neural network is trained using historical aging test data. The training samples include time-domain feature vectors, frequency-domain feature vectors, and corresponding degradation labels from historical moments. The degradation labels are obtained by labeling the actual degradation level of the diaphragm pump. The training process employs supervised learning, and the optimization objective is to minimize the mean squared error loss function between the predicted degradation state and the actual degradation label.

[0060] in, The number of training samples. For the first The predicted degradation state vector of each sample, For the first The true degradation label of each sample Let L2 norm be denoted. The Adam optimizer is used to update the network parameters, with a learning rate of 0.001 and a training batch size of 32.

[0061] In step 400, the composite degenerate state vector is... As an additional input, it is used to adaptively adjust the weighting coefficients. and First, adjust the parameter vector using weights. and Calculate the weighted score:

[0062]

[0063] in, and The weighting parameter vector is obtained by regression analysis using the optimal weight coefficients corresponding to different degradation stages in historical aging test data.

[0064] In this embodiment of the application, in order to achieve closed-loop adaptive optimization, the following steps are included after step 500: Step 600: Perform stress adjustment and acquire new temperature-vibration feedback data, update the frequency domain coherence model and performance boundary parameters. Specifically, after the stress loading device performs stress adjustment according to the stress loading command sequence, a time interval is elapsed. Temperature and vibration acceleration data were reacquired to obtain a new temperature spectrum. and vibration spectrum .

[0065] Based on the new spectral data, the frequency domain coherence coefficients were recalculated. Coherence coefficient with history Compare the results and calculate the rate of change in coherence. .

[0066] like ,in If the coherence change threshold is set, then a significant change in the frequency domain coupling relationship is determined, and the coherence threshold is updated. and peak discrimination coefficient The updated formula is:

[0067]

[0068] in, and To update the learning rate, This represents the change in the resonance peak value.

[0069] Simultaneously, monitor the performance boundary parameters of the diaphragm pump, including the maximum allowable temperature. and maximum permissible vibration acceleration If the measured temperature or vibration acceleration exceeds 90% of the performance boundary, then adjust the upper limit value in the optimization constraint conditions. and target vibration acceleration amplitude This ensures that the stress loading does not exceed the safe range.

[0070] Step 700: Based on the updated model parameters and the evolution law of the resonant frequency, iteratively optimize the frequency-stress joint loading strategy for the next cycle. Specifically, the updated coherence threshold Peak discrimination coefficient Using the performance boundary parameters as new input parameters, steps 100 to 500 are re-executed to generate the stress loading command sequence for the next cycle.

[0071] Statistical continuity Resonance frequency evolution data within a period Using LSTM time series forecasting models to predict the future The resonant frequencies of each period are predicted to obtain the predicted resonant frequency sequence. The LSTM time series prediction model takes a historical resonance frequency sequence as input and outputs a predicted resonance frequency value for future time periods.

[0072] The aforementioned LSTM time series prediction model is trained using the resonant frequency evolution sequence in historical aging test data, and training sample pairs are generated using a sliding window approach. The training process employs supervised learning, with the optimization objective being to minimize the mean squared error loss function between the predicted and true values.

[0073] in, The number of training samples. For the first The predicted value for each sample, These are the corresponding true values. The Adam optimizer is used to update the model parameters, with the learning rate set to 0.001.

[0074] In the multi-objective optimization step 400, the predicted resonance frequency sequence is incorporated into the constraints to pre-plan the stress frequency trajectory for multiple future cycles, avoiding the stress frequency adjustment process from passing through the predicted resonance frequency region. The stress frequency sequence obtained through multi-cycle joint optimization is then calculated. This ensures that the stress loading strategy is forward-looking and continuous.

[0075] This implementation method, by introducing a temperature-vibration frequency domain coherence analysis mechanism and quantifying the coupling relationship between temperature and vibration signals in the frequency domain through the cross-spectral density function, can identify frequency domain coupling characteristics that traditional time domain analysis cannot capture, thus overcoming the problem that relying solely on time domain analysis cannot discover frequency domain coupling patterns. Because it employs a peak tracking algorithm to identify and track the resonance frequency and its drift trend in real time, it can dynamically capture the evolution of the diaphragm pump's inherent frequency during aging, thus overcoming the problem of ignoring dynamic changes in resonance characteristics when applying stress at a fixed frequency. Because it constructs a multi-objective optimization function based on resonance frequency constraints, balancing aging acceleration efficiency and resonance risk, and forcing the stress frequency away from the resonance frequency band through constraints, it overcomes the problem of amplified resonance effects caused by stress application frequencies close to the resonance frequency. Because it utilizes an adaptive PID controller to generate stress loading commands based on the optimization results, it achieves refined stress control that avoids the resonance region, thus solving the problem of local fatigue acceleration and non-uniform degradation caused by resonance, and realizing adaptive stress loading control optimized across the entire frequency band.

[0076] By introducing a dual-stream neural network to fuse time-domain and frequency-domain features, a multi-dimensional representation of the degradation state of the hidden membrane pump can be extracted. The weight coefficients of the optimization target are adaptively adjusted, enabling the stress loading strategy to dynamically adjust according to the degree of degradation, thus improving the intelligence and adaptability of stress control. Through a closed-loop feedback mechanism to update the frequency-domain coherence model parameters and performance boundary parameters, and by performing multi-cycle forward-looking optimization based on the resonant frequency evolution law, continuous optimization and long-term stability of the stress loading strategy are achieved, ensuring that the aging test process is always under optimal control.

Claims

1. A dual-parameter intelligent closed-loop diaphragm pump aging test method, characterized in that, Includes the following steps: Acquire real-time temperature and vibration acceleration data of the diaphragm pump, perform fast Fourier transform, and generate temperature and vibration spectra; Based on the temperature spectrum and the vibration spectrum, the cross-spectral density function of the temperature-vibration signal is calculated. The frequency domain coherence coefficient is calculated by the ratio of the square of the modulus of the cross-spectral density function to the product of the temperature signal self-spectral density function and the vibration signal self-spectral density function. Frequency points that satisfy the coherence threshold are extracted to form a set of coherence peak frequencies. For the frequency points in the coherent peak frequency set, the spectral peaks in the neighborhood of the frequency points are extracted in the vibration spectrum. When the spectral peaks meet the peak discrimination conditions, they are marked as potential resonant frequencies. The potential resonant frequencies identified in multiple consecutive time windows are linearly fitted to obtain the resonant frequency drift trend, and the center frequency and frequency bandwidth of the resonant frequency band are determined. A multi-objective optimization function is constructed, simultaneously considering aging acceleration efficiency and resonance risk. The aging acceleration efficiency function is based on the coupling effect of temperature stress and vibration stress frequency, while the resonance risk function characterizes the proximity of the stress frequency to the resonance center frequency. Constraints are set to require that the distance between the stress frequency and the resonance center frequency is greater than the safe frequency interval. The optimal temperature stress amplitude and the optimal vibration stress frequency are obtained by solving the function. The optimal temperature stress amplitude and the optimal vibration stress frequency are then input into an adaptive PID controller to generate a stress loading command sequence that avoids the resonance zone. Perform stress adjustment and acquire new temperature-vibration feedback data. Recalculate the frequency domain coherence coefficient based on the new spectral data. When the coherence change rate exceeds the threshold, update the coherence threshold and peak discrimination coefficient. Iterate and optimize the stress loading strategy for the next cycle. The multi-objective optimization function includes: defining stress adjustment parameters including temperature stress amplitude and vibration stress frequency; constructing an optimization objective function by weighted summation of the negative value of the aging acceleration efficiency function and the resonance risk function, wherein the aging acceleration efficiency function is calculated based on the Arrhenius model and the vibration fatigue cumulative damage theory, characterizing the multiple relationship between the aging rate under given stress parameters and the reference state; the Arrhenius model part calculates the difference between the reciprocal of the temperature stress and the reciprocal of the reference temperature, and the exponential function of the product of the difference and the activation energy; the vibration fatigue cumulative damage part calculates the power of the vibration stress frequency.

2. The aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to claim 1, characterized in that, The steps for calculating the frequency domain coherence coefficients include: Based on the temperature spectrum and vibration spectrum, the temperature-vibration cross-spectral density function is calculated, which characterizes the correlation between the temperature signal and the vibration signal in the frequency domain. The autospectral density function of the temperature signal is calculated, which is obtained by multiplying the temperature spectrum by its conjugate. The self-spectral density function of the vibration signal is calculated, which is obtained by multiplying the vibration spectrum by its conjugate. Calculate the frequency domain coherence coefficient.

3. The aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to claim 1, characterized in that, The steps for determining the center frequency and bandwidth of the resonant frequency band include: For each frequency point in the coherent peak frequency set, extract the spectral amplitude in the neighborhood of that frequency point in the vibration spectrum, and calculate the spectral peak value in that neighborhood and the precise frequency corresponding to the peak value. Determine whether the peak value of the spectrum is greater than the product of the peak discrimination coefficient and the average amplitude of the vibration spectrum. If the condition is met, the corresponding frequency is marked as the potential resonance frequency. The potential resonant frequencies identified within multiple consecutive time windows are arranged in chronological order to form a resonant frequency time series. The resonant frequency time series is then linearly fitted to obtain the resonant frequency drift trend. Based on the current resonance frequency and the resonance frequency drift trend, the center frequency of the resonance frequency band is determined as the current resonance frequency; Find the frequency points on both sides of the resonant peak frequency in the vibration spectrum where the amplitude drops to half of the resonant peak frequency. The frequency bandwidth is the frequency difference between the two frequency points.

4. The aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to claim 1, characterized in that, The multi-objective optimization function also includes: Construct a resonance risk function, which is an exponential function of the square of the difference between the stress frequency and the resonance center frequency divided by the negative value of the frequency risk bandwidth parameter. This function represents the greater the risk value when the stress frequency is closer to the resonance center frequency. Set optimization constraints, including upper and lower limits for temperature stress amplitude, upper and lower limits for vibration stress frequency, and a constraint that the distance between the stress frequency and the resonance center frequency is greater than the safe frequency interval. The objective function is solved using the particle swarm optimization algorithm to obtain the optimal temperature stress amplitude and the optimal vibration stress frequency.

5. The aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to claim 4, characterized in that, The execution process of the particle swarm optimization algorithm includes: Initialize the particle swarm. The position vector of each particle represents a set of candidate stress parameters, including temperature stress amplitude and vibration stress frequency. The velocity vector represents the direction and magnitude of parameter adjustment. Randomly initialize the particle position and velocity within the constraints. Record the individual optimal position and global optimal position of each particle. During the iteration process, the particle velocity update is based on the product of the inertia weight coefficient and the current velocity, plus the product of the first acceleration coefficient, the first random number, and the difference between the individual's optimal position and the current position, plus the product of the second acceleration coefficient, the second random number, and the difference between the global optimal position and the current position. The particle position is updated to the current position plus the updated velocity; For the updated particle position, if it exceeds the constraint range, it is projected back to the constraint boundary; if it violates the constraint between the stress frequency and the resonance center frequency, a penalty term is added to the objective function using the penalty function method. Calculate the fitness value of each particle and update the individual optimal position and the global optimal position; The iterative process continues until the objective function converges or the maximum number of iterations is reached, at which point the global optimal solution is output.

6. The aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to claim 1, characterized in that, Following the step of determining the center frequency and bandwidth of the resonant frequency band, the method further includes: Extract time-domain degradation features, including the mean, standard deviation, and trend coefficient of the temperature time-domain sequence, and the mean, standard deviation, and peak factor of the vibration acceleration time-domain sequence, to form a time-domain feature vector; Extract frequency domain features, including resonance frequency, resonance peak, peak frequency domain coherence coefficient, and frequency bandwidth, to form a frequency domain feature vector; The time-domain feature vector and the frequency-domain feature vector are respectively input into the time-domain feature processing branch and the frequency-domain feature processing branch of the two-stream neural network. Each branch contains two fully connected layers and an activation function layer, which output the time-domain hidden representation and the frequency-domain hidden representation, respectively. The time-domain hidden representation and the frequency-domain hidden representation are concatenated to obtain a fused feature vector, which is then transformed nonlinearly through a feature fusion layer to generate a comprehensive degradation state vector. Using the comprehensive degradation state vector as an additional input, a weighted score is calculated through a weight adjustment parameter vector, and the weight coefficients of aging acceleration efficiency and resonance risk in the multi-objective optimization function are adaptively adjusted.

7. The aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to claim 6, characterized in that, The data transmission process of the time-domain feature processing branch is as follows: The temporal feature vector is linearly transformed through the weight matrix and bias vector of the first fully connected layer, and then the first layer output is obtained by modifying the linear unit activation function; The output of the first layer undergoes a linear transformation using the weight matrix and bias vector of the second fully connected layer, and then the temporal hidden representation is obtained by modifying the linear unit activation function. The data transmission process of the frequency domain feature processing branch is as follows: The frequency domain feature vector is linearly transformed through the weight matrix and bias vector of the first fully connected layer, and then the first layer output is obtained by modifying the linear unit activation function; The output of the first layer is linearly transformed by the weight matrix and bias vector of the second fully connected layer, and then the frequency domain hidden representation is obtained by modifying the linear unit activation function.

8. The aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to claim 1, characterized in that, The steps for iteratively optimizing the stress loading strategy for the next cycle include: After the stress loading device performs stress adjustment according to the stress loading command sequence, it re-acquires temperature data and vibration acceleration data after a preset time interval to obtain new temperature spectrum and vibration spectrum. The frequency domain coherence coefficient is recalculated based on the new spectrum data, and the coherence change rate is calculated by comparing it with the historical coherence coefficient. If the absolute value of the coherence change rate is greater than the coherence change threshold, it is determined that the frequency domain coupling relationship has changed significantly, and the coherence threshold and peak discrimination coefficient are updated. The update of the coherence threshold is the old threshold plus the product of the first update learning rate and the coherence change rate, and the update of the peak discrimination coefficient is the old coefficient plus the product of the second update learning rate and the resonant peak change rate. The performance boundary parameters of the diaphragm pump are monitored, including the maximum allowable temperature and the maximum allowable vibration acceleration. If the measured temperature or vibration acceleration exceeds the preset ratio of the performance boundary, the upper limit value in the optimization constraint condition is adjusted.

9. The aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to claim 1, characterized in that, Also includes: By statistically analyzing the resonance frequency evolution data over multiple consecutive periods, and using a long short-term memory time series prediction model, the resonance frequencies for future periods are predicted to obtain a predicted resonance frequency sequence. In multi-objective optimization, the predicted resonance frequency sequence is incorporated into the constraints to plan the stress frequency trajectory for multiple future cycles in advance, avoiding passing through the predicted resonance frequency region during the stress frequency adjustment process, and calculating the stress frequency sequence for multi-cycle joint optimization.

10. A system for performing the aging test method for a dual-parameter intelligent closed-loop diaphragm pump according to any one of claims 1-9, characterized in that, include: Temperature sensors are used to collect the time-domain temperature sequence of key components in diaphragm pumps. Vibration acceleration sensor, used to acquire time-domain sequences of vibration acceleration; Stress loading device, used to perform stress adjustment according to stress loading command sequence; The data processing unit is used to perform fast Fourier transform, calculate frequency domain coherence coefficients, identify resonant frequency bands, solve multi-objective optimization functions, generate stress loading command sequences, and update model parameters.

Images