Fault prediction model construction method, fault prediction method, equipment and medium

Abstract

The invention discloses a fault prediction model construction method, a fault prediction method, equipment and a medium, and relates to the field of water pump equipment fault prediction. Wherein the historical operation data comprises historical vibration signals, historical temperature parameters and historical current parameters; decomposing the historical vibration signal into a plurality of mode functions by adopting a variational mode decomposition method, and extracting a frequency spectrum characteristic of each mode function, a temperature characteristic of a historical temperature parameter and a current characteristic of a historical current parameter; fusing the spectrum feature, the temperature feature and the current feature to obtain a comprehensive feature vector of the water pump equipment; and training a pre-constructed long-short term memory neural network by using the comprehensive feature vector to obtain a fault prediction model which can be used for predicting the fault of the water pump equipment. According to the invention, the problem of insufficient water pump equipment fault prediction accuracy in related technical methods is solved.

Description

Technical Field

[0001] The present invention relates to the field of pump equipment fault prediction, and more specifically, it relates to a method for constructing a fault prediction model, a fault prediction method, a device, and a medium. Background Art

[0002] Currently, pumps, as important industrial equipment (such as in industrial processes, water supply systems, petrochemicals, etc.), are affected by various factors such as mechanical wear, abnormal vibration, temperature rise, and fluid medium corrosion during long-term operation, which may cause failures or performance degradation.

[0003] Traditional equipment monitoring methods mainly rely on single vibration spectrum analysis, single temperature monitoring, or machine learning algorithms for data fusion, and have the following deficiencies:

[0004] Traditional vibration spectrum analysis methods mainly use sensors to collect pump vibration data, calculate the spectrum distribution through fast Fourier transform (FFT), and then judge whether there is an abnormality according to a preset threshold. Its defects are: FFT can only analyze stationary signals, and for non-stationary and severely noise-interfered vibration signals, the recognition accuracy is low; and a single vibration signal is not sufficient to comprehensively reflect the health status of the equipment.

[0005] Traditional single temperature monitoring methods mainly use temperature sensors to monitor the working temperature of the pump, and an alarm is issued when the set temperature threshold is exceeded. Its defects are: the temperature signal is greatly affected by the environment, and false alarms are likely to occur when used alone, and the temperature change itself cannot directly reflect the mechanical state of the equipment.

[0006] In data fusion and machine learning methods, multi-sensor data fusion is adopted, and the data is comprehensively evaluated through statistical methods, decision trees, or traditional neural networks. Its defects are: lack of self-adaptability and dynamic adjustment mechanisms, failure to fully utilize the long-term dependence of deep learning models on time series data, and inability to meet the requirements of high-precision fault prediction.

[0007] In summary, the existing technical solutions have deficiencies in multi-dimensional data fusion, noise processing, dynamic adaptability, and fault prediction accuracy. Therefore, a new method is urgently needed to improve the accuracy of pump equipment fault prediction. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for constructing a fault prediction model, a fault prediction method, a device, and a medium, which solves the problem of insufficient accuracy of pump equipment fault prediction existing in related technical methods.

[0009] The above technical purpose of the present invention is achieved through the following technical solutions:

[0010] In the first aspect of the present invention, a method for constructing a fault prediction model of a water pump device is provided. The method includes:

[0011] Obtain the historical operation data of the water pump device; wherein, the historical operation data includes historical vibration signals, historical temperature parameters, and historical current parameters;

[0012] Use the variational mode decomposition method to decompose the historical vibration signal into multiple mode functions, and extract the spectral features of each mode function, the temperature features of the historical temperature parameters, and the current features of the historical current parameters;

[0013] Fuse the spectral features, temperature features, and current features to obtain a comprehensive feature vector of the water pump device;

[0014] Use the comprehensive feature vector to train a pre-constructed long short-term memory neural network to obtain a fault prediction model that can be used to predict the faults of the water pump device.

[0015] In one implementation, before using the variational mode decomposition method to decompose the vibration signal into multiple mode functions, the method further includes:

[0016] Perform offset compensation on the central frequency of each iteration of the variational mode decomposition method through the noise sensitivity coefficient and the frequency offset;

[0017] Perform regularization processing on the objective function of each iteration of the variational mode decomposition method through the noise sensitivity coefficient and the frequency change amount; and

[0018] Dynamically adjust the number of modes of each iteration of the variational mode decomposition method through the mode number penalty term, and introduce an energy threshold constraint condition to filter out invalid mode functions in the objective function of each iteration; wherein, the mode number penalty term is the product of the mode penalty coefficient and the total number of mode functions.

[0019] In one implementation, before performing offset compensation on the central frequency of each iteration of the variational mode decomposition method through the noise sensitivity coefficient and the frequency offset, the method further includes:

[0020] Divide the noise into high-frequency noise and low-frequency noise according to the frequency;

[0021] Assign different adjustment weights to the high-frequency noise and the low-frequency noise based on the proportion of the high-frequency components of the wavelet transform;

[0022] Perform weighted calculation on the high-frequency noise and the low-frequency noise according to the adjustment weights to obtain a noise sensitivity coefficient that compensates for noise interference of different scales in the kth mode in the nth iteration.

[0023] In one implementation, the method further includes:

[0024] Obtain the real-time signal-to-noise ratio of the water pump equipment during the transient process;

[0025] Dynamically adjust the noise sensitivity coefficient that compensates for noise interference of different scales through the real-time signal-to-noise ratio to obtain a noise sensitivity coefficient that compensates for noise interference of different scales and time-varying characteristics.

[0026] In one implementation, after offset compensation is performed on the central frequency of each iteration of the variational mode decomposition method through the noise sensitivity coefficient and the frequency offset, the method further includes: optimizing the offset-compensated central frequency through the rotational speed influence coefficient and rotational speed of the water pump equipment.

[0027] In one implementation, before dynamically adjusting the number of modes of each iteration of the variational mode decomposition method through the mode number penalty term, the method further includes: introducing the modal cosine similarity into the mode number penalty term.

[0028] In one implementation, introducing the modal cosine similarity into the mode number penalty term specifically includes:

[0029] Calculate the modal cosine similarity according to the i-th modal function, the j-th modal function, the norm of the i-th modal function, and the norm of the j-th modal function;

[0030] Multiply the modal penalty coefficient, the total number of modal functions, and the modal cosine similarity to obtain the mode number penalty term containing the modal cosine similarity.

[0031] In a second aspect of the present invention, a fault prediction method for a water pump equipment is provided, and the method includes:

[0032] Obtain the real-time operation data of the water pump equipment to be predicted; wherein, the historical operation data includes real-time vibration signals, real-time temperature parameters, and real-time current parameters

[0033] Input the real-time operation data into the fault prediction model constructed by the fault prediction model construction method for a water pump equipment provided in the first aspect of the present invention, and output the fault prediction result of the water pump equipment.

[0034] In a third aspect of the present invention, a computer device is provided, including a processor and a memory storing a computer program. When the computer program is run by the processor, it executes the fault prediction model construction method for a water pump equipment provided in the first aspect of the present invention and the fault prediction method for a water pump equipment provided in the second aspect of the present invention.

[0035] In a fourth aspect of the present invention, a computer-readable storage medium stores instructions that, when run on a computer, cause the computer to execute a method for constructing a fault prediction model of a water pump device provided in the first aspect of the present invention and a method for predicting faults of a water pump device provided in the second aspect of the present invention.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] In a fault prediction model construction method, a fault prediction method, a device, and a medium provided by the present invention, first, the present invention uses the variational mode decomposition method (VMD) to decompose the historical vibration signals of the water pump device to obtain a plurality of mode functions, and then extracts the features of the mode functions, historical temperature parameters, and historical current parameters respectively. Based on the fusion of these features, a comprehensive feature vector of the water pump device is obtained. This comprehensive feature vector not only uses vibration data but also introduces auxiliary parameters such as temperature and current, which can comprehensively evaluate the health state of the water pump device. On this basis, in order to strengthen the attention to the key time steps of fault development, the present invention combines the attention mechanism, enabling the long short-term memory neural network LSTM to focus on the importance of different time steps, thereby enhancing the accuracy of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, form a part of this application, and do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0039] Figure 1 It is a schematic flowchart of a method for constructing a fault prediction model provided by an embodiment of the present invention;

[0040] Figure 2 It is a schematic flowchart of a method for predicting faults of a water pump device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in combination with embodiments and drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and do not limit the present invention.

[0042] It should be noted that the term "comprising" or "may comprise" that can be used in various embodiments of the present application indicates the presence of the claimed functions, operations, or elements, and does not limit the addition of one or more functions, operations, or elements. In addition, as used in various embodiments of the present application, the terms "comprising", "having" and their cognates are only intended to mean specific features, numbers, steps, operations, elements, components, or combinations of the foregoing items, and should not be construed as precluding the existence or addition of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing items first.

[0043] It should be understood that terms such as "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality of" means two or more, unless otherwise specifically defined.

[0044] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a method for constructing a fault prediction model provided by an embodiment of the present invention. As Figure 1 shown, the method includes:

[0045] S101, obtaining historical operation data of the water pump device; wherein, the historical operation data includes historical vibration signals, historical temperature parameters, and historical current parameters.

[0046] In this embodiment, vibration signals and temperature parameters are collected by vibration sensors and temperature sensors arranged at different positions of the water pump device, and the current parameters of the water pump device can be collected by a current sensor. These are all conventional technical means well known to those skilled in the art, and no redundant description is made in this embodiment.

[0047] Secondly, the vibration signals (including vibration frequency, amplitude, etc.) collected by the sensors have different dimensions and amplitude ranges. In order to ensure data consistency and processing efficiency, Min-Max normalization processing is used to preprocess the historical operation data.

[0048] The expression of Min-Max normalization is: x′ i is the data after normalization processing (normalized to the interval [0, 1]). x i is the original data value, coming from the sensor. x min is the minimum value of all samples in the dataset. x maxis the maximum value of all samples in the dataset. This formula is a well-known Min-Max normalization formula, which can be used to standardize the original data of vibration sensors to ensure the unified dimension of data from different sensors. Its innovation can be reflected in combining the statistical characteristics of specific data (such as dynamic window adaptive normalization), and continuously optimizing the updates of x min and x max so as to dynamically adjust the normalization parameters in different time periods and scenarios.

[0049] Divide the continuous time series data into multiple windows according to a fixed length T for subsequent model training. Each window of data is represented as: where: X (j) : the jth time window, T: the window length (for example, data of 48 hours).

[0050] S102. Use the variational mode decomposition method to decompose the historical vibration signal into multiple mode functions, and extract the spectral features of each mode function, the temperature features of historical temperature parameters, and the current features of historical current parameters.

[0051] In this embodiment, an improved adaptive variational mode decomposition (IVMD) algorithm is used to decompose the vibration signal into several mode functions (IMFs) to eliminate the influence of noise.

[0052] The basic variational mode decomposition formula is:

[0053] where, u k (t) is the kth mode function, ω k is the corresponding central frequency, K is the number of mode functions, * represents convolution, δ(t) is the Dirac function, and ||·||2 is the second norm.

[0054] The above describes the basic situation of the traditional VDM algorithm. This embodiment proposes an adjustment mechanism for adaptively adjusting the central frequency and the number of modes on the basis of the traditional VMD algorithm, so that the effective components of the vibration signal can be stably decomposed under different working conditions, and the noise interference in the vibration signal can be reduced.

[0055] In one embodiment, before decomposing the vibration signal into a plurality of mode functions by using the variational mode decomposition method, the method further includes: performing offset compensation on the central frequency of each iteration of the variational mode decomposition method through a noise sensitivity coefficient and a frequency offset; performing regularization processing on the objective function of each iteration of the variational mode decomposition method through a noise sensitivity coefficient and a frequency change amount; and dynamically adjusting the number of modes of each iteration of the variational mode decomposition method through a mode number penalty term, and introducing an energy threshold constraint condition to filter out invalid mode functions in the objective function of each iteration; wherein, the mode number penalty term is the product of a mode penalty coefficient and the total number of mode functions.

[0056] Specifically, 1. The adjustment mechanism of the adaptive central frequency is as follows:

[0057] In the Lagrangian function of the traditional VMD, the central frequency ω k is usually updated by fixed iteration. The improved adaptive mechanism introduces a dynamic frequency offset compensation term and updates the central frequency to:

[0058] where: γ is the noise sensitivity coefficient (adaptive adjustment parameter), Δω noise is the frequency offset based on noise estimation, and the superscript (n) represents the nth iteration.

[0059] 2. The adjustment mechanism of the adaptive number of modes is as follows:

[0060] A mode number penalty term is added to the objective function of the traditional VMD, and a dynamic termination criterion is introduced:

[0061]

[0062] The constraint condition becomes: and where: λ is the mode number penalty coefficient, σ noise is the estimated value of the noise standard deviation, and β is the signal-noise energy ratio threshold

[0063] The variational problem of the improved VDM algorithm with the adjustment mechanisms of adaptively adjusting the central frequency and the adaptive number of modes can be expressed as:

[0064]

[0065] Constraint condition:

[0066] In this embodiment, through the noise sensitivity coefficient γ and the frequency offset Δω noise , the adaptive correction of the central frequency to noise interference is realized.

[0067] In this embodiment, the number of modes K and the noise level σ noiseCoupling is used to avoid over - decomposition or under - decomposition.

[0068] In this embodiment, the effective modes are dynamically screened through the β parameter to ensure the physical meaning of the decomposition result.

[0069] Introduce the frequency change amount of the regularization term to prevent frequency oscillation.

[0070] This expression realizes a dual - adaptive mechanism through mathematical modeling, and has significant improvements in noise suppression ability and decomposition stability compared with the traditional VMD.

[0071] The core logic of the adaptive mechanism provided in this embodiment is: through the noise - sensitive coefficient γ and the frequency offset Δω noise , the frequency estimation deviation caused by noise is corrected in real - time to ensure that the modal center frequency fits the true vibration characteristics (such as the characteristic frequency of the water - pump bearing fault). Through the modal number penalty term λ·K·σ noise couple K with the noise level, and the stronger the noise, the fewer the allowed modal numbers. Also, through the energy threshold constraint dynamically filter out invalid modes to avoid under - decomposition or over - decomposition.

[0072] Through the collaborative optimization of the objective function and the constraint conditions, the two major adaptive mechanisms enable the I - VMD to achieve in the decomposition of the water - pump vibration signal: the noise suppression ability is improved by 40% (compared with the traditional VMD, the proportion of the noise mode drops from 35% to 21%); the decomposition stability is increased by 50% (the amplitude of the frequency oscillation is reduced by 60%).

[0073] According to the description of the improved adaptive variational mode decomposition (I - VMD) algorithm, for further fitting the operating characteristics of the water - pump equipment, further optimization is carried out as follows:

[0074] Before offset - compensating the center frequency of each iteration of the variational mode decomposition method through the noise - sensitive coefficient and the frequency offset, the method further includes: dividing the noise into high - frequency noise and low - frequency noise according to the frequency; giving different adjustment weights to the high - frequency noise and the low - frequency noise based on the proportion of the high - frequency components of the wavelet transform; performing weighted calculation on the high - frequency noise and the low - frequency noise according to the adjustment weights to obtain the noise - sensitive coefficient that compensates for the noise interference of different scales in the k - th mode in the n - th iteration.

[0075] Specifically, this embodiment introduces multi - scale noise separation estimation, divides the noise into high - frequency and low - frequency components according to the frequency, and gives different adjustment weights respectively. The center - frequency update formula is:

[0076] Among them, is the noise - sensitive coefficient (adaptive adjustment parameter) of the k - th mode in the n - th iteration; ∈ k，highis the high-frequency noise sensitivity coefficient; ∈ k，low is the low-frequency noise sensitivity coefficient; H(f) is the high-frequency noise weight function (such as the high-frequency component ratio function based on wavelet transform).

[0077] This embodiment enables the center frequency to more accurately compensate for noise interference at different scales and is applicable to the complex noise environment where the pump has the coupling of fluid impact (high-frequency noise) and mechanical low-frequency vibration.

[0078] Secondly, the method further includes: obtaining the real-time signal-to-noise ratio of the pump equipment during the transient process; dynamically adjusting the noise sensitivity coefficient with the ability to compensate for noise interference at different scales through the real-time signal-to-noise ratio to obtain a noise sensitivity coefficient with the ability to compensate for noise interference at different scales and time-varying characteristics.

[0079] Specifically, for the time-varying characteristics of the pump operating conditions (such as load changes), a time-varying noise sensitivity coefficient is designed, and its expression is as follows: where ∈0 is the initial noise sensitivity coefficient; both α and β are adjustment coefficients; t is time; SNR(t) is the real-time signal-to-noise ratio. This embodiment enables the real-time signal-to-noise ratio to be dynamically adjusted with time t and signal quality, avoiding the decomposition failure caused by fixed parameters during transient processes such as pump startup and shutdown, and improving the robustness of the algorithm.

[0080] After offsetting and compensating the center frequency of each iteration of the variational mode decomposition method through the noise sensitivity coefficient and the frequency offset, the method further includes: optimizing the offset-compensated center frequency through the speed influence coefficient and speed of the pump equipment.

[0081] Specifically, by integrating the inherent physical parameters of the pump equipment (such as the speed R), the update formula of the center frequency described above is improved: where is the center frequency of the k-th mode in the (n + 1)-th iteration; is the value of the k-th mode function at time t in the n-th iteration; Δω noise is the frequency offset based on noise estimation; γ is the speed influence coefficient; R is the pump speed.

[0082] This embodiment enables the center frequency to conform to the actual physical vibration characteristics of the pump (such as the impeller rotation frequency being strongly correlated with the speed), avoiding the confusion of the modal physical meaning caused by pure data-driven decomposition, and enhancing the correlation between the spectral characteristics and the fault type.

[0083] Finally, before dynamically adjusting the number of modes in each iteration of the variational mode decomposition method through the mode number penalty term, the method further includes: introducing modal cosine similarity into the mode number penalty term, specifically: calculating the modal cosine similarity according to the i-th mode function, the j-th mode function, the norm of the i-th mode function, and the norm of the j-th mode function; multiplying the mode penalty coefficient, the total number of mode functions, and the modal cosine similarity to obtain the mode number penalty term containing the modal cosine similarity.

[0084] Specifically, introducing modal cosine similarity into the mode number penalty term suppresses over-decomposition of the variational mode decomposition method:

[0085] where λ is the mode number penalty coefficient; K is the total number of mode functions; IMF i 、IMF j are the i-th and j-th mode functions respectively; ||IMF i ||、||IMF j || are the norms (energy magnitudes) of IMF i 、IMF j respectively. When the newly generated mode is highly similar to the existing modes (such as redundant modes caused by noise), the penalty term increases significantly, forcing the algorithm to terminate the meaningless decomposition and ensuring that each IMF corresponds to a real physical feature (such as independent fault modes like bearing faults and pump body imbalance).

[0086] The above embodiments form a mechanism of "data-driven + physical constraint + time-varying adaptation", which closely combines the operating characteristics of the water pump equipment, so that the variational mode decomposition method can stably decompose the effective signal components under different working conditions and reduce noise interference.

[0087] Secondly, regarding the extraction of the spectral features of the mode functions, the specific implementation is as follows: For each mode function u k (t), the fast Fourier transform (FFT) is used to extract the spectral features: where: X(k) is the amplitude of the spectrum at the k-th frequency point, u k (n) is the value of the mode function at the n-th moment, N is the number of sampling points, and j is the imaginary unit.

[0088] The spectral features include: Peak Frequency: The frequency point with the largest amplitude in the spectrum, representing the main vibration component.

[0089] Spectral Energy: where E is the spectral energy of the mode function.

[0090] Spectral Centroid: This index reflects the center of gravity position of the vibration signal spectrum. These spectral features reflect the physical characteristics of the water pump vibration state and are important inputs for subsequent fault prediction.

[0091] For the feature extraction of historical temperature parameters and historical current parameters, it is realized by normalization. The extracted temperature features include average temperature, temperature volatility, etc., and the extracted current features include mean value, standard deviation, etc.

[0092] S103, fuse the spectral features, temperature features and current features to obtain the comprehensive feature vector of the water pump equipment.

[0093] In this embodiment, the attention mechanism is introduced to automatically assign different weights to different features and fuse multi-dimensional features. The formula is as follows: Where: e i is the attention score of the i-th feature, F i is the i-th feature vector, W i is the feature weight matrix, b i is the bias, v i is the context vector.

[0094] Normalized attention weight:

[0095] Finally, the comprehensive feature vector is obtained:

[0096] In this way, the importance of each data source can be automatically learned, and the overall prediction performance can be improved.

[0097] S104, use the comprehensive feature vector to train the pre-constructed long short-term memory neural network to obtain a fault prediction model that can be used to predict the faults of water pump equipment.

[0098] In this embodiment, the LSTM model is used to process the fused time series data to realize the prediction of the health state of the water pump. Assume that the input time series is {F fusion (t)}, and the model output is the health score or fault probability within a future period of time.

[0099] The main calculation steps of LSTM are as follows:

[0100] LSTM forget gate: f t =σ(W f ·[h t-1 , x t +b f ), where f t is the output of the forget gate at the current time step, indicating how much past information to retain. The output range is [0,1]. σ is the activation function, representing the switch of the gating mechanism. The formula is: W f is the weight matrix that controls how the input data and the hidden state from the previous time step are combined. h t-1 is the hidden state from the previous time step. x t is the input at the current time step (vibration data or other sensor data). b f is the bias term used to adjust the output of the neural network.

[0101] This formula is the classical LSTM forget gate formula, but it can be optimized according to the present invention by adopting a local gating mechanism, that is, the W f and b f are adaptively adjusted according to the different positions of the device and the vibration data, rather than fixed weights, to improve the prediction ability for different positions of the water pump device.

[0102] LSTM input gate: i t = σ(W i · [h t-1 , x t + b i ), where i t is the output of the input gate at the current time step, which determines how important the current input information is. W i is the weight matrix of the input gate that controls how the input data and the state from the previous time step are combined. b i is the bias term of the input gate.

[0103] In this embodiment, the output of the input gate can be combined with real-time monitored external environmental variables (such as temperature, humidity, etc.), and by introducing external environmental factors, the weight W i of the input gate is dynamically adjusted to make the fault prediction under different environmental conditions more accurate.

[0104] LSTM output gate: o t = σ(W o · [h t-1 , x t + b o ), where o t is the output of the output gate at the current time step, which determines the output of the current memory cell to the next step. W o is the weight matrix of the output gate. b o is the bias term of the output gate.

[0105] LSTM cell state update: where C t is the cell state at the current time step, storing long-term memory. C t-1 is the cell state from the previous time step. is the new candidate cell state for the current input, usually calculated by a tanh function. f t and i t are the outputs of the forget gate and the input gate respectively, determining how much past information and current information to store.

[0106] The final output of the LSTM: h t = o t ·tanh(C t ), where h t is the hidden state at the current time step, serving as the output value and passed to the next time step or used for final prediction.

[0107] To strengthen the attention to the critical time steps for fault prediction development, the attention mechanism is combined, enabling the LSTM to focus on the importance of different time steps, thereby enhancing the accuracy of fault prediction.

[0108] Based on the LSTM model, the attention mechanism is combined to make the model more focused on the critical time steps and enhance the accuracy of crack prediction. The attention mechanism highlights the important time periods affecting crack occurrence by weighting the influence of each time step.

[0109] Attention weighting for the LSTM hidden state:

[0110] Attention score: e t = v T tanh(W a ·h t + b a ), where e t is the attention score for the current time step, indicating the importance of this time step in crack prediction. v T is the attention weight vector. h t is the hidden state output of the LSTM. W a is the weight matrix of the attention layer for processing the output of each time step, b a is the bias term.

[0111] Normalized attention weights: where α t is the normalized attention weight, indicating the importance of time step t in the entire sequence, and the attention mechanism makes this step more prominent.

[0112] Through the attention mechanism, the LSTM can automatically focus on key features, thereby improving the accuracy of fault prediction.

[0113] Weighted hidden state:

[0114] After being processed by the LSTM and attention mechanism, the final prediction result is the probability of fault occurrence calculated based on the weighted hidden state \(A_t\).

[0115] Predicted output: Among them, is the predicted probability of fault occurrence, usually in the range of \([0, 1]\), \(A\) t is the weighted hidden state (obtained by weighted calculation of the LSTM output and the attention mechanism). \(W\) o is the output layer weight matrix, \(b\) o is the bias term.

[0116] For the training of the LSTM neural network, the mean squared error (MSE) is used as the loss function, and its expression is: Among them, \(N\) is the number of samples, is the predicted value, \(\hat{y}\) i is the actual state label.

[0117] The model parameters are continuously updated through the backpropagation algorithm to optimize the prediction accuracy. The larger the value, the higher the probability of fault occurrence.

[0118] Based on the content described in the above embodiments, in the method provided by the embodiments of the present invention, through the improved adaptive VMD, FFT spectrum analysis, LSTM and attention mechanism, and the multi-dimensional data fusion and weighting method, a full-process closed-loop method from data acquisition, preprocessing to fault prediction and decision support is constructed. This method still maintains high-precision fault prediction ability in a high-noise environment, can early warn of equipment faults, and provide a scientific basis for maintenance decisions, thereby reducing equipment failure rates and maintenance costs, and improving the overall operation efficiency of equipment. The present invention overcomes the defects of traditional single-signal analysis, fixed-threshold monitoring, and static scheduling methods, and has remarkable real-time performance, accuracy, and self-adaptability. This system is not only applicable to pump equipment, but also can be extended to the health monitoring field of other industrial equipment, and has broad application prospects.

[0119] Please refer to Figure 2 , Figure 2 which is a schematic flow diagram of a fault prediction method for a pump equipment provided by an embodiment of the present invention. As Figure 2 shown, the method includes:

[0120] S201, obtaining the real-time operation data of the pump equipment to be predicted; among them, the historical operation data includes real-time vibration signals, real-time temperature parameters, and real-time current parameters.

[0121] In this embodiment, vibration signals and temperature parameters are collected by vibration sensors and temperature sensors arranged at different positions of the water pump equipment, and the current parameters of the water pump equipment can be collected by current sensors. These are all conventional technical means well-known to those skilled in the art, and no redundant description is given in this embodiment.

[0122] S202, input the real-time operation data into the fault prediction model constructed by the fault prediction model construction method of a water pump equipment described in the above embodiment, and output the fault prediction result of the water pump equipment.

[0123] In this embodiment, based on the fault prediction result output by the fault prediction model of LSTM and attention mechanism, a health score and a fault warning are generated, and the prediction result is normalized and mapped to the 0-100 score range. If the predicted value exceeds the set safety threshold, the system automatically triggers an alarm and generates a detailed report, including the prediction time, fault probability, recommended maintenance measures, etc.

[0124] For example, the health score calculation formula is: where is the predicted fault probability, and S is the health score. If S is lower than 80 points, a warning is issued.

[0125] The embodiment of the present invention also provides a computer device. Among them, the computer device includes a processor, a memory, a communication interface, and at least one communication bus for connecting the processor, the memory, and the communication interface. Among them, the memory includes but is not limited to a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (PROM), or a portable read-only memory (CD-ROM), and this memory is used for relevant instructions and data.

[0126] The communication interface is used to receive and send data. The processor can be one or more CPUs. In the case where the processor is a single CPU, the CPU can be a single-core CPU or a multi-core CPU. The processor in the electronic device is used to read one or more programs stored in the memory and perform the following operations: obtain the historical operation data of the water pump equipment; wherein, the historical operation data includes historical vibration signals, historical temperature parameters, and historical current parameters; decompose the historical vibration signals into multiple modal functions by using the variational mode decomposition method, and extract the spectral features of each modal function, the temperature features of the historical temperature parameters, and the current features of the historical current parameters; fuse the spectral features, temperature features, and current features to obtain a comprehensive feature vector of the water pump equipment; use the comprehensive feature vector to train a pre-constructed long short-term memory neural network with an attention mechanism to obtain a fault prediction model that can be used to predict the faults of the water pump equipment.

[0127] It should be noted that the specific implementation of each operation can be the above Figure 1For the corresponding descriptions of the method embodiments shown, the electronic device can be used to execute a method for constructing a fault prediction model of a water pump device and a fault prediction method of a water pump device in the above method embodiments of the present application, which will not be elaborated here.

[0128] The embodiments of the present invention also provide a computer-readable storage medium. The computer-readable storage medium is a memory device in a computer device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and, of course, the extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space, and this storage space stores the operating system of the terminal. And, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps in the above embodiments regarding a method for constructing a fault prediction model of a water pump device and a fault prediction method of a water pump device. Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can adopt the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0129] The embodiments of the present invention also provide a computer program product containing program instructions. The computer program product can be software or a program product containing program instructions that can run on a computing device or be stored in any available medium. When the computer program product runs on at least one computer device, at least one computer device is enabled to execute a method for constructing a fault prediction model of a water pump device and a fault prediction method of a water pump device.

[0130] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for constructing a fault prediction model of a water pump device, characterized in that the method Including: Obtaining historical operation data of a water pump device; wherein, the historical operation data includes historical vibration signals, historical temperature parameters, and historical current parameters; Using variational mode decomposition to decompose the historical vibration signal into multiple mode functions, and extracting the spectral features of each mode function, the temperature features of the historical temperature parameters, and the current features of the historical current parameters; Fusing the spectral features, temperature features, and current features to obtain a comprehensive feature vector of the water pump device; Using the comprehensive feature vector to train a pre-constructed long short-term memory neural network with an attention mechanism to obtain a fault prediction model that can be used to predict faults of the water pump device.

2. The method for constructing a fault prediction model of a water pump device according to claim 1, characterized in that Before using variational mode decomposition to decompose the vibration signal into multiple mode functions, the method further includes: Performing offset compensation on the central frequency of each iteration of variational mode decomposition through a noise sensitivity coefficient and a frequency offset; Regularizing the objective function of each iteration of variational mode decomposition through a noise sensitivity coefficient and a frequency change; and Dynamically adjusting the number of modes of each iteration of variational mode decomposition through a mode number penalty term, and introducing an energy threshold constraint condition to filter out invalid mode functions in the objective function of each iteration; wherein, the mode number penalty term is the product of a mode penalty coefficient and the total number of mode functions.

3. The method for constructing a fault prediction model of a water pump device according to claim 2, characterized in that, Before performing offset compensation on the central frequency of each iteration of variational mode decomposition through a noise sensitivity coefficient and a frequency offset, the method further includes: Dividing the noise into high-frequency noise and low-frequency noise according to frequency; Assigning different adjustment weights to the high-frequency noise and the low-frequency noise based on the proportion of high-frequency components of wavelet transform; Performing weighted calculation on the high-frequency noise and the low-frequency noise according to the adjustment weights to obtain a noise sensitivity coefficient that compensates for noise interference of different scales in the k-th mode of the n-th iteration.

4. A method for constructing a fault prediction model of a water pump device according to claim 3, characterized in that, The method further includes: Obtaining the real-time signal-to-noise ratio of the water pump device in the transient process; Dynamically adjusting the noise sensitivity coefficient that compensates for noise interference of different scales through the real-time signal-to-noise ratio to obtain a noise sensitivity coefficient that compensates for noise interference of different scales and has time-varying characteristics.

5. A method for constructing a fault prediction model of a water pump device according to claim 2, characterized in that, After performing offset compensation on the central frequency of each iteration of variational mode decomposition through a noise sensitivity coefficient and a frequency offset, the method further includes: optimizing the offset-compensated central frequency through the rotational speed influence coefficient and rotational speed of the water pump device.

6. The method for constructing a fault prediction model of a water pump device according to claim 2, wherein, Before dynamically adjusting the number of modes of each iteration of variational mode decomposition through a mode number penalty term, the method further includes: introducing modal cosine similarity into the mode number penalty term.

7. A method for constructing a fault prediction model of a water pump device according to claim 6, characterized in that, Introducing modal cosine similarity into the mode number penalty term specifically includes: Calculating the modal cosine similarity according to the i-th mode function, the j-th mode function, the norm of the i-th mode function, and the norm of the j-th mode function; Multiplying the mode penalty coefficient, the total number of mode functions, and the modal cosine similarity to obtain a mode number penalty term containing modal cosine similarity.

8. A fault prediction method for a water pump device, characterized in that the method Including: Obtaining real-time operation data of the water pump device to be predicted; wherein, the historical operation data includes real-time vibration signals, real-time temperature parameters, and real-time current parameters; Input the real-time operation data into the fault prediction model constructed by the fault prediction model construction method of a water pump device according to any one of claims 1 to 7, and output the fault prediction result of the water pump device.

9. A computer device, characterized in that, It includes a processor and a memory storing a computer program. When the computer program is run by the processor, it executes the fault prediction model construction method of a water pump device according to any one of claims 1 to 7 and the fault prediction method of a water pump device according to claim 8.

10. A computer-readable storage medium, characterized in that, Instructions are stored. When the instructions are run on a computer, the computer is made to execute the fault prediction model construction method of a water pump device according to any one of claims 1 to 7 and the fault prediction method of a water pump device according to claim 8.

Images