Data-driven floating ring seal multi-source feature fusion life evaluation method and device

Abstract

The present application relates to the technical field of aero-engine sealing, and particularly relates to a data-driven floating ring seal multi-source feature fusion life evaluation method and device, wherein the method comprises: obtaining and preprocessing multi-channel operation data of a floating ring graphite seal to obtain a smoothed signal; extracting, reducing dimensions and fusing multi-domain degradation features in the smoothed signal to obtain a fused health index; respectively adopting a support vector regression model, an exponential degradation model and a polynomial regression model to fit the fused health index to obtain a final mixed prediction model; obtaining a real-time health index sequence and inputting the real-time health index sequence into the final mixed prediction model to predict a degradation trajectory; extrapolating the degradation trajectory until the degradation trajectory intersects with a preset health index threshold to determine a predicted failure time point, and calculating a remaining service life and a confidence interval thereof according to the predicted failure time point. Thus, the problems that the floating ring graphite seal cannot efficiently and accurately quantify the nonlinear influence of actual complex and changeable working conditions on the sealing life in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of aero-engine sealing, and in particular to a data-driven floating ring seal multi-source feature fusion life evaluation method and device based on data driving and multi-source feature fusion. BACKGROUND

[0002] Floating ring graphite sealing is an indispensable core component in high-end rotating machinery such as aero-engines, gas turbines and centrifugal compressors, and undertakes the key missions of isolating high and low pressure chambers, preventing process medium leakage and ensuring the efficient and stable operation of the unit. The floating ring graphite sealing is an integral, inlaid and non-contact circumferential graphite sealing, and is a non-contact type shaft seal combining gap sealing and end face sealing, which is composed of a floating graphite sealing ring, a gasket, a wave spring, a retainer and a housing. As an advanced non-contact fluid dynamic pressure sealing, its working principle is ingenious and complex: between the stationary graphite ring and the high-speed rotating shaft neck, a shallow groove (such as a spiral groove, a stepped groove, etc.) on the shaft is used to generate a pumping effect, and a small amount of sealing medium is compressed to form a rigid gas film with a thickness of only a few microns. The opening force and closing force (spring force and medium pressure) generated by the gas film reach a dynamic microsecond balance, so that the floating ring can be "suspended" on the rotating shaft, thereby realizing the running state of almost zero wear. It is this unique working mechanism that gives it the remarkable advantages of small leakage and long theoretical life.

[0003] However, the long life vision in theory faces great challenges in the extremely harsh working environment in practice. The floating ring graphite sealing does not operate under ideal conditions, and its location is often one of the most severe areas in the entire unit. First, it needs to withstand extremely high linear speed, and in modern high-speed machinery, the relative speed of the sealing surface can easily exceed 150 meters per second, even reaching more than 200 meters per second, and the huge shearing action and friction heat are the ultimate test of the material properties. Second, high temperature is another severe test, and the sealing medium is often hot gas or high-temperature synthesis gas at several hundred degrees Celsius, with an ambient temperature of 400°C to 600°C, which not only accelerates the oxidation and wear of the graphite material, but also causes the thermal physical properties of the material to degrade and produce thermal deformation, damaging the precise sealing gap. Third, the high pressure difference environment is its intrinsic characteristic, and the sealing needs to stably withstand a pressure difference of dozens of atmospheres, which is the driving force for forming the gas film and the potential risk source leading to instability. In addition, the medium itself may be corrosive (such as containing H2S, CO2, etc.), or carry fine solid particles, or be flammable and explosive hydrogen or hydrocarbon gas, and these factors all have complex physical and chemical reactions with the sealing material, exacerbating its performance degradation. More complex is that the start-stop cycle, load fluctuation, rotor vibration, surge and other transient conditions in the unit operation, in addition to this, the service life of the floating ring graphite sealing is mainly affected by the following aspects:

[0004] (1) The rubbing of the main sealing surface will cause the inner diameter of the graphite ring to wear, the leakage will increase, and the service life of the floating ring will decrease;

[0005] (2) The fretting wear of the secondary sealing surface will increase the volume wear, the boss height will decrease, the wave spring elasticity will attenuate, the secondary sealing surface will decrease the fit, the leakage will increase, and the sealing will fail;

[0006] (3) The wave spring elasticity will attenuate, the secondary sealing surface will decrease the fit, the leakage will increase, and the sealing will fail;

[0007] (4) The inlay ring will produce creep displacement at high temperature, the inner diameter of the graphite ring will increase, the leakage will increase, and the sealing will fail.

[0008] (5) The aging of graphite material, runway wear, aging deformation of shell material, and aging and wear of gasket material will destroy the stability of the sealing gap and the dynamic tracking ability of the floating ring, causing the leakage to increase and the friction and wear to intensify, resulting in sealing failure.

[0009] Under the influence of these factors, the balance of the sealing will be constantly impacted and broken, causing the floating ring to vibrate and rub, and leading to gradual performance degradation and ultimately failure.

[0010] However, accurate life assessment and prediction of floating ring graphite seals has always been a major challenge for the industry, with the main difficulties being: (1) The cost of experimental verification is extremely high. Building a test bench that can simulate real extreme conditions (such as high speed, high temperature, and high pressure) requires significant economic investment, and the single test cycle is long, making it difficult to obtain sufficient failure data to support traditional life assessment based on physical models. (2) The failure mechanism is complex and strongly coupled with multiple factors. The degradation of the seal is the result of the combined action of mechanical stress, thermal load, chemical corrosion, material wear, and other multi-physical fields, and any single theoretical model or empirical formula is difficult to fully and accurately describe the complex nonlinear degradation process, resulting in low accuracy and poor universality of traditional theoretical calculation methods. Therefore, it is urgent to develop a new intelligent life assessment method to break through the limitations of traditional testing and physical models.

[0011] It is the cornerstone of the transition from traditional "time-based maintenance" to advanced "condition-based maintenance", and through accurate life prediction, it can scientifically plan the maintenance window, maximize the effective life of the component, avoid the waste of spare parts and manpower caused by "over maintenance", and prevent unplanned shutdowns and serious accidents caused by "insufficient maintenance", thereby significantly reducing the life cycle cost of equipment and improving operational efficiency.

[0012] Despite the urgent need, the existing traditional life assessment methods have many inherent defects that are difficult to overcome and cannot meet the requirements of modern industry for prediction accuracy. The first type is the empirical estimation method, which relies on historical maintenance records or manufacturer recommendations and gives a fixed operating hours as the life indicator. This method is extremely rough and conservative, completely ignoring the uniqueness and volatility of the actual operating conditions of each device. To ensure "absolute safety", the life value set is often much shorter than the actual physical life of the seal, resulting in a large number of healthy components being replaced too early, causing huge waste; on the contrary, if the device operating conditions deteriorate, this method cannot provide any warning. The second type is the bench test method, which infers the life by simulating the operating conditions on the test bench to conduct accelerated life tests. This method can provide a reference, but it is extremely costly and time-consuming, and the biggest problem is "simulation distortion" - the laboratory environment is almost impossible to completely reproduce all the complex transient, variable and boundary conditions on site, resulting in a significant reduction in the accuracy of the test results when extrapolated to actual applications. The third type is the wear calculation method based on physical models, which tries to describe the wear process through mathematical formulas. However, the degradation of the floating ring seal is a highly nonlinear process involving strong coupling of multiple physical fields (flow, solid, heat, force, and chemistry), with many influencing factors and complex interaction mechanisms. Any simplified model is difficult to fully and truly reflect its nature, resulting in a large error in the prediction results and limited engineering practicability.

[0013] In summary, the core defect of the existing technology is that they cannot efficiently and accurately quantify the nonlinear impact of actual complex and variable operating conditions on seal life. The prediction results are either overly conservative or lack accuracy, and they always fail to find the best balance between "avoiding waste" and "ensuring safety". This long-standing technical bottleneck has seriously restricted the improvement of intelligent operation and maintenance of high-end equipment, and an innovative methodology that can deeply integrate real-time operating data and autonomously learn and capture complex nonlinear degradation rules from data is urgently needed to achieve a breakthrough. SUMMARY

[0014] The present application provides a data-driven floating ring seal multi-source feature fusion life assessment method and device to solve the problem that the existing floating ring graphite seal cannot efficiently and accurately quantify the nonlinear impact of actual complex and variable operating conditions on seal life.

[0015] The first aspect embodiment of the present application provides a data-driven floating ring seal multi-source feature fusion life assessment method, comprising the following steps:

[0016] The multi-channel operation data of the target floating ring graphite seal is acquired and pre-processed to obtain a smoothed signal; multi-domain degradation features in the smoothed signal are extracted, and the multi-domain degradation features are reduced in dimension and fused to obtain a fused health index; a support vector regression model, an exponential degradation model and a polynomial regression model are respectively used to fit the fused health index to obtain a final mixed prediction model; a real-time health index sequence of the target floating ring graphite seal is acquired, and the real-time health index sequence is input into the final mixed prediction model to predict a degradation trajectory; the degradation trajectory is extrapolated until it intersects with a preset health index threshold to determine a predicted failure time point, and the remaining useful life and its confidence interval are calculated according to the predicted failure time point.

[0017] Optionally, the acquiring and pre-processing the multi-channel operation data of the target floating ring graphite seal to obtain a smoothed signal comprises:

[0018] The multi-channel operation data of the target floating ring graphite seal is acquired, wherein the multi-channel operation data comprises vibration acceleration, gas temperature, gas high and low pressure side force, dynamic displacement between the floating ring seal and the eddy current sensor probe; the multi-channel operation data is subjected to mean zero processing to obtain a normalized signal; the normalized signal is subjected to convolution operation by using a least square smoothing filter algorithm to obtain the smoothed signal.

[0019] Optionally, the vibration acceleration is acquired by using a vibration acceleration sensor arranged at a low pressure side pressure measuring hole and a high pressure side pressure measuring hole of the target floating ring graphite seal, the gas temperature is acquired by using a temperature sensor arranged at a first threaded hole of the target floating ring graphite seal, the gas high and low pressure side force is acquired by using a pressure sensor arranged at the low pressure side pressure measuring hole and the high pressure side pressure measuring hole of the target floating ring graphite seal, and the dynamic displacement between the floating ring seal and the eddy current sensor probe is acquired by using an eddy current sensor arranged at a second threaded hole of the target floating ring graphite seal.

[0020] Optionally, the extracting the multi-domain degradation features in the smoothed signal and reducing the multi-domain degradation features in dimension and fusing the multi-domain degradation features to obtain a fused health index comprises:

[0021] The multi-domain degradation features in the smoothed signal are extracted, and the multi-domain degradation features comprise time domain features, frequency domain features and time-frequency domain features; a principal component analysis algorithm is used to reduce the multi-domain degradation features in dimension and fuse the multi-domain degradation features to obtain the fused health index.

[0022] Optionally, the multi-domain degradation features in the smoothed signal comprise:

[0023] obtain a time sequence waveform in the smoothed signal, calculate the time domain feature according to the time sequence waveform, perform fast Fourier transform on the smoothed signal to generate a frequency spectrum, and calculate the frequency domain feature according to the frequency spectrum, perform wavelet packet transform on the smoothed signal L obtain a wavelet packet transform decomposition tree, calculate the energy of each sub-band node corresponding to the signal at the first layer of the wavelet packet transform decomposition tree, combine the energy of all sub-band nodes to obtain the time-frequency domain feature, and combine the time domain feature, the frequency domain feature and the time-frequency domain feature to obtain a high-dimensional initial feature vector, and take the high-dimensional initial feature vector as the multi-domain degradation feature. L

[0024] Optionally, the multi-domain degradation feature is reduced in dimension and fused by using a principal component analysis algorithm to obtain a fused health indicator, including:

[0025] construct an original data matrix of the smoothed signal, normalize the original data matrix to obtain a normalized data matrix, calculate a covariance matrix of the normalized data matrix, perform eigenvalue decomposition on the covariance matrix to obtain a plurality of eigenvalues and unit eigenvectors corresponding to the plurality of eigenvalues, calculate the cumulative contribution rate of the first K principal components according to the plurality of eigenvalues based on a principal component analysis algorithm to construct a new subspace composed of the first K principal components, project the normalized data matrix onto the new subspace composed of the first K principal components to obtain reduced data, and fuse the reduced data to obtain the fused health indicator. k k k

[0026] Optionally, the fused health indicator is fitted by using a support vector regression model, an exponential degradation model and a polynomial regression model respectively to obtain a trained mixed model, including:

[0027] the support vector regression model, the exponential degradation model and the polynomial regression model are used to fit the fused health indicator respectively, and grid search and K-fold cross-validation are used to optimize the hyperparameters of each fitted model to determine the weight of each fitted model, and the final mixed prediction model is constructed according to the weight of each fitted model based on a weighted average method.

[0028] The second aspect embodiment of the present application provides a data-driven floating ring seal multi-source feature fusion life evaluation device, including:

[0029] ​​​​The preprocessing module is used for obtaining and preprocessing multi-channel operation data of a target floating ring graphite seal to obtain a smoothed signal; the extraction module is used for extracting multi-domain degradation features in the smoothed signal, and performing dimension reduction and fusion on the multi-domain degradation features to obtain a fused health index; the fitting module is used for fitting the fused health index by using a support vector regression model, an exponential degradation model and a polynomial regression model respectively to obtain a final hybrid prediction model; the prediction module is used for obtaining a real-time health index sequence of the target floating ring graphite seal, and inputting the real-time health index sequence into the final hybrid prediction model to predict a degradation trajectory; and the calculation module is used for extrapolating the degradation trajectory until the degradation trajectory intersects with a preset health index threshold to determine a predicted failure time point, and calculating a remaining useful life and a confidence interval thereof according to the predicted failure time point.

[0030] The third aspect of the present application provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the data-driven floating ring seal multi-source feature fusion life evaluation method as described in the above embodiments.

[0031] The fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the data-driven floating ring seal multi-source feature fusion life evaluation method as described above.

[0032] The data-driven floating ring seal multi-source feature fusion life evaluation method and device provided by the embodiments of the present application abandon the precise physical model which is difficult to establish, and instead start from a large amount of data actually obtained during the operation of the equipment, automatically mine features capable of representing the degradation state of the seal through advanced signal processing and machine learning technology, and construct a hybrid prediction model, so as to finally realize high-precision and personalized evaluation of the remaining useful life (RUL), thereby solving the problems of insufficient precision, dependence on experience and lack of uncertainty management in the life evaluation of key components such as floating ring graphite seals in high-end equipment, realizing high-precision and high-reliability prediction of the remaining useful life (RUL) of the floating ring graphite seal, and accurately evaluating the individualized health state and remaining life of each seal, so that the maintenance decision has a basis, greatly improving the safety and reliability of the operation of the equipment, and bringing significant improvement in precision, economic benefit, safety guarantee and industry promotion value.

[0033] The method not only effectively overcomes the excessive dependence on expensive tests, but also deeply mines failure precursors through surface data, realizes online, accurate and intelligent evaluation of the remaining useful life (RUL) of the seal, and provides core decision support for predictive maintenance and reliability management of high-end equipment.

[0034] Additional aspects and advantages of the present application will be partially apparent and partially described in the following description. BRIEF DESCRIPTION OF DRAWINGS

[0035] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the appended drawings.

[0036] Figure 1 A flow chart of a data-driven floating ring seal multi-source feature fusion life evaluation method according to an embodiment of the present application is provided.

[0037] Figure 2 A specific execution schematic diagram of a data-driven floating ring seal multi-source feature fusion life evaluation method according to an embodiment of the present application is provided.

[0038] Figure 3 A whole structure schematic diagram of a floating ring graphite seal life test according to an embodiment of the present application is provided.

[0039] Figure 4 A partial structure schematic diagram of a floating ring graphite seal life test according to an embodiment of the present application is provided.

[0040] Figure 5 A block schematic diagram of a data-driven floating ring seal multi-source feature fusion life evaluation device according to an embodiment of the present application is provided.

[0041] Figure 6 A structure schematic diagram of an electronic device according to an embodiment of the present application is provided.

[0042] Explanation of reference signs:

[0043] 1-cavity, 2-first positioning sleeve, 3-shaft sleeve, 4-end cover, 5-runway, 6-floating ring assembly, 7-second positioning sleeve, 8-housing, S1-electric eddy current sensor, S2-temperature sensor, S3-vibration acceleration sensor, S4-pressure sensor, P1-low pressure side pressure measuring hole, P2-high pressure side pressure measuring hole, T1-first temperature measuring hole, T2-second temperature measuring hole, N1-low pressure side oil inlet, N2-low pressure side oil outlet, N3-high pressure side oil outlet, N4-low pressure side exhaust port, N5-high pressure side air inlet, 50-data-driven floating ring seal multi-source feature fusion life evaluation device, 501-preprocessing module, 502-extraction module, 503-fitting module, 504-prediction module, 505-computation module, 601-memory, 602-processor, 603-communication interface. DETAILED DESCRIPTION

[0044] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0045] The following description, with reference to the accompanying drawings, illustrates a data-driven floating ring seal multi-source feature fusion lifetime assessment method and apparatus according to embodiments of the present invention.

[0046] Specifically, Figure 1 This is a schematic flowchart of a data-driven floating ring seal multi-source feature fusion lifetime assessment method provided in an embodiment of the present invention.

[0047] like Figure 1 As shown, this data-driven floating ring seal multi-source feature fusion lifetime assessment method includes the following steps:

[0048] In step S101, the multi-channel operating data of the target floating ring graphite seal is acquired and preprocessed to obtain a smooth signal.

[0049] In some embodiments, acquiring and preprocessing multi-channel operating data of the target floating ring graphite seal to obtain a smoothed signal includes:

[0050] Acquire multi-channel operating data of the target floating ring graphite seal, including vibration acceleration, gas temperature, high and low pressure side forces of gas, and dynamic displacement between the floating ring seal and the eddy current sensor probe.

[0051] The average value of the multi-channel operating data is zeroed out to obtain a normalized signal;

[0052] The least squares smoothing filter algorithm is used to perform convolution operation on the normalized signal to obtain a smooth signal.

[0053] In actual implementation, such as Figure 2 As shown, multiple sensors, including vibration acceleration sensors, temperature sensors, pressure sensors, and eddy current sensors, installed on the target floating ring graphite seal, are used to acquire multi-channel, long-term series raw operating data (i.e., multi-channel operating data) throughout the entire life cycle. The specific expression is as follows:

[0054] (1)

[0055] In the formula, For the sampling time point, In time Multi-channel operation data, In time The vibration acceleration below, is the gas temperature at time , is the gas high-low side force at time , is the dynamic displacement between the floating ring seal and the eddy current sensor probe at time .

[0056] Specifically, as shown in Figure 3 , the vibration acceleration of the object is measured by using the vibration acceleration sensors arranged at the low-pressure side pressure measuring hole and the high-pressure side pressure measuring hole of the target floating ring graphite seal, the mechanical vibration and impact state of the equipment can be monitored for fault diagnosis and health assessment; the temperature sensor arranged at the first threaded hole of the target floating ring graphite seal is used to monitor the temperature change of the sealing cavity and the gas (i.e. the gas temperature), so as to prevent material failure, lubrication deterioration or thermal deformation caused by overheating; the pressure sensors arranged at the low-pressure side pressure measuring hole and the high-pressure side pressure measuring hole of the target floating ring graphite seal are used to measure the gas high-low side force, the gas high-low side force in the sealing cavity is monitored, and the load working condition and sealing performance of the sealing are evaluated; the eddy current sensor arranged at the second threaded hole of the target floating ring graphite seal is used to measure the dynamic displacement between the floating ring seal and the eddy current sensor probe, the radial vibration, axial displacement or radial position of the rotating shaft is accurately measured, and it is a key sensor for monitoring the dynamic characteristics of the rotor.

[0057] Further, since the multi-channel operating data usually contains high-frequency random noise introduced by electromagnetic interference, mechanical vibration transmission and environmental noise, it must be effectively pretreated to extract the signal components that truly reflect the sealing degradation trend. However, the traditional filtering method (such as low-pass filtering) is easy to distort the signal waveform and weaken the key features while smoothing the noise, therefore, the embodiment of the present application adopts the Savitzky-Golay (SG) filter (also known as the least square smoothing filter) to perform smoothing pretreatment on the multi-channel operating data, the core of the algorithm is to use local polynomial least square fitting to smooth the noise, which can maximize the preservation of the original shape and high-order moment information (such as the width and height of the peak) of the signal in the time domain, and is particularly suitable for preserving the key details in the mechanical signal degradation trend, thereby effectively filtering out the high-frequency noise interference that is difficult to avoid in the field environment, while preserving the real change characteristics in the signal that can reflect the degradation trend of the equipment, providing a high-quality data basis for subsequent feature extraction.

[0058] Specifically, first, for any sampling point , the Savitzky-Golay filtering algorithm uses a +1, i.e. taking points on the left and right) within the local window centered on it, and uses a a polynomial of order as the smoothed output.

[0059] Further, let the sequence of data points within the current filter window be . Fit these points with a polynomial of order

[0060] (2)

[0061] where i is the relative index (0 i = -m,..., 0,..., m ) of the data point within the window, characterizes the complexity and sharpness of the local waveform of the signal. The goal of the fitting is to minimize the sum of squared fitting errors :

[0062] (3)

[0063] Further, by solving the least squares problem above, it can be found that the smoothed output value (i.e. the fitted value at the center point) can be expressed as a linear combination of all the original data points within the window:

[0064] (4)

[0065] where is a set of fixed convolution coefficients (or filter kernel). This set of coefficients h is determined solely by the window size 2 +1 and the polynomial order , and is independent of the specific signal data. The coefficients can be pre-computed by solving the generalized inverse of a Vandermonde matrix.

[0066] Based on the above process, the embodiments of the present application perform mean zero processing on the collected original signal to eliminate the influence of the direct current component. The normalized signal is input into the Savitzky-Golay filter configured according to the above parameters to perform convolution operation, and the smoothed signal is obtained. The smoothed signal will be used as the input of subsequent multi-domain feature extraction, significantly improving the signal-to-noise ratio of the extracted features and the correlation with the degradation process.

[0067] In step S102, multi-domain degradation features are extracted from the smoothed signal, and the multi-domain degradation features are dimensionally reduced and fused to obtain a fused health index.

[0068] In some embodiments, multi-domain degradation features in the smoothed signal are extracted, and the multi-domain degradation features are reduced in dimension and fused to obtain a fusion health indicator, including:

[0069] Multi-domain degradation features in the smoothed signal are extracted, including time-domain features, frequency-domain features, and time-frequency domain features.

[0070] The multi-domain degradation features are reduced in dimension and fused using a principal component analysis algorithm to obtain the fusion health indicator.

[0071] In some embodiments, the multi-domain degradation features in the smoothed signal include:

[0072] A time series waveform in the smoothed signal is obtained, and time-domain features are calculated based on the time series waveform.

[0073] The smoothed signal is subjected to a fast Fourier transform to generate a frequency spectrum, and frequency-domain features are calculated based on the frequency spectrum.

[0074] The smoothed signal is subjected to a wavelet packet transform algorithm L to obtain a wavelet packet transform decomposition tree.

[0075] The energy of the signal corresponding to each sub-band node at the first L layer of the wavelet packet transform decomposition tree is calculated, and the energies of all sub-band nodes are combined to obtain time-frequency domain features.

[0076] The time-domain features, frequency-domain features, and time-frequency domain features are combined to obtain a high-dimensional initial feature vector, and the high-dimensional initial feature vector is taken as the multi-domain degradation features.

[0077] In some embodiments, the multi-domain degradation features are reduced in dimension and fused using a principal component analysis algorithm to obtain the fusion health indicator, including:

[0078] An original data matrix of the smoothed signal is constructed, and the original data matrix is standardized to obtain a standardized data matrix.

[0079] A covariance matrix of the standardized data matrix is calculated, and the covariance matrix is subjected to eigenvalue decomposition to obtain a plurality of eigenvalues and unit eigenvectors corresponding thereto.

[0080] Based on the principal component analysis algorithm, the cumulative contribution rate of the first k principal components is calculated based on the plurality of eigenvalues to construct a new subspace composed of the first k principal components.

[0081] The standardized data matrix is projected onto the new subspace composed of the first k principal components to obtain reduced data.

[0082] The dimensionality-reduced data is then fused to obtain a fused health index.

[0083] In actual implementation, such as Figure 2 As shown, this invention makes a breakthrough by performing in-depth analysis of the preprocessed high-quality signal from multiple dimensions, including the time domain, frequency domain, and time-frequency domain, to extract a set of initial features that comprehensively characterize the degradation of the sealing state. Time-domain features include, but are not limited to, root mean square (RMS) value, kurtosis, and amplitude; frequency-domain features include centroid frequency and mean square frequency; time-frequency domain features are extracted by wavelet packet transform to obtain energy values ​​for each frequency band. Subsequently, principal component analysis (PCA) is used to process the high-dimensional initial feature set. Dimensionality reduction and fusion are performed. PCA transforms multiple correlated features into a few linearly independent principal components (PCs) through linear transformation. While retaining most of the original information (such as more than 95% variance), it significantly reduces data dimensionality, eliminates redundancy, and finally outputs a low-dimensional, sensitive, and most representative fusion health index (HI) that best represents the sealing degradation process. Specifically, the time-domain characteristics are calculated directly from the time-series waveform of the smoothed signal to reflect the signal's amplitude statistical properties and distribution changes. The specific solution process is as follows:

[0084] The root mean square (RMS) value characterizes the average energy level of a signal and is sensitive to long-term wear.

[0085] (5)

[0086] in, It is a discrete signal sequence. This represents the number of sampling points within the window.

[0087] Kurtosis characterizes the sharpness of the signal probability density distribution and is exceptionally sensitive to early faults and shock components.

[0088] (6)

[0089] in, μ The mean of the signal. σ The standard deviation is denoted as .

[0090] Crest Factor is the ratio of peak value to RMS value, used to detect the presence of prominent spikes in a signal.

[0091] (7)

[0092] Furthermore, embodiments of the present invention provide preprocessed signals. Performing a Fast Fourier Transform (FFT) converts the signal from the time domain to the frequency domain, obtaining its spectrum. The following characteristics are calculated based on the spectrum:

[0093] The Spectral Centroid (SC) frequency reflects the frequency component in which the spectral energy is concentrated, and its drift often indicates a change in dynamic characteristics.

[0094] (8)

[0095] Mean square frequency (MSF) characterizes the average position and width of the spectrum:

[0096] (9)

[0097] Furthermore, since seal degradation is a non-stationary process, this embodiment of the invention employs wavelet packet transform (WPT) to extract signal energy features at a finer time-frequency resolution. The specific extraction process is as follows:

[0098] (1) Decomposition process: Select a wavelet basis function (such as Db4 wavelet) and the number of decomposition levels. L , for signal conduct L Layer-wide subband wavelet packet decomposition yields a complete... WPT Decomposition tree. (The first...) L Layers will be generated 2 L Different sub-band nodes .

[0099] (2) Energy feature extraction: Calculate the first L The energy of the signal corresponding to each sub-band node in the layer forms a time-frequency energy feature vector:

[0100] (10)

[0101] in, It is a node Upper j Wavelet packet coefficients, M The coefficient length of this node.

[0102] (3) Constructing a feature vector: Combine the energy of all sub-bands (or their normalized values) into a time-frequency domain feature vector:

[0103] (11)

[0104] Finally, all time domain, frequency domain, time-frequency domain features are combined to form an initial feature vector with high dimension and redundancy and correlation among features :

[0105] (12)

[0106] The present application adopts principal component analysis (PCA) for dimension reduction and fusion, and automatically constructs an optimal and low-dimensional health index.

[0107] Further, there are N samples at time points, each sample has d features, forming an original data matrix . First, each dimension feature (each column) is standardized to eliminate the dimension effect, and a normalized data matrix is obtained:

[0108] (13)

[0109] wherein, mean and standard deviation of the i-th feature on all samples are and respectively. j

[0110] Further, the covariance matrix is calculated: the covariance matrix is calculated according to the normalized data matrix C , and the matrix reflects the correlation between features:

[0111] (14)

[0112] The eigenvalue decomposition is performed on the covariance matrix C to obtain the eigenvalue and the corresponding unit eigenvector , and the specific decomposition process is as follows:

[0113] (15)

[0114] The eigenvector defines the direction of the new feature space, which is called principal component (PC), and the importance is determined by the size of the corresponding eigenvalue .

[0115] Therefore, the cumulative contribution rate of the first k principal components is calculated based on the principal component analysis algorithm:

[0116] (16) ​

[0117] select the minimum k , so that ≥ 95% (i.e. retain more than 95% of the original information).

[0118] The original normalized data is projected onto a new subspace composed of the first k principal components to obtain the reduced dimension data (score matrix):

[0119] (17)

[0120] wherein, is a projection matrix composed of the first k principal components.

[0121] The score of the first principal component (PC1) is taken as the final fusion health index of the embodiment of the application. Because PC1 is the direction with the largest data variance, it best represents the overall degradation trend of the sealing performance:

[0122] (18)

[0123] The is a one-dimensional time series, and the clear downward or upward trend thereof clearly depicts the continuous degradation process of the floating ring graphite seal from the healthy state to complete failure.

[0124] In step S103, the support vector regression model, the exponential degradation model and the polynomial regression model are respectively used to fit the fusion health index to obtain a final mixed prediction model.

[0125] In some embodiments, the support vector regression model, the exponential degradation model and the polynomial regression model are respectively used to fit the fusion health index to obtain a trained mixed model, including:

[0126] The support vector regression model, the exponential degradation model and the polynomial regression model are respectively used to fit the fusion health index, and the grid search and K-fold cross-validation are used to optimize the hyperparameters of each fitted model to determine the weight of each fitted model.

[0127] Based on the weighted average method, the final mixed prediction model is constructed according to the weight of each fitted model.

[0128] In step S104, the real-time health index sequence of the target floating ring graphite seal is obtained, and the health index sequence is input into the final mixed prediction model to predict the degradation trajectory.

[0129] In actual execution, the embodiment of the application does not rely on a single model, but constructs a hybrid prediction model, and the fusion health index obtained in S102 is predicted in two stages Fitting and prediction are performed.

[0130] The first stage (offline training): known fusion health indexes are fitted by using a support vector regression (SVR) model and various fitting models (such as an exponential model, a polynomial model, etc.) The SVR model is good at learning the degradation law under complex working conditions due to its strong nonlinear mapping capability and good generalization, and the exponential model and the like provide a model basis conforming to the physical law of mechanical component wear. Through an optimization algorithm (such as grid search), the superparameters of each model are optimized, and the fitting degree (such as R 2 ) is used as an evaluation index to select the optimal single model or determine the weight of each model, and the offline training of the hybrid model is completed.

[0131] The second stage (online updating and prediction): for the equipment under online monitoring, real-time multi-channel operation data are obtained to calculate the latest health index value, and the trained hybrid model is input. The model can dynamically and accurately fit the current degradation trajectory of the sealing performance, and extrapolate the future trend.

[0132] Specifically, the SVR model maps the data to a high-dimensional feature space, finds a regression hyperplane with the largest interval, and the optimization problem is expressed as:

[0133] (19)

[0134] The constraint condition is:

[0135] (20)

[0136] (21)

[0137] (22)

[0138] wherein, w is a weight vector, b is a bias term, φ(·) is a nonlinear function for mapping time t to a high-dimensional space, C>0 is a penalty parameter, ε is an insensitive loss parameter, and are slack variables. The final regression function is:

[0139] (23)

[0140] where, , is the Lagrange multiplier, is the number of support vectors, is the kernel function (RBF kernel is preferred in embodiments of the present invention .

[0141] Further, exponential degradation model and polynomial regression model that conform to the physical laws of mechanical wear are used for fitting, specifically as follows:

[0142] Exponential degradation model:

[0143] (24)

[0144] Polynomial regression model:

[0145] (25)

[0146] The parameters of the exponential degradation model and the polynomial regression model are solved by the least square method (such as or ):

[0147] (26)

[0148] Further, grid search (Grid Search) and K-fold cross-validation (K-Fold Cross-Validation) are used to optimize the hyperparameters (such as the C,γ,ε of SVR; the order p of the polynomial model) of all sub-models. The goodness of fit R 2 is used as the core evaluation index:

[0149] (27)

[0150] where, is the average value of the health index.

[0151] After optimization, a weighted average method is used to construct the final hybrid prediction model :

[0152] (28)

[0153] where, represents the optimal fitting function of the corresponding model (i.e. , , ). The weight w can be allocated according to the score of each model (for example , or determined by a meta-learner. Finally, the reserved weights and all the optimal sub-model parameters are retained, and the offline training is completed.

[0154] For the new device under online monitoring, real-time multi-channel operation data are acquired to calculate the real-time health index sequence as .

[0155] The latest k data points are input into the trained hybrid model , which uses its comprehensive advantages to perform high-precision fitting on the current degradation trajectory. The function obtained by fitting is used to predict the health index values at future time points , to generate the degradation trajectory .

[0156] In step S105, the degradation trajectory is extrapolated until it intersects with the preset health index threshold to determine the predicted failure time point, and the remaining useful life and its confidence interval are calculated according to the predicted failure time point.

[0157] In actual execution, the embodiment of the present application sets a health index threshold representing sealing failure. The degradation trajectory predicted by the hybrid model in step S104 is extrapolated to intersect with the threshold, and the time point corresponding to the intersection point is the predicted failure time . The remaining useful life RUL is the difference between the current time and : In addition, the embodiment of the present application can also give the confidence interval (such as 90% confidence interval) of RUL by analyzing the distribution of prediction error, thereby providing a more robust reference for maintenance decision-making and realizing the management of prediction uncertainty.

[0158] Specifically, a health index threshold representing complete sealing failure is defined. The threshold can be determined according to historical failure data, expert experience or device specifications. The predicted trajectory obtained in step three is extrapolated until it intersects with the threshold . The equation is solved as follows:

[0159] (29)

[0160] The solution is the predicted failure time point. Then the remaining useful life at the current time (RUL) is:

[0161] (30)

[0162] Further, to quantify the uncertainty of prediction, embodiments of the present application construct the confidence interval of prediction by analyzing the prediction error distribution of the model in the offline training phase. RUL

[0163] On the historical multi-channel operation data, the prediction error of the final hybrid prediction model at each time point is calculated .

[0164] Assuming that the error obeys normal distribution , the mean and standard deviation of the error are estimated.

[0165] For the online prediction RUL , the confidence interval of ×100% (such as 90%) of the prediction can be expressed as:

[0166] (31)

[0167] Wherein, is the quantile of the standard normal distribution , and is the standard deviation of the prediction RUL , which can be derived by the error propagation law from the variance of the health index prediction .

[0168] The data-driven multi-source feature fusion life evaluation method for floating ring seals proposed by the embodiments of the present application is further described below through a specific embodiment.

[0169] As Figure 3 and 4 ​As shown, first, the servo motor is connected with the shaft sleeve 3 through a high-precision flexible coupling, and the motor as a whole is fixed on the rigid test platform. The motor output shaft is fixed in the circumferential direction by key groove cooperation with the shaft sleeve 3, and is axially locked by a locking nut. The shaft sleeve 3 provides an installation reference and power transmission for the raceway 5. The floating ring assembly 6 is carefully sleeved outside the raceway 5. Subsequently, the first positioning sleeve 2 and the second positioning sleeve 7 are respectively loaded into the cavity 1 from both sides. The core function of the two positioning sleeves is to accurately determine the axial and radial position of the floating ring assembly 6 in the cavity, and to ensure that it is concentric with the raceway 5, which is the key to the success of the test. The end cover 4 is aligned with the end face of the cavity 1, the positioning pin is inserted to ensure the correct circumferential angle, and then the high-strength bolt group around the periphery is uniformly tightened to press the end cover on the cavity 1. In this process, the end cover 4 will press the first positioning sleeve 2 and the second positioning sleeve 7, thereby finally fixing the floating ring assembly 6. An O-ring is installed at the joint surface of the end cover 4 and the cavity 1 to achieve static sealing. The core of the present application is to integrate a complete condition monitoring system for real-time acquisition of multi-source characteristic data related to sealing life evaluation. The specific monitoring point arrangement is as follows:

[0170] (1) Working condition parameter monitoring:

[0171] Through the high-pressure side pressure measuring hole P2 located on the high-pressure side and the low-pressure side pressure measuring hole P1 located on the low-pressure side, the pressure sensor S4 is connected to monitor the pressure of the upstream and downstream of the seal in real time, so as to accurately control the pressure difference of the seal working.

[0172] Through the first temperature measuring hole T1 and the second temperature measuring hole T2 located at different positions of the cavity, the temperature sensor S2 is installed for real-time monitoring of the temperature of the seal area inlet and the environment.

[0173] The servo motor is built-in with a rotational speed encoder to collect and feedback the actual rotational speed signal of the main shaft in real time.

[0174] (2) Performance and state parameter monitoring:

[0175] The eddy current sensor S1 is installed on the end cover 4 at a specific position, with its probe facing the back or end face of the floating ring assembly 6, for non-contact accurate measurement of the radial displacement and vibration trajectory of the floating ring in operation, directly obtaining key parameters such as eccentricity and eccentricity rate of the dynamic stability.

[0176] The vibration acceleration sensor S3 is installed near the bearing seat of the end cover 4 or the cavity 1, for collecting vibration signals caused by rotor imbalance, floating ring rubbing and other reasons, and diagnosing early faults by analyzing vibration spectrum characteristics.

[0177] The leaked medium is discharged from the low-pressure side exhaust port N4. A high-precision mass flow meter is installed in series in the pipeline to accurately measure the leakage rate of the seal, which is the most direct and important indicator for evaluating the sealing performance.

[0178] (3) Medium circulation system:

[0179] During the test, the driving motor drives the shaft sleeve 3 and the raceway 5 to rotate at high speed. The sealing medium is introduced from the high-pressure side inlet N5, and after sealing, most of the medium is sealed, and a small amount of leaked medium flows out from the low-pressure side exhaust port N4 and enters the flow meter for measurement. The low-pressure side oil inlet N1, the low-pressure side oil outlet N2, and the high-pressure side oil outlet N3 are used to connect the lubrication system to forcibly lubricate and cool the bearings and other components.

[0180] The signals of all sensors (S1, S2, S3, S4) are connected to a high-speed data acquisition system and finally transmitted to the host computer. This system synchronously collects multi-source data such as speed, pressure, temperature, displacement, vibration, and flow to construct a complete information map of the sealing working state, providing a solid data foundation for subsequent feature fusion, health state evaluation, and residual useful life (RUL) prediction using data-driven algorithms.

[0181] As shown in Figure 2 , the method calculates the data collected as features. First, after multi-source data acquisition and synchronization, SG filtering algorithm is used for smoothing and denoising to effectively suppress high-frequency noise while retaining the true degradation trend information in the signal. Then, steps such as outlier processing, missing value filling, and signal normalization are performed to finally output a high-quality preprocessed data set . The high-quality data is then input into a multi-domain feature extractor to mine information from multiple dimensions such as time domain, frequency domain, and time-frequency domain. All extracted features are aggregated and standardized to form an initial high-dimensional feature set . Principal component analysis (PCA) algorithm is used to reduce the dimensionality and fuse the high-dimensional feature set to extract the most important degradation trend information from the data, forming a one-dimensional, monotonicity good fusion health index HI. The historical health index sequence is used to train a hybrid model in parallel. The hybrid model combines prediction algorithms of different principles: support vector regression (SVR) model (using RBF kernel function, good at capturing complex nonlinear laws), exponential degradation model (consistent with the physical law of mechanical component wear), and polynomial regression model (with strong curve fitting ability). The real-time calculated health index sequence is input into the deployed hybrid prediction model for dynamic trajectory fitting and prediction, and the future degradation trajectory is extrapolated. By comparing the predicted trajectory with the preset failure threshold , the predicted failure time , and further obtain a current remaining useful life (RUL) point estimate value (RUL) ).

[0182] In summary, the data-driven floating ring seal multi-source feature fusion life evaluation method according to the embodiment of the present application has the following beneficial effects:

[0183] (1) Through the multi-source feature fusion and mixed modeling strategy, the prediction accuracy is fundamentally improved, the high-precision and high-reliability prediction of the floating ring graphite seal remaining useful life (RUL) is realized, and the problem that the traditional method relies on a single model or empirical estimation and is difficult to cope with the nonlinear degradation process under complex working conditions is solved. The prediction result is often greatly deviated. Specifically, the original data is preprocessed by using the SG filter, the noise interference is effectively removed, and a high-quality data basis is provided for subsequent analysis. By extracting comprehensive features representing the seal degradation state from multiple dimensions such as time domain, frequency domain and time-frequency domain, and using principal component analysis (PCA) for dimensionality reduction fusion, a low-dimensional, sensitive and good monotonicity fusion health index (HI) is constructed. The index can clearly and stably reveal the degradation trend of the seal performance, avoiding the randomness and instability of a single feature. More importantly, a mixed prediction model combining the support vector regression (SVR) model and the exponential model and the polynomial model is innovatively used. The model fully utilizes the respective advantages of the strong nonlinear fitting capability of SVR and the physical interpretability of the empirical model, realizes complementary advantages, enables it to learn the deep law in complex data, and conforms to the physical trend of mechanical wear. Finally, the prediction error of the method for RUL is much lower than that of the traditional empirical estimation method and the single model prediction method, and the certainty of the prediction result is significantly enhanced;

[0184] (2) The individualized health status and remaining life of each seal can be accurately evaluated, so that the maintenance decision has a basis. Specifically, the operator can scientifically and reasonably plan the maintenance window according to the predicted RUL and its confidence interval, and perform maintenance and replacement at the most appropriate time, avoiding "over maintenance" based on fixed cycle, maximizing the utilization of seal material life, and greatly saving the cost of expensive spare parts and the cost of frequent replacement. At the same time, "insufficient maintenance" caused by underestimating the life is completely avoided, and unplanned shutdown caused by seal sudden failure is effectively prevented. For continuous production process industry, the loss of minutes of unplanned shutdown can be as high as hundreds of thousands of yuan, therefore, the method guarantees the continuity and stability of production, creates huge indirect economic benefits, and brings significant operation and maintenance mode change, realizing the key technical support from traditional "timely maintenance" to advanced "condition-based maintenance";

[0185] (3) It acts as a reliable "early warning system" that can predict the risk of seal failure in advance for a long enough time to provide the operation and maintenance team with sufficient response time. Through continuous online monitoring and prediction, the degradation state of the seal can be grasped in real time. Once the prediction trajectory shows that the health index is about to approach the failure threshold, the system can automatically issue a warning to prompt intervention for inspection or repair. This proactive and predictive safety control mode eliminates accidents at the embryonic stage, fundamentally avoiding major safety risks and equipment damage caused by seal failure, providing a solid guarantee for the safe, long-term, and high-reliability operation of high-end equipment such as aircraft engines and gas turbines (i.e., avoiding medium leakage, environmental pollution, and even catastrophic accidents such as fires and explosions caused by floating ring graphite seal failure), greatly improving the safety and reliability of equipment operation;

[0186] (4) The "data preprocessing-multi-domain feature extraction and fusion-hybrid model prediction-uncertainty management" framework is not limited to floating ring graphite seals and can be extended to the life prediction and health management (PHM) of other mechanical key components such as rolling bearings, gears, and turbine blades. With the popularity of industrial Internet of Things (IIoT) and big data technology, it is easier to obtain equipment operation data, and the advantages of this data-driven method will become more apparent. It provides a replicable and generalizable example for solving the intelligent operation and maintenance problems of various complex equipment, has broad industry application prospects and industrialization potential, and is of great significance to the intelligent upgrading of the entire high-end equipment manufacturing industry.

[0187] Second, the data-driven floating ring seal multi-source feature fusion life evaluation device according to the embodiment of the present application is described with reference to the accompanying drawings.

[0188] Figure 5 A block diagram of a data-driven floating ring seal multi-source feature fusion life evaluation device according to an embodiment of the present application is provided.

[0189] As shown in Figure 5 , the data-driven floating ring seal multi-source feature fusion life evaluation device 50 includes a preprocessing module 501, an extraction module 502, a fitting module 503, a prediction module 504, and a calculation module 505.

[0190] The preprocessing module 501 is configured to obtain and preprocess multi-channel operation data of the target floating ring graphite seal to obtain a smoothed signal. The extraction module 502 is configured to extract multi-domain degradation features in the smoothed signal, and perform dimension reduction and fusion on the multi-domain degradation features to obtain a fused health index. The fitting module 503 is configured to respectively adopt a support vector regression model, an exponential degradation model and a polynomial regression model to fit the fused health index to obtain a final hybrid prediction model. The prediction module 504 is configured to obtain a real-time health index sequence of the target floating ring graphite seal, and input the real-time health index sequence into the final hybrid prediction model to predict a degradation trajectory. The calculation module 505 is configured to extrapolate the degradation trajectory until the degradation trajectory intersects with a preset health index threshold to determine a predicted failure time point, and calculate a remaining service life and a confidence interval thereof according to the predicted failure time point.

[0191] In some embodiments, the preprocessing module 501 comprises:

[0192] The acquisition unit is configured to acquire multi-channel operation data of the target floating ring graphite seal, wherein the multi-channel operation data comprises vibration acceleration, gas temperature, gas high-low pressure side force, dynamic displacement between the floating ring seal and an eddy current sensor probe;

[0193] The normalization processing unit is configured to perform mean zero processing on the multi-channel operation data to obtain a normalized signal;

[0194] The operation unit is configured to perform convolution operation on the normalized signal by using a least square smoothing filter algorithm to obtain a smoothed signal.

[0195] In some embodiments, the vibration acceleration is acquired by using vibration acceleration sensors arranged at low-pressure side pressure holes and high-pressure side pressure holes of the target floating ring graphite seal, the gas temperature is acquired by using a temperature sensor arranged at a first threaded hole of the target floating ring graphite seal, the gas high-low pressure side force is acquired by using pressure sensors arranged at the low-pressure side pressure holes and the high-pressure side pressure holes of the target floating ring graphite seal, and the dynamic displacement between the floating ring seal and the eddy current sensor probe is acquired by using an eddy current sensor arranged at a second threaded hole of the target floating ring graphite seal.

[0196] In some embodiments, the extraction module 502 comprises:

[0197] The extraction unit is configured to extract multi-domain degradation features in the smoothed signal, wherein the multi-domain degradation features comprise time domain features, frequency domain features and time-frequency domain features;

[0198] The fusion unit is configured to perform dimension reduction and fusion on the multi-domain degradation features by using a principal component analysis algorithm to obtain a fused health index.

[0199] In some embodiments, the extraction unit comprises:

[0200] The acquisition subunit is configured to acquire a time sequence waveform in the smoothed signal, and calculate a time domain feature according to the time sequence waveform.

[0201] The transformation subunit is configured to perform fast Fourier transform on the smoothed signal to generate a frequency spectrum, and calculate a frequency domain feature according to the frequency spectrum.

[0202] The first decomposition subunit is configured to perform wavelet packet transform on the smoothed signal by using a wavelet packet transform algorithm. L The first decomposition subunit is configured to perform wavelet packet transform on the smoothed signal by using a wavelet packet transform algorithm.

[0203] The calculation subunit is configured to calculate energy of a signal corresponding to each sub-band node at the first layer of the wavelet packet transform decomposition tree. L The calculation subunit is configured to calculate energy of a signal corresponding to each sub-band node at the first layer of the wavelet packet transform decomposition tree.

[0204] The combination subunit is configured to combine the time domain feature, the frequency domain feature, and the time-frequency domain feature to obtain a high-dimensional initial feature vector, and use the high-dimensional initial feature vector as the multi-domain degradation feature.

[0205] In some embodiments, the fusion unit includes:

[0206] The normalization subunit is configured to construct an original data matrix of the smoothed signal, and normalize the original data matrix to obtain a normalized data matrix.

[0207] The second decomposition subunit is configured to calculate a covariance matrix of the normalized data matrix, and perform eigenvalue decomposition on the covariance matrix to obtain a plurality of eigenvalues and unit eigenvectors corresponding to the plurality of eigenvalues.

[0208] The construction subunit is configured to calculate a cumulative contribution rate of the first principal components according to the plurality of eigenvalues based on a principal component analysis algorithm. k The construction subunit is configured to calculate a cumulative contribution rate of the first principal components according to the plurality of eigenvalues based on a principal component analysis algorithm. k The construction subunit is configured to calculate a cumulative contribution rate of the first principal components according to the plurality of eigenvalues based on a principal component analysis algorithm.

[0209] The projection subunit is configured to project the normalized data matrix onto the new subspace formed by the first principal components to obtain reduced data. k The projection subunit is configured to project the normalized data matrix onto the new subspace formed by the first principal components to obtain reduced data.

[0210] The fusion subunit is configured to fuse the reduced data to obtain a fusion health index.

[0211] In some embodiments, the fitting module 503 includes:

[0212] a fitting subunit configured to fit the fused health index using a support vector regression model, an exponential degradation model and a polynomial regression model respectively, and to perform hyperparameter optimization on the model in each fitting using grid search and K-fold cross-validation to determine the weight of each fitted model;

[0213] a construction subunit configured to construct a final hybrid prediction model according to the weight of each fitted model based on a weighted average method.

[0214] It should be noted that the foregoing explanation of the data-driven multi-source feature fusion life evaluation method for floating ring seals also applies to the data-driven multi-source feature fusion life evaluation device for floating ring seals, which will not be described here again.

[0215] The data-driven multi-source feature fusion life evaluation device for floating ring seals according to the embodiment of the present application has the following beneficial effects:

[0216] (1) The multi-source feature fusion and hybrid modeling strategy fundamentally improve the prediction accuracy, achieve high-precision and high-reliability prediction of the residual useful life (RUL) of the floating ring graphite seal, and solve the problem that the traditional method relies on a single model or empirical estimation and is difficult to cope with the nonlinear degradation process under complex working conditions, and the prediction result often has a large deviation. Specifically, the original data is preprocessed using an SG filter to effectively remove noise interference and provide a high-quality data basis for subsequent analysis. Comprehensive features representing the degradation state of the seal are extracted from multiple dimensions such as time domain, frequency domain and time-frequency domain, and principal component analysis (PCA) is used for dimensionality reduction fusion to construct a low-dimensional, sensitive and monotonic fusion health index (HI). This index can clearly and stably reveal the degradation trend of the seal performance, avoiding the randomness and instability of a single feature. More importantly, a hybrid prediction model combining a support vector regression (SVR) model and an exponential model and a polynomial model is innovatively used. This model takes full advantage of the respective advantages of the strong nonlinear fitting capability of SVR and the physical interpretability of empirical models, realizes complementary advantages, enables it to learn the deep laws in complex data, and conforms to the physical trend of mechanical wear. Finally, the prediction error of the method for RUL is much lower than that of the traditional empirical estimation method and the single model prediction method, and the certainty of the prediction result is significantly enhanced;

[0217] (2) The individualized health state and remaining useful life of each seal can be accurately evaluated, so that the maintenance decision has a basis. Specifically, the operator can scientifically and reasonably plan the maintenance window according to the predicted RUL and its confidence interval, and perform maintenance and replacement at the most appropriate time, avoiding "over maintenance" based on a fixed cycle, maximizing the use of the material life of the seal, greatly saving the cost of expensive spare parts and the cost of frequent replacement. At the same time, "under maintenance" caused by underestimating the life is completely avoided, and unplanned shutdown caused by seal sudden failure is effectively prevented. For continuous production process industry, the loss of minutes of unplanned shutdown can be as high as hundreds of thousands of yuan, therefore, the method creates huge indirect economic benefits by ensuring the continuity and stability of production, and brings significant operation and maintenance mode change, and realizes the key technical support for the transformation from traditional "timely maintenance" to advanced "condition-based maintenance";

[0218] (3) It serves as a reliable "early warning system" and can predict the failure risk of the seal in advance for a long enough time to provide sufficient response time for the operation and maintenance team. Through continuous online monitoring and prediction, the degradation state of the seal can be grasped in real time, and once the prediction trajectory shows that the health index is about to approach the failure threshold, the system can automatically issue a warning and prompt the need for intervention inspection or maintenance. This proactive and predictive safety control mode eliminates accidents at the embryonic stage, fundamentally avoids major safety risks and equipment damage caused by seal failure, and provides a solid guarantee for the safe, long-period and high-reliability operation of high-end equipment such as aircraft engines and gas turbines (i.e. avoids medium leakage, environmental pollution, and even disastrous accidents such as fire and explosion caused by floating ring graphite seal failure), greatly improving the safety and reliability of equipment operation;

[0219] (4) The "data preprocessing-multi-domain feature extraction and fusion-hybrid model prediction-uncertainty management" framework is not limited to the floating ring graphite seal, and the technical idea can be extended to the life prediction and health management (PHM) of other mechanical key components such as rolling bearings, gears and turbine blades. With the popularization of industrial internet of things (IIoT) and big data technology, it is easier to obtain equipment operation data, and the advantages of this data-driven method will become more and more obvious. It provides a replicable and generalizable excellent example for solving the intelligent operation and maintenance problems of various complex equipment, has broad industry application prospects and industrialization potential, and has important significance for promoting the intelligent upgrading of the entire high-end equipment manufacturing industry.

[0220] Figure 6 A structural schematic diagram of an electronic device according to an embodiment of the present application is provided.

[0221] The electronic device can include a memory 601, a processor 602, and a computer program stored on the memory 601 and executable on the processor 602.

[0222] The processor 602 implements the data-driven floating ring seal multi-source feature fusion life evaluation method provided in the above embodiments when executing the program.

[0223] Further, the electronic device further includes:

[0224] A communication interface 603 for communication between the memory 601 and the processor 602.

[0225] The memory 601 is used to store the computer program executable on the processor 602.

[0226] The memory 601 can include a high-speed RAM memory, and can also include a non-volatile memory, such as at least one disk memory.

[0227] If the memory 601, the processor 602 and the communication interface 603 are implemented independently, the communication interface 603, the memory 601 and the processor 602 can be connected to each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 Only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0228] Optionally, in specific implementation, if the memory 601, the processor 602 and the communication interface 603 are integrated on a chip, the memory 601, the processor 602 and the communication interface 603 can complete communication between each other through an internal interface.

[0229] The processor 602 can be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0230] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to realize the data-driven floating ring seal multi-source feature fusion life assessment method as above.

[0231] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0232] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include at least one of the features. In the description of the present application, the meaning of "N" is at least two, for example, two, three, etc., unless otherwise specifically limited.

[0233] Any process or method descriptions in flow charts or otherwise described herein represent embodiments that can be understood as a module, segment, or portion of code that includes one or N executable instructions for implementing the specified logical function or process, and the scope of the preferred embodiments of the present application includes additional implementation in which the functions are performed in different orders, including substantially simultaneously, or in reverse order, depending on the functionality involved, which will be understood by those skilled in the art.

[0234] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of instructions to implement logic functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this specification, a "computer-readable medium" can be any means that can contain, store, communicate, propagate or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be a machine-readable storage device (e.g., magnetic, optical or other) a machine-readable storage diskette (e.g., floppy, flexible or other), a machine-readable storage card (e.g., ROM, EEPROM, flash memory or other), a machine- readable storage tape (e.g., magnetic, optical or other), a machine-readable storage medium (e.g., a portable electronic device, a computer diskette, a computer memory stick, a computer hard drive, a computer tape, a computer readable storage medium, or other), or a machine-readable wireless transmission (e.g., a radio frequency signal, an infrared signal, a microwave signal, or other). More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via, for example, optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0235] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware and in another embodiment, any of the following technologies known in the art or their combinations can be used: discrete logic circuitry having logic gates for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), etc.

[0236] Those skilled in the art of the present technology can understand that all or part of the steps carried out by the above-mentioned embodiment methods can be completed by programs instructing related hardware, and the programs can be stored in a computer-readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0237] In addition, each function unit in each embodiment of the present application can be integrated in one processing module, or each unit can be physically present separately, or two or more units can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software function module. When the integrated module is realized in the form of a software function module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0238] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A data-driven floating ring seal multi-source feature fusion life assessment method, characterized in that, The method comprises the following steps: Obtaining and preprocessing the multi-channel operation data of the target floating ring graphite seal to obtain a smoothed signal, specifically comprising: Obtaining the multi-channel operation data of the target floating ring graphite seal, wherein the multi-channel operation data comprises vibration acceleration, gas temperature, gas high-low pressure side force, dynamic displacement between the floating ring seal and the eddy current sensor probe; Performing mean zero processing on the multi-channel operation data to obtain a normalized signal; Performing convolution operation on the normalized signal by using a least square smoothing filter algorithm to obtain the smoothed signal; Extracting multi-domain degradation features from the smoothed signal, and performing dimension reduction and fusion on the multi-domain degradation features to obtain a fusion health index, specifically comprising: Extracting multi-domain degradation features from the smoothed signal, and performing dimension reduction and fusion on the multi-domain degradation features to obtain a fusion health index, comprising: Extracting multi-domain degradation features from the smoothed signal, wherein the multi-domain degradation features comprise time domain features, frequency domain features and time-frequency domain features; Performing dimension reduction and fusion on the multi-domain degradation features by using a principal component analysis algorithm to obtain a fusion health index; Fitting the fusion health index by using a support vector regression model, an exponential degradation model and a polynomial regression model respectively to obtain a final hybrid prediction model, specifically comprising: Fitting the fusion health index by using the support vector regression model, the exponential degradation model and the polynomial regression model respectively, and performing hyperparameter optimization on the model in each fitting by using grid search and K-fold cross-validation to determine the weight of each fitted model; Constructing the final hybrid prediction model based on the weighted average method according to the weight of each fitted model; Obtaining a real-time health index sequence of the target floating ring graphite seal, and inputting the time health index sequence into the final hybrid prediction model to predict a degradation trajectory; Extrapolating the degradation trajectory until it intersects with a preset health index threshold to determine a predicted failure time point, and calculating the remaining useful life and its confidence interval according to the predicted failure time point.

2. The data-driven floating ring seal multi-source feature fusion life assessment method according to claim 1, characterized in that, The vibration acceleration is obtained by using a vibration acceleration sensor arranged at the low-pressure side pressure hole and the high-pressure side pressure hole of the target floating ring graphite seal, the gas temperature is obtained by using a temperature sensor arranged at the first threaded hole of the target floating ring graphite seal, the gas high-low pressure side force is obtained by using a pressure sensor arranged at the low-pressure side pressure hole and the high-pressure side pressure hole of the target floating ring graphite seal, and the dynamic displacement between the floating ring seal and the eddy current sensor probe is obtained by using an eddy current sensor arranged at the second threaded hole of the target floating ring graphite seal.

3. The data-driven floating ring seal multi-source feature fusion life assessment method of claim 1, wherein, The multi-domain degradation features in the smoothed signal comprise: Obtaining a time series waveform in the smoothed signal, and calculating the time domain features according to the time series waveform; Performing fast Fourier transform on the smoothed signal to generate a frequency spectrum, and calculating the frequency domain features according to the frequency spectrum; The smooth signal is processed by using a wavelet packet transform algorithm L layer full sub-band wavelet packet decomposition to obtain a wavelet packet transform decomposition tree; a first node of the wavelet packet transform decomposition tree is calculated L energy of a signal corresponding to each sub-band node of a layer, to combine the energy of all sub-band nodes to obtain the time-frequency domain feature The time domain feature, the frequency domain feature and the time-frequency domain feature are combined to obtain a high-dimensional initial feature vector, and the high-dimensional initial feature vector is taken as the multi-domain degradation feature.

4. The data-driven floating ring seal multi-source feature fusion life assessment method of claim 1, wherein, The principal component analysis algorithm is used to reduce the dimension and fuse the multi-domain degradation feature to obtain a fused health index. An original data matrix of the smoothed signal is constructed, and the original data matrix is standardized to obtain a standardized data matrix; The covariance matrix of the standardized data matrix is calculated, and the covariance matrix is subjected to eigenvalue decomposition to obtain a plurality of eigenvalues and unit eigenvectors corresponding to the plurality of eigenvalues; Based on the principal component analysis algorithm, the cumulative contribution rate of the first K principal components is calculated according to the plurality of characteristic values, to construct a new subspace composed of the first K principal components. k k principal components.​ projecting the standardized data matrix onto a new subspace composed of the first k principal components to obtain reduced dimension data; The reduced data are fused to obtain the fused health index.

5. A data-driven floating ring seal multi-source feature fusion life assessment device, characterized in that, It comprises: The preprocessing module is used for obtaining and preprocessing the multi-channel running data of the target floating ring graphite seal to obtain a smoothed signal, specifically comprising: The acquisition unit is used for acquiring the multi-channel running data of the target floating ring graphite seal, wherein the multi-channel running data includes vibration acceleration, gas temperature, gas high and low pressure side force, and dynamic displacement between the floating ring seal and the eddy current sensor probe; The normalization processing unit is used for mean zero processing of the multi-channel running data to obtain a normalized signal; The operation unit is used for convolution operation of the normalized signal by using the least square smoothing filter algorithm to obtain a smoothed signal; The extraction module is used for extracting the multi-domain degradation feature in the smoothed signal, and reducing the dimension and fusing the multi-domain degradation feature to obtain a fused health index, specifically comprising: The extraction unit is used for extracting the multi-domain degradation feature in the smoothed signal, and the multi-domain degradation feature includes time domain feature, frequency domain feature and time-frequency domain feature; The fusion unit is used for reducing the dimension and fusing the multi-domain degradation feature by using the principal component analysis algorithm to obtain the fused health index; The fitting module is used for fitting the fused health index by using the support vector regression model, the exponential degradation model and the polynomial regression model respectively to obtain a final mixed prediction model, specifically comprising: The fitting subunit is used for fitting the fused health index by using the support vector regression model, the exponential degradation model and the polynomial regression model respectively, and the grid search and K-fold cross-validation are used to optimize the hyperparameters of each fitted model to determine the weight of each fitted model; The construction subunit is used for constructing the final mixed prediction model based on the weighted average method according to the weight of each fitted model; The prediction module is used for acquiring a real-time health index sequence of the target floating ring graphite seal, and inputting the real-time health index sequence into the final mixed prediction model to predict a degradation trajectory; The calculation module is used for extrapolating the degradation trajectory until it intersects with a preset health index threshold to determine a predicted failure time point, and calculating the remaining useful life and its confidence interval according to the predicted failure time point.

6. An electronic device, comprising: It comprises: A memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the data-driven floating ring seal multi-source feature fusion life evaluation method according to any one of claims 1-4.

7. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor for implementing the data-driven floating ring seal multi-source feature fusion life assessment method as claimed in any one of claims 1-4.

Images