Method for evaluating equipment health based on mechanism model and non-parameter diagnosis fusion

Abstract

The application provides a device health evaluation method based on a mechanism model and a non-parameter diagnosis fusion, comprising the following steps: constructing a three-dimensional entity dynamics model of a rotating device, applying excitation loads of different types of faults to generate simulation vibration signals, calculating theoretical characteristic frequencies of each fault type, constructing a mapping relationship between the fault types and the theoretical characteristic frequencies, and forming a mechanism characteristic library; performing non-linear dynamics non-parameter decomposition on the measured vibration signals, extracting intrinsic dynamic characteristics, and forming a non-parameter characteristic library; fusing the theoretical characteristic frequencies and the intrinsic dynamic characteristics to generate a fusion evaluation characteristic set; determining the health grade of the rotating device according to the deviation degree between the fusion evaluation characteristic set and a preset health benchmark characteristic space, and positioning and identifying the fault type and the fault position of the rotating device. The method improves the health diagnosis accuracy of the rotating device in an industrial scene with scarce fault samples.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology for electromechanical equipment, and in particular relates to a method for equipment health assessment based on the fusion of mechanistic models and parameterless diagnostics. Background Technology

[0002] Rotating equipment in factories, such as gearboxes, motors, fans, and pumps, are characterized by long-term continuous operation and complex structures and operating conditions. Sudden equipment failures and shutdowns can cause significant losses. Currently, fault diagnosis methods for rotating equipment are mainly divided into two categories: model-driven methods based on physical mechanisms and data-driven methods. Traditional model-driven methods based on physical mechanisms utilize the physical laws of the equipment (such as mechanics and thermodynamics) to establish mathematical models, and determine whether a fault has occurred by comparing the residuals between the actual output and the model predictions. However, this type of method highly relies on the accurate dynamic equations and physical parameters of the equipment. For rotating equipment with complex structures and variable operating conditions, accurate modeling is difficult, and the model's generalization ability is limited.

[0003] Data-driven approaches learn the mapping between normal and failure modes directly from historical operational data. Commonly used techniques include traditional machine learning methods such as support vector machines and random forests, as well as deep learning methods such as convolutional neural networks and long short-term memory networks. Although data-driven approaches do not rely on physical models, their diagnostic performance is highly dependent on a large number of labeled training samples. However, in real-world industrial scenarios, failure samples, especially early failure samples, are scarce, leading to insufficient model training and inadequate generalization ability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a device health assessment method based on the fusion of mechanistic models and non-parametric diagnostics. This method integrates mechanistic models and non-parametric diagnostics to achieve device status assessment and prediction, thereby improving the accuracy, interpretability, and adaptability of diagnostics.

[0005] To achieve the above objectives, this invention provides a device health assessment method based on the fusion of mechanistic models and parametric diagnostics, comprising: Acquire measured vibration signals generated during the operation of rotating equipment; A three-dimensional solid dynamic model of the rotating equipment is constructed. Excitation loads of different types of faults are applied according to the operating conditions of the equipment to generate simulated vibration signals corresponding to each fault type. The theoretical characteristic frequencies of each fault type are calculated, and the mapping relationship between fault types and theoretical characteristic frequencies is constructed to form a mechanism feature library. The measured vibration signal is subjected to nonlinear dynamic nonparametric decomposition to extract intrinsic dynamic features and form a nonparametric feature library. The theoretical characteristic frequencies in the mechanistic feature library are fused with the intrinsic dynamic features in the parameterless feature library to generate a fused evaluation feature set; Based on the degree of deviation between the fused evaluation feature set and the preset health benchmark feature space, the health level of the rotating equipment is determined, and the fault type and fault location of the rotating equipment are located and identified.

[0006] In one embodiment of the present invention, the steps of constructing a three-dimensional solid dynamic model of the rotating equipment, applying excitation loads of different fault types according to the equipment's operating conditions to generate simulated vibration signals corresponding to each fault type, and forming a virtual dataset include: Based on the geometric dimensions, material properties, and component assembly relationships of the rotating equipment, a three-dimensional solid geometric model of the rotating equipment is established. The material properties include elastic modulus, density, and Poisson's ratio. The component assembly relationships include bearing fit, gear meshing, and coupling connection. According to the operating conditions of the equipment, different fault types of excitation loads are applied to the three-dimensional solid dynamic model to simulate the vibration response under different fault types, and the corresponding simulated vibration signal is extracted at at least one key measuring point for each fault type. The excitation load includes at least one of unbalanced excitation force, misalignment excitation torque, bearing local defect impact excitation, and gear local defect impact excitation.

[0007] In one embodiment of the present invention, the steps of calculating the theoretical characteristic frequencies of each fault type, constructing the mapping relationship between fault types and theoretical characteristic frequencies, and forming a mechanism feature library include: Based on the simulated vibration signal, the theoretical characteristic frequencies of each fault type are calculated according to the structural parameters of each component in the rotating equipment. The structural parameters include bearing type, number of gear teeth, number of blades, and shaft speed. The theoretical characteristic frequencies include bearing inner ring fault characteristic frequency, bearing outer ring fault characteristic frequency, bearing rolling element fault characteristic frequency, bearing cage fault characteristic frequency, gear meshing frequency, gear sideband frequencies, blade passing frequency, and rotor speed. The mapping relationship between fault types and theoretical characteristic frequencies is constructed based on the theoretical characteristic frequencies corresponding to each fault type and added to the mechanism feature library.

[0008] In one embodiment of the present invention, the step of performing nonlinear dynamic nonparametric decomposition on the measured vibration signal, extracting intrinsic dynamic features, and forming a nonparametric feature library includes: The measured vibration signal is subjected to empirical mode decomposition, which decomposes it into several intrinsic mode function components. Each intrinsic mode function component represents a local feature signal at a different time scale. Perform a Hilbert transform on each of the intrinsic mode function components to obtain a time-frequency spectrum, and extract time-frequency domain features from the time-frequency spectrum as the first type of intrinsic dynamic features; The time-frequency domain features include instantaneous frequency and instantaneous amplitude.

[0009] In one embodiment of the present invention, the measured vibration signal is decomposed into a local mean decomposition into a plurality of product function components, each of which is obtained by multiplying an envelope signal and a pure frequency-modulated signal. Time-frequency domain features are extracted from each component of the product function as second-type intrinsic dynamic features; The time-frequency domain features include instantaneous frequency and instantaneous amplitude.

[0010] In one embodiment of the present invention, the step of fusing the theoretical characteristic frequencies in the mechanistic feature library with the intrinsic dynamic characteristics in the parameterless feature library to generate a fused evaluation feature set includes: The time-frequency domain features of each function component in the parameterless feature library are matched with the theoretical characteristic frequencies of each fault type in the mechanism feature library. When the deviation between the time-frequency domain features of a certain function component and a certain theoretical characteristic frequency is less than a preset frequency tolerance threshold, the function component is determined to be the associated component of the fault type. Based on the similarity between the time-frequency domain features of the associated components and the theoretical characteristic frequencies of the corresponding fault types in the mechanism feature library, the confidence weight of the associated components is determined. The confidence weight is then used to weight and combine the intrinsic dynamic features in the parameterless feature library and the theoretical characteristic frequencies in the mechanism feature library to obtain the fusion evaluation feature set. The preset frequency tolerance threshold is adaptively determined based on the rotational speed fluctuation range and spectral resolution of the rotating device.

[0011] In one embodiment of the present invention, the frequency centroid of each intrinsic mode function component is calculated based on the instantaneous frequency and instantaneous amplitude. The frequency centroids of each intrinsic modulus function component are matched with the theoretical characteristic frequencies of each fault type in the mechanism feature library. When the deviation between the frequency centroid of a certain function component and a certain theoretical characteristic frequency is less than a preset frequency tolerance threshold, the function component is determined to be the associated component of the fault type.

[0012] In one embodiment of the present invention, the step of determining the health level of the rotating device based on the degree of deviation between the fused evaluation feature set and the preset health benchmark feature space includes: The fusion evaluation feature set of the rotating equipment under healthy operating conditions is collected for multiple consecutive time windows. The mean vector and covariance matrix of the fusion evaluation feature set are calculated. The mean vector is used as the center of the healthy baseline feature space, and the covariance matrix is ​​used as the dispersion measure of the healthy baseline feature space. Calculate the distance between the current fusion evaluation feature set and the health benchmark feature space, and use the distance as a quantitative indicator of the degree of deviation to determine the health level of the rotating equipment. The health level includes healthy, sub-healthy, faulty, and dangerous. The distances include Mahalanobis distance and / or Hausdorff distance.

[0013] In one embodiment of the present invention, the step of locating and identifying the fault type and fault location of the rotating device includes: When the health level of the rotating equipment is faulty or dangerous, the dynamic time warping distance between the current fusion evaluation feature set and the theoretical characteristic frequency of each fault type in the mechanism feature library is calculated. The fault type with the smallest dynamic time warping distance is taken as the current fault type, and the fault location corresponding to the fault type is taken as the current fault location.

[0014] In one embodiment of the present invention, Mahalanobis distance is used as the first quantification index, and Hausdorff distance is used as the second quantification index. When the health level determined by the Mahalanobis distance and the Hausdorff distance are consistent, the health level shall be used as the final health level. When the health level determined by the Mahalanobis distance and the Hausdorff distance are inconsistent, the lower of the two levels shall be used as the final health level.

[0015] As can be seen from the above solutions, the advantages of the present invention are: This invention provides a device health assessment method based on the fusion of mechanistic models and parametric diagnostics. By constructing a three-dimensional solid dynamic model of the rotating equipment and simulating fault conditions, it generates theoretical characteristic frequencies for each fault type, forming a mechanistic feature library. This solves the problem of pure data-driven methods relying on a large number of labeled samples. Simultaneously, it introduces a parametric diagnostic algorithm to adaptively decompose measured vibration signals and extract intrinsic dynamic features, avoiding the dependence of traditional mechanistic methods on precise mathematical models. The fusion of these two approaches achieves complementarity between physical mechanisms and data features, improving the accuracy of rotating equipment health diagnosis in industrial scenarios where fault samples are scarce. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a device health assessment method based on the fusion of mechanistic models and parametric diagnostics, provided in an embodiment of the present invention. Figure 2 It shows Figure 1 A flowchart illustrating the empirical mode decomposition method used in step S3. Figure 3 It shows Figure 1 A flowchart illustrating the process of using local mean decomposition in step S3. Figure 4 It shows Figure 1 A flowchart illustrating step S4 in the middle section; Figure 5 This is a schematic diagram of the structure of a device health assessment apparatus provided in another embodiment of the present invention.

[0017] The attached figures are labeled as follows: 400: Equipment health assessment device; 410: Data Acquisition Module; 420: Mechanism module; 430: Parameterless decomposition module; 440: Fusion Module; 450: Identification module. Detailed Implementation

[0018] To make the above features and effects of the present invention clearer and easier to understand, specific embodiments are described below, and detailed descriptions are provided in conjunction with the accompanying drawings.

[0019] refer to Figure 1 As shown, Figure 1 The diagram shows a flowchart of a device health assessment method based on the fusion of mechanistic model and parametric diagnosis provided in an embodiment of the present invention.

[0020] A device health assessment method based on the fusion of mechanistic models and parametric diagnostics includes the following steps: Step S1: Obtain the measured vibration signal generated by the rotating equipment during operation and perform preprocessing.

[0021] In one specific implementation, the rotating device includes at least one of the following rotating devices: gearbox, motor, fan, water pump, reducer, hoist, and belt conveyor; however, this invention is not limited thereto.

[0022] In practical engineering, vibration signals are typically time-domain vibration acceleration signals. Most vibration signals are aperiodic (such as impact vibration) or random, requiring sampling by an accelerometer to obtain discrete data. Accelerometers can be installed as needed on frequently failing components such as bearing housings, gearbox housings, motor housings, equipment bases, or foundation surfaces. For example, bearings are among the most prone to failure in rotating machinery; an accelerometer installed on a bearing can detect bearing fault conditions. An accelerometer mounted on the surface of a gearbox housing can detect gear meshing conditions. An accelerometer mounted on the surface of a motor housing can be used to monitor the motor's operating status.

[0023] The measured vibration acceleration time-domain signal of the rotating equipment during operation is acquired by an accelerometer at a sampling frequency, resulting in a discrete sampling point sequence. Where N is the number of sampling points, determined by the product of the sampling frequency and the sampling duration. This represents the vibration acceleration value measured at the i-th sampling point. Wherein, the sampling frequency is... The Nyquist criterion must be satisfied, i.e. , If the sampling frequency is too low (to the highest frequency of the signal), it will cause high-frequency signal aliasing. For random signals, a sufficiently long sampling duration is required, typically covering the statistically stable range of the signal; otherwise, the average value will be unstable and the effective value will fluctuate greatly.

[0024] In one embodiment, vibration acceleration time-domain signals from all sampling points can be synchronously collected at set time intervals to obtain continuous operating status data of the equipment under normal working conditions. Alternatively, data can be collected periodically according to preset time periods (such as every hour, every 12 hours, 24 hours, etc.). When the equipment alarms or detects abnormal characteristics, the data collection frequency is automatically increased, etc.

[0025] Furthermore, the acquired vibration acceleration time-domain signals undergo preprocessing, including at least one of the following: boundary check automatically removes out-of-range data; abrupt change check filters electromagnetic interference spikes; and removal of invariant data. Specifically, the boundary check uses preset upper and lower limits of the physical range to determine the amplitude of each sample point. If the amplitude of a sample point exceeds the sensor's range, it is determined to be out-of-range data and removed to prevent misdiagnosis due to out-of-range data entering subsequent analysis stages. The abrupt change check calculates the rate of change of amplitude between adjacent sample points. If the rate of change of a sample point exceeds a preset abrupt change threshold, it is determined to be an electromagnetic interference spike and filtered out to prevent interference signals from being misjudged as equipment failure impact characteristics. Removing invariant data involves calculating the variance or standard deviation of the data within the continuous acquisition window. If the variance or standard deviation is below a preset threshold, the data is determined to be invariant and removed.

[0026] Step S2: Construct a three-dimensional solid dynamic model of the rotating equipment, apply excitation loads of different types of faults according to the equipment operating conditions to generate simulated vibration signals corresponding to each fault type, calculate the theoretical characteristic frequency of each fault type, construct the mapping relationship between fault types and theoretical characteristic frequencies, and form a mechanism feature library.

[0027] A three-dimensional solid dynamic model that reflects the real physical structure of rotating equipment is constructed to simulate the dynamic response of the equipment under normal and abnormal conditions.

[0028] Specifically, based on the geometric dimensions, material properties, and component assembly relationships of the rotating equipment, a three-dimensional solid geometric model of the rotating equipment is established using finite element analysis tools (such as ANSYS, ABAQUS, and COMSOL). Material properties are set, including elastic modulus, density, and Poisson's ratio. Component assembly relationships include bearing fits, gear meshing, and coupling connections. Boundary conditions (including fixed and free ends) are set. Then, the three-dimensional solid geometric model is meshed, displacement constraint boundary conditions and load boundary conditions are set, and finite element modal analysis is performed to obtain key parameters such as the first few natural frequencies, mode shapes, and modal damping ratios of the rotating equipment. These parameters are used to identify the resonance region of the equipment.

[0029] Based on the equipment's operating conditions, different fault-type excitation loads are applied to the three-dimensional solid dynamic model to simulate the vibration response under different fault types (such as bearing spalling, rotor eccentricity, and shaft cracks). For each fault type, the corresponding simulated vibration signal is extracted at at least one key measuring point. The excitation load includes at least one of unbalanced excitation force, misalignment excitation torque, bearing local defect impact excitation, and gear local defect impact excitation.

[0030] The vibration mechanism is established based on the vibration generation mechanism corresponding to the typical fault types of the rotating equipment. Specifically, it includes: for imbalance faults, a vibration response model dominated by the 1st rotational frequency is established, whose spectral characteristics are peaks appearing at the 1st rotational frequency in the radial or axial direction, including static imbalance, couple imbalance, vertically mounted rotor imbalance, and cantilever rotor imbalance. For misalignment faults, a vibration response model with the 2nd rotational frequency as a significant feature is established, including parallel misalignment and angular misalignment. For bearing faults, a frequency modulation response model characterized by the characteristic frequencies of the bearing outer ring fault, inner ring fault, rolling element fault, and cage fault is established. For gear faults, a vibration response model characterized by the gear meshing frequency and its sidebands is established. For mechanical loosening faults, a vibration response model characterized by broadband energy increase and nonlinear response is established.

[0031] Finally, based on the simulated vibration signals, the theoretical characteristic frequencies of each fault type are calculated according to the structural parameters of each component in the rotating equipment, such as bearing type, number of gear teeth, number of blades, and shaft speed. These theoretical characteristic frequencies include the characteristic frequencies of bearing inner ring faults, bearing outer ring faults, bearing rolling element faults, bearing cage faults, gear meshing frequencies, gear sideband frequencies, blade passing frequencies, and rotor rotational frequencies. A mapping relationship between fault types and theoretical characteristic frequencies is constructed based on the theoretical characteristic frequencies corresponding to each fault type and added to the mechanism feature library. The mechanism feature library contains no fewer than 11 major equipment category mechanism models, including no fewer than 4 bearing fault mechanism models, no fewer than 3 gear fault mechanism models, and no fewer than 4 power frequency fault mechanism models.

[0032] Step S3: Perform nonlinear dynamic nonparametric decomposition on the measured vibration signal, extract intrinsic dynamic features, and form a nonparametric feature library.

[0033] In vibration fault diagnosis, parameter-free diagnostic algorithms are a class of diagnostic methods that do not rely on precise mathematical models of the equipment, prior statistical distributions, or a large number of labeled samples. They are typically based on the structural characteristics, pattern similarity, or nonlinear dynamic behavior of the signal itself, and are suitable for practical industrial scenarios with few samples, variable operating conditions, and difficult modeling. The key to achieving classification and judgment using parameter-free diagnostic algorithms lies in extracting discriminative features from the original vibration signal that do not require parameter assumptions, and completing the classification and identification of fault categories through unsupervised or weakly supervised methods. "Parameter-free" does not mean that no parameters are required at all, but rather that: it does not assume that the data follows a specific probability distribution (such as a Gaussian distribution); it does not rely on the precise dynamic equations or physical parameters of the equipment; it does not require a large amount of labeled training data to fit complex models; and the algorithm is robust to noise and changes in operating conditions.

[0034] One of the core inventive points of this invention is the integration of nonparametric diagnostics on the basis of the mechanism model. Nonparametric diagnostics is used to perform nonlinear dynamic nonparametric decomposition on the measured vibration signal, extract intrinsic dynamic features, and form a nonparametric feature library.

[0035] In one embodiment, the parameterless diagnostic algorithm employs the empirical mode decomposition method, as detailed in the following reference. Figure 2 As shown, Figure 2 The flowchart of step S3 using empirical mode decomposition is shown.

[0036] Empirical Mode Decomposition (EMD) is an adaptive signal decomposition method that can decompose complex non-stationary and nonlinear signals into several intrinsic mode function (IMF) components.

[0037] Step S311: Perform empirical mode decomposition on the measured vibration signal to decompose it into several intrinsic mode function components. Each intrinsic mode function component represents a local feature signal at a different time scale.

[0038] Let the measured vibration signal at time t be... After EMD decomposition, the following was obtained:

[0039] in, Let be the i-th intrinsic modulo function component, and n be the total number of intrinsic modulo function components. The residual components represent the average trend of the signal. Each intrinsic modulus function component satisfies two conditions: over the entire signal length, the number of extrema is equal to or differs from the number of zero-crossings by at most one; and at any given time, the mean of the upper envelope formed by the local maxima and the lower envelope formed by the local minima is zero.

[0040] Step S312: Perform a Hilbert transform on each of the intrinsic modulus function components to obtain a time-frequency spectrum, and extract time-frequency domain features from the time-frequency spectrum as first-type intrinsic dynamic features; wherein, the time-frequency domain features include instantaneous frequencies. and instantaneous amplitude .

[0041] For each eigenmode function component Perform Hilbert transform:

[0042] in, The result of the Hilbert transform; Constructing analytic signals :

[0043] Where j is the imaginary unit; It is an instantaneous amplitude function. ; For instantaneous phase function, Instantaneous frequency for ; Arrange the instantaneous amplitude and instantaneous frequency of all components according to time and frequency to obtain the Hilbert time spectrum:

[0044] in, The instantaneous angular frequency, This indicates taking the real part.

[0045] Integrating the Hilbert time spectrum on the time axis yields the marginal spectrum. This reflects the energy accumulation of each frequency component over the entire time length:

[0046] Where T is the total duration of the signal. ω is the angular frequency.

[0047] In another embodiment, the parameter-free diagnostic algorithm employs a local mean decomposition method, as detailed in [reference needed]. Figure 3 As shown, Figure 3 The flowchart of step S3 using the local mean decomposition method is shown.

[0048] Local Mean Decomposition (LMD) is an adaptive signal decomposition method. Similar to EMD, LMD can decompose complex non-stationary and nonlinear signals into several product function (PF) components, each of which is obtained by multiplying an envelope signal and a pure frequency-modulated signal.

[0049] Step S321: Perform local mean decomposition on the measured vibration signal to decompose it into several product function components. Each product function component is obtained by multiplying an envelope signal and a pure frequency-modulated signal.

[0050] Let the measured vibration signal at time t be... After LMD decomposition, the following results were obtained:

[0051] in, Let p be the p-th component of the product function, and k be the total number of product function components. This is the residual component.

[0052] For the p-th component of the product function, its instantaneous amplitude is the envelope signal. :

[0053] in This represents the number of iterations required to decompose the p-th PF component. Let be the envelope estimation function for the r-th iteration.

[0054] pure frequency modulation signal instantaneous phase The following is obtained by performing an inverse cosine transform on the pure frequency-modulated signal:

[0055] Alternatively, the phase can be determined by constructing an analytic signal from the pure frequency-modulated signal:

[0056] instantaneous frequency for:

[0057] Arrange the instantaneous amplitudes and frequencies of all PF components according to time and frequency to obtain the LMD time spectrum. :

[0058] in Let be the instantaneous angular frequency of the p-th PF component. For the Dirac function, ω is the angular frequency.

[0059] Step S322: Extract time-frequency domain features from each of the product function components as a second type of intrinsic dynamic features, wherein the time-frequency domain features include instantaneous frequency and instantaneous amplitude.

[0060] In this embodiment, Empirical Mode Decomposition (EMD) or Local Means Decomposition (LMD) is used to perform parameter-free adaptive decomposition of the measured vibration signal. This eliminates the need for preset parameters such as filter bands and wavelet basis functions, automatically decomposing the complex vibration signal into several components with clear physical meanings based entirely on the signal's time-domain characteristics. Each component corresponds to a local oscillation mode of the signal at different time scales. It can simultaneously characterize the local energy distribution of the signal in both time and frequency dimensions, making it particularly suitable for capturing the non-stationary and nonlinear impact characteristics caused by early faults in rotating equipment.

[0061] Step S4: Fuse the theoretical characteristic frequencies in the mechanism feature library with the intrinsic dynamic characteristics in the parameterless feature library to generate a fusion evaluation feature set.

[0062] In one embodiment, reference Figure 4 As shown, Figure 4 A schematic diagram of the specific process for step S4 is shown.

[0063] Step S41: Match the time-frequency domain features of each function component in the parameterless feature library with the theoretical characteristic frequencies of each fault type in the mechanism feature library. When the deviation between the time-frequency domain features of a certain function component and a certain theoretical characteristic frequency is less than a preset frequency tolerance threshold, the function component is determined to be the associated component of the fault type.

[0064] Specifically, the frequency centroids of each intrinsic mode function component are calculated based on the instantaneous frequency and instantaneous amplitude. : ; The frequency centroids of each intrinsic modulus function component are matched with the theoretical characteristic frequencies of each fault type in the aforementioned mechanism feature library. When the frequency centroid of a certain function component... With a certain theoretical characteristic frequency When the deviation between them is less than the preset frequency tolerance threshold, that is... This function component is identified as an associated component of the k-th fault type. The preset frequency tolerance threshold can be adaptively determined based on the rotational speed fluctuation range and spectral resolution of the rotating equipment. For example, the maximum rotational speed can be calculated by collecting rotational speed data of the equipment under rated operating conditions for a certain period of time using a speed sensor. Minimum value ,get: ,in, Rotational speed The average value. Based on the time-frequency analysis method used, the spectral resolution is calculated. for: ,in Where is the signal sampling rate, and N is the time-frequency analysis window length or the number of FFT points. The adaptive tolerance threshold for the k-th type of fault. It can be set as follows: .

[0065] Step S42: Based on the similarity between the time-frequency domain features of the associated component and the theoretical characteristic frequencies of the corresponding fault type in the mechanism feature library, determine the confidence weight of the associated component, and use the confidence weight to weight and combine the intrinsic dynamic features in the parameterless feature library and the theoretical characteristic frequencies in the mechanism feature library to obtain the fusion evaluation feature set.

[0066] Specifically, based on the frequency centroid of the associated components With theoretical characteristic frequency Calculate the similarity to obtain the confidence weights of the associated components. :

[0067] With the confidence weight The intrinsic dynamic features in the parameterless feature vector and the theoretical feature frequencies in the mechanistic feature library are weighted and combined to obtain the fusion evaluation feature set. :

[0068] in, Let be the intrinsic dynamic eigenvector of the i-th function component, and m be the total number of function components.

[0069] In this embodiment, the intrinsic dynamic features obtained from parametric decomposition are fused across domains with the theoretical characteristic frequencies calculated by the mechanistic model, achieving complementarity between physical mechanisms and data features. Specifically, a frequency centroid matching mechanism is used to associate EMD / LMD decomposition components with fault types in the mechanistic feature library, and confidence weights are calculated based on an exponential decay function of the matching deviation. The smaller the deviation, the higher the weight, thereby adaptively highlighting feature components highly correlated with known fault modes and suppressing noise components unrelated to faults. The fused feature set retains the physical structure information of the signal itself while introducing prior constraints from mechanistic knowledge, effectively improving the accuracy of subsequent health level assessment and fault location.

[0070] Step S5: Determine the health level of the rotating equipment based on the degree of deviation between the fused evaluation feature set and the preset health benchmark feature space, and locate and identify the fault type and fault location of the rotating equipment.

[0071] In one embodiment, a fusion evaluation feature set is collected for n consecutive time windows when the rotating device is in a healthy operating state. Calculate the mean vector of the fusion evaluation feature set. Covariance Matrix The mean vector is used as the center of the health baseline feature space, and the covariance matrix is ​​used as the dispersion measure of the health baseline feature space.

[0072] Calculate the distance between the current fusion evaluation feature set and the health benchmark feature space, and use the distance as a quantitative indicator of the degree of deviation to determine the health level of the rotating equipment. The health level includes healthy, sub-healthy, faulty, and dangerous.

[0073] In one specific implementation, Mahalanobis distance and / or Hausdorff distance are used as quantitative indicators of the degree of deviation.

[0074] Mahalanobis distance is a metric for measuring the distance between a sample point and a distribution. Unlike Euclidean distance, Mahalanobis distance considers the covariance structure of the data, thus eliminating the influence of dimensions and correlations between features. Calculate the current fusion evaluation feature set. Mahalanobis distance between the health benchmark feature space and the for:

[0075] By calculating the Mahalanobis distances corresponding to the fused evaluation feature sets for each time window, a sample set of Mahalanobis distances for health status is obtained. The Mahalanobis distance of the health status sample set as a mean Standard deviation ,in Let be the value of the i-th Mahalanobis distance sample.

[0076] With interval Corresponding to health levels, in intervals Corresponding to the sub-health level, in intervals Corresponding to the fault level, in interval Corresponding hazard level; among which For an incremental threshold factor, for example, using the common 3σ principle, set... .

[0077] Hausdorff distance is a metric for measuring the similarity between two sets; it reflects the maximum degree of mismatch between the two sets. The current fusion evaluation feature set is calculated. The Hausdorff distance H between the health baseline feature space and the health baseline feature space is:

[0078] Then, similar to Mahalanobis distance as a quantitative indicator, Hausdorff distance is used as a quantitative indicator. The Hausdorff distances corresponding to the fusion evaluation feature sets of each time window are statistically analyzed to obtain the Hausdorff distance sample set of the health status. The mean and standard deviation of the Mahalanobis distance sample set of the health status are calculated, and different levels are set according to the mean and standard deviation. The specific settings are similar to those of Mahalanobis distance, and will not be elaborated here.

[0079] Furthermore, Mahalanobis distance can be used simultaneously as the first quantification indicator and Hausdorff distance as the second quantification indicator. When the health levels determined by the Mahalanobis distance and the Hausdorff distance are consistent, that health level is taken as the final health level. When the health levels determined by the Mahalanobis distance and the Hausdorff distance are inconsistent, the lower of the two levels is taken as the final health level.

[0080] Furthermore, when the health level of the rotating equipment is faulty or dangerous, the dynamic time warping distance between the current fusion evaluation feature set and the theoretical characteristic frequency of each fault type in the mechanism feature library is calculated. The fault type with the smallest dynamic time warping distance is taken as the current fault type, and the fault location corresponding to the fault type is taken as the current fault location.

[0081] In this embodiment, Dynamic Time Warping (DTW) is used as an algorithm to measure the similarity between two time series. Unlike point-by-point matching methods such as Euclidean distance, DTW allows the two series to undergo non-linear scaling on the time axis, thereby finding the optimal alignment path between the two series. Therefore, it has better adaptability to sequences with different lengths or phase shifts.

[0082] Calculate the current fusion evaluation feature set The theoretical characteristic frequency of the k-th type of fault Dynamic time-normalized distance between them:

[0083] in Let be the number of theoretical characteristic frequencies of the k-th type of fault. Let p be the p-th sample in the current fusion evaluation feature set, where p is the feature dimension of the current fusion evaluation feature set; For the k-th type of fault One theoretical characteristic frequency; C is the cumulative distance matrix. Let C(i, j) be the last element of the matrix, where the (i, j)th element C(i, j) represents the minimum cumulative distance from the starting point (1, 1) to the point (i, j), i.e.:

[0084] Traverse all K types of faults in the mechanism feature library, calculate their respective DTW distances, and select the fault type with the smallest DTW distance as the current fault type:

[0085] in, The identified fault type number corresponds to the fault location that is pre-mapped in the mechanism feature library for that fault type.

[0086] In this embodiment, Mahalanobis distance and / or Hausdorff distance are used as indices. Mahalanobis distance is used to evaluate the overall distribution deviation dimension, while Hausdorff distance can provide a complementary evaluation from the perspective of local maximum deviation, thus increasing the accuracy of health level identification. When a fault or hazard is determined, DTW distance is further used to accurately locate the fault type. The DTW algorithm allows for nonlinear scaling and alignment of theoretical characteristic frequencies with actual characteristics on the frequency axis, overcoming the sensitivity of traditional point-by-point matching methods to frequency offset, and can accurately identify the fault type and corresponding fault location.

[0087] In summary, the equipment health assessment method based on the fusion of mechanistic modeling and parametric diagnostics provided by this invention constructs a three-dimensional solid dynamic model of rotating equipment and performs fault condition simulation to generate theoretical characteristic frequencies for each fault type, forming a mechanistic feature library. This solves the problem of pure data-driven methods relying on a large number of labeled samples. Simultaneously, it introduces a parametric diagnostic algorithm to adaptively decompose measured vibration signals and extract intrinsic dynamic features, avoiding the dependence of traditional mechanistic methods on precise mathematical models. The fusion of these two approaches achieves complementarity between physical mechanisms and data features, improving the accuracy of rotating equipment health diagnosis in industrial scenarios where fault samples are scarce.

[0088] In one embodiment, reference Figure 5 As shown, Figure 5 A schematic diagram of the overall structure of a device health assessment apparatus 400 is shown.

[0089] A device health assessment device 400 based on the fusion of mechanistic model and parametric diagnostics includes: Acquisition module 410: Used to acquire measured vibration signals generated by rotating equipment during operation.

[0090] Mechanism module 420 is used to construct a three-dimensional solid dynamic model of the rotating equipment, apply excitation loads of different types of faults according to the operating conditions of the equipment to generate simulated vibration signals corresponding to each fault type, calculate the theoretical characteristic frequency of each fault type, construct the mapping relationship between fault type and theoretical characteristic frequency, and form a mechanism feature library.

[0091] The non-parametric decomposition module 430 is used to perform non-linear dynamic non-parametric decomposition on the measured vibration signal, extract intrinsic dynamic features, and form a non-parametric feature library. The fusion module 440 is used to fuse the theoretical characteristic frequencies in the mechanism feature library with the intrinsic dynamic characteristics in the parameterless feature library to generate a fusion evaluation feature set. The identification module 450 is used to determine the health level of the rotating equipment based on the degree of deviation between the fused evaluation feature set and the preset health benchmark feature space, and to locate and identify the fault type and fault location of the rotating equipment.

[0092] It should be noted that this device implementation can be combined with the above-described method implementation. The relevant technical details mentioned in the above method implementation remain valid in this device implementation, and will not be repeated here to avoid repetition.

[0093] In addition, embodiments of the present invention also provide a readable storage medium storing a program or instructions that, when executed by a processor, implement the steps of the above-described device health assessment method based on the fusion of mechanistic model and parameterless diagnosis, and achieve the same technical effect.

[0094] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described device health assessment method based on the fusion of mechanistic model and parameterless diagnosis, and achieves the same technical effect.

[0095] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of the present invention is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be applied, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0096] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of the present invention.

Claims

1. A device health assessment method based on the fusion of mechanistic models and parametric diagnostics, characterized in that, include: Acquire measured vibration signals generated during the operation of rotating equipment; A three-dimensional solid dynamic model of the rotating equipment is constructed. Excitation loads of different types of faults are applied according to the operating conditions of the equipment to generate simulated vibration signals corresponding to each fault type. The theoretical characteristic frequencies of each fault type are calculated, and the mapping relationship between fault types and theoretical characteristic frequencies is constructed to form a mechanism feature library. The measured vibration signal is subjected to nonlinear dynamic nonparametric decomposition to extract intrinsic dynamic features and form a nonparametric feature library. The theoretical characteristic frequencies in the mechanistic feature library are fused with the intrinsic dynamic features in the parameterless feature library to generate a fused evaluation feature set; Based on the degree of deviation between the fused evaluation feature set and the preset health benchmark feature space, the health level of the rotating equipment is determined, and the fault type and fault location of the rotating equipment are located and identified.

2. The method according to claim 1, characterized in that, The steps of constructing a three-dimensional solid dynamic model of the rotating equipment, applying excitation loads of different fault types according to the equipment's operating conditions to generate simulated vibration signals corresponding to each fault type, and forming a virtual dataset include: Based on the geometric dimensions, material properties, and component assembly relationships of the rotating equipment, a three-dimensional solid geometric model of the rotating equipment is established. The material properties include elastic modulus, density, and Poisson's ratio. The component assembly relationships include bearing fit, gear meshing, and coupling connection. According to the operating conditions of the equipment, different fault types of excitation loads are applied to the three-dimensional solid dynamic model to simulate the vibration response under different fault types, and the corresponding simulated vibration signal is extracted at at least one key measuring point for each fault type. The excitation load includes at least one of unbalanced excitation force, misalignment excitation torque, bearing local defect impact excitation, and gear local defect impact excitation.

3. The method according to claim 2, characterized in that, The steps of calculating the theoretical characteristic frequencies of each fault type, constructing the mapping relationship between fault types and theoretical characteristic frequencies, and forming a mechanism feature library include: Based on the simulated vibration signal, the theoretical characteristic frequencies of each fault type are calculated according to the structural parameters of each component in the rotating equipment. The structural parameters include bearing type, number of gear teeth, number of blades, and shaft speed. The theoretical characteristic frequencies include bearing inner ring fault characteristic frequency, bearing outer ring fault characteristic frequency, bearing rolling element fault characteristic frequency, bearing cage fault characteristic frequency, gear meshing frequency, gear sideband frequencies, blade passing frequency, and rotor rotation frequency. The mapping relationship between fault types and theoretical characteristic frequencies is constructed based on the theoretical characteristic frequencies corresponding to each fault type and added to the mechanism feature library.

4. The method according to claim 3, characterized in that, The steps of performing nonlinear dynamic nonparametric decomposition on the measured vibration signal, extracting intrinsic dynamic features, and forming a nonparametric feature library include: The measured vibration signal is subjected to empirical mode decomposition, which decomposes it into several intrinsic mode function components. Each intrinsic mode function component represents a local feature signal at a different time scale. Perform a Hilbert transform on each of the intrinsic mode function components to obtain a time-frequency spectrum, and extract time-frequency domain features from the time-frequency spectrum as the first type of intrinsic dynamic features; The time-frequency domain features include instantaneous frequency and instantaneous amplitude.

5. The method according to claim 3, characterized in that, The steps of performing nonlinear dynamic nonparametric decomposition on the measured vibration signal, extracting intrinsic dynamic features, and forming a nonparametric feature library further include: The measured vibration signal is decomposed into a local mean decomposition into several product function components. Each product function component is obtained by multiplying an envelope signal and a pure frequency-modulated signal. Time-frequency domain features are extracted from each component of the product function as second-type intrinsic dynamic features; The time-frequency domain features include instantaneous frequency and instantaneous amplitude.

6. The method according to claim 4 or 5, characterized in that, The step of fusing the theoretical characteristic frequencies in the mechanistic feature library with the intrinsic dynamic features in the parameterless feature library to generate a fused evaluation feature set includes: The time-frequency domain features of each function component in the parameterless feature library are matched with the theoretical characteristic frequencies of each fault type in the mechanism feature library. When the deviation between the time-frequency domain features of a certain function component and a certain theoretical characteristic frequency is less than a preset frequency tolerance threshold, the function component is determined to be the associated component of the fault type. Based on the similarity between the time-frequency domain features of the associated components and the theoretical characteristic frequencies of the corresponding fault types in the mechanism feature library, the confidence weight of the associated components is determined. The confidence weight is then used to weight and combine the intrinsic dynamic features in the parameterless feature library and the theoretical characteristic frequencies in the mechanism feature library to obtain the fusion evaluation feature set. The preset frequency tolerance threshold is adaptively determined based on the rotational speed fluctuation range and spectral resolution of the rotating device.

7. The method according to claim 6, characterized in that, Calculate the frequency centroid of each intrinsic mode function component based on the instantaneous frequency and instantaneous amplitude; The frequency centroids of each intrinsic modulus function component are matched with the theoretical characteristic frequencies of each fault type in the mechanism feature library. When the deviation between the frequency centroid of a certain function component and a certain theoretical characteristic frequency is less than a preset frequency tolerance threshold, the function component is determined to be the associated component of the fault type.

8. The method according to claim 6, characterized in that, The step of determining the health level of the rotating equipment based on the deviation between the fused evaluation feature set and the preset health benchmark feature space includes: The fusion evaluation feature set of the rotating equipment under healthy operating conditions is collected for multiple consecutive time windows. The mean vector and covariance matrix of the fusion evaluation feature set are calculated. The mean vector is used as the center of the healthy baseline feature space, and the covariance matrix is ​​used as the dispersion measure of the healthy baseline feature space. Calculate the distance between the current fusion evaluation feature set and the health benchmark feature space, and use the distance as a quantitative indicator of the degree of deviation to determine the health level of the rotating equipment. The health level includes healthy, sub-healthy, faulty, and dangerous. The distances include Mahalanobis distance and / or Hausdorff distance.

9. The method according to claim 8, characterized in that, The steps for locating and identifying the fault type and location of the rotating equipment include: When the health level of the rotating equipment is faulty or dangerous, the dynamic time warping distance between the current fusion evaluation feature set and the theoretical characteristic frequency of each fault type in the mechanism feature library is calculated. The fault type with the smallest dynamic time warping distance is taken as the current fault type, and the fault location corresponding to the fault type is taken as the current fault location.

10. The method according to claim 8, characterized in that, Mahalanobis distance was used as the first quantification index, and Hausdorff distance was used as the second quantification index. When the health level determined by the Mahalanobis distance and the Hausdorff distance are consistent, the health level shall be used as the final health level. When the health level determined by the Mahalanobis distance and the Hausdorff distance are inconsistent, the lower of the two levels shall be used as the final health level.

Images