Health assessment method and system based on electrical characteristics and electric fingerprint identification of industrial equipment

Abstract

The invention relates to the technical field of equipment health assessment, and discloses a health assessment method and system based on industrial equipment electrical characteristics and electric fingerprint identification. The method comprises the steps of obtaining an electrical signal of industrial equipment, extracting a time domain statistical feature, a frequency domain harmonic feature, bearing fault feature frequency sideband energy, a wavelet packet energy entropy and a bispectrum amplitude feature in the electrical signal, and constructing an original feature vector; performing standardization processing on the original feature vector to obtain a comprehensive electric fingerprint feature vector, and calculating a distance value from the comprehensive electric fingerprint feature vector to a sphere center through a pre-established health reference model; and mapping the distance value into an equipment health degree score, positioning a fault feature component causing health degree reduction, and generating a fault diagnosis suggestion report. According to the invention, through deep fusion of multi-dimensional electrical fingerprint features, unsupervised health benchmark modeling and continuous quantitative evaluation, accurate evaluation and early fault early warning of the health state of the industrial equipment are realized.

Description

Technical Field

[0001] This invention relates to the field of equipment health assessment technology, and in particular to a health assessment method and system based on the electrical characteristics and electronic fingerprint recognition of industrial equipment. Background Technology

[0002] Predictive maintenance of industrial equipment has become a key technology for ensuring production safety and reducing operation and maintenance costs. Traditional equipment health assessment methods rely on periodic inspections or monitoring based on vibration and temperature sensors, which suffer from problems such as over-maintenance, high installation costs, and difficulty in adapting to complex industrial environments. Furthermore, simple current threshold alarms cannot identify early performance degradation of equipment and cannot achieve quantitative assessment of its health status.

[0003] As an emerging non-invasive monitoring method, electronic fingerprint technology identifies the status of equipment by analyzing electrical signals during operation. However, the feature extraction of existing technologies mostly relies on simple features in the time or frequency domains, making it difficult to capture early and weak fault signs. Supervised learning is used for binary classification, which cannot provide a continuously quantified health score. Furthermore, it heavily relies on a large amount of fault sample data for training, while actual industrial field fault data is extremely scarce, resulting in weak model generalization ability and poor engineering practicality. Summary of the Invention

[0004] The main objective of this invention is to provide a health assessment method and system based on the electrical characteristics and electronic fingerprint recognition of industrial equipment. This invention achieves accurate assessment of the health status of industrial equipment and early fault warning through deep fusion of multi-dimensional electronic fingerprint features, unsupervised health benchmark modeling, and continuous quantitative assessment.

[0005] To achieve the above objectives, the present invention provides a health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment, comprising the following steps: Acquire electrical signals from industrial equipment, and extract time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signals to construct the original feature vector; The original feature vector is standardized to obtain a comprehensive electronic fingerprint feature vector, and the distance from the comprehensive electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health benchmark model. The distance value is mapped to a device health score, and the fault characteristic components that cause the health decline are located, generating a fault diagnosis suggestion report.

[0006] Optionally, in a first implementation of the first aspect of the present invention, an electrical signal from an industrial device is acquired, and time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features are extracted from the electrical signal to construct an original feature vector, including: Anti-aliasing filtering is performed on the three-phase current signal and three-phase voltage signal of industrial equipment to obtain the anti-aliasing filtered signal; Pulse noise is removed from the anti-aliasing filtered signal to obtain a denoised signal, and DC components with frequencies below a preset frequency and low-frequency drift are removed from the denoised signal to obtain an electrical signal. Extract the time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signal to construct the original feature vector.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features of the electrical signal are extracted to construct an original feature vector, including: Extract the time-domain statistical features of the electrical signal; Perform a fast Fourier transform on the electrical signal to obtain the frequency domain harmonic characteristics and bearing fault characteristic frequency sideband energy; The electrical signal is decomposed into wavelet packets to obtain the wavelet packet energy entropy. The electrical signal is subjected to bispectral calculation to obtain bispectral amplitude characteristics; The original feature vector is constructed based on the time-domain statistical features, the frequency-domain harmonic features, the bearing fault feature frequency sideband energy, the wavelet packet energy entropy, and the bispectral amplitude features.

[0008] Optionally, in a third implementation of the first aspect of the present invention, bispectral calculation is performed on the electrical signal to obtain bispectral amplitude characteristics, including: Calculate the bispectral matrix of an electrical signal; The amplitude of the bispectral matrix is ​​extracted, and the peak value, mean, and variance of the amplitude are used as bispectral amplitude features.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the original feature vector is standardized to obtain a comprehensive electronic fingerprint feature vector, and the distance from the comprehensive electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health benchmark model, including: Calculate the standardized parameters for each feature dimension in the historical health dataset; Based on the standardization parameters of each feature dimension, each feature component in the original feature vector is standardized to obtain the comprehensive electronic fingerprint feature vector. The distance from the integrated electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health benchmark model.

[0010] Optionally, in a fifth implementation of the first aspect of the present invention, before calculating the distance value from the integrated electronic fingerprint feature vector to the center of the sphere, the method further includes: When industrial equipment is in a healthy state, a health benchmark feature dataset is constructed; The support vector data description algorithm is used to train the health benchmark feature dataset to solve for the center and radius of the minimum radius hypersphere; The healthy boundary radius is obtained by multiplying the radius by the boundary expansion coefficient, and a healthy baseline model is constructed based on the sphere center, the healthy boundary radius, the support vector set, the Lagrange multipliers, and the kernel parameters.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, calculating the distance from the integrated electronic fingerprint feature vector to the center of the sphere using a pre-established health benchmark model includes: Extract the sphere center, support vector set, Lagrange multipliers, and kernel parameters from a pre-established health benchmark model; The first kernel function value of the integrated electronic fingerprint feature vector itself, the second kernel function value of the integrated electronic fingerprint feature vector and each support vector, and the third kernel function value between each support vector are calculated using the radial basis kernel function. The distance from the integrated electronic fingerprint feature vector to the center of the sphere is calculated based on the first kernel function value, the second kernel function value, the third kernel function value, and the Lagrange multiplier.

[0012] Optionally, in the seventh implementation of the first aspect of the present invention, the calculation of the first kernel function value of the integrated electronic fingerprint feature vector itself, the second kernel function value between the integrated electronic fingerprint feature vector and each support vector, and the third kernel function value between each support vector using the radial basis function includes: Calculate the sum of squared differences between each standard feature component in the integrated electronic fingerprint feature vector and its own standard feature components, and substitute the sum of squared differences into the radial basis kernel function to calculate the first kernel function value; Calculate the sum of squared differences between each standard feature component and each support vector in the integrated electronic fingerprint feature vector, and substitute the sum of squared differences into the radial basis kernel function to calculate the value of the second kernel function; Calculate the sum of squared differences between each support vector and substitute it into the radial basis function to obtain the value of the third kernel function.

[0013] Optionally, in an eighth implementation of the first aspect of the present invention, the distance value is mapped to a device health score and the fault feature components causing the health decline are located, generating a fault diagnosis suggestion report, including: Map the distance value to a device health score; The device health score is compared with the first threshold, the second threshold, and the third threshold, respectively. When the device health score is between the first threshold and the second threshold, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the sphere is calculated. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated as an early warning. When the device health score is between the second threshold and the third threshold, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the ball is calculated respectively. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated as a serious alarm. When the device health score is less than the third threshold, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the sphere is calculated. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated, which suggests shutdown and maintenance.

[0014] This invention also provides a health assessment system based on the electrical characteristics and electronic fingerprint recognition of industrial equipment, comprising: The feature extraction module is used to acquire electrical signals from industrial equipment and extract time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signals to construct the original feature vector. The calculation module is used to standardize the original feature vector to obtain a comprehensive electronic fingerprint feature vector, and to calculate the distance from the comprehensive electronic fingerprint feature vector to the center of the sphere using a pre-established health benchmark model. The fault diagnosis module is used to map the distance value to a device health score, locate the fault characteristic components that cause the health decline, and generate a fault diagnosis suggestion report.

[0015] In summary, this invention achieves accurate assessment of the health status of industrial equipment and early fault warning through deep fusion of multi-dimensional electrical fingerprint features, unsupervised health benchmark modeling, and continuous quantitative evaluation. This invention extracts time-domain statistical features, fast Fourier transform frequency-domain harmonic features, bearing fault feature frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features to construct a comprehensive electrical fingerprint feature vector. Compared to existing methods that rely solely on simple time-domain or frequency-domain features, this invention's multi-dimensional feature fusion comprehensively characterizes the time-domain, frequency-domain, time-frequency-domain, and higher-order statistical properties of equipment electrical signals, exhibiting higher sensitivity and identification accuracy for early, subtle faults. This invention employs a support vector data description algorithm to establish a health benchmark model, requiring only equipment health status data for training, thus solving the problem of scarce fault samples in industrial settings and improving engineering practicality and scalability. This invention achieves continuous quantitative scoring of equipment health status through distance measurement and exponential decay function mapping, transforming health metrics into intuitive scores from 0 to 100. It also establishes a multi-level threshold early warning mechanism. By calculating the deviation of characteristic components and generating fault diagnosis suggestion reports, it can accurately locate fault characteristics leading to a decline in health, tracing the root cause of the fault to specific physical components and fault types, and providing maintenance personnel with precise repair guidance. This invention only requires the installation of sensors on the power supply side, without modifying the equipment itself, and has the advantages of low cost, simple deployment, and wide applicability. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the steps of a health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment in one embodiment of the present invention; Figure 2 This is a block diagram of a health assessment system based on the electrical characteristics and electronic fingerprint recognition of industrial equipment in one embodiment of the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] Reference Figure 1 This embodiment provides a health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment, including the following steps: S1. Acquire the electrical signals of industrial equipment and extract the time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signals to construct the original feature vector; S2, standardize the original feature vector to obtain the comprehensive electronic fingerprint feature vector, and calculate the distance from the comprehensive electronic fingerprint feature vector to the center of the ball through a pre-established health benchmark model; S3 maps the distance value to the device health score and locates the fault characteristic components that cause the health decline, generating a fault diagnosis suggestion report.

[0020] In one example, the electrical signals of industrial equipment are acquired, and the time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features are extracted from the electrical signals to construct the original feature vector, including: Anti-aliasing filtering is performed on the three-phase current signal and three-phase voltage signal of industrial equipment to obtain the anti-aliasing filtered signal; Pulse noise is removed from the anti-aliasing filtered signal to obtain a denoised signal. The DC component with a frequency lower than the preset frequency and low-frequency drift in the denoised signal are then removed to obtain an electrical signal. Extract time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from electrical signals to construct the original feature vector.

[0021] In this example, anti-aliasing filtering is performed on the three-phase current and three-phase voltage signals. Since the data acquisition process uses a high-frequency sampling card for synchronous measurement, high-frequency interference components are introduced into the signal. Based on the Nyquist sampling theorem, a low-pass filter with a reasonable cutoff frequency is designed to perform anti-aliasing filtering on the original signal, preserving the true signal frequency band within the sampling bandwidth while suppressing high-frequency noise exceeding the bandwidth limit, resulting in an anti-aliasing filtered signal. Impulse noise removal is then performed on the anti-aliasing filtered signal. Median filtering or a variable-scale discrimination threshold algorithm is used to identify and eliminate spike-like impulse noise introduced by transient arcs, relay jumps, coupling interference, etc., obtaining a denoised signal with good continuity and stable structure. To eliminate the bias in feature calculation caused by the superposition of the power frequency fundamental wave with DC bias or low-frequency drift caused by the environment, which is common in equipment power supply networks, a high-pass filtering operation is performed on the denoised signal. By setting a reasonable cutoff frequency, DC components below a preset frequency and low-frequency baseline drift terms are removed, ensuring that the electrical signal reflects the true energy and frequency characteristics of the equipment dynamically changing under operating conditions. Based on the purified electrical signal, the signal is grouped into segments using a sliding time window. Within each segment, its time-domain statistical characteristics (such as mean, variance, RMS value, peak value, kurtosis, etc.) are calculated to characterize the intensity and impact of current fluctuations. A Fast Fourier Transform (FFT) is performed on each segment to extract frequency-domain harmonic features, including the fundamental frequency amplitude, 2nd to 25th harmonic amplitudes, and phase information, reflecting changes in equipment load characteristics or power supply harmonic pollution. Simultaneously, bearing fault characteristic frequencies (such as outer ring fault frequency BPFO, inner ring fault frequency BPFI, rolling element pass frequency BSF, etc.) are calculated based on the structural parameters of the target equipment, and energy density changes in adjacent frequency bands are statistically analyzed to reflect potential bearing degradation trends. Wavelet packet decomposition is performed on the electrical signal to obtain the energy distribution of sub-signals in each frequency band. Energy entropy indices are calculated to capture the transient behavior and frequency transitions of non-stationary waveforms. Third-order spectral analysis is used to extract bispectral amplitude features to reveal non-Gaussian components and nonlinear coupling effects in the signal. All these feature values ​​are then combined to form the original feature vector.

[0022] In one example, time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features are extracted from the electrical signal to construct the original feature vector, including: Extracting the time-domain statistical features of electrical signals; By performing a fast Fourier transform on the electrical signal, the frequency domain harmonic characteristics and bearing fault characteristic frequency sideband energy are obtained. Wavelet packet decomposition is performed on the electrical signal to obtain the wavelet packet energy entropy; Bispectral calculations are performed on electrical signals to obtain bispectral amplitude characteristics; The original feature vector is constructed based on time-domain statistical features, frequency-domain harmonic features, bearing fault feature frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features.

[0023] In this example, the statistical characteristics of the signal are analyzed in the time domain. By performing statistical calculations on the signal within each sampling time window, multiple time-domain statistical features are obtained, including mean, root mean square value, peak value, skewness, kurtosis, crest factor, and margin factor. These features reflect the energy distribution, fluctuation pattern, and impact degree of the electrical signal over time, which helps to identify load abrupt changes and mechanical clearance problems. The signal within the same window is subjected to a Fast Fourier Transform to obtain its spectral structure and extract frequency domain harmonic features, including the fundamental frequency component amplitude, the amplitudes of each higher harmonic, and their phase change characteristics. This reveals the energy periodicity and harmonic distortion of the equipment under steady-state operation. Simultaneously, common bearing fault frequencies (such as BPFO, BPFI, BSF, etc.) are calculated based on the equipment bearing structural parameters. Then, the amplitude energy of these characteristic frequencies and their sideband regions is statistically analyzed in the frequency domain to uncover subtle signs of rotational faults. To capture the dynamic changes of electrical signals during non-steady-state operation, wavelet packet decomposition is performed. A multi-scale decomposition tree is constructed to divide the signal into several frequency band sub-signals, and the energy proportion of each node is calculated to obtain the wavelet packet energy entropy. This reflects the energy complexity and uncertainty of the signal in the frequency dimension, identifying the startup process or sudden anomalies. To extract the nonlinear and non-Gaussian structures in the signal, the third-order bispectral or bispectral amplitude map of the electrical signal is calculated. From this, bispectral amplitude features reflecting coupling phase and amplitude clustering are extracted, revealing potential higher-order modulation effects and non-stationary interference sources. All time-domain statistical features, frequency-domain harmonic features, bearing characteristic frequency band energy, wavelet packet energy entropy, and bispectral amplitude features are uniformly arranged and spliced ​​to form the original feature vector.

[0024] In one example, bispectral calculation is performed on an electrical signal to obtain bispectral amplitude characteristics, including: Calculate the bispectral matrix of an electrical signal; The amplitude of the bispectral matrix is ​​extracted, and the peak value, mean, and variance of the amplitude are used as bispectral amplitude features.

[0025] In this example, a bispectral analysis is performed on the electrical signal to calculate the corresponding bispectral matrix. Based on the Fourier transform of the third-order cumulants, the coupling relationship between frequency components in the signal is analyzed in a two-dimensional frequency domain. By traversing the frequency pairs (f1, f2), the corresponding third-order spectral values ​​are calculated, forming a two-dimensional complex matrix, the bispectral matrix. The real and imaginary parts of the bispectral matrix reflect the phase coupling characteristics and amplitude coupling strength between different frequency components, respectively, while its complex modulus length is the bispectral amplitude, an indicator revealing nonlinear modulation effects and higher-order harmonic interactions. Feature compression and statistical index extraction are performed on the amplitude portion of the bispectral matrix, extracting the peak value, mean, and variance of the amplitude distribution from the entire amplitude matrix. The peak value is used to identify the presence of significant frequency interaction enhancement phenomena, the mean reflects the overall intensity level of the bispectral energy, and the variance reflects the dispersion and complexity of the bispectral amplitude distribution, thus collectively forming a three-dimensional bispectral amplitude feature vector to describe the higher-order coupling characteristics of the electrical signal. The bispectral amplitude feature vector has strong anomaly sensitivity when the fault characteristics are extremely weak and the changes in traditional frequency domain indicators are not obvious. It is suitable for capturing modulation signal components caused by factors such as bearing inner ring peeling, eccentric disturbance, and intermittent arc discharge.

[0026] In one example, the original feature vector is standardized to obtain a comprehensive electronic fingerprint feature vector. The distance from the comprehensive electronic fingerprint feature vector to the center of the sphere is then calculated using a pre-established health benchmark model, including: Calculate the standardized parameters for each feature dimension in the historical health dataset; Based on the standardization parameters of each feature dimension, the feature components in the original feature vector are standardized to obtain the comprehensive electronic fingerprint feature vector. The distance from the integrated electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health benchmark model.

[0027] In this example, standardized parameters are calculated for each feature dimension in the historical health dataset. When equipment is confirmed to be in a healthy operating condition (e.g., new equipment leaving the factory, initial trial operation, or a manually verified stable operating phase), multi-source signals such as current and voltage are continuously collected, and raw feature vectors from multiple time periods are extracted to form a representative health dataset. The mean and standard deviation of each feature dimension in the health dataset are calculated. The mean characterizes the central trend of the feature under healthy conditions, while the standard deviation measures its fluctuation range, constituting statistical reference parameters for standardization. Before performing a health assessment on the current raw feature vector collected from online operating equipment, each component is standardized according to the mean and standard deviation in the health baseline data. Specifically, a zero-mean, unit variance method is used to subtract the historical health mean of the corresponding dimension from each feature value and then divide by the corresponding standard deviation. This maps all features to a dimensionless, standard normal space with a mean of 0 and a standard deviation of 1, ensuring that different features have equivalent weight contributions in distance calculation and preventing certain dimensions from dominating the final result due to differences in numerical scale. This generates a standardized comprehensive electronic fingerprint feature vector. The integrated electronic fingerprint feature vector is input into a pre-established health benchmark model to calculate the deviation of the health status. The model is constructed using a support vector description (SVDD) method, which fits historical health feature vector data to form a minimum bounding sphere or hypersphere structure, where the center of the sphere represents the central distribution location of the health data. By calculating the Euclidean distance or kernel-based high-dimensional mapping distance between the standardized feature vector and the sphere center coordinates in a high-dimensional feature space, the deviation of the real-time vector from the center of the health status is obtained, denoted as the distance value D. The smaller the distance value, the higher the similarity between the current operating status of the device and the historical health status; if the distance value is larger, it indicates that the current device has exhibited a certain degree of electrical behavioral abnormality, indicating performance degradation or early potential failure.

[0028] Before standardizing the original feature vector, the process includes: detecting whether there are missing values ​​or non-numerical values ​​in each feature component of the original feature vector; when abnormal data is detected, the feature vector is marked as invalid and discarded; calculating the Euclidean modulus of the original feature vector and comparing the Euclidean modulus with the statistical range of the modulus of the original feature vector in the historical health dataset; when the Euclidean modulus exceeds five times the statistical range, it is determined as a data acquisition anomaly and an emergency warning is triggered; calculating the cosine similarity between the original feature vector at the current time and the original feature vector at the previous time; when the cosine similarity is lower than a preset similarity threshold, it is determined as a drastic change in equipment operating conditions and is marked as abnormal operating condition data.

[0029] In one example, before calculating the distance from the composite electronic fingerprint feature vector to the center of the sphere, the following steps are also included: When industrial equipment is in a healthy state, a health benchmark feature dataset is constructed; The support vector data description algorithm is used to train on the health benchmark feature dataset to solve for the center and radius of the minimum radius hypersphere; The healthy boundary radius is obtained by multiplying the radius by the boundary expansion coefficient, and a healthy baseline model is constructed based on the sphere center, the healthy boundary radius, the set of support vectors, the Lagrange multipliers, and the kernel parameters.

[0030] In this example, operational data is fully collected and features extracted during the healthy operating phase of the equipment. The process involves continuously collecting three-phase current and voltage signals during periods when the equipment has just completed factory commissioning, is running smoothly, or has been confirmed by experts to be free of abnormalities. Following a predetermined process, multi-dimensional information such as time-domain statistical features, frequency-domain harmonic features, bearing characteristic frequency energy, wavelet packet energy entropy, and bispectral amplitude are extracted to construct a feature vector sequence, thus forming a health baseline feature dataset. The data distribution of the health baseline feature dataset constitutes the statistical regularity and boundary range of the equipment's "electrical fingerprint" features in a healthy state. Unsupervised training is performed on the health feature dataset using the Support Vector Data Description (SVM) algorithm to find the minimum radius hypersphere that can cover most healthy samples in the dataset. In the specific implementation, the SVM algorithm determines a minimum sphere by solving a constrained minimization optimization problem. Its core objective is to contain all sample points within the sphere's boundary as much as possible, while maintaining maximum compactness in the distances between points on the boundary. The optimization problem is solved using the Lagrange multiplier method, and combined with kernel methods such as the radial basis function (RBF) kernel, the input feature space is nonlinearly mapped to a high-dimensional space. In this high-dimensional space, the coordinates of the sphere center and the initial radius are determined, ensuring that the feature distribution of healthy samples is separable and has controllable boundaries in the kernel space. The sphere center position and the corresponding minimum envelope radius are calculated. To enhance the health baseline model's capacity to accommodate future edge state samples and prevent oversensitivity to natural fluctuations under normal operating conditions, a boundary expansion coefficient, slightly greater than 1, is introduced based on the minimum envelope radius. This coefficient is multiplied by the original radius to obtain the healthy boundary radius, which defines the model's healthy state tolerance limit in the high-dimensional kernel space. The solved sphere center vector, healthy boundary radius, Lagrange multiplier values, the set of support vectors used to define the healthy sample boundaries, and kernel function parameters (such as the γ value in the RBF kernel) are stored together to form the health baseline model.

[0031] The construction of the health benchmark model also includes: collecting electrical signals and extracting comprehensive electrical fingerprint feature vectors when industrial equipment is under different operating conditions; classifying the comprehensive electrical fingerprint feature vectors into light load condition feature datasets, medium load condition feature datasets, and heavy load condition feature datasets according to the load rate range; training the light load condition feature datasets, medium load condition feature datasets, and heavy load condition feature datasets respectively using the support vector data description algorithm to construct light load health benchmark models, medium load health benchmark models, and heavy load health benchmark models; during real-time health assessment, calculating the real-time load rate based on the active power and rated power of the current electrical signals, and automatically selecting the corresponding health benchmark model based on the real-time load rate for distance value calculation and health assessment.

[0032] In one example, the distance from the composite electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health baseline model, including: Extract the sphere center, support vector set, Lagrange multipliers, and kernel parameters from a pre-established health benchmark model; The first kernel function value of the integrated electronic fingerprint feature vector itself, the second kernel function value between the integrated electronic fingerprint feature vector and each support vector, and the third kernel function value between each support vector are calculated using the radial basis kernel function. The distance from the center of the sphere to the integrated electronic fingerprint feature vector is calculated based on the first kernel function value, the second kernel function value, the third kernel function value, and the Lagrange multiplier.

[0033] In this example, structural parameters for constructing a high-dimensional kernel-space hypersphere are extracted from a pre-established health baseline model. These parameters include the description of the sphere's center in kernel space, the set of support vectors, the Lagrange multipliers corresponding to each support vector, and the kernel function type and parameter values ​​used in model construction. The sphere's center is a functional representation formed by combining multiple support vectors and their corresponding Lagrange multipliers through kernel mapping. To calculate the distance between the current device's comprehensive electronic fingerprint feature vector and the sphere's center, it is indirectly measured using a kernel function. The radial basis function kernel is selected as the mapping tool, defined as K(x, x′) = exp( γ‖x x′‖ 2 ), where γ is the kernel width parameter, determined during model training. Three kernel function values ​​are calculated using the radial basis function: the first kernel function value between the current integrated electronic fingerprint feature vector and itself, i.e., K(x, x), which is always 1; the second kernel function value set K(x, s) between the feature vector and all support vectors in the healthy model; and the third kernel function value set K(x, s). i ), used to characterize the similarity between the current state and the healthy boundary samples; the pairwise third kernel function value K(s) between each support vector. i , s jThis is used to maintain the consistency of the model structure when constructing the sphere center description. The distance formula D in the support vector data description algorithm is utilized. 2 (x) = K(x, x) 2∑α i K(x, s i ) + ∑α i α j K(s i , s j Substitute the values ​​of the three types of kernel functions mentioned above into the model, and combine them with the Lagrange multiplier weights α stored in the model. i We combine the support vector index with the support vector index to obtain the squared distance D between the current comprehensive electronic fingerprint feature vector and the center of the kernel space sphere. 2 Then, the distance value D is obtained through the square root operation.

[0034] In one example, the radial basis function kernel is used to calculate the first kernel function value of the integrated electronic fingerprint feature vector itself, the second kernel function value between the integrated electronic fingerprint feature vector and each support vector, and the third kernel function value between each support vector, including: Calculate the sum of squares of the first difference between each standard feature component in the integrated electronic fingerprint feature vector and its own standard feature components, and substitute the sum of squares of the first difference into the radial basis kernel function to calculate the value of the first kernel function; Calculate the second sum of squared differences between each standard feature component and each support vector in the integrated electronic fingerprint feature vector, and substitute the second sum of squared differences into the radial basis kernel function to calculate the value of the second kernel function; Calculate the sum of squared third differences between each support vector, and substitute the sum of squared third differences into the radial basis function to obtain the value of the third kernel function.

[0035] In this example, the first difference between each standard feature component in the comprehensive electronic fingerprint feature vector and its own standard feature components is calculated. This is equivalent to subtracting the value of each feature component in the same dimension from its own value and then squared. Mathematically, this is equivalent to a zero vector. Therefore, the sum of the squares of the first difference is zero. Substituting this result into the radial basis function K(x, x) = exp( γ‖x x‖ 2 The first kernel function value obtained in the process is always 1, serving as the first constant term in the distance calculation of the support vector data description model. The difference between the comprehensive electronic fingerprint feature vector and each support vector is calculated dimension by dimension. This involves taking the difference between each standard feature component and the standard feature value of the support vector in the corresponding dimension, squaring the difference, and summing the squared differences across all dimensions to obtain the second sum of squared differences. This sum reflects the overall distance between the support vector and the current state in the feature space. Each second sum of squared differences is then substituted into the radial basis function K(x, s).i ) = exp( γ‖x s i || 2 This yields a set of second kernel function values. Simultaneously, to construct the formal representation of the sphere center vector in the kernel space, the pairwise differences between all support vectors are calculated, i.e., for any two support vectors s... i With s j The standardized differences of these features are calculated dimension by dimension and squared. The sum of these squares over all dimensions is then used to obtain the third sum of squared differences. This result is then substituted into the kernel function K(s). i , s j ) = exp( γ‖s i s j || 2 This yields the set of third kernel function values ​​for each pair of support vectors.

[0036] In one example, distance values ​​are mapped to device health scores, and fault characteristic components causing a decline in health are located, generating a fault diagnosis suggestion report, including: Map the distance value to a device health score; The device health score is compared with the first threshold, the second threshold, and the third threshold, respectively. When the device health score is between the first threshold and the second threshold, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the sphere is calculated. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated as an early warning. When the device health score is between the second and third thresholds, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the sphere is calculated. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated as a serious alarm. When the device health score is less than the third threshold, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the sphere is calculated. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated, which suggests shutdown and maintenance.

[0037] In this example, the distance value is converted into a device health score between 0 and 100% based on a defined nonlinear mapping function, where an exponential decay function is used to map the distance value D to a health score HS = 100 × exp( k × D), or choose an S-shaped function with boundary control capability HS = 100 / (1 + exp(a × (D)). b) Both parameters are calibrated based on field experience and historical data to adapt to different equipment categories. After completing the health mapping, the current health score is compared with the preset first threshold of 80%, the second threshold of 60%, and the third threshold of 40% in turn, dividing the equipment status into normal fluctuation zone, early performance degradation zone, obvious abnormal zone, and severe failure zone. If the health score is between the first threshold of 80% and the second threshold of 60%, it means that the equipment is still within the working tolerance range but has shown early performance degradation signs. At this time, the key reasons for the decline in health are analyzed. The difference between each standard feature component in the current comprehensive electronic fingerprint feature vector and the corresponding component of the sphere center vector in the health benchmark model is calculated dimension by dimension, and the absolute value of the difference is taken for each dimension. By comparing the magnitude of the absolute value of the deviation of all feature dimensions, the fault feature component with the largest deviation is located. For example, if the frequency domain harmonic energy deviates significantly in the high-order frequency band, it indicates that there is a power waveform distortion or load imbalance problem. Based on this, a fault diagnosis suggestion report is generated and marked as "early warning" level to prompt maintenance personnel to pay attention to the changing trend of key indicators. If the health score drops to between the second threshold of 60% and the third threshold of 40%, it indicates that the equipment's current state has significantly deviated from the healthy state boundary and is in the moderate fault development stage. The source of abnormal features is determined by a dimension-by-dimensional comparison with the sphere's center vector. Similarly, the absolute value of the deviation is sorted to identify the most severely deviated feature indicators. A diagnostic recommendation report at the "critical alarm" level is generated, combining equipment type, operating parameters, and the fault knowledge base, suggesting that inspection and maintenance should be arranged as soon as possible. If the health score is below the third threshold of 40%, it indicates that the equipment's feature distribution has deviated far from the normal area and is on the verge of a severe fault or failure. In this case, the same method is used to compare the absolute value of the difference between the comprehensive electrical fingerprint vector and the sphere's center feature, locking in the highest deviation dimension. Combined with historical fault cases, the structural or electrical system failure mode is inferred, and a fault diagnosis recommendation report of "suggested shutdown for maintenance" is output to avoid uncontrollable secondary faults or safety accidents during equipment operation.

[0038] After generating the remaining service life prediction report, the process also includes: extracting the health score time series and corresponding fault feature component deviation series of the current equipment within the most recent preset period to construct the current health degradation trajectory features; extracting the health degradation trajectory features of confirmed faulty equipment from the historical fault case database, and using a dynamic time warping algorithm to calculate the similarity distance between the current health degradation trajectory features and each historical fault trajectory feature to obtain a similarity ranking list; selecting the top three historical fault cases with the highest similarity in the similarity ranking list, extracting the final fault type, fault location, and effective maintenance measures of the historical fault cases, generating a similar fault mode reference report, and pushing it to maintenance personnel.

[0039] After mapping the distance value to the device health score, the process also includes: continuously recording the device health score sequence within a preset time period; performing linear regression fitting on the health score sequence to obtain the health decline rate; determining whether the health decline rate exceeds the rapid degradation threshold; when the health decline rate exceeds the rapid degradation threshold, raising the current warning level by one level and sending an accelerated degradation alarm; and calculating the remaining time required for the device health score to drop to the third threshold based on the health decline rate, the current device health score, and the third threshold, and generating a remaining service life prediction report.

[0040] After generating the fault diagnosis suggestion report, the process also includes: mapping the located fault feature components to the corresponding fault types; when the fault feature component is a deviation of the bearing fault characteristic frequency sideband energy, it is determined to be a mechanical wear fault; when the fault feature component is a deviation of the frequency domain harmonic characteristics, it is determined to be an electrical anomaly fault; and when the fault feature component is a deviation of the time domain statistical characteristics, it is determined to be a load anomaly fault. Based on the fault type and equipment health score, the system generates equipment operating parameter control instructions; when the fault is determined to be a mechanical wear fault and the health score is below the second threshold, it generates instructions to reduce equipment speed and start the cooling system; when the fault is determined to be an electrical anomaly fault, it generates instructions to adjust the power supply voltage and power factor compensation; and when the fault is determined to be a load anomaly fault, it generates instructions for load balancing. The equipment operating parameter control instructions are then sent to the equipment control system via an industrial communication protocol, whereby the equipment control system executes speed adjustment, cooling start-up, or electrical parameter adjustment, achieving protective control of the equipment based on health assessment.

[0041] When the monitoring targets are multiple related devices in the production line, the process also includes: assessing the health of upstream devices, current devices, and downstream devices in the production line to obtain the health scores of upstream devices, current devices, and downstream devices; calculating the time delay correlation coefficient between the health scores of upstream devices and current devices; determining that there is a health status transmission correlation between upstream devices and current devices when the time delay correlation coefficient is greater than a preset correlation threshold, and recording the delay time corresponding to the time delay correlation coefficient; and predicting the time when the health score of upstream devices is lower than a second threshold based on the delay time, sending an early warning for related devices, and generating a collaborative diagnostic report that includes the fault type of upstream devices and preventive measures for current devices.

[0042] Reference Figure 2 This embodiment provides a health assessment system based on the electrical characteristics and electronic fingerprint recognition of industrial equipment, including: Feature extraction module 1 is used to acquire electrical signals from industrial equipment and extract time-domain statistical features, frequency-domain harmonic features, bearing fault feature frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signals to construct the original feature vector; Calculation module 2 is used to standardize the original feature vector to obtain the comprehensive electronic fingerprint feature vector, and calculate the distance value 3 from the comprehensive electronic fingerprint feature vector to the center of the ball through a pre-established health benchmark model; The fault diagnosis module 4 is used to map distance values ​​to equipment health scores and locate fault feature components that cause a decline in health, and generate a fault diagnosis suggestion report.

[0043] In this embodiment, the specific implementation of each unit in the above system embodiment is described in the above method embodiment, and will not be repeated here.

[0044] 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, system, article, or method 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, system, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, system, article, or method that includes that element.

[0045] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment, characterized in that, include: Acquire electrical signals from industrial equipment, and extract time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signals to construct the original feature vector; The original feature vector is standardized to obtain a comprehensive electronic fingerprint feature vector, and the distance from the comprehensive electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health benchmark model. The distance value is mapped to a device health score, and the fault characteristic components that cause the health decline are located, generating a fault diagnosis suggestion report.

2. The health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment according to claim 1, characterized in that, Acquire electrical signals from industrial equipment, and extract time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signals to construct an original feature vector, including: Anti-aliasing filtering is performed on the three-phase current signal and three-phase voltage signal of industrial equipment to obtain the anti-aliasing filtered signal; Pulse noise is removed from the anti-aliasing filtered signal to obtain a denoised signal, and DC components with frequencies below a preset frequency and low-frequency drift are removed from the denoised signal to obtain an electrical signal. Extract the time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signal to construct the original feature vector.

3. The health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment according to claim 2, characterized in that, Extract the time-domain statistical features, frequency-domain harmonic features, bearing fault characteristic frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signal to construct the original feature vector, including: Extract the time-domain statistical features of the electrical signal; Perform a fast Fourier transform on the electrical signal to obtain the frequency domain harmonic characteristics and bearing fault characteristic frequency sideband energy; The electrical signal is decomposed into wavelet packets to obtain the wavelet packet energy entropy. The electrical signal is subjected to bispectral calculation to obtain bispectral amplitude characteristics; The original feature vector is constructed based on the time-domain statistical features, the frequency-domain harmonic features, the bearing fault feature frequency sideband energy, the wavelet packet energy entropy, and the bispectral amplitude features.

4. The health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment according to claim 3, characterized in that, The electrical signal is subjected to bispectral calculation to obtain bispectral amplitude characteristics, including: Calculate the bispectral matrix of an electrical signal; The amplitude of the bispectral matrix is ​​extracted, and the peak value, mean, and variance of the amplitude are used as bispectral amplitude features.

5. The health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment according to claim 1, characterized in that, The original feature vector is standardized to obtain a comprehensive electronic fingerprint feature vector. The distance from the comprehensive electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health benchmark model, including: Calculate the standardized parameters for each feature dimension in the historical health dataset; Based on the standardization parameters of each feature dimension, each feature component in the original feature vector is standardized to obtain the comprehensive electronic fingerprint feature vector. The distance from the integrated electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health benchmark model.

6. The health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment according to claim 5, characterized in that, Before calculating the distance from the integrated electronic fingerprint feature vector to the center of the sphere, the method further includes: When industrial equipment is in a healthy state, a health benchmark feature dataset is constructed; The support vector data description algorithm is used to train the health benchmark feature dataset to solve for the center and radius of the minimum radius hypersphere; The healthy boundary radius is obtained by multiplying the radius by the boundary expansion coefficient, and a healthy baseline model is constructed based on the sphere center, the healthy boundary radius, the support vector set, the Lagrange multipliers, and the kernel parameters.

7. The health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment according to claim 6, characterized in that, The distance from the integrated electronic fingerprint feature vector to the center of the sphere is calculated using a pre-established health benchmark model, including: Extract the sphere center, support vector set, Lagrange multipliers, and kernel parameters from a pre-established health benchmark model; The first kernel function value of the integrated electronic fingerprint feature vector itself, the second kernel function value of the integrated electronic fingerprint feature vector and each support vector, and the third kernel function value between each support vector are calculated using the radial basis kernel function. The distance from the integrated electronic fingerprint feature vector to the center of the sphere is calculated based on the first kernel function value, the second kernel function value, the third kernel function value, and the Lagrange multiplier.

8. The health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment according to claim 7, characterized in that, The first kernel function value of the integrated electronic fingerprint feature vector itself, the second kernel function value of the integrated electronic fingerprint feature vector and each support vector, and the third kernel function value of each support vector are calculated using the radial basis function, including: Calculate the sum of squared differences between each standard feature component in the integrated electronic fingerprint feature vector and its own standard feature components, and substitute the sum of squared differences into the radial basis kernel function to calculate the first kernel function value; Calculate the sum of squared differences between each standard feature component and each support vector in the integrated electronic fingerprint feature vector, and substitute the sum of squared differences into the radial basis kernel function to calculate the value of the second kernel function; Calculate the sum of squared differences between each support vector and substitute it into the radial basis function to obtain the value of the third kernel function.

9. The health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment according to claim 1, characterized in that, The distance values ​​are mapped to device health scores, and fault characteristic components causing a decline in health are located. A fault diagnosis suggestion report is generated, including: Map the distance value to a device health score; The device health score is compared with the first threshold, the second threshold, and the third threshold, respectively. When the device health score is between the first threshold and the second threshold, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the sphere is calculated. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated as an early warning. When the device health score is between the second threshold and the third threshold, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the ball is calculated respectively. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated as a serious alarm. When the device health score is less than the third threshold, the absolute value of the difference between each standard feature component in the comprehensive electronic fingerprint feature vector and each component in the center of the sphere is calculated. Based on the absolute value of the difference, the fault feature component with the largest deviation is located and a fault diagnosis suggestion report is generated, which suggests shutdown and maintenance.

10. A health assessment system based on the electrical characteristics and electronic fingerprint recognition of industrial equipment, characterized in that, The steps for implementing the health assessment method based on the electrical characteristics and electronic fingerprint recognition of industrial equipment as described in any one of claims 1 to 9 include: The feature extraction module is used to acquire electrical signals from industrial equipment and extract time-domain statistical features, frequency-domain harmonic features, bearing fault feature frequency sideband energy, wavelet packet energy entropy, and bispectral amplitude features from the electrical signals to construct the original feature vector. The calculation module is used to standardize the original feature vector to obtain a comprehensive electronic fingerprint feature vector, and to calculate the distance from the comprehensive electronic fingerprint feature vector to the center of the sphere using a pre-established health benchmark model. The fault diagnosis module is used to map the distance value to a device health score, locate the fault characteristic components that cause the health decline, and generate a fault diagnosis suggestion report.

Images