Equipment health degree assessment method and system based on industrial equipment electrical characteristics

Abstract

The invention relates to an equipment health degree assessment method and system based on industrial equipment electrical characteristics, and the method comprises the steps: collecting a current signal and a voltage signal of target industrial equipment, and carrying out the anti-aliasing filtering and power frequency interference suppression, and obtaining regular waveform data; respectively extracting a time domain feature, a frequency domain feature and a time-frequency domain feature in the regular waveform data, and carrying out vectorization splicing to generate a multi-dimensional feature vector; a plurality of continuous multi-dimensional feature vectors arranged according to a time sequence are constructed into a feature sequence through a preset health degree evaluation model; and the equipment health index is compared with a preset early warning threshold value in real time, and when it is judged that the equipment health index is continuously lower than the early warning threshold value, an equipment health early warning signal is output. According to the method, the equipment state can be comprehensively represented through multi-dimensional feature fusion, precise health index calculation is achieved in combination with the health degree evaluation model, early warning is achieved through a dynamic threshold mechanism, and the accuracy and timeliness of equipment fault prediction are remarkably improved.

Description

Technical Field

[0001] This invention relates to the technical field of industrial equipment evaluation, and in particular to a method and system for evaluating the health of industrial equipment based on its electrical characteristics. Background Technology

[0002] As core assets of modern production systems, the operational status of industrial equipment directly impacts production safety, efficiency, and cost. Analyzing the electrical characteristics of equipment, such as current and voltage, during operation to achieve accurate health assessment and early warning has become a crucial research direction in the field of intelligent industrial operation and maintenance. However, existing assessment methods based on electrical characteristics largely rely on simple threshold judgments of single-dimensional or static features, failing to fully explore the dynamic evolution of multi-dimensional features over time. This simplified approach struggles to accurately capture the gradual degradation trend of equipment performance, potentially leading to delayed warnings or false alarms, thus failing to provide timely and reliable decision-making basis for predictive maintenance. Summary of the Invention

[0003] The main objective of this invention is to provide a method and system for assessing the health of industrial equipment based on its electrical characteristics, which can accurately capture the temporal gradual change pattern of equipment status and achieve early insight into performance degradation trends.

[0004] To achieve the above objectives, the present invention provides a method for assessing equipment health based on the electrical characteristics of industrial equipment, comprising: The current and voltage signals of the target industrial equipment are collected and subjected to anti-aliasing filtering and power frequency interference suppression to obtain regular waveform data. The time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the regularized waveform data, and the time-domain features, frequency-domain features, and time-frequency-domain features are vectorized and concatenated to generate a multi-dimensional feature vector. The pre-defined health assessment model constructs a feature sequence from multiple consecutive multi-dimensional feature vectors arranged in chronological order. The equipment operation feature trend of the feature sequence is analyzed, and the equipment operation feature trend is compared and similarity calculated with the predefined equipment health benchmark information item by item to obtain the equipment health index. The device health index is compared with a preset warning threshold in real time. When the device health index is determined to be continuously lower than the warning threshold, a device health warning signal is generated and output.

[0005] Preferably, the current and voltage signals of the target industrial equipment are subjected to anti-aliasing filtering and power frequency interference suppression to obtain regular waveform data, including: The current signal of the power supply circuit in the target industrial equipment is obtained by a current transformer, and the voltage signal of the equipment input terminal in the target industrial equipment is obtained by a voltage probe. The current signal and the voltage signal are respectively input into a preset anti-aliasing filter to filter out frequency components higher than 5kHz in the signal, thereby obtaining a first voltage signal and a first current signal; The first voltage signal and the first current signal are band-stop filtered at a first frequency and a second frequency to suppress power frequency interference, thereby obtaining the second voltage signal and the second current signal. The waveform data of the second voltage signal and the waveform data of the second current signal are sampled synchronously and integrated to output the regularized waveform data.

[0006] Preferably, the step of extracting time-domain features, frequency-domain features, and time-frequency-domain features from the regularized waveform data, and then vectorizing and concatenating the three types of features to generate a multi-dimensional feature vector includes: Identify and separate the steady-state operation data segment and the startup transient data segment in the regularized waveform data; A set of steady-state time-domain parameters are extracted from the steady-state operation data segment, and a set of transient time-domain parameters are extracted from the startup transient data segment. The steady-state time-domain parameters and the transient time-domain parameters are integrated into the time-domain feature. Perform a fast Fourier transform on the steady-state operating data segment to obtain the spectrum, extract the fundamental amplitude from the spectrum, and statistically analyze the frequency band energy ratio to obtain the frequency domain characteristics; Perform multi-level wavelet packet decomposition on the data segment of the startup transient process and calculate the node energy entropy to obtain the time-frequency domain features; The time-domain features, frequency-domain features, and time-frequency-domain features are sequentially concatenated and combined to generate the multidimensional feature vector.

[0007] Preferably, a set of steady-state time-domain parameters is extracted from the steady-state operation data segment, and a set of transient time-domain parameters is extracted from the startup transient data segment. The steady-state time-domain parameters and the transient time-domain parameters are integrated into the time-domain feature, including: The steady-state time-domain parameters are obtained by calculating and integrating the effective value, peak factor and waveform distortion rate of the current signal based on the steady-state operating data segment. Based on the aforementioned transient data segment, the current integral value and current rise time of the current signal are calculated and integrated to obtain the transient time-domain parameters. The steady-state time-domain parameters and the transient time-domain parameters are combined to form the time-domain features.

[0008] Preferably, the step of constructing a feature sequence from multiple consecutive multidimensional feature vectors arranged in chronological order using a preset health assessment model, analyzing the equipment operation feature trend of the feature sequence, and comparing and calculating the similarity between the equipment operation feature trend and predefined equipment health benchmark information item by item to obtain the equipment health index includes: The multidimensional feature vector is input into a preset health assessment model, and the multidimensional feature vector of the current time and multiple consecutive time steps in the preceding time step are constructed into a feature sequence through the sequence construction layer of the health assessment model. The feature sequence is input into the time-series feature extraction layer of the health assessment model, and multi-scale local feature extraction and trend recognition are performed on the feature sequence to obtain the equipment operation feature trend; The dynamic trend analysis of the device's operating characteristics is performed through the context encoding layer of the health assessment model to obtain an instantaneous deviation sequence, and the dynamic deviation between the instantaneous deviation sequence and the device's health baseline information is calculated. Through the health mapping layer of the health assessment model, the dynamic deviation is associated with the multidimensional feature vector in the feature sequence and weighted by time step to obtain the time decay weight. The time decay weight is then used to calculate the dynamic deviation and the multidimensional feature vector to form and output the device health index.

[0009] Preferably, the step of inputting the feature sequence into the time-series feature extraction layer of the health assessment model, and performing multi-scale local feature extraction and trend recognition on the feature sequence to obtain the equipment operation feature trend includes: The temporal feature extraction layer extracts feature subsequences of corresponding lengths sequentially from the feature sequence using multiple preset continuous time intervals of different lengths. The mean, variance, and peak features of each extracted feature subsequence are calculated separately. The mean, variance, and peak features at the same time point are combined to form an enhanced feature vector. The enhanced feature vector is compared with a predefined device health baseline pattern to calculate the pattern matching degree at each time point. Target feature points are selected based on the pattern matching degree, and the selected target feature points are connected in chronological order to form the device operation feature trend.

[0010] Preferably, the step of performing dynamic trend analysis on the device operating characteristic trends through the context encoding layer of the health assessment model to obtain an instantaneous deviation sequence, and calculating the dynamic deviation between the instantaneous deviation sequence and the device health baseline information, includes: The difference between feature values ​​at adjacent time steps in the device operation feature trend is calculated through the context coding layer, and the difference is used as the instantaneous rate of change to form a feature rate of change sequence. The instantaneous rate of change at each time step in the characteristic rate of change sequence is compared with the baseline rate of change in the device health baseline information to calculate the difference, thus obtaining the instantaneous deviation value at each time step. The instantaneous deviation values ​​of each time step are arranged in chronological order to form an instantaneous deviation sequence; The instantaneous deviation sequence is truncated by sliding a window according to a preset window length, and the window deviation index of all instantaneous deviation values ​​within each window is calculated; The window deviation indices are arranged in chronological order to form a deviation sequence, and the window deviation index at the end of the deviation sequence is taken as the dynamic deviation degree.

[0011] Preferably, the health mapping layer of the health assessment model associates the dynamic deviation with the multidimensional feature vector in the feature sequence and assigns time-step weights to obtain time decay weights. The time decay weights are then used to weight the dynamic deviation and the multidimensional feature vector to form and output the device health index, including: The health mapping layer of the health assessment model arranges the dynamic deviation and the feature vectors of each time step in the feature sequence in chronological order to form a triplet sequence. Based on the principle of prioritizing the closest time step, time decay weights are assigned to the time steps in the triplet sequence, wherein the time decay weight at the current time step is set to the maximum value and decreases sequentially in reverse chronological order. Traverse the triplet sequence, multiply the feature vector by the time decay weight to obtain the weighted feature value, and multiply the dynamic deviation by the time decay weight to obtain the weighted deviation. The weighted feature value and the weighted deviation are concatenated along the feature dimension to form and output the device health index.

[0012] Preferably, the step of comparing the device health index with a preset warning threshold in real time, and generating and outputting a device health warning signal when it is determined that the device health index is continuously lower than the warning threshold, includes: The device health indices at the current moment and multiple consecutive moments preceding it are integrated to form a health index sequence; Each device health index in the health index sequence is compared with the warning threshold one by one. If the device health index is lower than the warning threshold, the device health index is marked as an abnormal state point. The system counts whether the number of consecutive occurrences of the abnormal state points within a preset period reaches or exceeds a preset threshold for the number of abnormal occurrences. If the threshold for the number of abnormal occurrences is not reached, it is determined that the early warning triggering condition is not met, and the system maintains normal monitoring status. If the number of abnormal occurrences reaches or exceeds the threshold, the warning triggering condition is determined to be met, and the device health warning signal is generated and output.

[0013] The present invention also provides an equipment health assessment system based on the electrical characteristics of industrial equipment, applied to any of the above-described equipment health assessment methods based on the electrical characteristics of industrial equipment, comprising: A high-frequency acquisition module is used to acquire current and voltage signals of the target industrial equipment, perform anti-aliasing filtering and power frequency interference suppression, and obtain regular waveform data. The signal processing module, wherein the analysis module is used to extract the time-domain features, frequency-domain features and time-frequency-domain features from the regularized waveform data respectively, and to vectorize and concatenate the time-domain features, the frequency-domain features and the time-frequency-domain features to generate a multi-dimensional feature vector; The feature calculation module, the association module is used to construct a feature sequence from multiple consecutive multidimensional feature vectors arranged in chronological order through a preset health assessment model, analyze the equipment operation feature trend of the feature sequence, compare the equipment operation feature trend with predefined equipment health benchmark information item by item and calculate the similarity to obtain the equipment health index; The health assessment module, the processing module is used to compare the device health index with a preset warning threshold in real time, and when it is determined that the device health index is continuously lower than the warning threshold, generate and output a device health warning signal.

[0014] The present invention provides a method and system for assessing the health of industrial equipment based on its electrical characteristics, which has the following beneficial effects: By extracting time-domain, frequency-domain, and time-frequency-domain features in parallel from current and voltage signals and then vectorizing and concatenating them, a multi-dimensional feature vector comprehensively characterizing the equipment's operating status can be constructed. This overcomes the limitations of single-dimensional feature information and provides a richer and more reliable data foundation for health assessment. By constructing a feature sequence from multi-dimensional feature vectors across multiple consecutive time steps and calculating its dynamic deviation from the equipment's health baseline pattern, the temporal gradual change pattern of the equipment's status can be accurately captured, enabling early insight into performance degradation trends. Mapping the dynamic deviation to a health index based on the principle of prioritizing time proximity significantly improves the real-time performance and sensitivity of health status assessment, making the assessment results more reflective of the equipment's current condition. By setting early warning thresholds and continuously comparing the health index, online and automatic monitoring and early warning of the equipment's health status can be achieved, providing timely and accurate decision support for predictive maintenance and effectively avoiding production interruptions and economic losses caused by downtime due to faults. Attached Figure Description

[0015] Figure 1 This is a flowchart of a method for assessing the health of industrial equipment based on its electrical characteristics, provided by the present invention. Figure 2 This is a structural diagram of an equipment health assessment system based on the electrical characteristics of industrial equipment, provided by the present invention.

[0016] 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

[0017] 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.

[0018] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0019] Reference Figure 1 As shown, the present invention provides a method for assessing the health of industrial equipment based on its electrical characteristics, comprising: Step S1: Collect the current and voltage signals of the target industrial equipment, perform anti-aliasing filtering and power frequency interference suppression to obtain regular waveform data; Step S2: Extract the time-domain features, frequency-domain features, and time-frequency-domain features from the regularized waveform data respectively, and then vectorize and concatenate the time-domain features, frequency-domain features, and time-frequency-domain features to generate a multi-dimensional feature vector; Step S3: Construct a feature sequence from multiple consecutive multidimensional feature vectors arranged in chronological order using a preset health assessment model; analyze the equipment operation feature trend of the feature sequence; compare the equipment operation feature trend with the predefined equipment health benchmark information item by item and calculate the similarity to obtain the equipment health index. Step S4: Compare the equipment health index with the preset warning threshold in real time. When it is determined that the equipment health index is continuously lower than the warning threshold, generate and output the equipment health warning signal.

[0020] Based on the steps described above, the detailed process is as follows: Step S1: In the power supply circuit of the target industrial equipment, a current transformer with an accuracy of 0.5 is connected in series to obtain the current signal, while a voltage probe with a bandwidth of not less than 100kHz is connected in parallel to the equipment's input terminal to collect the voltage signal. The signal acquisition device synchronously samples the original current and voltage waveforms at a sampling rate of not less than 10kHz to ensure complete capture of transient characteristics during equipment start-up, shutdown, and load changes.

[0021] The acquired raw waveform data first enters the anti-aliasing filtering stage. A low-pass filter with a cutoff frequency of 5kHz is used to filter out high-frequency noise components in the signal that are higher than the Nyquist frequency, so as to prevent spectral aliasing from interfering with subsequent analysis.

[0022] The filtered signal then enters the power frequency interference suppression stage, where a notch filter with a center frequency of 50Hz is used to specifically attenuate the fundamental frequency of the power grid and its main harmonic components, effectively eliminating the masking effect of the power frequency environment on weak fault characteristics.

[0023] The well-formed waveform data output after the above two-stage preprocessing maintains a complete time-domain waveform and pure frequency-domain components, laying a high-quality data foundation for subsequent feature extraction. This step, through the synergy of hardware acquisition and software filtering, achieves the key transformation for extracting effective signals from the complex electromagnetic environment of industrial sites.

[0024] Step S2: Time-domain feature extraction focuses on the basic statistical properties and transient response of waveforms, calculating the effective value, peak factor and waveform distortion rate of the current signal within a complete power frequency cycle. At the same time, it calculates the time integral value and rise time of the current during the equipment startup transient process. These parameters together constitute a time-domain feature subset that reflects the equipment's operating energy level and dynamic response.

[0025] Frequency domain analysis employs Fast Fourier Transform (FFT) to convert the current signal in the steady-state operating range to the frequency domain, extracting the fundamental amplitude and the amplitude ratios of the third, fifth, and seventh harmonics relative to the fundamental. It also statistically analyzes the energy proportion in the 2kHz to 5kHz high-frequency band, forming a frequency domain feature subset characterizing electrical anomalies and high-frequency resonance phenomena. Time-frequency domain processing targets non-stationary start-up transient signals. It calculates the energy entropy of each node through four-layer wavelet packet decomposition and uses Hilbert-Huang transform to obtain the instantaneous frequency variance, thereby capturing transient time-frequency joint features that traditional frequency domain analysis cannot identify.

[0026] The three feature subsets mentioned above are concatenated in a predetermined order to form a 21-dimensional feature vector. This vector integrates comprehensive state information of the equipment in steady state and transient state, low frequency and high frequency, and time domain and frequency domain, providing a quantitative description in a high-dimensional feature space for health assessment. The feature vector construction process reflects the technical approach of characterizing the degradation law of equipment state from multiple physical dimensions.

[0027] Step S3: Multidimensional feature vectors from the current moment and ten consecutive time steps prior are selected and arranged chronologically to construct a feature sequence, forming a time-series data sample characterizing the evolution of the device's state. This feature sequence is then input into a pre-trained health assessment model, and the temporal dependencies between feature vectors at each time step within the sequence are analyzed using a long short-term memory network layer to extract dynamic pattern features that reflect the trend of state degradation.

[0028] The dynamic deviation between the feature sequence and a predefined equipment health baseline pattern is calculated. This baseline pattern is established and standardized by collecting operational data from new equipment or fully healthy equipment. The dynamic deviation is weighted according to the principle of prioritizing the most recent time point, assigning higher weight to deviations with more recent timestamps to highlight the dominant role of recent status changes in health assessment.

[0029] By using a fully connected layer and a sigmoid activation function, the weighted dynamic deviation is mapped to a health assessment value within the range of zero to one. Finally, this value is multiplied by 100% to convert it into a device health index within the range of zero to 100%. This index comprehensively considers the temporal evolution of device status and recent status change trends, achieving a continuous quantitative assessment of device health.

[0030] Step S4: A multi-level early warning threshold system is established, with 85% as the primary early warning threshold to identify significant degradation in equipment health. The system monitors the trend of equipment health index changes in real time, and initiates an early warning trigger judgment process when the health index consistently falls below the early warning threshold. This process integrates the health index values ​​from the current moment and several consecutive previous moments to form a health index sequence, and counts the number of consecutive occurrences of abnormal state points below the early warning threshold in this sequence.

[0031] When the number of consecutive occurrences of an abnormal state point reaches a preset threshold, the warning trigger condition is met, and a device health warning signal is generated. The warning signal includes device identifier information, a warning trigger timestamp, the current health index value, and the determined warning level, encapsulated according to a preset message format. The warning signal is distributed to the human-machine interface of the device monitoring system for visual alert display via an industrial communication protocol, and simultaneously stored persistently in a historical database. This warning mechanism effectively avoids false alarms through multi-condition joint judgment, ensuring that the warning signal is only triggered when the device's health status continues to deteriorate, providing a reliable decision-making basis for predictive maintenance.

[0032] This invention provides a method for assessing the health of industrial equipment based on its electrical characteristics. By extracting time-domain, frequency-domain, and time-frequency-domain features from current and voltage signals in parallel and then vectorizing and concatenating them, a multi-dimensional feature vector comprehensively characterizing the equipment's operating status can be constructed. This overcomes the limitations of single-dimensional feature information and provides a richer and more reliable data foundation for health assessment. By constructing a feature sequence from the multi-dimensional feature vectors of multiple consecutive time steps and calculating its dynamic deviation from the equipment's health baseline pattern, the method can accurately capture the temporal gradual change pattern of the equipment's status, enabling early insight into performance degradation trends. Mapping the dynamic deviation to a health index based on the principle of prioritizing time proximity significantly improves the real-time performance and sensitivity of health status assessment, making the assessment results more reflective of the equipment's current condition. By setting early warning thresholds and continuously comparing the health index, online and automatic monitoring and early warning of the equipment's health status can be achieved, providing timely and accurate decision support for predictive maintenance and effectively avoiding production interruptions and economic losses caused by downtime due to faults.

[0033] In one embodiment, the current and voltage signals of the target industrial equipment are acquired and subjected to anti-aliasing filtering and power frequency interference suppression to obtain well-formed waveform data, including: A toroidal current transformer with an accuracy class of 0.5 is selected and connected in series to the three-phase power supply circuit of the target industrial equipment to ensure accurate sensing of load current changes without affecting the normal operation of the equipment.

[0034] Voltage acquisition employs a high-impedance differential voltage probe, directly connected in parallel between the phase and neutral lines at the equipment's input terminal. The measurement point is selected downstream of the equipment's main switch to reflect the actual operating voltage. The secondary output range of the current transformer is designed to be a 0-5 amp standard signal, and the voltage probe's measurement bandwidth is no less than 100 kHz to ensure the acquisition capability of high-frequency components.

[0035] Signal transmission uses shielded twisted-pair cables connected to the data acquisition card, with the cable length controlled within 10 meters and grounded for protection, effectively suppressing electromagnetic interference. The acquisition system synchronously acquires current and voltage signals at a sampling rate of no less than 10 kHz, ensuring the integrity of the time-domain waveform correspondence. This hardware deployment scheme enables non-intrusive monitoring of the equipment's electrical parameters, providing raw data for subsequent signal processing.

[0036] A Butterworth low-pass filter is used as the anti-aliasing filter implementation scheme, with its cutoff frequency set to 5 kHz and the transition band attenuation slope set to 40 dB per decade. An impedance matching circuit is configured in the filter input stage to ensure that the current signal output from the current transformer and the voltage signal output from the voltage probe can be input without distortion.

[0037] For the current signal channel, the 0-5 ampere current signal is first converted into a 0-5 volt voltage signal by a current-to-voltage conversion circuit, and then sent to an anti-aliasing filter for processing; the voltage signal channel is directly filtered. During the filtering process, the signal waveform is monitored in real time to ensure that effective spectral information in the range from the fundamental frequency to 5 kHz is preserved while eliminating components above the Nyquist frequency.

[0038] The amplitude attenuation of the filtered signal is controlled within 3 dB, and phase distortion is corrected using group delay compensation technology. A signal conditioning circuit is configured at the output to adjust the filtered signal to the optimal input range of the data acquisition card, resulting in a complete waveform and a bandwidth-limited first voltage and current signal.

[0039] A multi-stage cascaded notch filter bank is used, with the center frequencies set as the first frequency (50 Hz fundamental frequency) and the second frequency (150 Hz, 250 Hz, and other major harmonic frequencies). Each notch filter is designed as a second-order infinite impulse response filter with a stopband width of ±2 Hz, achieving an attenuation depth of no less than 40 dB within this narrow band.

[0040] The filter employs a recursive digital filter structure, using a direct form (Type II) to reduce rounding errors. During processing, the spectral characteristics of the input signal are monitored in real time. When grid frequency fluctuations exceed the 49 Hz to 51 Hz range, the notch filter center frequency is automatically adjusted for tracking compensation. For the voltage signal channel, an additional phase compensation circuit is added after notch filtering to correct the phase delay caused by filtering; for the current signal channel, the focus is on maintaining the linearity of the amplitude-frequency response.

[0041] After processing, the fundamental frequency and third and fifth harmonic components are effectively suppressed, while high-frequency components related to equipment fault characteristics are preserved. A level detection circuit is set at the output terminal to ensure that the dynamic range of the second voltage signal and the second current signal meets the requirements for subsequent sampling.

[0042] A 24-bit analog-to-digital converter with synchronous sampling function is used to synchronously quantize the second voltage signal and the second current signal, with the sampling rate strictly maintained at 10 kHz. The synchronization mechanism is implemented through hardware triggering, with sampling pulses generated from the same clock source to ensure that the timing deviation of voltage and current sampling is less than 100 nanoseconds.

[0043] Within each sampling period, 2048 voltage and current data points are simultaneously acquired to form a complete waveform record. During the data integration phase, a timestamp is added to each sampling point, and the voltage and current data are combined into data frames according to the principle of aligning data at the same time. The data frame header contains metadata such as sampling rate, timestamp, and channel identifier, while the payload area stores the instantaneous values ​​of voltage and current in an alternating manner.

[0044] Before output, data integrity is verified using a cyclic redundancy check (CRC) code to ensure error-free data transmission. The final output regularized waveform data adopts a standardized binary format, containing complete voltage and current waveform information and their corresponding relationships, providing a structured data foundation for subsequent feature extraction. This synchronous sampling scheme effectively maintains the phase consistency of voltage and current signals, providing accurate data support for power calculation and harmonic analysis.

[0045] This embodiment achieves non-invasive and accurate acquisition of electrical parameters of industrial equipment through the coordinated arrangement of current transformers and voltage probes, providing a high-quality raw data foundation for subsequent analysis. A multi-stage processing scheme employing anti-aliasing filtering and power frequency interference suppression effectively eliminates high-frequency noise and power grid interference, significantly improving the signal-to-noise ratio and ensuring the accuracy of feature extraction. Synchronous sampling technology ensures the phase consistency of voltage and current signals, providing reliable data support for power calculation and harmonic analysis. The standardized output format of regularized waveform data enables effective integration of multi-source data, creating a structured data processing foundation for subsequent feature extraction and health assessment. The entire signal preprocessing flow, through the organic combination of hardware filtering and digital processing, minimizes environmental interference while maintaining signal integrity, laying a solid technical foundation for accurate assessment of equipment health status.

[0046] In one embodiment, time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the regularized waveform data, and the three types of features are vectorized and concatenated to generate a multi-dimensional feature vector, including: A sliding window variance analysis was performed on the current signal in the regular waveform data. The window length was set to ten power frequency cycles. When the fluctuation range of the effective value of the current within five consecutive windows is less than five percent, the interval is determined to be a steady-state operating data segment.

[0047] The identification of the transient data segment at startup is achieved by detecting the rate of change of current amplitude. When the current value rises from the no-load state to more than 80% of the rated value and the duration is within the range of 0.5 seconds to 3 seconds, it is determined to be a valid startup process.

[0048] A dynamic threshold adjustment algorithm is employed to adaptively set the judgment threshold based on historical equipment operating data, avoiding misjudgments caused by load fluctuations. Timestamps are added to identified data segments; steady-state data segments are marked as operating baseline modes, while transient data segments are stored numbered according to their startup sequence. Windowing smoothing is applied at data segment boundaries to prevent spectral leakage from affecting the accuracy of subsequent feature extraction. Through state recognition and data separation operations, a classification processing foundation is provided for feature extraction under different operating conditions.

[0049] Steady-state time-domain parameter extraction targets the continuous stable operating range. It calculates the root mean square value of the current signal within a complete power frequency cycle to reflect the load level, the peak factor to characterize the waveform impact characteristics, and the waveform distortion rate to quantify the degree of harmonic distortion.

[0050] Transient time-domain parameters are calculated for the startup process. The time integral of the current characterizes the startup energy accumulation, and the rise time reflects the response speed of the electromechanical system. Parameter calculation employs a sliding window iterative algorithm, with the window width dynamically adjusted according to signal characteristics to ensure the statistical significance of eigenvalues. Steady-state parameters are updated every ten power frequency cycles, while transient parameters are fully recorded during each startup process.

[0051] In the feature integration stage, the two types of parameters are normalized to eliminate the influence of dimensions, and then combined in a predetermined order to form a time-domain feature vector. The positions of each parameter in the feature vector are fixed, which facilitates subsequent model recognition and pattern matching. This time-domain feature set comprehensively covers the key parameters of the equipment's steady-state operation and dynamic processes, providing multi-angle time-domain observation basis for health status assessment.

[0052] A windowed Fast Fourier Transform (FFT) was applied to the labeled steady-state data segment. A Hanning window was chosen to reduce spectral leakage, and the transform length was set to 2048 points to ensure sufficient frequency resolution. After spectral calculation, normalization was performed, converting the amplitude of each frequency component into a percentage representation relative to the fundamental amplitude. In harmonic component analysis, the third, fifth, and seventh harmonic components were extracted as a key focus.

[0053] To accurately locate the fundamental frequency in the spectrum, the maximum amplitude point is searched within ±0.5 Hz, centered at 50 Hz, to determine the precise location of the fundamental frequency. Based on the determined fundamental frequency, the amplitudes of the third harmonic (150 Hz), fifth harmonic (250 Hz), and seventh harmonic (350 Hz) are calculated. Considering that the actual power grid frequency may have slight fluctuations, the determination of harmonic frequency points is dynamically adjusted based on multiples of the fundamental frequency to ensure the accuracy of harmonic component extraction.

[0054] The harmonic amplitude ratio is calculated using precise mathematical methods. The third harmonic component amplitude is divided by the fundamental frequency amplitude to obtain the third harmonic amplitude ratio; the fifth and seventh harmonic amplitude ratios are calculated using the same method. This calculation process effectively eliminates the influence of fundamental frequency amplitude fluctuations on the harmonic analysis results, allowing the harmonic amplitude ratio to stably reflect the degree of electrical characteristic distortion of the equipment. The third harmonic amplitude ratio mainly reflects the symmetrical distortion of the power supply waveform, the fifth harmonic amplitude ratio characterizes the magnetic circuit saturation characteristics, and the seventh harmonic amplitude ratio is closely related to the winding distribution.

[0055] High-frequency feature extraction targets the 2kHz to 5kHz frequency band, which is evenly divided into ten sub-bands. The energy value of each sub-band is calculated, and then the percentage of total energy to the full spectrum energy is statistically analyzed. All frequency domain parameters are processed by moving average filtering to eliminate interference from instantaneous fluctuations. The final generated frequency domain feature vector includes multiple dimensions of parameters such as fundamental amplitude, third harmonic amplitude ratio, fifth harmonic amplitude ratio, seventh harmonic amplitude ratio, and high-frequency energy percentage, comprehensively characterizing the operating status of the equipment within the frequency domain.

[0056] Wavelet basis functions are used to perform four-level wavelet packet decomposition on the initial transient data segment, generating decomposition coefficients for sixteen terminal nodes. Each node coefficient is reconstructed using an algorithm to obtain the time-domain signal of the corresponding frequency band. The energy value of each node signal is calculated and normalized to form a node energy distribution spectrum. Based on the energy distribution spectrum, the wavelet packet energy entropy is calculated; the entropy value reflects the concentration of energy in the frequency band distribution during the transient process. A Hilbert-Huang transform is applied to the original transient signal, and intrinsic mode function components are obtained through empirical mode decomposition. The instantaneous frequency sequence is solved, and its variance is calculated to characterize the stability of the electromechanical system during the transient process.

[0057] The time-frequency domain feature extraction uses an overlapping sliding window, with the window length adaptively adjusted based on the startup duration to ensure the completeness of feature extraction. The final time-frequency domain feature vector contains six feature parameters, including wavelet packet energy entropy and instantaneous frequency variance, effectively capturing transient characteristics that are difficult to identify using traditional time-domain or frequency-domain analysis.

[0058] Outlier detection and data smoothing are performed on the eight parameters in the time-domain feature subset, the seven parameters in the frequency-domain feature subset, and the six parameters in the time-frequency-domain feature subset, respectively. A min-max normalization method is used to map each feature parameter to a numerical range of zero to one, eliminating the influence of different physical dimensions on feature weights. Feature concatenation is performed according to a predetermined order of time-domain features, frequency-domain features, and time-frequency-domain features to form a 21-dimensional feature vector.

[0059] The positions of each parameter in the feature vector are fixed, and an index mapping table is established to ensure that the physical meaning of each feature parameter can be accurately identified during subsequent model processing. After concatenation, correlation analysis is performed on the feature vectors. Highly correlated feature dimensions are marked but not deleted, preserving the integrity of the original feature set. The final generated multidimensional feature vector serves as a holographic digital portrait of the equipment's health status, providing standardized input data for subsequent health assessment models. This feature fusion scheme achieves the technical goal of comprehensively depicting the equipment's operating status from multiple physical dimensions.

[0060] This embodiment achieves targeted feature extraction for different operating conditions through precise separation of steady-state and transient data segments, effectively improving the comprehensiveness and accuracy of condition monitoring. Employing a multi-dimensional feature fusion scheme in the time domain, frequency domain, and time-frequency domain, it can comprehensively characterize the health status of equipment from different physical dimensions, overcoming the limitations of single-parameter evaluation. Fine-grained analysis of harmonic components, especially the extraction of the amplitude ratios of the third, fifth, and seventh harmonics, provides sensitive detection indicators for early fault diagnosis. The introduction of wavelet packet energy entropy and instantaneous frequency variance enables the effective capture of subtle feature changes during transient processes, greatly improving the timeliness of fault warnings. The standardized concatenation of multi-dimensional feature vectors constructs a unified health status assessment framework, providing high-quality feature input for subsequent intelligent assessment models.

[0061] In one embodiment, a set of steady-state time-domain parameters is extracted from the steady-state operation data segment, and a set of transient time-domain parameters is extracted from the startup transient data segment. The steady-state time-domain parameters and transient time-domain parameters are integrated into a time-domain feature, including: The RMS value is calculated using the true RMS algorithm. Current sampling data from ten consecutive power frequency cycles is squared, the average value is calculated, and the square root is taken to obtain the root mean square value reflecting the equipment load level. The peak factor is calculated by locating the maximum value of the current waveform within a complete cycle and using its ratio to the RMS value as a quantitative indicator. This parameter has high sensitivity to impact load changes caused by bearing damage, etc. The waveform distortion rate is calculated using the full harmonic distortion algorithm. The fundamental component is extracted through digital filtering, and then the sum of the squares of the ratios of each harmonic component to the fundamental component is calculated. Finally, the square root is taken to obtain a quantitative indicator characterizing the degree to which the current waveform deviates from a sine wave.

[0062] All three parameters were calculated using a sliding window technique, with the window width dynamically adjusted based on device characteristics to ensure the statistical stability of the feature values. A real-time data verification mechanism was implemented during the calculation process; when data anomalies were detected, the recalculation process was automatically triggered to guarantee the reliability of parameter extraction. The resulting steady-state time-domain parameter set characterizes the electrical characteristics of the device under stable operating conditions from different dimensions, providing a fundamental criterion for health status assessment.

[0063] The current integral value is calculated using a numerical integration method. Starting from the moment the equipment is energized, and ending at 95% of the current's steady-state value, a definite integral is performed on the current-time curve. This integral value reflects the total electrical energy consumed during startup and is closely related to the equipment's rotor inertia and mechanical friction state.

[0064] The current rise time is calculated by identifying the time interval required for the current to rise from 10% to 90% of its rated value, and a linear interpolation algorithm is used to improve the time resolution. Adaptive baseline correction technology is employed during feature extraction to eliminate the influence of zero-point drift in the measurement system on the calculation results.

[0065] For the startup process under different load conditions, a correlation model between startup parameters and load rate is established to achieve normalization of characteristic values. The transient time-domain parameter set provides important indicators of the equipment's dynamic performance and can effectively identify early fault symptoms such as changes in mechanical resistance and rotor imbalance. By comparing and analyzing with steady-state parameters, the operating status of the equipment under different working conditions can be comprehensively evaluated.

[0066] The effective values, peak factor, and waveform distortion rate in the steady-state time-domain parameter set are normalized, and the maximum-minimum scaling method is used to map each parameter to a numerical range of zero to one. The current integral value and rise time in the transient time-domain parameter set are also normalized. The current integral value is standardized based on the equipment's rated power, and the rise time is scaled proportionally to the normal range of historical startup data. Parameter merging employs a feature-level fusion strategy, arranging parameters in a predetermined order: steady-state parameters first, followed by transient parameters. The positions of each parameter in the feature vector are fixed, and an index mapping relationship is established to ensure accurate identification of the physical meaning of each parameter during subsequent processing. After merging, correlation analysis and redundancy checks are performed on the feature vectors, preserving the integrity of the original parameter set while labeling highly correlated feature dimensions.

[0067] The constructed time-domain feature vector serves as a complete digital representation of the device's operating state within the time domain, providing standardized data input for subsequent multi-domain feature fusion. This feature integration scheme achieves the technical goal of comprehensively characterizing the device's time-domain characteristics from both steady-state and transient dimensions.

[0068] This embodiment, through the synergistic analysis of current RMS value, peak factor, and waveform distortion rate, accurately characterizes the steady-state operating characteristics of equipment from three dimensions: load level, impact characteristics, and waveform quality. Introducing the current integral value and rise time parameter during startup makes quantitative evaluation of equipment dynamic performance possible, providing an effective means for early mechanical fault diagnosis. The organic integration of steady-state and transient parameters constructs a complete feature profile in the time domain, overcoming the limitations of single-condition analysis. Parameter normalization ensures the comparability of feature values ​​with different physical dimensions, laying the foundation for multi-dimensional feature fusion. This technical solution significantly improves the accuracy and reliability of equipment health status monitoring through refined extraction and intelligent integration of time-domain features.

[0069] In one embodiment, a feature sequence is constructed from multiple consecutive multidimensional feature vectors arranged in chronological order using a preset health assessment model. The device operation feature trend of the feature sequence is analyzed, and the device operation feature trend is compared and similarity calculated item by item with predefined device health benchmark information to obtain a device health index, including: The sequence construction layer of the health assessment model receives a multidimensional feature vector stream from the feature extraction module, with these vectors input sequentially in chronological order. The sequence construction layer determines the size of the time window, i.e., the number of consecutive time steps to be included, which is preset based on the device's operating cycle and state change characteristics. The selection of time steps traces backward from the current moment, ensuring coverage of a sufficiently long observation period to capture patterns in state evolution.

[0070] Each time step corresponds to a multidimensional feature vector containing 21 dimensions of feature data. This data has been standardized to eliminate the influence of differences in dimensions. During sequence construction, data integrity checks are performed to verify the dimensional consistency and numerical validity of each feature vector. Missing or outlier data are repaired using interpolation between adjacent time steps or based on historical patterns.

[0071] The sequence construction layer arranges the validated feature vectors in chronological order, forming a feature sequence with a clear time orientation. This sequence serves as the foundational data structure for subsequent time series analysis, and its construction quality directly impacts the accuracy of health assessment. After sequence construction is complete, timestamp metadata and sequence identifiers are added to facilitate tracking and management in subsequent processing stages.

[0072] The health assessment model's context encoding layer performs dynamic trend analysis on the equipment's operational characteristics to obtain an instantaneous deviation sequence. The dynamic deviation between this instantaneous deviation sequence and the equipment's health baseline information is then calculated. This step implements context-aware dynamic trend parsing and deviation quantification. The context encoding layer receives equipment operational characteristic trend data from the temporal feature extraction layer, which includes local features and global trend information obtained from multi-scale analysis.

[0073] Dynamic trend analysis employs a sliding window technique, with the window width adaptively adjusted based on equipment operating characteristics. Within each window, differential features such as the gradient and curvature of the characteristic trend are calculated. The instantaneous deviation sequence is generated by comparing the current characteristic trend with the historical baseline pattern, which is derived from the statistical learning results of long-term monitoring data under normal equipment operating conditions.

[0074] Deviation calculation is performed for each time step, measuring the difference between the current feature vector and the baseline feature vector at the corresponding time step dimension by dimension. The difference measurement comprehensively considers both the absolute value difference and the relative rate of change. The calculation of dynamic deviation not only considers the instantaneous deviation value, but also analyzes the time evolution characteristics of the deviation sequence, including indicators such as the continuity, directionality, and acceleration of deviation changes.

[0075] An outlier suppression mechanism is implemented during the calculation process to smooth out sudden instantaneous deviations and avoid the impact of interference terms on the overall deviation assessment. The final generated dynamic deviation is a comprehensive quantitative indicator that reflects both the absolute degree of deviation at the current moment and the dynamic characteristics of deviation development, providing an accurate deviation measurement basis for health status assessment.

[0076] The health mapping layer first establishes a correlation mapping between dynamic deviation and feature sequences, and uses a time alignment algorithm to ensure that the deviation value at each time step is precisely matched with the corresponding multidimensional feature vector. The time step weight allocation is based on the principle of prioritizing the closest time step, and a monotonically decreasing function is used to generate the weight sequence. The weight at the current time step is set to the maximum value, and the weight gradually decays as time goes back.

[0077] The specific form of the weighting function is selected according to the equipment condition assessment requirements, with linear decay and exponential decay being the most common modes. The weighting calculation process is performed separately for the dynamic deviation sequence and the feature sequence: dynamic deviation weighting uses scalar multiplication, where the deviation value at each time step is multiplied by its corresponding weight; feature vector weighting uses element-wise multiplication, where each dimension value of each feature vector is multiplied by its respective weight value.

[0078] The weighted deviation sequence and feature sequence are fused at the feature level, and a fully connected neural network is used to achieve a non-linear mapping from high-dimensional features to the health index. During the mapping process, a sigmoid activation function is used to compress the output value to the range of zero to one, and then a percentage conversion is performed to obtain the device health index between zero and one hundred.

[0079] This index comprehensively reflects the overall health status of the equipment in the current and historical periods. It includes quantitative information on status deviations and integrates multi-dimensional feature information to achieve a comprehensive and accurate health assessment. Results are validated before final output to ensure the health index is within a reasonable range and consistent with the actual condition of the equipment.

[0080] This embodiment constructs a feature sequence from multi-dimensional feature vectors in chronological order through the sequence construction layer of the health assessment model, ensuring the continuity and integrity of time-series data and providing a structured data foundation for equipment status analysis. The time-series feature extraction layer performs multi-scale local feature extraction and trend identification on the feature sequence, simultaneously capturing both short-term fluctuations and long-term evolution patterns of equipment operation, thus comprehensively improving the accuracy and timeliness of status assessment. The context encoding layer performs dynamic trend analysis on equipment operation characteristics and calculates instantaneous deviation sequences, achieving a refined measurement of the deviation between equipment status and health benchmarks, significantly enhancing the sensitivity of early fault identification. The health mapping layer weights and fuses dynamic deviation with feature vectors, mapping them to a health index that comprehensively reflects the overall health status of the equipment, providing a reliable basis for predictive maintenance decisions. The entire assessment process, through the collaborative work of multiple processing mechanisms, achieves intelligent and quantitative assessment of equipment health status, effectively improving equipment management efficiency.

[0081] 6. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 5, characterized in that, the step of inputting the feature sequence into the time-series feature extraction layer of the health assessment model, and performing multi-scale local feature extraction and trend recognition on the feature sequence to obtain the equipment operation feature trend, includes: The time-series feature extraction layer presets multiple time intervals of different lengths. These interval lengths are set according to the equipment's operating cycle and state change characteristics, typically including short-term, medium-term, and long-term scales. The short-term time interval is set to several sampling periods to capture rapid fluctuation characteristics; the medium-term time interval covers dozens of sampling periods to analyze trend changes; and the long-term time interval contains hundreds of sampling periods to identify macroscopic operating patterns.

[0082] During the extraction process, each preset time interval serves as a sliding window, sliding sequentially along the time axis of the feature sequence. The window moves by one sampling interval at a time, ensuring complete coverage of the sequence. At each window position, a feature subsequence of the corresponding time length is extracted, preserving the dimensional structure of the original feature vector. Boundary processing is implemented during the extraction operation; when the window is close to the ends of the sequence, mirror expansion or constant padding methods are used to maintain consistent subsequence lengths.

[0083] Each extracted feature subsequence is marked with a timestamp and a window scale identifier to establish a correspondence with the original sequence. The extraction of multi-scale feature subsequences provides rich temporal context information for subsequent analysis, enabling feature extraction to take into account both short-term dynamics and long-term trends.

[0084] For each extracted feature subsequence, its statistical characteristics are calculated: the mean characteristic reflects the average level of the feature values ​​within the time period, and the average value of each feature dimension in the subsequence is calculated by the arithmetic mean method; the variance characteristic characterizes the degree of fluctuation of the feature values, and the dispersion of each dimension value relative to the mean is calculated; the peak characteristic identifies the extreme value characteristics in the subsequence, including the maximum value, the minimum value and their occurrence position.

[0085] The calculation process employs a sliding window technique, with the window size matched to the subsequence length to ensure the accuracy of statistical features. For multi-dimensional feature vectors, statistical values ​​are calculated independently for each dimension, maintaining the integrity of the feature space. Statistical features of subsequences at different scales corresponding to the same time point are aligned to ensure temporal consistency.

[0086] In the feature combination stage, the mean, variance, and peak features at the same time point are concatenated in a predetermined order to form an enhanced feature vector. This vector not only contains the original feature information but also incorporates multi-scale statistical properties, enhancing the feature's representational ability. The enhanced feature vector undergoes normalization processing to eliminate the dimensional differences between different statistical quantities, providing standardized input for subsequent pattern matching.

[0087] The predefined equipment health baseline model is derived from the statistical learning results of long-term monitoring data under normal equipment operation, and includes the typical value range and distribution pattern of multi-scale statistical features. Similarity calculation adopts the distance metric method in multi-dimensional feature space to calculate the degree of difference between the enhanced feature vector and the corresponding feature vector of the baseline model.

[0088] The calculation process is performed independently for the enhanced feature vector at each time point, evaluating the degree of matching between the device status and the health baseline at that time point. The similarity metric comprehensively considers multiple indicators such as the Euclidean distance and cosine similarity of the feature vectors, and obtains a comprehensive matching score through weighted fusion. The matching score is normalized to the range of zero to one; the closer the score is to one, the more similar it is to the health baseline, and the closer it is to zero, the greater the deviation.

[0089] A dynamic benchmark adjustment mechanism is implemented during the calculation process, adaptively correcting the health benchmark model based on equipment runtime and maintenance history to ensure the timeliness and accuracy of the benchmark. The model matching degree results at each time point form a time series, which reflects the dynamic degree of conformity between the equipment status and the health benchmark, providing a quantitative basis for subsequent feature point selection. The matching degree calculation also considers the influence of operating conditions such as equipment load, introducing a compensation factor to eliminate matching degree fluctuations caused by non-fault factors.

[0090] The selection of target feature points is based on a pattern matching degree threshold. Upper and lower thresholds are set to select feature points that represent the typical state of the device. The selection strategy employs an adaptive threshold mechanism, dynamically adjusting the threshold based on the device's operating status: during stable operation, the threshold selection standard is increased to obtain more representative feature points; during periods of state change, the threshold is appropriately relaxed to ensure the integrity of trend changes. The selection process also considers the uniformity of feature point distribution to avoid excessive concentration or sparseness of feature points on the timeline.

[0091] The selected target feature points are arranged in chronological order, and a curve fitting algorithm is used to connect adjacent feature points to construct a smooth trend line for equipment operation. During the trend line generation process, filtering techniques are applied to eliminate noise interference and retain the true trend change characteristics.

[0092] The resulting equipment operation characteristic trends are represented in time series form, containing both precise information on key state points and reflecting the continuous pattern of state changes. This trend line serves as an important basis for equipment health status assessment, providing refined and optimized feature inputs for subsequent health index calculations. The trend line data includes a confidence index, characterizing the reliability of the trend analysis, for reference by subsequent processing modules.

[0093] This embodiment extracts feature subsequences across multiple time windows, enabling simultaneous capture of both short-term fluctuations and long-term evolution patterns in equipment operation, thus achieving comprehensive monitoring of equipment health status. A combined analysis of mean, variance, and peak features delineates equipment operating characteristics from multiple statistical dimensions, significantly improving the richness and accuracy of feature representation. By enhancing the similarity calculation between feature vectors and equipment health benchmark patterns, a refined comparison between the current state and the health benchmark is achieved, providing a reliable basis for anomaly detection. A target feature point selection mechanism based on pattern matching degree effectively extracts state feature points, resulting in equipment operation feature trends that retain important state information while eliminating redundant interference.

[0094] 7. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 5, characterized in that, the step of performing dynamic trend analysis on the equipment operating characteristic trend through the context coding layer of the health assessment model to obtain an instantaneous deviation sequence, and calculating the dynamic deviation between the instantaneous deviation sequence and the equipment health benchmark information, includes: The context coding layer receives device operation feature trend data from the temporal feature extraction layer. This data contains a sequence of feature vectors arranged in chronological order. During the calculation, starting from the second time step of the sequence, the difference between the feature vector at each time step and the corresponding feature vector at the previous time step is calculated sequentially. The difference calculation is performed independently for each dimension of the feature vector, using the backward differencing method, that is, subtracting the feature value of the previous time step from the feature value of the current time step.

[0095] For multidimensional feature vectors, the difference calculations for each dimension are performed simultaneously to maintain the integrity of the feature space. The calculated difference vector contains the change in each feature dimension per unit time, and these changes constitute the instantaneous rate of change. The instantaneous rate of change reflects the speed and direction of change in the device's operating state; positive values ​​indicate feature enhancement, and negative values ​​indicate feature weakening.

[0096] The instantaneous rates of change at all time steps are arranged in chronological order to form a characteristic rate of change sequence. Data smoothing is performed during the construction of this sequence, and a moving average method is used to eliminate noise interference from instantaneous fluctuations. The characteristic rate of change sequence serves as a quantitative representation of the dynamic characteristics of state changes, providing fundamental data for subsequent deviation analysis. After the sequence is constructed, timestamp metadata and rate of change statistics are added for use in subsequent processing stages.

[0097] Equipment health baseline information includes baseline change rate data, which comes from the statistical results of characteristic change rates obtained from long-term monitoring under normal equipment conditions, including the typical change range and distribution pattern of each characteristic dimension.

[0098] The difference calculation is performed independently for each time step in the feature rate of change sequence, comparing the instantaneous rate of change vector at each time step with the baseline rate of change vector dimension by dimension. The calculation uses the absolute difference method, that is, subtracting the baseline rate of change value from the current instantaneous rate of change value to obtain the deviation of each feature dimension. For multi-dimensional feature vectors, the deviation calculation of each dimension is completed simultaneously, forming an instantaneous deviation vector.

[0099] The sign of the deviation value indicates the direction of the deviation of the current rate of change relative to the benchmark level, while the absolute value reflects the degree of deviation. An adaptive threshold mechanism is introduced during the calculation process. When the deviation value exceeds a preset range, a data verification process is initiated to ensure the reliability of the calculation results.

[0100] The calculation of instantaneous deviation values ​​also considers the influence of equipment operating conditions. The baseline change rate is dynamically adjusted based on parameters such as load rate and ambient temperature to eliminate deviations caused by non-fault factors. Instantaneous deviation values ​​for all time steps are arranged chronologically to form a complete deviation dataset, providing input for subsequent sequence analysis. This dataset comprehensively records the dynamic deviation between the equipment condition change rate and the health baseline, serving as a crucial basis for health assessment.

[0101] The instantaneous deviation values ​​calculated in the previous step are arranged according to their corresponding timestamps, with the deviation value of the earliest time step placed at the beginning of the sequence and the deviation value of the current time step placed at the end of the sequence, forming a complete instantaneous deviation sequence. Data integrity verification is performed during the arrangement process to check whether there are any missing or abnormal deviation values ​​at each time step.

[0102] For time steps with missing data, linear interpolation or estimation methods based on adjacent data are used to fill in the gaps, ensuring the continuity of the sequence. Time dimension metadata, including sampling interval, sequence length, and start timestamp, is added during sequence construction. The numerical range of the instantaneous deviation sequence is standardized, mapping the deviation values ​​of each dimension to a uniform numerical interval, eliminating dimensional differences between different feature dimensions. The sequence data is stored using a circular buffer, with the latest data overwriting the earliest data, maintaining a constant sequence length. This instantaneous deviation sequence accurately records the temporal deviation of the rate of change of equipment operating status relative to a health baseline, providing well-structured input data for subsequent window analysis.

[0103] After the sequence is constructed, data quality is assessed, and the integrity index and confidence level of the sequence are calculated to provide a reliability reference for subsequent processing.

[0104] The preset window length is set according to the device's operating characteristics and monitoring needs, and typically includes several to dozens of time steps to ensure that the window can cover meaningful runtime periods. The window slides using a fixed step size, sliding one time step at a time to form a sequence of continuously overlapping windows.

[0105] For each captured window, the statistical characteristics of all instantaneous deviation values ​​within it are calculated: the average deviation value reflects the overall deviation level, the standard deviation characterizes the degree of deviation fluctuation, and the maximum value indicates extreme deviations. The window deviation index is calculated by combining the above statistical characteristics using a weighted fusion algorithm, where the average value is given a higher weight to highlight the overall deviation trend.

[0106] Outlier handling is implemented during the calculation process. Deviation values ​​that significantly exceed the normal range are truncated or weighted down to avoid excessive influence on the window index. After the calculation of each window is completed, a corresponding window deviation index is generated, which quantifies the overall deviation of the equipment status change rate within that time period. The deviation indices of all windows are arranged in chronological order to form a window deviation index sequence. This sequence smooths the fluctuations of instantaneous deviations, highlights the continuous changes in the deviation trend, and provides a stable and reliable data foundation for the final extraction of dynamic deviation.

[0107] All calculated window deviation indices are arranged in chronological order according to their corresponding windows to form a complete deviation sequence. This sequence arrangement ensures continuity over time, with the index of the earliest time window placed at the beginning of the sequence and the index of the most recent time window at the end. Data consistency checks are performed during the deviation sequence construction process to verify the continuity of timestamps and the reasonableness of the values ​​for each window index.

[0108] The dynamic deviation is extracted by selecting the latest window deviation index at the end of the deviation sequence as the final result. This index represents the comprehensive deviation of the equipment status change rate within the most recent time window. The reason for selecting the end index is that it best reflects the current latest status of the equipment and meets the needs of real-time monitoring.

[0109] Numerical verification is implemented during the extraction process to check whether the end index is within a reasonable range. If an anomaly is found, a review mechanism is initiated. The final determined dynamic deviation is a standardized value, whose magnitude directly reflects the degree of deviation of the equipment's current state from the health baseline; a larger value indicates a worse health status. This dynamic deviation serves as a key indicator for equipment health status assessment, providing accurate input for subsequent health index mapping.

[0110] This embodiment calculates the instantaneous rate of change of equipment operating characteristics and forms a characteristic rate of change sequence, thereby achieving precise quantification of the dynamic characteristics of equipment status and providing real-time change basis for health assessment. By calculating the difference between the instantaneous rate of change and the baseline rate of change in the equipment health baseline information, abnormal deviations in the rate of change can be accurately identified, effectively improving the sensitivity of early fault identification. By constructing an instantaneous deviation sequence and employing sliding window analysis, the temporal characteristics of the deviation data are extracted, enhancing the stability of the status assessment. Calculating the window deviation index and extracting the end value of the sequence as the dynamic deviation degree ensures that the assessment results reflect both historical trends and focus on the latest status changes. The entire analysis process, through the collaborative work of multi-level processing mechanisms, achieves dynamic and accurate assessment of equipment health status, providing reliable technical support for predictive maintenance of industrial equipment.

[0111] In one embodiment, the health mapping layer of the health assessment model associates the dynamic deviation with the multidimensional feature vector in the feature sequence and assigns time-step weights to obtain the time decay weight. The time decay weight is then used to weight the dynamic deviation and the multidimensional feature vector to form and output the device health index, including: The health mapping layer of the health assessment model receives the dynamic deviation sequence and feature sequence output from the context encoding layer. The temporal alignment of the two sequences is verified to ensure strict temporal consistency between the dynamic deviation and the corresponding feature vector at each time step. Alignment verification is implemented using a timestamp matching algorithm to check whether the feature vector and dynamic deviation at each time step originate from the same sampling time window.

[0112] After successful verification, the feature vector, corresponding dynamic deviation value, and time step identifier for each time step are combined into a triplet data structure. The time step identifier contains timestamp information and sequence position index, ensuring the traceability of the triplet in the time dimension. The triplets are arranged strictly according to chronological order, with the triplet of the earliest time step placed at the beginning of the sequence and the triplet of the current time step placed at the end of the sequence.

[0113] During the permutation process, data integrity checks are performed, and missing or abnormal triples are marked and repaired using interpolation methods. The final triple sequence serves as the input data for weighted processing, providing a complete data foundation for subsequent time-weighted feature fusion.

[0114] Weight allocation is based on the principle of prioritizing the closest time step, assigning the maximum weight to the triple at the current time step. This maximum value is standardized according to the sequence length and weight distribution requirements. Weight decay is implemented using a monotonically decreasing function, with the function type selected as either linear or exponential decay depending on the specific application scenario.

[0115] For the linear decay mode, the weight value decreases uniformly as the time step moves further away from the current time; for the exponential decay mode, the weight value decreases exponentially with increasing time distance, highlighting the importance of recent data. The weight allocation process also considers the equipment's operating characteristics. For equipment with periodic operating characteristics, the weight decay pattern can be adjusted accordingly to adapt to its operating features.

[0116] Each triple is assigned a weight value that is bound to its time step identifier, ensuring an accurate correspondence between weights and data. After weight allocation, the weight sequence is normalized so that the sum of all weight values ​​is a single unit, guaranteeing the proportional consistency of the weighted calculation. This weight allocation mechanism ensures that the equipment health index more sensitively reflects recent changes in condition, meeting the real-time requirements of equipment condition assessment.

[0117] The traversal operation processes each triplet sequentially in chronological order, performing a weighted calculation on the feature vector at each time step. The weighting of the feature vector uses element-wise multiplication, multiplying each dimension value of the feature vector by the corresponding time decay weight to obtain the weighted feature vector. The weighted feature vector retains the original feature dimension structure, but the values ​​of each dimension are recalibrated according to their temporal importance.

[0118] The weighted calculation of dynamic deviation employs scalar multiplication, multiplying the dynamic deviation value by the time decay weight of the corresponding time step to obtain the weighted deviation value. A numerical range check is performed during the calculation process to ensure that the weighted value is within a reasonable range. For outlier values, a smoothing mechanism is activated, using interpolation correction based on the weighted values ​​of adjacent time steps.

[0119] After weighted calculations are performed at all time steps, a weighted feature value sequence and a weighted deviation sequence are generated, maintaining a strict temporal correspondence between the two sequences. This weighting process effectively enhances the contribution of recent data to health assessment while retaining the reference value of historical data, achieving intelligent weight allocation across the time dimension.

[0120] The weighted feature value sequence and the weighted deviation sequence are first aligned in dimension to ensure that the two sequences have the same time step length and correspondence. Feature concatenation adopts a channel merging method, connecting the weighted feature value vector of each time step with the corresponding weighted deviation value along the feature dimension to form an enhanced feature vector.

[0121] The concatenated feature dimensions contain both original feature information and state deviation information, providing a more comprehensive representation of the device state. The fused feature sequence is input into a fully connected layer for dimensionality reduction, where a linear transformation maps high-dimensional features to a low-dimensional space. During the transformation, a feature selection mechanism is employed to retain information-rich feature dimensions and remove redundant information.

[0122] The output value is compressed to the range of zero to one using a sigmoid activation function, and then scaled by a percentage to obtain a device health index between zero and one hundred. This index comprehensively reflects the overall health status of the device in the current and historical time periods, considering both the absolute values ​​of characteristic parameters and the relative changes in state deviation, thus achieving a comprehensive and accurate health assessment.

[0123] This embodiment achieves spatiotemporal alignment of feature vectors and dynamic deviation by constructing a triplet sequence data structure, providing a precise data foundation for subsequent weighted fusion. A time decay weighting mechanism is employed to make health assessment more focused on recent state changes, effectively improving the timeliness of fault warnings. Through separate weighting of feature vectors and dynamic deviations, the integrity of the original feature information is preserved while highlighting the temporal evolution of state deviation. The weighted feature values ​​and deviations are then concatenated and fused along the feature dimensions to form a comprehensive index reflecting the equipment's health status. The entire weighted transformation process, through the refined application of the time proximity priority principle, ensures that the equipment health index maintains both the ability to perceive long-term trends and the rapid response characteristics to recent state changes, significantly improving the accuracy and practicality of health assessment. In one embodiment, the device health index is compared with a preset warning threshold in real time. When it is determined that the device health index is continuously lower than the warning threshold, a device health warning signal is generated and output, including: The health index sequence is constructed based on the device health index values ​​of the current moment and several consecutive time steps prior to it. These index values ​​are derived from the output of the health assessment model, and each index value represents the quantitative result of the device's health status at a specific point in time.

[0124] The length of the sequence is pre-set based on the equipment's operating characteristics and early warning requirements, typically covering a sufficient time range to capture trends in status changes. During integration, the health index values ​​for the current and historical moments are first retrieved from the data storage module to ensure data integrity and timeliness. Data retrieval is sorted based on timestamps, with the earliest index value placed at the beginning of the sequence and the current index value at the end, forming a linear structure in chronological order.

[0125] Data verification checks are performed during sequence construction. Missing or abnormal health index values ​​are repaired using interpolation or adjacent value substitution methods to ensure sequence continuity. The health index sequence serves as the data foundation for subsequent early warning judgments, and its structure is designed to facilitate time series analysis, such as sliding window calculations or trend detection. This sequence not only contains numerical information of the health index but also includes timestamp metadata to track the temporal dimension characteristics of status changes. By integrating index values ​​from multiple moments, the sequence can reflect the dynamic evolution of equipment health status, providing comprehensive temporal context for anomaly detection.

[0126] The warning threshold is preset based on equipment type, operating environment, and historical data, and is usually expressed as a percentage value of the health index, such as 85%, indicating that when the health index is below this value, it may indicate an abnormal equipment condition. The comparison operation starts from the first element of the health index sequence and processes the health index value of each device in chronological order.

[0127] For each index value, a numerical comparison is performed: if the health index value is less than the warning threshold, the time point is marked as an abnormal state point; otherwise, it is considered a normal state point. The marking process includes assigning an identifier to each abnormal state point and recording its timestamp and health index value to form a list of abnormal state points. The comparison process considers the fluctuation characteristics of the health index. To avoid mismarking caused by instantaneous fluctuations, smoothing processing can be introduced, such as using a moving average of the health index value for comparison.

[0128] The labeling of abnormal state points is based not only on single-point values ​​but also on the trends of adjacent points. For example, if the current point is below a threshold but the points before and after it are normal, further verification may be needed. All comparison results are stored in a temporary data structure for subsequent statistical use. The output of this step is the state label (normal or abnormal) for each point in the health index sequence, which provides input for subsequent statistics on the number of consecutive abnormalities. By comparing each point one by one, the state at each time point is accurately assessed, improving the accuracy of early warnings.

[0129] The preset cycle time is set according to the equipment's operating characteristics and maintenance requirements, typically covering several complete operating cycles or specific monitoring periods. The counting of consecutive occurrences of abnormal state points is based on chronological order, starting from the first abnormal state point and checking whether subsequent time points are also marked as abnormal. Continuity is determined by requiring abnormal state points to be temporally adjacent with no normal state point intervals; the length of the longest consecutive abnormal sequence is recorded during the counting process. The preset abnormal occurrence threshold is set based on equipment reliability requirements and historical data; this threshold represents the minimum number of consecutive abnormalities required to trigger an alert.

[0130] The statistical operation employs a sliding window method, with the window width corresponding to a preset period. The window slides along the time axis to detect consecutive anomalies within different time periods. Statistical results include the number of consecutive anomalies, the duration of the anomaly, and the start and end timestamps of the anomaly sequence. When the number of consecutive occurrences of an anomaly is detected to reach or exceed a threshold, the event is recorded, and preparations are made to initiate subsequent early warning procedures. During the statistical process, the temporal distribution characteristics of the anomaly sequence are simultaneously monitored to ensure the accuracy of anomaly detection.

[0131] When the statistical results show that the number of consecutive occurrences of abnormal status points has not reached the preset threshold, the system determines that the current device status is abnormal but has not yet reached the warning level. The determination is based on a comparison of the number of consecutive abnormal occurrences with the threshold value, and also takes into account the time distribution characteristics of the abnormal sequence for auxiliary verification. The system maintains normal monitoring status and continues to perform routine monitoring tasks such as health index collection, sequence integration, and abnormality marking.

[0132] Monitoring data is updated in real time, with newly generated health index values ​​continuously added to the health index sequence, replacing the oldest historical data to maintain sequence length. The system records the current assessment result and related statistical information for subsequent trend analysis and threshold optimization. Under normal monitoring conditions, the system interface displays the device's current health index and operating status without triggering any warning signals. Monitoring data is continuously stored in the historical database, providing data support for long-term device performance evaluation. The system also monitors other relevant parameters to ensure a comprehensive understanding of the device's operating status.

[0133] When the number of consecutive occurrences of an abnormal state point reaches or exceeds a preset threshold, the system determines that the early warning trigger condition is met. The determination process verifies the continuity and persistence of the abnormal sequence, eliminating misjudgments caused by transient interference. The early warning signal generation includes elements such as device identification information, early warning trigger timestamp, health index value, abnormal statistics, and early warning level. The early warning level is determined based on the severity and duration of the abnormality, and is usually divided into multiple levels to distinguish different levels of urgency.

[0134] The warning signals are encapsulated according to standard communication protocols to ensure compatibility with existing monitoring systems. Output methods include displaying warning information on the interface, storing records in the database, and pushing messages via the network. The warning signals contain detailed descriptions of the anomalies and suggested handling measures to support equipment maintenance decisions. The system simultaneously activates a warning confirmation mechanism, awaiting operator response or automatically executing preset emergency procedures. After the warning signal is output, the system continues to monitor changes in equipment status and track the development trend of the warning event.

[0135] This embodiment integrates health indices from multiple consecutive time points to form a sequence, achieving continuous monitoring and trend analysis of equipment health status, significantly improving the comprehensiveness of status assessment. The mechanism of comparing early warning thresholds with health indices one by one accurately identifies abnormal state points, providing a reliable basis for early fault diagnosis. By statistically analyzing the consecutive occurrences of abnormal state points, false alarms caused by transient interference are effectively avoided, improving the reliability of the early warning system. The setting of multi-level early warning trigger conditions ensures timely alarms for important abnormal situations while preventing unnecessary frequent alarm interference. The standardized output format of the early warning signal ensures compatibility with existing monitoring systems, facilitating unified equipment health management. The entire early warning mechanism, through the organic combination of time-series analysis and threshold judgment, achieves intelligent monitoring and early warning of equipment health status, providing an effective technical means for predictive maintenance.

[0136] In one embodiment, if the number of abnormal occurrences reaches or exceeds a threshold, the warning triggering condition is determined to be met, and a device health warning signal is generated and output, including: The identifier information of the target industrial equipment is extracted from the equipment registration database, which stores basic parameters such as the equipment's unique identification code, model specifications, and installation location. Identifier extraction is achieved through equipment identifier indexing, ensuring that the acquired equipment information perfectly matches the current monitoring target. The current timestamp is generated using a high-precision clock source, with time information including year, month, day, hour, minute, second, and millisecond precision, and is synchronized with a standard time source.

[0137] The equipment health index is obtained from the real-time monitoring data buffer. This index undergoes data validity verification to ensure that the value is within a reasonable range and consistent with the current equipment status. Multiple verification mechanisms are implemented during data extraction: verifying the validity of the equipment identifier to confirm that the equipment is in normal monitoring status; checking the continuity of timestamps to avoid data anomalies caused by time jumps; and verifying the up-to-dateness of the health index to ensure that the most recent assessment results are used. All extracted data items are temporarily stored in the early warning processing cache and undergo standardized format processing to provide standardized data input for subsequent early warning level calculations.

[0138] The numerical ratio is calculated using precise division, dividing the actual number of consecutive occurrences of abnormal state points obtained from statistics by a preset abnormality threshold to obtain a standardized ratio. This ratio reflects the severity of the current abnormal situation relative to the warning threshold, and its value is typically greater than or equal to one. The preset level threshold ranges are divided according to the criticality of the equipment and maintenance requirements; for example, level one, level two, and level three warning ranges are set, each corresponding to different levels of urgency.

[0139] The comparison operation proceeds sequentially, starting from the lowest threshold level. The calculated numerical ratios are compared to the upper and lower bounds of each level interval to determine its corresponding threshold range. The determination of the warning level also considers auxiliary indicators such as the duration of the anomaly and the rate of decline in the health index for a comprehensive assessment. The level determination result is related to the equipment type; different types of equipment can use different threshold range settings.

[0140] This early warning level mechanism enables a tiered differentiation of abnormal situations, providing a basis for subsequent differentiated handling. The level determination process records a detailed judgment log, including ratio calculation results, threshold range parameters, and judgment criteria, ensuring the traceability of early warning levels.

[0141] The default message format adopts a layered structure, consisting of a message header, a data body, and a checksum. The message header sets the start identifier, message length, and protocol version number to ensure that the receiving end can correctly identify and parse it.

[0142] The data body is organized according to a fixed field order: the timestamp field uses an international standard time format, accurate to milliseconds; the identifier information field contains the device's unique identification code and type code; the device health index field retains two decimal places with precision; and the warning level field uses numeric codes to represent different levels. Separators are set between fields to ensure the accuracy of data parsing. Data integrity checks are performed during the encapsulation process to verify that the length and format of each field meet the specifications.

[0143] The checksum calculation employs a cyclic redundancy check (CRC) algorithm and is appended to the end of the message for transmission error detection. The encapsulated warning signal is temporarily stored in a transmission buffer, awaiting output instructions. The output stage supports multiple communication protocols, selecting the appropriate output method based on the monitoring system's interface requirements. For monitoring systems supporting the Modbus TCP protocol, the warning signal is converted into a standard Modbus data frame format and transmitted via the Ethernet interface.

[0144] For systems using the OPC UA protocol, a corresponding OPC UA node structure is constructed to enable data subscription and publishing. During the output process, transmission status is monitored, and a retransmission mechanism is automatically initiated when communication interruption or transmission failure is detected to ensure reliable delivery of warning signals. Simultaneously, warning logs are generated, saving message content, sending time, and reception confirmation status for subsequent auditing and analysis. The warning signal is ultimately delivered to the target monitoring system's human-machine interface, presented as a visual alert, and triggers relevant alarm processing procedures.

[0145] This embodiment extracts multi-dimensional information such as device identifiers, timestamps, and health indices to ensure that the warning signal contains a complete state context, providing comprehensive data support for accurate diagnosis. A mechanism employing numerical ratio calculation and multi-level threshold range comparison enables refined grading of anomaly severity, effectively distinguishing warning situations of different urgency levels. Structured encapsulation through a preset message format ensures the standardization and cross-system compatibility of the warning signal, facilitating integration into existing device management platforms. The judgment logic linking the warning level to specific device parameters allows for differentiated strategies in the warning response based on the characteristics of different devices. The entire warning generation process, through the orderly connection of data verification, level determination, and standard output, achieves intelligent identification and graded warning of abnormal device states, significantly improving the accuracy and timeliness of device health management.

[0146] Reference Figure 2 As shown, this invention provides an equipment health assessment system based on the electrical characteristics of industrial equipment, applicable to any of the above-mentioned equipment health assessment methods based on the electrical characteristics of industrial equipment, comprising: The acquisition module is used to acquire the current and voltage signals of the target industrial equipment, perform anti-aliasing filtering and power frequency interference suppression, and obtain regular waveform data. The analysis module is used to extract time-domain features, frequency-domain features, and time-frequency-domain features from the regularized waveform data, and then vectorize and concatenate the three types of features to generate a multi-dimensional feature vector. The association module is used to construct a feature sequence from multiple consecutive multidimensional feature vectors arranged in chronological order, calculate the dynamic deviation between the feature sequence and the predefined equipment health benchmark information, and map the dynamic deviation to the equipment health index according to the principle of prioritizing the closest time point. The processing module compares the device health index with the preset warning threshold in real time. When the device health index is determined to be continuously lower than the warning threshold, the module generates and outputs a device health warning signal.

[0147] This invention provides an equipment health assessment system based on the electrical characteristics of industrial equipment. By extracting time-domain, frequency-domain, and time-frequency-domain features from current and voltage signals in parallel and then vectorizing and concatenating them, a multi-dimensional feature vector comprehensively characterizing the equipment's operating status can be constructed. This overcomes the limitations of single-dimensional feature information and provides a richer and more reliable data foundation for health assessment. By constructing a feature sequence from multi-dimensional feature vectors across multiple consecutive time steps and calculating its dynamic deviation from the equipment health baseline pattern, the system can accurately capture the temporal gradual change pattern of equipment status, enabling early insight into performance degradation trends. Mapping the dynamic deviation to a health index based on the principle of prioritizing time proximity significantly improves the real-time performance and sensitivity of health status assessment, making the assessment results more reflective of the equipment's current condition. By setting early warning thresholds and continuously comparing the health index, online and automatic monitoring and early warning of equipment health status can be achieved, providing timely and accurate decision support for predictive maintenance and effectively avoiding production interruptions and economic losses caused by downtime due to faults.

[0148] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the system and each module described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0149] 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 method for assessing equipment health based on the electrical characteristics of industrial equipment, characterized in that, include: The current and voltage signals of the target industrial equipment are collected and subjected to anti-aliasing filtering and power frequency interference suppression to obtain regular waveform data. The time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the regularized waveform data, and the time-domain features, frequency-domain features, and time-frequency-domain features are vectorized and concatenated to generate a multi-dimensional feature vector. The pre-defined health assessment model constructs a feature sequence from multiple consecutive multi-dimensional feature vectors arranged in chronological order. The equipment operation feature trend of the feature sequence is analyzed, and the equipment operation feature trend is compared and similarity calculated with the predefined equipment health benchmark information item by item to obtain the equipment health index. The device health index is compared with a preset warning threshold in real time. When the device health index is determined to be continuously lower than the warning threshold, a device health warning signal is generated and output.

2. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 1, characterized in that, The current and voltage signals of the target industrial equipment are subjected to anti-aliasing filtering and power frequency interference suppression to obtain regular waveform data, including: The current signal of the power supply circuit in the target industrial equipment is obtained by a current transformer, and the voltage signal of the equipment input terminal in the target industrial equipment is obtained by a voltage probe. The current signal and the voltage signal are respectively input into a preset anti-aliasing filter to filter out frequency components higher than 5kHz in the signal, thereby obtaining a first voltage signal and a first current signal; The first voltage signal and the first current signal are band-stop filtered at a first frequency and a second frequency to suppress power frequency interference, thereby obtaining the second voltage signal and the second current signal. The waveform data of the second voltage signal and the waveform data of the second current signal are sampled synchronously and integrated to output the regularized waveform data.

3. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 1, characterized in that, The time-domain features, frequency-domain features, and time-frequency-domain features are extracted from the regularized waveform data, and the three types of features are vectorized and concatenated to generate a multi-dimensional feature vector, including: Identify and separate the steady-state operation data segment and the startup transient data segment in the regularized waveform data; A set of steady-state time-domain parameters are extracted from the steady-state operation data segment, and a set of transient time-domain parameters are extracted from the startup transient data segment. The steady-state time-domain parameters and the transient time-domain parameters are integrated into the time-domain feature. Perform a fast Fourier transform on the steady-state operating data segment to obtain the spectrum, extract the fundamental amplitude from the spectrum, and statistically analyze the frequency band energy ratio to obtain the frequency domain characteristics; Perform multi-level wavelet packet decomposition on the data segment of the startup transient process and calculate the node energy entropy to obtain the time-frequency domain features; The time-domain features, frequency-domain features, and time-frequency-domain features are sequentially concatenated and combined to generate the multidimensional feature vector.

4. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 3, characterized in that, A set of steady-state time-domain parameters is extracted from the steady-state operation data segment, and a set of transient time-domain parameters is extracted from the startup transient data segment. The steady-state time-domain parameters and the transient time-domain parameters are integrated into the time-domain feature, including: The steady-state time-domain parameters are obtained by calculating and integrating the effective value, peak factor and waveform distortion rate of the current signal based on the steady-state operating data segment. Based on the aforementioned transient data segment, the current integral value and current rise time of the current signal are calculated and integrated to obtain the transient time-domain parameters. The steady-state time-domain parameters and the transient time-domain parameters are combined to form the time-domain features.

5. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 1, characterized in that, The process involves constructing a feature sequence from multiple consecutive multidimensional feature vectors arranged in chronological order using a preset health assessment model, analyzing the equipment operation feature trends of the feature sequence, and comparing and calculating the similarity between the equipment operation feature trends and predefined equipment health benchmark information item by item to obtain an equipment health index, including: The multidimensional feature vector is input into a preset health assessment model, and the multidimensional feature vector of the current time and multiple consecutive time steps in the preceding time step are constructed into a feature sequence through the sequence construction layer of the health assessment model. The feature sequence is input into the time-series feature extraction layer of the health assessment model, and multi-scale local feature extraction and trend recognition are performed on the feature sequence to obtain the equipment operation feature trend; The dynamic trend analysis of the device's operating characteristics is performed through the context encoding layer of the health assessment model to obtain an instantaneous deviation sequence, and the dynamic deviation between the instantaneous deviation sequence and the device's health baseline information is calculated. Through the health mapping layer of the health assessment model, the dynamic deviation is associated with the multidimensional feature vector in the feature sequence and weighted by time step to obtain the time decay weight. The time decay weight is then used to calculate the dynamic deviation and the multidimensional feature vector to form and output the device health index.

6. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 5, characterized in that, The step of inputting the feature sequence into the time-series feature extraction layer of the health assessment model, performing multi-scale local feature extraction and trend recognition on the feature sequence to obtain the equipment operation feature trend includes: The temporal feature extraction layer extracts feature subsequences of corresponding lengths sequentially from the feature sequence using multiple preset continuous time intervals of different lengths. The mean, variance, and peak features of each extracted feature subsequence are calculated separately. The mean, variance, and peak features at the same time point are combined to form an enhanced feature vector. The enhanced feature vector is compared with a predefined device health baseline pattern to calculate the pattern matching degree at each time point. Target feature points are selected based on the pattern matching degree, and the selected target feature points are connected in chronological order to form the device operation feature trend.

7. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 5, characterized in that, The step of performing dynamic trend analysis on the equipment operating characteristic trends through the context encoding layer of the health assessment model to obtain an instantaneous deviation sequence, and calculating the dynamic deviation between the instantaneous deviation sequence and the equipment health baseline information, includes: The difference between feature values ​​at adjacent time steps in the device operation feature trend is calculated through the context coding layer, and the difference is used as the instantaneous rate of change to form a feature rate of change sequence. The instantaneous rate of change at each time step in the characteristic rate of change sequence is compared with the baseline rate of change in the device health baseline information to calculate the difference, thus obtaining the instantaneous deviation value at each time step. The instantaneous deviation values ​​of each time step are arranged in chronological order to form an instantaneous deviation sequence; The instantaneous deviation sequence is truncated by sliding a window according to a preset window length, and the window deviation index of all instantaneous deviation values ​​within each window is calculated; The window deviation indices are arranged in chronological order to form a deviation sequence, and the window deviation index at the end of the deviation sequence is taken as the dynamic deviation degree.

8. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 5, characterized in that, The health mapping layer of the health assessment model associates the dynamic deviation with the multidimensional feature vector in the feature sequence and assigns time-step weights to obtain time decay weights. These time decay weights are then used to weight and calculate the dynamic deviation and the multidimensional feature vector, respectively, to form and output the device health index, including: The health mapping layer of the health assessment model arranges the dynamic deviation and the feature vectors of each time step in the feature sequence in chronological order to form a triplet sequence. Based on the principle of prioritizing the closest time step, time decay weights are assigned to the time steps in the triplet sequence, wherein the time decay weight at the current time step is set to the maximum value and decreases sequentially in reverse chronological order. Traverse the triplet sequence, multiply the feature vector by the time decay weight to obtain the weighted feature value, and multiply the dynamic deviation by the time decay weight to obtain the weighted deviation. The weighted feature value and the weighted deviation are concatenated along the feature dimension to form and output the device health index.

9. The equipment health assessment method based on the electrical characteristics of industrial equipment according to claim 1, characterized in that, The step of comparing the device health index with a preset warning threshold in real time, and generating and outputting a device health warning signal when the device health index is determined to be continuously lower than the warning threshold, includes: The device health indices at the current moment and multiple consecutive moments preceding it are integrated to form a health index sequence; Each device health index in the health index sequence is compared with the warning threshold one by one. If the device health index is lower than the warning threshold, the device health index is marked as an abnormal state point. The system counts whether the number of consecutive occurrences of the abnormal state points within a preset period reaches or exceeds a preset threshold for the number of abnormal occurrences. If the threshold for the number of abnormal occurrences is not reached, it is determined that the warning triggering condition is not met, and the system remains in normal monitoring status. If the number of abnormal occurrences reaches or exceeds the threshold, the warning triggering condition is determined to be met, and the device health warning signal is generated and output.

10. A system for assessing the health of industrial equipment based on its electrical characteristics, characterized in that, The equipment health assessment method based on the electrical characteristics of industrial equipment as described in any one of claims 1-9 includes: A high-frequency acquisition module is used to acquire current and voltage signals of the target industrial equipment, perform anti-aliasing filtering and power frequency interference suppression, and obtain regular waveform data. The signal processing module, wherein the analysis module is used to extract the time-domain features, frequency-domain features and time-frequency-domain features from the regularized waveform data respectively, and to vectorize and concatenate the time-domain features, the frequency-domain features and the time-frequency-domain features to generate a multi-dimensional feature vector; The feature calculation module, the association module is used to construct a feature sequence from multiple consecutive multidimensional feature vectors arranged in chronological order through a preset health assessment model, analyze the equipment operation feature trend of the feature sequence, compare the equipment operation feature trend with predefined equipment health benchmark information item by item and calculate the similarity to obtain the equipment health index; The health assessment module, the processing module is used to compare the device health index with a preset warning threshold in real time, and when it is determined that the device health index is continuously lower than the warning threshold, generate and output a device health warning signal.

Images