A method, device and medium for evaluating health of a unit based on PHM

Abstract

The application discloses a kind of based on PHM's unit health assessment method, equipment and medium, it is related to unit health assessment technical field, including, the multiple-source heterogeneous operating data of unit, according to equipment hierarchical topological relationship is mapped to object, obtains component level correlation dataset;From component level correlation dataset, extract multi-dimensional state feature and carry out dimension reduction splicing, obtain health feature vector;Utilize physical constraint meta-learning algorithm to carry out few-sample operating condition self-adaptive update to health feature vector, generate dynamic health benchmark curve, and calculate three-dimensional degradation degree, pass through mapping function and convert degradation degree into basic probability distribution, obtain component level health evidence set;Using component level health evidence set, the causal topological structure between components is mined and evidence conflict coefficient is calculated, while deducing the causal credible weight of each evidence, generate adaptive conflict adjustment factor.The application realizes that unit health benchmark changes from static benchmark to operating condition self-adaptive dynamic benchmark.

Description

Technical Field

[0001] This invention relates to the field of unit health assessment technology, and in particular to a unit health assessment method, equipment and medium based on PHM. Background Technology

[0002] As large-scale generating units evolve towards higher parameterization, continuous operation, and intelligence, PHM (Prognostics and Health Management) technology is gradually expanding from simple fault diagnosis to include operational status perception, health assessment, trend prediction, and maintenance-assisted decision-making. Existing technologies typically rely on sensor networks, monitoring systems, and historical operation and maintenance data to collect multi-source operational information such as vibration, temperature, pressure, current, voltage, speed, and control response. This information is then combined with threshold rules, statistical analysis, or data-driven algorithms to identify anomalies, classify conditions, and score health for the generating units, thereby improving equipment operational safety, maintainability, and operational efficiency.

[0003] However, existing technologies still have some obvious shortcomings: First, multi-source heterogeneous operating data are mostly processed at the whole machine level or a small number of key measurement points, lacking component-level correlation modeling that combines the hierarchical topology of equipment, making it difficult to accurately reflect the structural correlation, functional coupling and local degradation differences between components; Second, under conditions of frequent unit operating condition switching, insufficient effective samples and conflicting health evidence of multiple components, problems such as health benchmark mismatch, fluctuation in state discrimination and insufficient fusion credibility are likely to occur. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a PHM-based unit health assessment method to solve the problems of insufficient component-level correlation modeling and low stability and reliability of health assessment under complex operating conditions in the prior art.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a PHM-based unit health assessment method, comprising: collecting multi-source heterogeneous operating data of the unit, mapping objects according to the hierarchical topology of the equipment to obtain a component-level associated dataset; extracting multi-dimensional state features from the component-level associated dataset and performing dimensionality reduction and splicing to obtain a health feature vector; using a physical constraint meta-learning algorithm to adaptively update the health feature vector under few-sample operating conditions, generating a dynamic health baseline curve, and calculating a three-dimensional degradation degree, converting the degradation degree into a basic probability allocation through a mapping function to obtain a component-level health evidence set; using the component-level health evidence set to mine the causal topological structure between components and calculate the evidence conflict coefficient, while inferring the causal credibility weight of each piece of evidence to generate an adaptive conflict adjustment factor; using the causal credibility weight and the adaptive conflict adjustment factor to perform weighted preprocessing on the component-level health evidence set, and executing the DS evidence synthesis rule according to the equipment hierarchical structure to obtain a unit health trust distribution; calculating the unit comprehensive health index and unit assessment confidence based on the unit health trust distribution, and identifying the degradation evolution trend based on the health index sequence over continuous time periods to generate risk warning levels and maintenance decision recommendations.

[0007] As a preferred embodiment of the PHM-based unit health assessment method of the present invention, the specific steps for obtaining the component-level association dataset are as follows: Collect multi-source heterogeneous operating data from the units to form a raw operating dataset, and construct a hierarchical topology relationship of the equipment based on the equipment object information, signal association information, and hierarchical affiliation information in the raw operating dataset; The original operational dataset is processed for timestamp alignment, unit unification, and quality verification to obtain a standardized operational dataset. Based on the equipment hierarchical topology, the standardized operational dataset is mapped to the corresponding measurement point nodes to form a measurement point-level mapping dataset. Based on the measurement point-level mapping dataset, the nodes are merged and associated according to the membership and connection relationships between the measurement point nodes and the key component nodes. Combined with the key component identifiers and topological association information, a component-level association dataset is generated.

[0008] As a preferred embodiment of the PHM-based unit health assessment method of the present invention, the specific steps for obtaining the health feature vector are as follows: Extract the time-domain state features, frequency-domain state features, topological coupling features, and control response features corresponding to each key component from the component-level associated dataset to form the original multidimensional feature set; The original multidimensional feature set is used to reduce the dimensionality of different types of state features, and then the features are spliced ​​and fused according to a preset arrangement order to obtain a health feature vector.

[0009] As a preferred embodiment of the PHM-based unit health assessment method of the present invention, the steps of using the physical constraint meta-learning algorithm to adaptively update the health feature vector under few operating conditions, generating a dynamic health baseline curve, and calculating the three-dimensional degradation degree are as follows: Extract a small number of operating condition features corresponding to the current operating state from the health feature vector, and combine them with preset physical constraints to input the physical constraint meta-learning algorithm to obtain adaptive update parameters under the current operating condition; The preset health baseline is corrected by adaptively updating parameters to form a dynamic health baseline curve that matches the current working conditions. Based on the dynamic health baseline curve and the corresponding health feature vector, the amplitude deviation, evolution rate deviation and physical consistency deviation are calculated to obtain the three-dimensional degradation degree.

[0010] As a preferred embodiment of the PHM-based unit health assessment method of the present invention, the specific steps for obtaining the component-level health evidence set are as follows: Based on the three-dimensional degradation degree, the state mapping range of each key component under the preset health status judgment framework is determined, and degradation degree state mapping data is formed. Based on the degradation status mapping data, the three-dimensional degradation of each key component is transformed into a basic probability allocation of the corresponding health status using a mapping function, forming component-level probability allocation data. By using component-level probability allocation data, the basic probability allocation of each health state corresponding to the same key component is collected to generate a component-level health evidence set.

[0011] As a preferred embodiment of the PHM-based unit health assessment method of the present invention, the specific steps for generating the adaptive conflict adjustment factor are as follows: Extract the health evidence evolution sequence of each key component in a continuous time period from the component-level health evidence set, and mine the causal relationship between each key component based on the health evidence evolution sequence to obtain the causal topology between components; The causal topology is used to calculate the evidence conflict coefficients corresponding to health evidence between adjacent key components. By using causal topology and evidence conflict coefficient, the credibility of health evidence for each key component in causal transmission is deduced, and the causal credibility weight corresponding to each piece of health evidence is obtained. The causal credibility weight is used to adaptively adjust the evidence conflict coefficient to obtain the adaptive conflict adjustment factor.

[0012] As a preferred embodiment of the PHM-based unit health assessment method of the present invention, the specific steps for obtaining the unit health trust distribution are as follows: Based on the component-level health evidence set, causal credibility weight, and adaptive conflict adjustment factor, the health evidence corresponding to each key component is weighted and corrected to form a preprocessed component-level health evidence set. Based on the preprocessed component-level health evidence set, the health evidence corresponding to the key components at the same level is grouped according to the equipment hierarchy, and the DS evidence synthesis rules are executed layer by layer to form a health trust distribution at each level. The health trust distributions at each level are recursively synthesized upwards along the equipment hierarchy to form the unit health trust distribution.

[0013] As a preferred embodiment of the PHM-based unit health assessment method of the present invention, the specific steps for generating risk warning levels and maintenance decision recommendations are as follows: The overall health index of the unit is calculated using the unit health trust distribution within the framework of health status discrimination, and the corresponding unit assessment confidence level is calculated based on the concentration of quality allocation in the unit health trust distribution. Based on the unit's comprehensive health index and the unit's assessment confidence level, a health index sequence for continuous time periods is constructed, and the deterioration and evolution trend of the unit is identified based on the magnitude, rate of change, and stage transition characteristics of the health index sequence in adjacent time periods. The risk quantification and classification of the deterioration evolution trend, the unit's comprehensive health index and the unit's assessment confidence level are used to form the corresponding risk warning level for the unit. Based on the risk warning level, the trend of deterioration and evolution, and the unit's comprehensive health index, the corresponding maintenance strategy for the unit is determined, and maintenance decision recommendations are formed.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein the computer program, when executed by the processor, implements any step of the PHM-based unit health assessment method as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the PHM-based unit health assessment method as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By using a physical constraint meta-learning algorithm to adaptively update the health feature vector under few operating conditions, the unit health benchmark is transformed from a static benchmark to an adaptive dynamic benchmark under operating conditions. This not only improves the accuracy and stability of health assessment under complex operating conditions, but also refines the degradation state from a single anomaly judgment to a multi-dimensional deterioration characterization. Furthermore, by using a component-level health evidence set to mine the causal topological structure between components and calculate the evidence conflict coefficient, the unit health analysis is transformed from a static and independent component evidence processing method to an evidence collaborative analysis method with causal propagation relationships and conflict adjustment capabilities, thereby improving the robustness and credibility of component-level health evidence fusion. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a PHM-based unit health assessment method.

[0019] Figure 2 A flowchart for constructing health feature vectors.

[0020] Figure 3 A flowchart for generating a set of component-level health evidence.

[0021] Figure 4 This is a comparison chart of the dynamic health baseline curve and the preset health baseline. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides a unit health assessment method based on PHM, including the following steps: S1. Collect multi-source heterogeneous operating data of the unit, perform object mapping based on the hierarchical topology of the equipment, and obtain a component-level associated dataset; extract multi-dimensional status features from the component-level associated dataset and perform dimensionality reduction and splicing to obtain a health feature vector.

[0026] Collect multi-source heterogeneous operating data from the units to form a raw operating dataset, and construct a hierarchical topology relationship of the equipment based on the equipment object information, signal association information and hierarchical affiliation information in the raw operating dataset.

[0027] The specific process includes periodically collecting electrical operation data, speed regulation process data, actuator status data, protection interlock data, sensor measurement data, control command data, and feedback response data through data acquisition terminals deployed at key parts of the unit using industrial communication protocols. After timestamp alignment and data cleaning to remove abnormal, missing, and duplicate data, the data is integrated and stored to form the original operation dataset.

[0028] Text identifiers containing equipment object information are extracted from the original operational dataset. String matching is used to identify the name and type of physical entities. Encoded fields containing hierarchical attribution information are extracted. Based on the hierarchical rules of the encoded fields, the attribution levels of physical entities are divided. A membership tree is established from unit to equipment set, from equipment set to component, and from component to measurement point. Signal flow direction fields containing signal association information are extracted. The signal flow direction fields are analyzed to determine the data connection paths between physical entities. Physical entities are mapped to nodes, and membership and connection relationships are mapped to edges to construct a hierarchical topology of equipment.

[0029] The original operational dataset is processed for timestamp alignment, unit unification, and quality verification to obtain a standardized operational dataset. Based on the hierarchical topology of the equipment, the standardized operational dataset is mapped to the corresponding measurement point nodes to form a measurement point-level mapping dataset.

[0030] The specific process includes resampling and aligning the timestamps in the original running dataset, unifying data from different collection frequencies to the same time base, performing unit conversion on the aligned data, converting data from different physical units to a unified standard unit, removing null values ​​and filtering outliers on the converted data, removing data records containing null values ​​and filtering outliers that exceed the preset physical range, and generating a standardized running dataset.

[0031] Based on the measurement point node identifiers contained in the equipment hierarchical topology, data columns corresponding to the measurement point node identifiers are extracted from the standardized operation dataset, and the data columns are attribute-bound to the measurement point nodes to form a measurement point-level mapping dataset containing measurement point node identifiers and corresponding standardized values.

[0032] It should be noted that the physical range is set based on the emission standard limits of the monitored object, the actual operating concentration range, and the sensor hardware characteristics. The upper and lower limits of the actual measurement of the equipment are set (usually a margin of 10%-20% is required) to ensure that it can cover data fluctuations under normal and fault conditions, while also taking into account the accuracy of the measurement.

[0033] Based on the measurement point-level mapping dataset, the nodes are merged and associated according to the membership and connection relationships between the measurement point nodes and the key component nodes. Combined with the key component identifiers and topological association information, a component-level association dataset is generated.

[0034] The specific process includes: based on the measurement point-level mapping dataset, according to the membership tree between measurement point nodes and key component nodes determined in the equipment hierarchical topology and the connection paths contained in the signal association information, classifying and merging the standardized values ​​corresponding to all measurement point nodes belonging to the same key component node to form measurement point data groups divided by key components; extracting the component identifier and signal flow direction field corresponding to each key component node from the equipment hierarchical topology, and binding the component identifier with the measurement point data groups divided by key components; and combining the data connection paths between measurement point nodes determined by the signal flow direction field to associate and label the standardized values ​​corresponding to different measurement point nodes within the same key component, thereby obtaining a component-level association dataset containing component identifiers, standardized measurement point values, and connection relationships between measurement points.

[0035] It should be noted that a key component node refers to a topological object in the equipment hierarchical topology that is mapped from a physical entity and represents a component (such as a bearing or valve) in the unit that has an independent function or requires key monitoring; a key component identifier refers to a unique code or name used to distinguish and identify a specific key component; and topology association information refers to structured data that describes the affiliation between measuring point nodes and key component nodes, as well as the data connection path between measuring point nodes.

[0036] Extract the time-domain state features, frequency-domain state features, topological coupling features, and control response features corresponding to each key component from the component-level associated dataset to form the original multidimensional feature set.

[0037] The specific process includes: based on the time series data contained in the component-level associated dataset, using a sliding window to extract continuous time series segments, performing statistical analysis on the time series segments to extract time-domain state features that characterize the data distribution pattern; performing spectral transformation processing on the time series segments to analyze the frequency domain energy distribution and extract frequency domain state features that characterize the periodic fluctuation pattern; simultaneously obtaining the cross-correlation coefficients and lag times between different measurement point node data from the component connection paths determined by the hierarchical topology of the equipment, constructing topological coupling features that characterize the dynamic correlation strength between components; and extracting overshoot, settling time, and steady-state error in the step response process from the time synchronization sequence of control command data and feedback response data, constructing control response features that characterize the dynamic performance of the actuator; and concatenating the time-domain state features, frequency-domain state features, topological coupling features, and control response features into vectors to form the original multidimensional feature set.

[0038] The original multidimensional feature set is used to reduce the dimensionality of different types of state features, and then the features are spliced ​​and fused according to a preset arrangement order to obtain a health feature vector.

[0039] The specific process includes: extracting feature data matrices corresponding to various types of state features from the original multidimensional feature set; standardizing the feature data matrices to obtain a standard feature matrix with zero mean and unit variance; using principal component analysis to decompose the standard feature matrix into covariance matrix and eigenvalues; selecting the top several order eigenvectors whose cumulative variance contribution rate reaches a preset proportion to form a projection matrix; mapping the standard feature matrix to the low-dimensional subspace spanned by the projection matrix; and obtaining a dimensionality-reduced feature set containing time-domain dimensionality reduction features, frequency-domain dimensionality reduction features, topological coupling dimensionality reduction features, and control response dimensionality reduction features.

[0040] According to the preset arrangement order, channel splicing operation is performed on the various types of dimensionality reduction features contained in the dimensionality reduction feature set, and dimensionality reduction features with different physical meanings are stacked and fused along the feature dimension to form a healthy feature vector containing complete state information.

[0041] It should be noted that the preset percentage is based on the information requirement after feature dimensionality reduction, setting the cumulative variance contribution rate threshold of the feature values ​​in principal component analysis (usually set to 85% to 95%) to ensure that the dimensionality-reduced features can cover the main change information in the original data; the arrangement order is based on the logical sequence of feature extraction and the classification level of different physical attributes, setting the order of time-domain dimensionality reduction features, frequency-domain dimensionality reduction features, topological coupling dimensionality reduction features, and control response dimensionality reduction features to ensure that the fused vector structure remains consistent.

[0042] S2. The physical constraint meta-learning algorithm is used to adaptively update the health feature vector under few working conditions to generate a dynamic health baseline curve. The three-dimensional degradation degree is calculated and the degradation degree is transformed into a basic probability assignment through a mapping function to obtain a set of component-level health evidence.

[0043] Extract a small number of operating condition features corresponding to the current operating state from the health feature vector, and combine them with preset physical constraints to input the physical constraint meta-learning algorithm to obtain adaptive update parameters under the current operating condition.

[0044] The specific process includes extracting time-series segments corresponding to the current operating state from the health feature vector, segmenting the time-series segments using a sliding window to obtain a subset of few-sample data, and extracting features from the subset of few-sample data to obtain few-sample operating condition features.

[0045] The few-sample operating condition features and physical constraints are fed into the physical constraint meta-learning algorithm. The physical constraints are used to limit the reasonable update boundaries of the few-sample operating condition features in terms of state evolution direction, control response matching relationship and component association change range. The physical constraint meta-learning algorithm iteratively adjusts the correction magnitude and correction direction of the health benchmark corresponding to the health feature vector based on the matching between the few-sample operating condition features and the physical constraints, and forms the adaptive update parameters under the current operating condition after convergence.

[0046] It should be noted that the physical constraints are predetermined based on the unit's structural composition, operating mechanism, control logic, and the normal correlation between various state variables in historical health condition data. The physical constraint meta-learning algorithm is a learning method that combines unit operating mechanism constraints with meta-learning adaptive capabilities. It is used to quickly adjust health baseline parameters based on few sample operating condition characteristics under conditions of limited samples and changing operating conditions, while ensuring that the update results conform to the actual physical laws of the unit. The adaptive update parameters are a set of parameters formed after convergence by iteratively adjusting the magnitude and direction of health baseline correction based on the matching between few sample operating condition characteristics and physical constraints. The adaptive update parameters refer to the set of parameters formed by the physical constraint meta-learning algorithm in the process of correcting the health baseline under the current operating condition based on few sample operating condition characteristics, including baseline offset parameters, rate of change correction parameters, and physical consistency adjustment parameters.

[0047] The preset health baseline is corrected by adaptively updating parameters to form a dynamic health baseline curve that matches the current working conditions.

[0048] The specific process includes: mapping the adaptive update parameters under the current operating condition to the feature dimension arrangement order in the health benchmark that is consistent with the health feature vector; using the benchmark offset parameter to correct the benchmark position of the health benchmark in the time domain dimensionality reduction feature, frequency domain dimensionality reduction feature, topological coupling dimensionality reduction feature, and control response dimensionality reduction feature; using the rate of change correction parameter to correct the change trend of the health benchmark in continuous time periods; using the physical consistency adjustment parameter to correct the correlation state between various features in the health benchmark, so that the corrected health benchmark simultaneously reflects the few-sample operating condition features corresponding to the current operating state and the reasonable change boundary limited by physical constraints; after completing various corrections, continuously expanding the corrected benchmark features at each time point according to the time arrangement order of the health feature vector, and smoothly connecting the corrected benchmarks at adjacent time points to form a dynamic health benchmark curve that matches the current operating condition.

[0049] Specifically, the baseline offset parameter, rate of change correction parameter, and physical consistency adjustment parameter are determined according to the time-domain dimensionality reduction feature, frequency-domain dimensionality reduction feature, topological coupling dimensionality reduction feature, and control response dimensionality reduction feature in the health feature vector, respectively. The baseline offset parameter is determined by comparing the average eigenvalue of each feature dimension within the current few-sample operating condition time window with the average baseline value of the health baseline. When the average eigenvalue is higher than the average baseline value, the baseline offset parameter is used to move the health baseline of the corresponding feature dimension in the direction of increasing value. When the average eigenvalue is lower than the average baseline value, the baseline offset parameter is used to move the health baseline of the corresponding feature dimension in the direction of decreasing value. The amount of movement is determined based on the difference between the average eigenvalue and the average baseline value.

[0050] The rate of change correction parameter is determined by obtaining the changes in health characteristic values ​​and health benchmark values ​​at adjacent times within the current few-sample operating condition time window, and comparing the average difference between the two changes over a continuous period. When the rate of change of the health characteristic value is higher than the rate of change of the health benchmark value, the slope of the corresponding health benchmark is increased; when the rate of change of the health characteristic value is lower than the rate of change of the health benchmark value, the slope of the corresponding health benchmark is decreased, so that the corrected health benchmark is consistent with the trend of health characteristic changes under the current operating condition.

[0051] After correcting the baseline position and trend, the corrected baseline values ​​for each feature dimension are compared with the physical lower limit, physical upper limit, control response matching range, and component-related change range defined by the physical constraints. When the corrected baseline value is within the range defined by the physical constraints, the physical consistency adjustment parameter is set to a smaller value or is not adjusted again. When the corrected baseline value exceeds the range defined by the physical constraints, the physical consistency adjustment parameter is increased according to the degree of exceeding the reasonable change boundary, and the corrected baseline value is moved towards the nearest reasonable change boundary. The greater the degree of exceeding the reasonable change boundary, the greater the movement, until the corrected baseline value is again within the reasonable change range defined by the physical constraints.

[0052] After the baseline offset parameter, rate of change correction parameter, and physical consistency adjustment parameter are determined, the baseline offset parameter is used to adjust the baseline position of each feature dimension at the corresponding time. Then, the rate of change correction parameter is used to adjust the magnitude and direction of change between adjacent baseline values ​​in a continuous period. The physical consistency adjustment parameter is used to revert baseline values ​​that exceed the reasonable change boundary, thus obtaining the corrected baseline features corresponding to each time. The corrected baseline features are arranged in the time order of the health feature vector, and the abrupt changes between adjacent baseline values ​​are weakened by using the moving average method of adjacent time baseline values. The smoothed baseline features of each time are continuously connected to form a dynamic health baseline curve that matches the current working condition.

[0053] The initial values ​​of the baseline offset parameter and the rate of change correction parameter are derived from the positional and rate of change differences between the current few-sample operating condition features and the corresponding health baselines in historical health operating condition data. The value of the physical consistency adjustment parameter is derived from the degree to which the corrected baseline value deviates from the physical constraint limit. As the few-sample operating condition features corresponding to the current operating condition are updated, the baseline offset parameter, the rate of change correction parameter, and the physical consistency adjustment parameter are reacquired, and the dynamic health baseline curve is updated synchronously.

[0054] It should be noted that the health benchmark is pre-established based on the normal distribution range, changing trend and interrelationship of the health feature vectors corresponding to each key component in historical health condition data.

[0055] like Figure 4The graph compares the dynamic health baseline curve with the preset health baseline, illustrating the difference between the dynamic health baseline curve (formed by adaptively updating the health feature vector using a physical constraint meta-learning algorithm) and the preset health baseline under complex and low-sample operating conditions. The dynamic health baseline curve adjusts synchronously with the phased changes in the health feature vector, while the preset health baseline remains relatively static. This visually reflects the technical effect of transforming the unit's health baseline from a static baseline to an adaptive dynamic baseline, and demonstrates the improved accuracy and stability of health assessment under complex operating conditions.

[0056] Based on the dynamic health baseline curve and the corresponding health feature vector, the amplitude deviation, evolution rate deviation and physical consistency deviation are calculated to obtain the three-dimensional degradation degree.

[0057] The specific process includes: based on the correspondence between the dynamic health baseline curve and the corresponding health feature vector in terms of feature dimension and time sequence, comparing the feature values ​​of each moment in the health feature vector with the baseline values ​​at the same position in the dynamic health baseline curve item by item, and normalizing the comparison differences to obtain the amplitude deviation degree, which reflects the degree of deviation of the current operating state at the feature value level; simultaneously extracting the changes between the health feature vector and the dynamic health baseline curve at adjacent moments, comparing and normalizing the two types of changes to obtain the evolution rate deviation degree, which reflects the degree of deviation of the current operating state at the change rhythm level; and combining the physical lower limit threshold, physical upper limit threshold, and change threshold for the health index evolution trend judgment to verify the feature value range and change direction of the health feature vector to obtain the physical consistency deviation degree; and combining the amplitude deviation degree, evolution rate deviation degree, and physical consistency deviation degree to obtain the three-dimensional degradation degree.

[0058] The expressions for calculating amplitude deviation, evolution rate deviation, and physical consistency deviation are as follows: ; ; ; in, Indicates time No. The amplitude deviation of each key component Indicates amplitude. Indicates the index of key components. Indicates time No. Health characteristic observations of key components Indicates time No. Dynamic health benchmark values ​​for key components Represents a very small positive number. Indicates time No. Deviation in the evolution rate of key components Indicates rate, Indicates time No. The rate of change of the observed health characteristics of key components Indicates time No. The rate of change of dynamic health benchmark values ​​of key components Indicates the first The maximum allowable rate of change threshold for each key component under historical healthy operating conditions. Indicates time No. Deviation in physical consistency of key components Represents physics. Indicates the physical lower limit threshold. Indicates the physical upper limit threshold. The threshold for determining the evolution trend of the health index.

[0059] It should be noted that, Through statistics The 95th to 99th percentile of the characteristic change rate under the historical healthy working conditions of each component is determined. The example value range depends on the characteristic type (e.g., vibration velocity change rate 0.5 to 5 mm / s, temperature change rate 0.1 to 2 ℃ / min). It is determined based on the 0.3% quantile of the equipment technical manual, material physical properties, or historical health data statistics. For example, the value range is 5~20℃ for bearing temperature and 0~0.1 mm for vibration amplitude. It is determined based on the 99.7th percentile of the material's heat resistance limit, insulation class, or historical health data. For example, the value range is 80~120 ℃ for bearing temperature and 0.5~2.0 mm for vibration amplitude. It is obtained by statistically analyzing the differences between adjacent time periods of historical health index series, and taking the 90th percentile of the stable range of change. The example value range is 0.02~0.10.

[0060] Based on the three-dimensional degradation degree, the state mapping range of each key component under the preset health status judgment framework is determined, and degradation degree state mapping data is formed.

[0061] The specific process includes: based on the amplitude deviation, evolution rate deviation, and physical consistency deviation included in the three-dimensional degradation degree, extracting the numerical boundary thresholds corresponding to healthy state, sub-healthy state, slightly degraded state, moderately degraded state, and severely degraded state from the health status discrimination framework; comparing the amplitude deviation, evolution rate deviation, and physical consistency deviation with the numerical boundaries item by item to determine the numerical range in which each dimension of deviation falls; combining the numerical ranges in which the amplitude deviation, evolution rate deviation, and physical consistency deviation of each key component fall to determine the corresponding state mapping interval of each key component under the health status discrimination framework; and associating and storing the state mapping intervals corresponding to each key component with the component identifier to form degradation degree state mapping data containing the component identifier, the interval to which the amplitude deviation belongs, the interval to which the evolution rate deviation belongs, and the interval to which the physical consistency deviation belongs.

[0062] It should be noted that the health status assessment framework is determined based on the technical specifications of the unit equipment, the statistical analysis results of historical health operation data, and industry standards. The health status assessment framework divides the numerical range into 0 to 0.2 corresponding to a healthy state, 0.2 to 0.4 corresponding to a sub-healthy state, 0.4 to 0.6 corresponding to a slightly deteriorated state, 0.6 to 0.8 corresponding to a moderately deteriorated state, and 0.8 to 1.0 corresponding to a severely deteriorated state. For example, if the amplitude deviation of a certain key component is 0.35, the evolution rate deviation is 0.25, and the physical consistency deviation is 0.15, then the deviation of each dimension falls into the sub-healthy state or healthy state range, respectively.

[0063] Based on the degradation status mapping data, the three-dimensional degradation of each key component is transformed into a basic probability allocation of the corresponding health status using a mapping function, forming component-level probability allocation data.

[0064] The specific process includes extracting the amplitude deviation, evolution rate deviation, and physical consistency deviation of each key component from the degradation state mapping data; using a mapping function (interval membership mapping function) to convert the amplitude deviation, evolution rate deviation, and physical consistency deviation into basic probability allocation values ​​corresponding to healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state, respectively; and combining the basic probability allocation values ​​corresponding to the amplitude deviation, evolution rate deviation, and physical consistency deviation of each key component to form component-level probability allocation data containing component identification, basic probability allocation of amplitude deviation, basic probability allocation of evolution rate deviation, and basic probability allocation of physical consistency deviation.

[0065] Specifically, the mapping function is a function that maps the interval membership degree of each state according to the numerical intervals in the health status discrimination framework. For any dimension deviation of amplitude deviation, evolution rate deviation and physical consistency deviation, the numerical interval in which the deviation of any dimension is located is determined, and then the interval membership degree of the deviation of any dimension to the corresponding health state in the numerical interval is determined to be 1, and the interval membership degree of the deviation of any dimension to the other health states is determined to be 0. The interval membership degree corresponding to each health state is used as the basic probability assignment value.

[0066] When the deviation of any dimension is between 0 and 0.2, the basic probability allocation value corresponding to the healthy state is set to 1, and the basic probability allocation values ​​corresponding to the sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state are set to 0; when the deviation of any dimension is greater than 0.2 but not greater than 0.4, the basic probability allocation value corresponding to the sub-healthy state is set to 1, and the basic probability allocation values ​​corresponding to the other healthy states are set to 0; when the deviation of any dimension is greater than 0.4 but not greater than 0.6, the basic probability allocation value corresponding to the slightly deteriorated state is set to 1; when the deviation of any dimension is greater than 0.6 but not greater than 0.8, the basic probability allocation value corresponding to the moderately deteriorated state is set to 1; when the deviation of any dimension is greater than 0.8 but not greater than 1.0, the basic probability allocation value corresponding to the severely deteriorated state is set to 1, and the basic probability allocation values ​​corresponding to the other healthy states are all set to 0 in the corresponding cases.

[0067] For any dimension deviation less than 0, map that dimension deviation to 0; for any dimension deviation greater than 1, map that dimension deviation to 1. The sum of the basic probability distribution values ​​for each dimension deviation corresponding to healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state is 1.

[0068] For example, when the amplitude deviation is 0.35, the amplitude deviation falls within the sub-healthy state range corresponding to 0.2 to 0.4, and the basic probability allocation value of the amplitude deviation to the sub-healthy state is determined to be 1; when the evolution rate deviation is 0.25, the evolution rate deviation falls within the sub-healthy state range corresponding to 0.2 to 0.4, and the basic probability allocation value of the evolution rate deviation to the sub-healthy state is determined to be 1; when the physical consistency deviation is 0.15, the physical consistency deviation falls within the healthy state range corresponding to 0 to 0.2, and the basic probability allocation value of the physical consistency deviation to the healthy state is determined to be 1.

[0069] By using component-level probability allocation data, the basic probability allocation of each health state corresponding to the same key component is collected to generate a component-level health evidence set.

[0070] The specific process includes extracting the basic probability distributions of amplitude deviation, evolution rate deviation, and physical consistency deviation corresponding to the same key component from the component-level probability distribution data; classifying and summarizing the basic probability distributions of amplitude deviation, evolution rate deviation, and physical consistency deviation according to healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state to obtain a set of basic probability distributions for each health state corresponding to the same key component; and associating and binding the set of basic probability distributions for each health state corresponding to the same key component with the component identifier of the key component to form a set of component-level health evidence.

[0071] S3. Use the component-level health evidence set to mine the causal topology between components and calculate the evidence conflict coefficient. At the same time, deduce the causal credibility weight of each piece of evidence and generate an adaptive conflict adjustment factor.

[0072] The evolutionary sequence of health evidence for each key component in a continuous time period is extracted from the component-level health evidence set, and the causal relationship between each key component is mined based on the evolutionary sequence of health evidence to obtain the causal topology between components.

[0073] The specific process includes extracting the basic probability allocation set corresponding to each key component in a continuous time period from the component-level health evidence set, arranging it according to component identification and time order, and continuously connecting the results of the basic probability allocation of the same key component in the healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state and severely deteriorated state over time to form the health evidence evolution sequence of the corresponding key component in a continuous time period.

[0074] Based on the evolution sequence of health evidence and the component connection paths already determined in the hierarchical topology of the equipment, the evolution sequences of health evidence between key components with connection relationships are compared. Cross-correlation analysis and lag time analysis are used to determine the order of change, consistency of change direction, and strength of continuous correlation between the evolution sequences of health evidence of different key components. The key component that changes first and can stably correspond to the change of another key component is determined as the preceding component, and the key component that changes later and maintains a correspondence with the preceding component is determined as the subsequent component, thus obtaining the causal topology between components.

[0075] The conflict coefficient of evidence corresponding to health evidence between adjacent key components is calculated using a causal topological structure, expressed as follows: ; in, Indicates adjacent key components and The coefficient of evidence conflict between them and This represents the index number of two adjacent critical components in a causal topology. and This represents mutually exclusive health state focal elements in the health state discrimination framework. Indicates components Focusing element The basic probability allocation value, Indicates components Focusing element The basic probability allocation value, Indicates adjacent key components and The causal adjacency coefficient between them.

[0076] It should be noted that, A causal topology graph was constructed by mining the causal relationships of the evolutionary sequences of health evidence from consecutive time periods for each key component; if the component and If there is a direct causal connection edge, the value is 1; otherwise, the value is 0.

[0077] The specific process includes: extracting pairs of adjacent key components with direct connections from the causal topology between components and determining the corresponding causal adjacency coefficients; extracting the set of basic probability assignments of adjacent key component pairs at the same time from the set of component-level health evidence; calculating and summing the basic probability assignments of the previous key component in each health state and the basic probability assignments of the subsequent key component in mutually exclusive health states to obtain the conflict quantity between health evidences; and using the causal adjacency coefficients to weight and correct the conflict quantity between health evidences to obtain the evidence conflict coefficients corresponding to the health evidences between adjacent key components.

[0078] By using causal topology and evidence conflict coefficients, the credibility of health evidence for each key component in causal transmission is deduced, and the causal credibility weight corresponding to each piece of health evidence is obtained.

[0079] The specific process includes: extracting the associated positions of each key component on the causal transmission path based on the preceding components, subsequent components, and causal adjacency relationships in the causal topology between components; extracting the basic probability allocation set of the corresponding key component and adjacent key components in continuous time periods from the component-level health evidence set; and comparing the evidence conflict coefficients between adjacent key components by combining the determined sequence of changes, consistency of change direction, and continuous correlation strength in the health evidence evolution sequence. When the health evidence is consistent with the causal transmission direction and the evidence conflict coefficient is low, the credibility of the corresponding health evidence in causal transmission is determined to be high. When the health evidence is inconsistent with the causal transmission direction or the evidence conflict coefficient is high, the credibility of the corresponding health evidence in causal transmission is determined to be low. The credibility of the same key component in each adjacent causal connection relationship is merged to obtain the causal credibility weight corresponding to each health evidence.

[0080] The causal credibility weight is used to adaptively adjust the evidence conflict coefficient to obtain the adaptive conflict adjustment factor.

[0081] The specific process includes: extracting the causal credibility weights corresponding to each key component from the component-level health evidence set; matching the causal credibility weights with the evidence conflict coefficients between adjacent key components to obtain a matching result containing the causal credibility weights and the evidence conflict coefficients; adjusting the evidence conflict coefficients based on the causal credibility weights in the matching result; using the causal credibility weights as adjustment coefficients to amplify or reduce the evidence conflict coefficients to obtain an adaptive conflict adjustment factor that reflects the combined influence of the credibility of evidence and the level of conflict.

[0082] Specifically, for any key component, the causal credibility weight corresponding to the key component is read, and the adjacent key components with a direct causal connection relationship with the key component are determined from the causal topology between components. Then, the evidence conflict coefficient between the key component and each adjacent key component is read. The evidence conflict coefficient between the key component and each adjacent key component is averaged and merged to obtain the average evidence conflict coefficient corresponding to the key component. When any key component has a direct causal connection relationship with only one adjacent key component, the evidence conflict coefficient between the key component and the adjacent key component is directly determined as the average evidence conflict coefficient.

[0083] The adaptive conflict adjustment factor is determined by the causal credibility weight and the average evidence conflict coefficient. Specifically, the unit value is used as the adjustment benchmark. The remaining proportion of the total unit quantity together with the causal credibility weight is determined as the inverse proportion of the credibility weight. Then, the conflict discount is determined based on the joint mapping result of the inverse proportion of the credibility weight and the average evidence conflict coefficient. The proportion retained after the unit adjustment benchmark is discounted is determined as the adaptive conflict adjustment factor corresponding to any key component. Both the causal credibility weight and the average evidence conflict coefficient are normalized and have a value range of 0 to 1. The value range of the adaptive conflict adjustment factor is also 0 to 1.

[0084] When the causal credibility weight is high, the reverse proportion of the credibility weight is low, and the constraint of the average evidence conflict coefficient on the adaptive conflict adjustment factor is reduced accordingly, and the adaptive conflict adjustment factor approaches 1; when the causal credibility weight is low and the average evidence conflict coefficient is high, the reverse proportion of the credibility weight and the evidence conflict level together form a large conflict discount, and the adaptive conflict adjustment factor approaches 0; when the average evidence conflict coefficient is low, it indicates that the health evidence between adjacent key components has high consistency, and the adaptive conflict adjustment factor maintains a high value.

[0085] For example, the causal credibility weight corresponding to any key component is 0.8, and the evidence conflict coefficients between any key component and two adjacent key components are 0.3 and 0.5, respectively. After mean merging, the average evidence conflict coefficient is 0.4. The credibility weight inverse ratio corresponding to the causal credibility weight of 0.8 is 0.2. The conflict discount corresponding to the credibility weight inverse ratio of 0.2 and the average evidence conflict coefficient of 0.4 is 0.08. The retention ratio corresponding to the unit adjustment benchmark after conflict discount is 0.92. Therefore, the adaptive conflict adjustment factor corresponding to any key component is 0.92.

[0086] When any critical component does not have a direct causal connection with an adjacent critical component in the causal topology between components, no conflict adjustment is performed on the health evidence corresponding to any critical component, and the adaptive conflict adjustment factor is set to 1; when the component-level health evidence set or the causal topology between components is updated with continuous time periods, the average evidence conflict coefficient and the adaptive conflict adjustment factor are re-determined.

[0087] S4. The component-level health evidence set is preprocessed using causal credibility weights and adaptive conflict adjustment factors, and the DS evidence synthesis rules are executed according to the equipment hierarchy to obtain the unit health trust distribution.

[0088] Based on the component-level health evidence set, causal credibility weight, and adaptive conflict adjustment factor, the health evidence corresponding to each key component is weighted and corrected to form a preprocessed component-level health evidence set.

[0089] The specific process includes: extracting the aggregated basic probability allocation set corresponding to each key component from the component-level health evidence set; matching the aggregated basic probability allocation set with the causal credibility weight and adaptive conflict adjustment factor corresponding to each key component to obtain a matching result containing the aggregated basic probability allocation set, causal credibility weight, and adaptive conflict adjustment factor; weighting and correcting the aggregated basic probability allocation set based on the causal credibility weight and adaptive conflict adjustment factor in the matching result; applying the causal credibility weight as the basic weighting coefficient to the aggregated basic probability allocation set; and using the adaptive conflict adjustment factor to perform a secondary adjustment on the weighted result to obtain the preprocessed component-level health evidence set after credibility weighting and conflict adjustment.

[0090] Specifically, for any key component, the basic probability allocations for healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state are retrieved from the aggregated basic probability allocation set according to the component identifier. The causal credibility weight and adaptive conflict adjustment factor corresponding to any key component are then matched one-to-one with the basic probability allocations corresponding to the five health states. Both the causal credibility weight and the adaptive conflict adjustment factor are represented by normalized values, with values ​​ranging from 0 to 1.

[0091] In the basic weighting process, based on the credibility of the evidence represented by the causal credibility weight, credibility discounts are applied to the basic probability allocations of the health status, sub-health status, slightly deteriorated status, moderately deteriorated status, and severely deteriorated status. The closer the causal credibility weight is to 1, the higher the retention rate of the basic probability allocations corresponding to the five health statuses; the closer the causal credibility weight is to 0, the lower the retention rate of the basic probability allocations corresponding to the five health statuses. The basic probability allocations that are not retained during the credibility discount process are transferred to the uncertain state corresponding to the health status discrimination framework, which is used to represent the uncertainty in the health status judgment caused by insufficient causal credibility.

[0092] During the secondary adjustment process, conflict discounting is applied to the basic probability allocations of health status, sub-health status, slightly deteriorated status, moderately deteriorated status, and severely deteriorated status after credibility discounting, according to the degree of evidence conflict adjustment represented by the adaptive conflict adjustment factor. The closer the adaptive conflict adjustment factor is to 1, the smaller the impact of health evidence conflict between adjacent key components on the corresponding health evidence of any key component, and the higher the retention rate of the basic probability allocation after credibility discounting. The closer the adaptive conflict adjustment factor is to 0, the greater the impact of health evidence conflict between adjacent key components, and the lower the retention rate of the basic probability allocation after credibility discounting. Basic probability allocations that are not retained during the conflict discounting process also enter an uncertain state.

[0093] After completing the credibility discount and conflict discount, the basic probability assignments for healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, severely deteriorated state, and uncertain state are normalized so that the sum of all basic probability assignments corresponding to the same key component remains 1. The normalized basic probability assignments are then associated with the corresponding component identifiers to form a preprocessed set of component-level health evidence.

[0094] When the causal credibility weight and adaptive conflict adjustment factor corresponding to any key component are both 1, the basic probability allocation set after aggregation for any key component remains unchanged; when the causal credibility weight is less than 1, the basic probability allocation corresponding to insufficient causal credibility enters an uncertain state; when the adaptive conflict adjustment factor is less than 1, the basic probability allocation corresponding to the impact of evidence conflict enters an uncertain state; the lower the causal credibility weight and adaptive conflict adjustment factor, the higher the basic probability allocation corresponding to the uncertain state.

[0095] For example, in the aggregated basic probability allocation set corresponding to any key component, the basic probability allocations for healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state are 0.10, 0.50, 0.25, 0.10, and 0.05, respectively. The causal credibility weight is 0.80, the adaptive conflict adjustment factor is 0.92, and after credibility discounting and conflict discounting, the basic probability allocations for the five healthy states are 0.0736, 0.3680, 0.1840, 0.0736, and 0.0368, respectively. The basic probability allocation for the uncertain state is 0.2640, and the sum of all basic probability allocations is 1.

[0096] When the causal credibility weight, adaptive conflict adjustment factor, or basic probability assignment set after aggregation corresponding to any key component is updated over consecutive time periods, credibility discount, conflict discount, and normalization processing are re-executed, and the basic probability assignment of any key component in the pre-processed component-level health evidence set is updated.

[0097] Based on the preprocessed component-level health evidence set, the health evidence corresponding to key components at the same level is grouped according to the equipment hierarchy, and the DS evidence synthesis rules are executed layer by layer to form a health trust distribution at each level.

[0098] The specific process includes extracting the basic probability allocation set corresponding to each key component after credibility weighting and conflict adjustment from the preprocessed component-level health evidence set, and classifying and grouping the basic probability allocation sets corresponding to key components under the same level according to the determined affiliation level from unit to equipment set and from equipment set to component in the equipment hierarchical topology relationship, forming the same-level key component health evidence group divided according to the equipment hierarchical structure.

[0099] For each group of health evidence for key components at the same level, the DS evidence synthesis rule is used to fuse the basic probability distribution set of each key component in the group layer by layer. The basic probability distribution of different key components in the group in the healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state and severely deteriorated state is orthogonally combined and normalized to obtain the health trust distribution of each level that reflects the overall health trust level of that level.

[0100] It should be noted that the DS evidence synthesis rule refers to the method of combining the basic probability assignments of two or more independent evidence sources under the same identification framework, and obtaining the comprehensive basic probability assignment by calculating the intersection product and normalizing it using the conflict coefficient.

[0101] Specifically, "critical components at the same level" refers to critical components that have the same hierarchical level and correspond to the same direct superior node in the device hierarchical topology. For the preprocessed component-level health evidence set, the component identifier, hierarchical level, and direct superior node identifier of each critical component are read. Critical components with the same hierarchical level and direct superior node identifier are grouped into the same critical component health evidence group at the same level, while critical components with different direct superior node identifiers are grouped into different critical component health evidence groups at the same level. When a critical component health evidence group at the same level contains only one critical component, the basic probability allocation set of the corresponding critical component is directly determined as the corresponding level health trust distribution.

[0102] For health evidence groups of key components on the same layer that contain two or more key components, the basic probability allocation set after aggregation is read according to the arrangement order of key components in the equipment hierarchical topology. Each basic probability allocation set after aggregation contains the basic probability allocation corresponding to the healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state and severely deteriorated state. The basic probability allocation corresponding to each health state is normalized after preprocessing.

[0103] When executing the DS evidence synthesis rule, the basic probability allocation sets corresponding to the two first-ranked key components are paired, and the key elements corresponding to healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state are combined in pairs. For key element combinations whose intersection is the corresponding healthy state, the joint quality corresponding to the key element combination is assigned to the healthy state corresponding to the intersection. For key element combinations with no intersection, the joint quality corresponding to the key element combination is assigned to the evidence conflict coefficient. The evidence conflict coefficient represents the degree of mutual exclusion between the health evidence of two key components, and the value ranges from 0 to 1. The higher the evidence conflict coefficient, the more obvious the difference in the judgment of the health state corresponding to the two key components.

[0104] After completing the coulomb pairing, the non-conflict joint quality corresponding to the healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state are collected separately. The total amount of all non-conflict joint quality is used as the normalization benchmark to recalibrate the proportion of the non-conflict joint quality corresponding to each healthy state, so that the total amount of trust in each healthy state after recalibration remains at the unit total amount, forming the intermediate health trust distribution corresponding to the two key components.

[0105] The intermediate health trust distribution and the basic probability allocation set corresponding to the next key component are further processed by focal element pairing, evidence conflict coefficient aggregation, and non-conflict joint quality normalization to form an updated intermediate health trust distribution. All the basic probability allocation sets of the health evidence group of the same level key component are processed in the order of key component arrangement until all key components in the health evidence group of the same level key component participate in the DS evidence synthesis rule to obtain the hierarchical health trust distribution of the corresponding health evidence group of the same level key component.

[0106] The hierarchical health trust distribution includes the trust level of healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state. The trust level of each health state is within the range of 0 to 1, and the total trust level of each health state remains a unit total. The higher the trust level of each health state, the higher the overall support level of the corresponding level belonging to the corresponding health state.

[0107] After completing the DS evidence synthesis at the critical component attribution level, the hierarchical health trust distribution corresponding to the health evidence group of each critical component at the same level is bound to the corresponding direct superior equipment set node, and the hierarchical health trust distribution is used as the health evidence of the corresponding equipment set node; the equipment set nodes with the same direct superior unit node are grouped together, and DS evidence synthesis is performed upward according to the same focal element pairing, evidence conflict coefficient aggregation and non-conflict joint quality normalization processing method, so as to form the critical component hierarchical health trust distribution and the equipment set hierarchical health trust distribution in sequence, thus obtaining the health trust distribution at each level.

[0108] The health trust distributions at each level are recursively synthesized upwards along the equipment hierarchy to form the unit health trust distribution.

[0109] The specific process includes extracting a hierarchical health trust distribution sequence from lower to higher levels from the health trust distribution at each level, and using the hierarchical health trust distribution at each lower level as the basic evidence for synthesis at the next higher level, based on the established hierarchical topology tree from components to equipment sets and from equipment sets to units. The DS evidence synthesis rules are then used to recursively orthogonally combine and normalize the basic evidence with the health trust distribution at each higher level. The health trust distribution results after fusion at lower levels are used as new evidence sources and iteratively synthesized with the health trust distribution at higher levels until the fusion of all levels from the bottom components to the top units is completed, forming a unit health trust distribution that reflects the overall health trust level of the unit.

[0110] Specifically, according to the hierarchical relationship tree in the device hierarchical topology, the health trust distribution is collected starting from the lowest level where the key components are located; for any device set node, the hierarchical health trust distribution corresponding to each key component directly belonging to any device set node is read, and each hierarchical health trust distribution is assigned to the basic evidence group corresponding to any device set node according to the direct superior node identifier; the hierarchical health trust distributions belonging to different device set nodes are respectively entered into different basic evidence groups and are not cross-processed in the same recursive synthesis process.

[0111] When the basic evidence group corresponding to any device set node contains only one level of health trust distribution, the level of health trust distribution is directly used as the health trust distribution corresponding to any device set node; when the basic evidence group corresponding to any device set node contains two or more levels of health trust distribution, the DS evidence synthesis rule is executed on each level of health trust distribution in turn according to the arrangement order of each key component in the device hierarchical topology.

[0112] When executing the DS evidence synthesis rule, the health state focal elements corresponding to the healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, severely deteriorated state, and uncertain state in the two hierarchical health trust distributions to be synthesized are combined accordingly. The joint quality corresponding to the health state focal elements with non-empty intersection is assigned to the health state corresponding to the intersection, and the joint quality corresponding to the health state focal elements with empty intersection is assigned to the evidence conflict coefficient. Then, the total amount of all non-conflicting joint quality is used as the normalization benchmark to recalibrate the proportion of non-conflicting joint quality corresponding to each health state, forming the intermediate health trust distribution corresponding to the current recursive stage.

[0113] After the synthesis of the first two levels of health trust distributions is completed, the intermediate health trust distribution is used as the new basic evidence. It is then combined with the next level of health trust distribution in the basic evidence group to perform health status focal element combination, evidence conflict coefficient aggregation and non-conflict joint quality normalization processing until all levels of health trust distributions in the basic evidence group corresponding to any device set node participate in the recursive synthesis to obtain the device set level health trust distribution corresponding to any device set node.

[0114] Each device set hierarchical health trust distribution is bound to the corresponding device set node identifier. Based on the hierarchical relationship between device set nodes and unit nodes in the device hierarchical topology, the device set hierarchical health trust distributions that directly belong to the same unit node are assigned to the next higher level basic evidence group corresponding to the unit node. Following the fixed arrangement order of the device set nodes, the same DS evidence synthesis rules are applied to the device set hierarchical health trust distributions in the next higher level basic evidence group to form the top-level health trust distribution corresponding to the unit node.

[0115] During the recursive synthesis process, after each device-level health trust distribution synthesis is completed, the health trust distribution corresponding to the current device level is used as the basic evidence of the direct superior node. The underlying health trust distribution that has already participated in the synthesis of the current device level is no longer written to higher device levels separately. In this way, the process is advanced sequentially from the key component level, the device set level, and the unit level, until the top unit node in the device hierarchical topology forms a health trust distribution, and the recursive synthesis ends.

[0116] The unit health trust distribution includes the degree of health trust of the unit in healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, severely deteriorated state, and uncertain state. The value of the health trust degree corresponding to each health state is between 0 and 1, and the total amount of all health trust degrees is kept as a unit total. The higher the health trust degree corresponding to any health state, the higher the comprehensive support of the health evidence at each equipment level for the unit being in that health state.

[0117] S5. Based on the unit health trust distribution, the computer unit comprehensive health index and unit assessment confidence level are calculated, and the deterioration evolution trend is identified based on the health index sequence over a continuous period of time, generating risk warning levels and maintenance decision recommendations.

[0118] The overall health index of the unit is calculated using the unit health trust distribution within the framework of health status discrimination, and the corresponding unit assessment confidence level is calculated based on the concentration of quality allocation in the unit health trust distribution. The expression is as follows: ; ; in, This indicates the overall health index of the generator set; Indicates the unit index; Represents the basic probability assignment function for the generator set; The health status proposition represents a definite state in which the unit is operating normally and has not experienced significant deterioration; This represents the basic probability assignment value of the crew's health status proposition. This represents the uncertain state proposition in the health status discrimination framework. This represents the basic probability assignment value of the unit for uncertain state propositions. This indicates the confidence level of the unit assessment.

[0119] It should be noted that, It is a function used to assign corresponding confidence qualities to the healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, severely deteriorated state, and uncertain state in the unit health confidence distribution.

[0120] The specific process includes extracting the basic probability allocation values ​​of the corresponding health status propositions and the basic probability allocation values ​​of the uncertain status propositions under the health status discrimination framework from the crew health trust distribution. The basic probability allocation values ​​of the health status propositions are used as the main representation of the crew being in a normal health state. The basic probability allocation values ​​of the uncertain status propositions are incorporated into the health degree representation process according to a preset reduction method. After weighted synthesis, the crew comprehensive health index is obtained. The crew comprehensive health index can reflect both the clarity of the current health status of the crew and the uncertainties that still exist in the crew health judgment.

[0121] The degree of concentration of quality allocation is judged based on the clustering of the basic probability distributions corresponding to each health status proposition in the crew health trust distribution. The degree of concentration is then corrected by combining the proportion of the basic probability distributions of uncertain status propositions, thus forming the corresponding crew assessment confidence level. When the basic probability distributions in the crew health trust distribution are mainly concentrated on a few clear health status propositions and the proportion of uncertain status propositions is low, the crew assessment confidence level is correspondingly increased. When the basic probability distributions in the crew health trust distribution are dispersed among multiple health status propositions or the proportion of uncertain status propositions is high, the crew assessment confidence level is correspondingly decreased. This allows the crew assessment confidence level to reflect the clarity and credibility of the judgment conclusions corresponding to the comprehensive health index of the crew.

[0122] It should be noted that the reduction method is a weighted reduction rule set according to the degree of influence of uncertain state propositions on the assessment of unit health. It is used to include the basic probability allocation of uncertain state propositions in the calculation of the unit's comprehensive health index at a lower proportion than the influence of healthy state propositions.

[0123] Based on the unit's comprehensive health index and the unit's assessment confidence level, a health index series for continuous time periods is constructed, and the deterioration and evolution trend of the unit is identified based on the magnitude, rate of change, and stage transition characteristics of the health index series in adjacent time periods.

[0124] The specific process includes extracting the unit comprehensive health index sequence and the unit assessment confidence score sequence for continuous time periods from the unit comprehensive health index and the unit assessment confidence score, and pairing and binding the unit comprehensive health index and the unit assessment confidence score at the same time according to the time sequence to form a continuous time period health index sequence containing the unit comprehensive health index and the unit assessment confidence score.

[0125] The magnitude of change is obtained by analyzing the numerical differences of the health index sequence over consecutive time periods in adjacent time periods, and the rate of change in adjacent time periods is obtained by combining the time interval. At the same time, the stage transition characteristics are determined by analyzing the range changes of the unit's comprehensive health index within the health status discrimination framework, which includes healthy status, sub-healthy status, slightly deteriorated status, moderately deteriorated status, and severely deteriorated status. The magnitude of change, the rate of change, and the stage transition characteristics are combined and analyzed to identify the deterioration evolution trend that reflects the direction of the evolution of the unit's health status.

[0126] Risk quantification and classification are performed on the deterioration evolution trend, the unit's comprehensive health index, and the unit's assessment confidence level to form a corresponding risk warning level for the unit.

[0127] The specific process includes extracting the deterioration evolution trend (reflecting the direction of the evolution of the unit's health status), the unit comprehensive health index (characterizing the current health level of the unit), and the unit assessment confidence level (characterizing the credibility of the unit's health assessment) from the deterioration evolution trend, the unit comprehensive health index, and the unit assessment confidence level. Based on the risk level classification rules corresponding to healthy state, sub-healthy state, slightly deteriorated state, moderately deteriorated state, and severely deteriorated state in the health status discrimination framework, the deterioration evolution trend, the unit comprehensive health index, and the unit assessment confidence level are comprehensively mapped to obtain the risk quantification and classification results reflecting the degree of unit health risk.

[0128] Based on the risk quantification and grading results, the corresponding risk warning level for the unit is determined. The risk level corresponding to the healthy state in the risk quantification and grading results is mapped to the low risk warning level, the risk level corresponding to the sub-healthy state in the risk quantification and grading results is mapped to the attention warning level, the risk level corresponding to the slightly deteriorated state in the risk quantification and grading results is mapped to the general warning level, the risk level corresponding to the moderately deteriorated state in the risk quantification and grading results is mapped to the severe warning level, and the risk level corresponding to the severely deteriorated state in the risk quantification and grading results is mapped to the emergency warning level, thus forming the corresponding risk warning level for the unit.

[0129] Based on the risk warning level, the trend of deterioration and evolution, and the unit's comprehensive health index, the corresponding maintenance strategy for the unit is determined, and maintenance decision recommendations are formed.

[0130] The specific process includes extracting the risk warning level (reflecting the degree of health risk of the unit), the deterioration evolution trend (reflecting the direction of the evolution of the unit's health status), and the comprehensive health index (characterizing the current health level of the unit) from the risk warning level, the deterioration evolution trend, and the comprehensive health index of the unit. Based on the urgency level corresponding to the risk warning level, the rate of change corresponding to the deterioration evolution trend, and the degree of health deviation corresponding to the comprehensive health index of the unit, the process matches routine inspection strategies, condition monitoring strategies, preventive maintenance strategies, targeted maintenance strategies, and emergency shutdown maintenance strategies from the preset maintenance strategy library. The matched maintenance strategies are then integrated with the risk warning level, the deterioration evolution trend, and the comprehensive health index of the unit to form maintenance decision recommendations corresponding to the unit.

[0131] It should be noted that the maintenance strategy library is formed by classifying and storing routine inspection strategies, condition monitoring strategies, preventive maintenance strategies, targeted overhaul strategies, and emergency shutdown maintenance strategies according to the corresponding relationship between risk level, deterioration rate, and degree of health deviation, based on the technical specifications of the unit equipment, historical maintenance experience, and industry standards.

[0132] This embodiment also provides a computer device applicable to the PHM-based unit health assessment method, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the PHM-based unit health assessment method proposed in the above embodiment.

[0133] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0134] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the PHM-based unit health assessment method as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0135] In summary, this invention achieves the transformation of the unit health benchmark from a static benchmark to a dynamic benchmark that adapts to operating conditions by using a physical constraint meta-learning algorithm to adaptively update the health feature vector with a small number of samples. This improves the accuracy and stability of health assessment under complex operating conditions and refines the degradation state from a single anomaly judgment to a multi-dimensional deterioration characterization. Furthermore, by using a component-level health evidence set to mine the causal topological structure between components and calculate the evidence conflict coefficient, this invention transforms the unit health analysis from a static and independent component evidence processing method to an evidence collaborative analysis method with causal propagation relationships and conflict adjustment capabilities, thereby enhancing the robustness and credibility of component-level health evidence fusion.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A unit health assessment method based on PHM, characterized in that, include: Collect multi-source heterogeneous operating data of the unit, perform object mapping based on the hierarchical topology of the equipment, and obtain a component-level associated dataset; Multi-dimensional state features are extracted from the component-level associated dataset and then dimensionality-reduced and concatenated to obtain a health feature vector. The physical constraint meta-learning algorithm is used to adaptively update the health feature vector under few working conditions, generating a dynamic health baseline curve. The three-dimensional degradation degree is then calculated, and a mapping function is used to transform the degradation degree into a basic probability assignment, resulting in a component-level health evidence set. The specific steps are as follows: Extract a small number of operating condition features corresponding to the current operating state from the health feature vector, and combine them with preset physical constraints to input the physical constraint meta-learning algorithm to obtain adaptive update parameters under the current operating condition; The preset health baseline is corrected by adaptively updating parameters to form a dynamic health baseline curve that matches the current working conditions. Based on the dynamic health baseline curve and the corresponding health feature vector, the amplitude deviation, evolution rate deviation and physical consistency deviation are calculated to obtain the three-dimensional degradation degree; Based on the three-dimensional degradation degree, the state mapping range of each key component under the preset health status judgment framework is determined, and degradation degree state mapping data is formed. Based on the degradation status mapping data, the three-dimensional degradation of each key component is transformed into a basic probability allocation of the corresponding health status using a mapping function, forming component-level probability allocation data. The basic probability distribution of each health status corresponding to the same key component is collected using component-level probability allocation data to generate a component-level health evidence set. The causal topology between components is mined using a component-level health evidence set, and the evidence conflict coefficient is calculated. At the same time, the causal credibility weight of each piece of evidence is deduced, and an adaptive conflict adjustment factor is generated. The specific steps are as follows: Extract the health evidence evolution sequence of each key component in a continuous time period from the component-level health evidence set, and mine the causal relationship between each key component based on the health evidence evolution sequence to obtain the causal topology between components; The causal topology is used to calculate the evidence conflict coefficients corresponding to health evidence between adjacent key components. By using causal topology and evidence conflict coefficient, the credibility of health evidence for each key component in causal transmission is deduced, and the causal credibility weight corresponding to each piece of health evidence is obtained. The adaptive conflict adjustment factor is obtained by adaptively adjusting the evidence conflict coefficient using causal credibility weights. The component-level health evidence set is preprocessed using causal credibility weights and adaptive conflict adjustment factors, and the DS evidence synthesis rules are executed according to the equipment hierarchy to obtain the unit health trust distribution. The specific steps are as follows: Based on the component-level health evidence set, causal credibility weight, and adaptive conflict adjustment factor, the health evidence corresponding to each key component is weighted and corrected to form a preprocessed component-level health evidence set. Based on the preprocessed component-level health evidence set, the health evidence corresponding to the key components at the same level is grouped according to the equipment hierarchy, and the DS evidence synthesis rules are executed layer by layer to form a health trust distribution at each level. The health trust distributions at each level are recursively synthesized upwards along the equipment hierarchy to form the unit health trust distribution. Based on the unit health trust distribution, the computer unit comprehensive health index and unit assessment confidence level are calculated, and the deterioration evolution trend is identified based on the health index sequence over a continuous period of time, generating risk warning levels and maintenance decision recommendations.

2. The unit health assessment method based on PHM as described in claim 1, characterized in that, The specific steps for obtaining the component-level associated dataset are as follows: Collect multi-source heterogeneous operating data from the units to form a raw operating dataset, and construct a hierarchical topology relationship of the equipment based on the equipment object information, signal association information, and hierarchical affiliation information in the raw operating dataset; The original operational dataset is processed for timestamp alignment, unit unification, and quality verification to obtain a standardized operational dataset. Based on the equipment hierarchical topology, the standardized operational dataset is mapped to the corresponding measurement point nodes to form a measurement point-level mapping dataset. Based on the measurement point-level mapping dataset, the nodes are merged and associated according to the membership and connection relationships between the measurement point nodes and the key component nodes. Combined with the key component identifiers and topological association information, a component-level association dataset is generated.

3. The unit health assessment method based on PHM as described in claim 1, characterized in that, The specific steps for obtaining the health feature vector are as follows: Extract the time-domain state features, frequency-domain state features, topological coupling features, and control response features corresponding to each key component from the component-level associated dataset to form the original multidimensional feature set; The original multidimensional feature set is used to reduce the dimensionality of different types of state features, and then the features are spliced ​​and fused in a preset order to obtain a health feature vector.

4. The unit health assessment method based on PHM as described in claim 1, characterized in that, The specific steps for generating risk warning levels and maintenance decision recommendations are as follows: The overall health index of the unit is calculated using the unit health trust distribution within the framework of health status discrimination, and the corresponding unit assessment confidence level is calculated based on the concentration of quality allocation in the unit health trust distribution. Based on the unit's comprehensive health index and the unit's assessment confidence level, a health index sequence for continuous time periods is constructed, and the deterioration and evolution trend of the unit is identified based on the magnitude, rate of change, and stage transition characteristics of the health index sequence in adjacent time periods. The risk quantification and classification of the deterioration evolution trend, the unit's comprehensive health index and the unit's assessment confidence level are used to form the corresponding risk warning level for the unit. Based on the risk warning level, the trend of deterioration and evolution, and the unit's comprehensive health index, the corresponding maintenance strategy for the unit is determined, and maintenance decision recommendations are formed.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the PHM-based unit health assessment method according to any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the PHM-based unit health assessment method according to any one of claims 1 to 4.

Images