Method and apparatus for condition assessment of industrial equipment

Abstract

The application provides a kind of industrial equipment state evaluation method and device, it is related to the technical field of industrial equipment monitoring, the industrial equipment state evaluation method and device provided by the present application can collect the input vector and actual output vector of industrial equipment, and output the health reference output vector under the current working condition based on the health baseline vector calibrated in advance through the health baseline model;Further calculate the residual vector of health reference output vector and actual output vector, determine whether the industrial equipment is abnormal state, if it is abnormal state, determine the offset degree of health parameter relative to health baseline vector, and quantitatively calculate the contribution degree of health parameter to the current abnormal state, and then generate a state evaluation report;It can be realized in the same physical parameter space Abnormal state detection and attribution interpretation, and provide more direct basis for maintenance decision and operation optimization of industrial equipment, and effectively meet the state evaluation needs of industrial equipment.

Description

Technical Field

[0001] This invention relates to the technical field of industrial equipment monitoring, and in particular to a method and apparatus for assessing the condition of industrial equipment. Background Technology

[0002] With the continuous improvement of automation and digitalization in industrial processes, more and more key equipment is equipped with the capability for online data acquisition. For example, various physical quantities such as temperature, pressure, flow rate, vibration, current, voltage, power, speed, and efficiency can be continuously monitored. How to use this data to accurately assess the operating status of equipment, promptly identify degradation and anomalies, and provide a basis for maintenance decisions has become an important issue in the field of industrial intelligent operation and maintenance.

[0003] Traditional methods for assessing the condition of industrial equipment mainly include threshold-based monitoring methods, mechanism-based methods, and data-driven methods.

[0004] Threshold-based monitoring methods typically compare individual measurements such as vibration values, bearing temperatures, current amplitudes, and power deviations with empirical thresholds, triggering an alarm when the threshold is exceeded. This method is simple to implement and has low engineering deployment costs, but it inherently relies on judging exceedances of single variables or a small number of indicators, making it difficult to adapt to the multi-factor coupling changes under complex industrial conditions. In actual operation, equipment load fluctuations, changes in environmental conditions, control strategy switching, and sensor noise can all cause changes in measured values. Relying solely on exceeding the limit of a single indicator often fails to distinguish between true health degradation and normal operating condition fluctuations, easily leading to false alarms or missed alarms.

[0005] Mechanism-based methods attempt to establish a relationship model between inputs and outputs based on the physical laws of the equipment, identifying anomalies by comparing theoretical outputs with actual measurements. This type of method has good physical interpretability and can reveal, to some extent, the relationship between changes in internal equipment parameters and external performance. However, in industrial settings, equipment structures are complex, boundary conditions are unstable, and individual differences are significant, making it difficult for idealized mechanistic models to fully adapt to real-world situations. Furthermore, if the parameters in the model are not explicitly organized into a unified set of parameters that can be used for health description, even if the model can generate some predictive bias, it is still difficult to directly map abnormal phenomena into explainable changes in health factors.

[0006] Data-driven approaches use statistical learning, machine learning, or deep learning techniques to directly train state classifiers, anomaly detectors, or degradation prediction models from historical data. These methods can achieve good results when there is sufficient data, complete labels, and adequate coverage of operating conditions. However, industrial environments often suffer from scarce fault samples, inaccurate fault labels, significant shifts in operating conditions, and insufficient model interpretability. Especially for large, critical equipment, real-world faults occur infrequently, often lacking a sufficient number of comprehensive fault samples, which limits the generalization ability and reliability of purely data-driven methods.

[0007] Therefore, traditional industrial equipment condition assessment methods are insufficient to meet the condition assessment needs of industrial equipment. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a method and apparatus for assessing the condition of industrial equipment, so as to alleviate the above-mentioned technical problems.

[0009] In a first aspect, embodiments of the present invention provide a method for assessing the condition of industrial equipment. The method includes: during the online operation phase of the industrial equipment, collecting the input vector and actual output vector of the industrial equipment under the current operating condition; inputting the input vector into a pre-established health baseline model, and outputting a health reference output vector under the current operating condition based on a pre-calibrated health baseline vector through the health baseline model; wherein the health baseline vector includes multiple physical factors of the industrial equipment in a healthy state, the physical factors being parameters characterizing the healthy level of the industrial equipment; calculating the residual vector between the health reference output vector and the actual output vector; constructing a comprehensive anomaly index based on the residual vector; and when the comprehensive anomaly index... When the combined abnormality index exceeds a preset threshold, the industrial equipment is determined to be in an abnormal state. After determining that the industrial equipment is in an abnormal state, the health baseline model is expanded based on the health baseline vector to establish a mapping relationship between the residual vector and the parameter offset. The parameter offset is the offset between the actual parameter vector of the industrial equipment in the current abnormal state and the health baseline vector. The mapping relationship is solved by an optimization algorithm to obtain the degree of offset of each health parameter in the current abnormal state relative to the health baseline vector, and the contribution of each health parameter to the current abnormal state is quantitatively calculated. A state assessment report containing the abnormal state, the degree of offset, and the degree of contribution is generated.

[0010] In conjunction with the first aspect, the present invention provides a first possible implementation of the first aspect, wherein the method further includes: calculating the similarity between the residual vector and each abnormal pattern in a pre-constructed abnormal pattern library, determining the abnormal patterns with similarity higher than a preset similarity threshold as candidate abnormal patterns corresponding to the current abnormal state; performing a consistency comparison between the parameter offset and the expected parameter offset direction corresponding to the candidate abnormal pattern; and if the comparison result satisfies the consistency condition, determining the candidate abnormal pattern as the abnormal pattern corresponding to the current abnormal state.

[0011] In conjunction with the first aspect, this invention provides a second possible implementation of the first aspect, wherein the method further includes: acquiring historical data of the industrial equipment in a healthy operating state, and a health baseline vector in the healthy operating state; wherein the historical data includes a historical input vector and a historical output vector corresponding to the historical input vector; establishing a health baseline mapping relationship from the historical input vector to the historical output vector based on the historical data; wherein the health baseline mapping relationship is used to characterize a health reference output vector that the industrial equipment should present when maintaining the healthy operating state corresponding to the health baseline vector under the operating conditions of the historical input vector; and constructing a health baseline function of the health baseline model based on the health baseline mapping relationship.

[0012] In conjunction with the second possible implementation of the first aspect, this embodiment of the invention provides a third possible implementation of the first aspect, wherein the above method further includes: constructing an optimization objective function, wherein the optimization objective function is used to minimize the error between the historical output vector in the historical data and the health reference output vector output based on the health baseline function; and calibrating the health baseline vector based on the optimization objective function.

[0013] In conjunction with the first aspect, this embodiment of the invention provides a fourth possible implementation of the first aspect, wherein the step of calculating the residual vector between the health reference output vector and the actual output vector includes: calculating the difference between the health reference output vector and the actual output vector, and determining the difference as the residual vector.

[0014] In conjunction with the first aspect, this embodiment of the invention provides a fifth possible implementation of the first aspect, wherein the step of constructing a comprehensive anomaly index based on the residual vector includes: standardizing the residual vector to obtain a standardized residual vector; and constructing a comprehensive anomaly index based on the standardized residual vector.

[0015] In conjunction with the first aspect, this embodiment of the invention provides a sixth possible implementation of the first aspect, wherein the step of expanding the health baseline model based on the health baseline vector to establish a mapping relationship between the residual vector and the parameter offset, so as to map the residual vector to the parameter offset, includes: constructing expressions for the true parameter vector of the industrial equipment under abnormal conditions and expressions for the actual output vector based on the health baseline vector; performing Taylor expansion processing on the expression of the actual output vector based on the health baseline vector to obtain the Taylor expansion corresponding to the actual output vector, and determining the sensitivity matrix based on the Taylor expansion; and establishing a mapping relationship between the residual vector and the parameter offset based on the sensitivity matrix, so as to map the residual vector to the parameter offset.

[0016] In conjunction with the first aspect, this embodiment of the invention provides a seventh possible implementation of the first aspect, wherein the step of solving the mapping relationship by an optimization algorithm to obtain the degree of offset of each health parameter in the current abnormal state relative to the health baseline vector includes: constructing an optimization function for estimating the degree of offset based on the mapping relationship; solving the optimization function to obtain an offset vector; wherein each element in the offset vector represents the degree of offset of the corresponding health parameter.

[0017] In conjunction with the sixth possible implementation of the first aspect, this embodiment of the invention provides an eighth possible implementation of the first aspect, wherein the step of quantifying the contribution of each of the health parameters to the current abnormal state includes: for each health parameter, extracting the matrix component corresponding to the health parameter from the sensitivity matrix, and calculating the contribution based on the offset corresponding to the health parameter and the matrix component; the contribution is used to represent the contribution of the health parameter to the current residual vector.

[0018] Secondly, embodiments of the present invention also provide a condition assessment device for industrial equipment. The device includes: a data acquisition module, used to acquire the input vector and actual output vector of the industrial equipment under current operating conditions during the online operation phase of the industrial equipment; a prediction module, used to input the input vector into a pre-established health baseline model, and output a health reference output vector under the current operating conditions based on a pre-calibrated health baseline vector; wherein the health baseline vector includes multiple physical factors of the industrial equipment in a healthy state, and the physical factors are parameters characterizing the healthy level of the industrial equipment; a calculation module, used to calculate the residual vector between the health reference output vector and the actual output vector; and an anomaly module, used to construct a comprehensive anomaly index based on the residual vector. When the comprehensive anomaly index exceeds a preset threshold, the industrial equipment is determined to be in an abnormal state. A mapping module, after determining the industrial equipment to be in an abnormal state, expands the health baseline model based on the health baseline vector to establish a mapping relationship between the residual vector and the parameter offset; wherein the parameter offset is the offset between the actual parameter vector of the industrial equipment in the current abnormal state and the health baseline vector. A quantization module, through an optimization algorithm, solves the mapping relationship to obtain the degree of offset of each health parameter relative to the health baseline vector in the current abnormal state, and quantifies the contribution of each health parameter to the current abnormal state. A generation module, which generates a state assessment report containing the abnormal state, the degree of offset, and the degree of contribution.

[0019] The embodiments of the present invention bring the following beneficial effects: This invention provides a method and apparatus for assessing the condition of industrial equipment. It can collect the input vector and actual output vector of the industrial equipment during its online operation phase, and output a health reference output vector under the current operating condition based on a pre-calibrated health baseline vector using a health baseline model. Further, it calculates the residual vector between the health reference output vector and the actual output vector, constructs a comprehensive anomaly index to determine whether the industrial equipment is in an abnormal state. If it is in an abnormal state, it establishes a mapping relationship between the residual vector and parameter offsets to determine the degree of offset of each health parameter relative to the health baseline vector under the current abnormal state, and quantifies the contribution of each health parameter to the current abnormal state, generating a condition assessment report containing the abnormal state, the degree of offset, and the degree of contribution. Since the detection of the abnormal state and the mapping relationship between the residual vector and parameter offsets are both constructed based on the health baseline vector, it is possible to realize the detection of the abnormal state and the explanation of the cause of the abnormal state in the same physical parameter space, forming a closed-loop assessment with physical consistency. This helps improve the engineering interpretability of the condition assessment and provides a more direct basis for the maintenance decision-making, maintenance priority ranking, spare parts preparation, and operation optimization of industrial equipment, thereby effectively meeting the condition assessment needs of industrial equipment.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 A flowchart of a condition assessment method for industrial equipment provided in an embodiment of the present invention; Figure 2 A comparative schematic diagram of average triggering time provided for an embodiment of the present invention; Figure 3 A comparative schematic diagram of average detection rate provided in an embodiment of the present invention; Figure 4This is a schematic diagram of the structure of a condition assessment device for industrial equipment provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] To mitigate the problems of false alarms, difficulty in interpreting abnormal states, and limited generalization ability and reliability inherent in traditional industrial equipment condition assessment methods, a class of methods based on multivariate statistical process control, such as principal component analysis, partial least squares, and contribution map analysis, are widely used in the existing field of industrial condition monitoring. These methods typically first establish a statistical model in the space of observed variables using healthy samples, and then detect and locate anomalies based on residuals, principal component projections, or contribution maps. While these methods have been widely applied in engineering, their main problem lies in the fact that their contribution analysis usually occurs at the level of observed variables, yielding results on "which measured variable contributes more to the statistic," rather than "which physical health parameter has changed." In other words, while these methods can indicate anomalies in certain measured quantities, they often cannot directly provide physically meaningful changes in health parameters such as decreased blade aerodynamic performance, reduced transmission efficiency, increased friction loss, and increased motor losses; maintenance personnel still need to perform secondary inferences.

[0026] Furthermore, a common approach in existing technologies involves first constructing a health baseline model to generate predictive output, then detecting anomalies based on prediction deviations, and subsequently introducing a separate diagnostic or explanatory model to analyze the causes of the anomalies. While this approach includes both "detection" and "explanation," they are often two independent methods linked sequentially: the first part generates residuals in the observation space, while the second part interprets them in another feature space or empirical rule space, lacking a shared physical parameter benchmark. Due to the lack of a unified parameter anchor, there is often a logical disconnect between the detection results and the attribution results; that is, the "anomaly" defined in the previous step and the "cause" explained in the subsequent step may not originate from the same model or the same set of parameters, resulting in insufficient physical consistency in the overall diagnostic chain.

[0027] Especially in complex industrial equipment, if the health references used for anomaly detection differ from the parameter systems used for attribution analysis, the following problems may arise: First, detection and attribution are independent of each other, making it difficult to form a self-consistent closed loop; second, attribution results rely on empirical mapping or posterior interpretation, lacking rigorous model support; third, even if a so-called "contribution" is given, it is often only a statistical variable contribution, rather than a physical health parameter that corresponds to the object being inspected; fourth, when there are few fault samples, relying solely on data-driven classification or empirical rules to infer causes has weak reliability.

[0028] Therefore, existing methods for assessing the condition of industrial equipment often have the following shortcomings: First, there is a lack of a unified parameter system that can simultaneously support the generation of health references and the attribution of abnormal causes; Second, anomaly detection and cause attribution usually rely on two separate methods, which lack physical consistency. Third, the contribution analysis of existing multivariate statistical methods mostly stays at the level of observed variables, making it difficult to directly correspond to the internal health parameters of the equipment; Fourth, many methods rely heavily on fault samples, making them unsuitable for situations where fault samples are scarce in real industrial settings.

[0029] Based on the above problems, there is an urgent need to propose a new method for assessing the condition of industrial equipment, so that the health baseline modeling and anomaly cause attribution can be carried out around the same set of physically meaningful health parameters, thereby achieving a physical consistency closed loop between anomaly detection, condition identification and cause attribution without relying on a large number of fault samples.

[0030] Based on this, the present invention provides a method and apparatus for assessing the condition of industrial equipment, in order to solve the problems in existing condition assessment methods such as the separation between health reference generation and attribution of abnormal causes, the lack of physical meaning in the attribution results, and the strong dependence on fault samples.

[0031] To facilitate understanding of this embodiment, a method for assessing the condition of industrial equipment disclosed in this embodiment of the invention will first be described in detail.

[0032] In one possible implementation, this invention provides a method for assessing the condition of industrial equipment. Specifically, the method does not simply involve stacking multiple steps such as health modeling, residual analysis, pattern recognition, and attribution analysis, but rather establishes a unified anchoring mechanism for physical parameters. Specifically, in this invention, the health baseline vector in the health baseline model of the industrial equipment can be used simultaneously in two different but related stages: on the one hand, as a parameter benchmark for the health baseline model, used to generate a health reference output for the industrial equipment under a given operating condition; on the other hand, as a local expansion base point during anomaly attribution analysis, used to establish a sensitivity mapping relationship between residuals and changes in health parameters. Through this dual utilization, anomaly detection and causal attribution are unified into the same physical parameter space, thus ensuring that the "detected anomaly" and the "explanation of the cause of the anomaly" are based on the same health definition and the same parameter system, forming a physically consistent closed loop.

[0033] Furthermore, another objective of this invention is to provide a condition assessment method that can be implemented without relying on a large number of fault samples. Since this invention uses a health state parameter system as the modeling basis, it focuses on characterizing the relationship between the input and output of industrial equipment in a healthy state, and its local response to changes in health parameters. Therefore, even under conditions of scarce fault samples, anomaly detection and cause attribution can still be achieved through health baselines and sensitivity analysis.

[0034] This invention also aims to improve the engineering interpretability of condition assessment results, moving beyond simply stating "some variables are abnormal" or "a certain type of statistical indicator exceeds limits." Instead, it provides a more quantifiable calculation of the contribution of each health parameter to the current abnormal state, indicating which health parameter deviates from the healthy state, the degree of deviation, and its contribution to the current anomaly. This provides a more direct basis for industrial equipment maintenance decisions, maintenance prioritization, spare parts preparation, and operational optimization.

[0035] Specifically, such as Figure 1 The flowchart shown illustrates a condition assessment method for industrial equipment, which includes the following steps: Step S102: During the online operation phase of the industrial equipment, collect the input vector and actual output vector of the industrial equipment under the current operating conditions. Step S104: Input the input vector into the pre-established health baseline model, and output the health reference output vector under the current working condition based on the pre-calibrated health baseline vector through the health baseline model; In this embodiment of the invention, the health baseline vector includes multiple physical factors of the industrial equipment in a healthy state, and the physical factors are parameters used to characterize the industrial equipment as being in a healthy state. Step S106: Calculate the residual vector between the health reference output vector and the actual output vector; Step S108: Construct a comprehensive anomaly index based on the residual vector. When the comprehensive anomaly index is greater than a preset threshold, determine that the industrial equipment is in an abnormal state. Step S110: After determining that the industrial equipment is in an abnormal state, expand the health baseline model based on the health baseline vector to establish the mapping relationship between the residual vector and the parameter offset. Wherein, the parameter offset is the offset between the actual parameter vector of the industrial equipment under the current abnormal state and the healthy baseline vector; Step S112: Solve the mapping relationship through optimization algorithm to obtain the offset of each health parameter in the current abnormal state relative to the health baseline vector, and quantify the contribution of each health parameter to the current abnormal state. Step S114: Generate a status assessment report that includes the abnormal state, the degree of deviation, and the degree of contribution.

[0036] In practical use, in order to realize the state assessment method of industrial equipment in the embodiments of the present invention, the present invention usually predefines a set of health parameters with physical meaning, and constructs a health baseline vector based on the set of health parameters. The health baseline vector can be used as a reference baseline for abnormal states, or as the expansion base point for the reverse analysis of abnormal attribution, so that the residual vector and the attribution analysis process can share the same physical parameter system.

[0037] In practice, this physical parameter system is a shared health parameter system for industrial equipment. Under this system, for the industrial equipment to be evaluated, its input vector and actual output vector in the online operating state can be defined; that is, the input vector and actual output vector collected in step S102 above. In this embodiment of the invention, the input vector represents the operating condition input at a certain moment, and the actual output vector represents the measurement output at that moment; for example, the input vector can be represented as:

[0038] in, Indicates time The operating condition input, i.e., the input vector in this embodiment of the invention, may include operating condition parameters such as load, ambient temperature, flow rate, control setpoint, and boundary conditions, where m represents the number of operating condition parameters; the corresponding actual output vector can be expressed as:

[0039] in, The actual output vector represents the measurement output at time t. The output variables included in this actual output vector may include measurable parameters such as temperature, pressure, power, current, rotational speed, vibration, differential pressure, flow rate, and efficiency at preset locations on the industrial equipment; n represents the number of parameters measured; and, in this embodiment of the invention, the actual output vector represents the physical quantity measured under the above-mentioned working conditions.

[0040] Further define the health parameter vector as follows:

[0041] in, For the first time t One health parameter, p The number of health parameters is used to represent physical factors related to the health status of industrial equipment in this embodiment of the invention. These physical factors may be efficiency coefficient, drag coefficient, loss coefficient, friction coefficient, heat transfer coefficient, aerodynamic performance coefficient, electromagnetic loss parameter, stiffness parameter, or other factors that can reflect the health level of industrial equipment.

[0042] Furthermore, the health parameter vector determined when the industrial equipment is in a healthy state is the aforementioned health baseline vector in this embodiment of the invention. In this embodiment of the invention, the health baseline vector is represented as:

[0043] The key point of this invention lies in the health baseline vector. It is not merely used as a static parameter, but as a unified physical reference point in the entire state assessment process, and is used simultaneously for forward health reference generation and reverse attribution analysis.

[0044] In practical applications, to achieve condition assessment of industrial equipment, a health baseline model needs to be pre-established in this embodiment of the invention. Specifically, the health baseline model in this embodiment of the invention is used to characterize the aforementioned input vector. To the actual output vector The invention establishes a health baseline mapping relationship based on historical data of industrial equipment in a healthy operating state. Specifically, in this embodiment, historical data of industrial equipment in a healthy operating state and a health baseline vector in the healthy operating state can be obtained. The historical data includes historical input vectors and historical output vectors corresponding to the historical input vectors. A health baseline mapping relationship from historical input vectors to historical output vectors is established based on the historical data. This health baseline mapping relationship is used to characterize the health reference output vector that the industrial equipment should present when maintaining the healthy operating state corresponding to the health baseline vector under the operating conditions of the historical input vector. Then, a health baseline function of the health baseline model is constructed based on this health baseline mapping relationship.

[0045] In practical implementation, historical data on the healthy operating state of industrial equipment under normal and fault-free conditions can be collected, and data pairs including historical input vectors and historical output vectors can be established. For example, taking motor operation as an example, the historical input vector can include operating condition data such as speed, load, and ambient temperature, while the historical output vector can include measurement data such as amplitude, motor temperature, and current response. These data pairs can be collected to form a healthy sample dataset. Then, the health baseline model in this embodiment of the invention is constructed, that is, a functional relationship between the historical input vector and the historical output vector is established, which can be expressed by the formula:

[0046] Where, f( ) represents the health baseline function corresponding to the health baseline model, y Let (t) be the output vector that the industrial equipment should present under the current operating condition u(t) and maintain its current healthy state; that is, the historical output vector corresponding to the health baseline vector. The health baseline function corresponding to the health baseline model can adopt a mechanistic model, a semi-mechanistic model, a data-driven model with physical constraints, or a fusion model of mechanistic and data aspects. The basic requirement is that the physical meaning of the health parameters be explicitly preserved in the model, so that the health state can be represented by the health baseline vector. 0 is used for a unified expression.

[0047] Furthermore, based on the health baseline function corresponding to the above-mentioned health baseline model, an optimization objective function can be further constructed in this embodiment of the invention. The optimization objective function is used to minimize the error between the historical output vector in the historical data and the health reference output vector output based on the health baseline function; then the health baseline vector is calibrated based on the optimization objective function.

[0048] That is, the health baseline vector in the embodiments of the present invention can be calibrated by the following expression:

[0049] in, denoted as the number of samples in the health sample dataset constructed based on historical data; k represents the k-th sample, i.e., the data pair containing the aforementioned historical input vector and historical output vector; W is the weight matrix, used to balance the scale and importance of different output variables. This represents a health parameter vector; through the above calibration process, a baseline parameter vector reflecting the individual health status of the industrial equipment can be obtained, namely, the health baseline vector in this embodiment of the invention. .

[0050] Furthermore, based on the constructed health baseline model and the calibrated health baseline vector, the state assessment method for industrial equipment in this embodiment of the invention can be executed.

[0051] In practice, steps S104 to S108 above constitute anomaly detection based on the health baseline vector. Specifically, step S102 is a real-time acquisition process, that is, the input vector is acquired in real time during the online operation phase of the industrial equipment. and actual output vector Furthermore, based on the health baseline model, the health reference output vector in this embodiment of the invention can be expressed as:

[0052] When constructing the residual vector, the difference between the healthy reference output vector and the actual output vector can be calculated, and this difference can be determined as the residual vector. Specifically, the formula for the residual vector can be expressed as:

[0053] That is, based on the above expression for the health reference output vector, the residual vector can also be expressed as:

[0054] Specifically, this residual vector This characterizes the vector of industrial equipment relative to a healthy baseline under current operating conditions. The degree of deviation is reflected in the health baseline vector, which serves to provide a center for health parameters. Since the health baseline model has absorbed the normal impact of operating condition changes on the output, the residual vector no longer mainly reflects load changes or environmental changes, but mainly reflects the abnormal performance caused by the deviation of the health status from the center of health parameters.

[0055] Furthermore, to facilitate unified comparison between different variables, the residual vector can also be standardized in this embodiment of the invention. That is, the residual vector is standardized to obtain a standardized residual vector; and then an anomaly comprehensive index is constructed based on the standardized residual vector.

[0056] Specifically, assume that the mean of the residuals in the healthy state is The covariance matrix is The standardized residual is: ; The comprehensive anomaly index is further constructed as follows: When comprehensive abnormal indicators When the value exceeds a preset threshold, it can be determined that the state of the industrial equipment has deviated from the health center. The corresponding health status is that the industrial equipment is in an abnormal or degrading state.

[0057] Furthermore, the processes S110 and S112 described above are processes for sensitivity expansion and anomaly attribution analysis based on the health baseline vector. The difference between this embodiment of the invention and the aforementioned conventional techniques lies in the fact that, after identifying an abnormal state, this embodiment does not require configuring a separate interpretation model. Instead, it continues to locally expand the health baseline model around the same health baseline vector, thereby directly mapping the residual vector to the offset of the health parameters.

[0058] Specifically, in this embodiment of the invention, expressions for the true parameter vector and the actual output vector of industrial equipment under abnormal conditions can be constructed based on the healthy baseline vector; then, the expression for the actual output vector is subjected to Taylor expansion based on the healthy baseline vector to obtain the Taylor expansion corresponding to the actual output vector, and the sensitivity matrix is ​​determined based on the Taylor expansion; a mapping relationship between the residual vector and the parameter offset is established based on the sensitivity matrix to map the residual vector to the parameter offset.

[0059] In specific implementation, the true parameter vector in this embodiment of the invention is the true health parameter vector of industrial equipment under abnormal conditions, which can be expressed as a parameter offset relative to the above-mentioned health baseline vector. Its formula expression can be expressed as: ;in, For the healthy baseline vector, Represents the vector relative to the health baseline. The parameter offset. At this point, the actual output vector can be expressed as: in, To measure noise and unmodeled disturbances.

[0060] Furthermore, in this embodiment of the invention, during the Taylor expansion process, the expression of the actual output vector can be modified... Performing a first-order Taylor expansion nearby, we obtain:

[0061] in, , is the sensitivity matrix in the embodiments of the present invention. because

[0062] Therefore:

[0063] This formula represents the mapping relationship between the residual vector and the parameter offset established based on the sensitivity matrix in this embodiment of the invention, and it indicates that the residual vector of the current anomaly... This is not an isolated observational bias, but can be interpreted as a bias relative to the center of health parameters. Health parameter offset The resulting output response. Due to the sensitivity matrix Also in the same The result is obtained by expanding the data at this point. Therefore, the residual generation and the causal attribution at this point share the same physical parameter anchor point, forming a unified physical explanation chain.

[0064] Based on the above mapping relationship, in this embodiment of the invention, an optimization function for estimating the degree of offset can also be constructed based on the mapping relationship; the offset vector is obtained by solving the optimization function; wherein, each element in the offset vector at this time represents the degree of offset of the corresponding health parameter.

[0065] In practice, the degree of parameter shift can be estimated using weighted regularized least squares, and the optimization function can then be expressed as:

[0066] in, The residual noise covariance matrix is... This is the regularization coefficient.

[0067] The analytical solution to this optimization function is obtained, and is expressed as:

[0068] The analytical solution is the offset vector in this embodiment of the invention, where each element represents an estimate of the degree of offset for each health parameter. Due to each health parameter Each corresponds to a health factor with a clear physical meaning, therefore This can be directly interpreted as the deviation of the corresponding health factor from the state of health.

[0069] Furthermore, in this embodiment of the invention, the contribution of each health factor to the current abnormal state can be further quantified based on the degree of offset. Specifically, for each health parameter, the matrix component corresponding to the health parameter can be extracted from the above sensitivity matrix, and the contribution degree can be calculated based on the offset degree corresponding to the health parameter and the matrix component; this contribution degree is used to represent the contribution of the health parameter to the current residual vector.

[0070] Specifically, in this embodiment of the invention, the first is defined as follows: The expression for the contribution of each health parameter is:

[0071] in, Represents the sensitivity matrix The j-th column; l Indicates the first lOne health parameter; Contribution level in the embodiments of the present invention This directly represents the proportion of the current residual explained by the j-th health parameter. Unlike traditional contribution plot methods, this embodiment of the invention does not determine which measured variable contributes more to the outlier statistic, but rather which physically significant health parameter contributes more to the performance of the current anomalous state.

[0072] Furthermore, in this embodiment of the invention, the process of identifying abnormal states is also included. That is, after determining the abnormal state, the similarity between the residual vector and each abnormal pattern in the pre-built abnormal pattern library can be calculated, and the abnormal patterns with similarity higher than a preset similarity threshold are determined as candidate abnormal patterns corresponding to the current abnormal state. Then, the parameter offset is compared with the expected parameter offset direction corresponding to the candidate abnormal pattern. If the comparison result meets the consistency condition, the candidate abnormal pattern at this time is determined as the abnormal pattern corresponding to the current abnormal state.

[0073] That is, the consistency identification process can further enhance the analysis of abnormal states, that is, further determine the specific abnormal pattern of the abnormal state at this time. In other words, in this embodiment of the invention, the abnormal state identification process is not carried out in isolation in the residual space, but preferably establishes a consistency mapping with the parameter offset estimation results.

[0074] Specifically, in this embodiment of the invention, several abnormal patterns can be predefined. For example, data or parameter offsets with known abnormal states can be extracted from historical data, and templates for the corresponding abnormal patterns can be established in the residual space. The similarity between the current residual vector and the template is calculated. Specifically, the standardized residual vector corresponding to the original residual vector can be used for the calculation. The similarity formula can be expressed as:

[0075] in, , indicating the number of exception patterns.

[0076] Furthermore, abnormal patterns with similarity scores higher than a preset similarity threshold can be identified as candidate abnormal patterns, and then consistency conditions can be determined. Specifically, based on a shared health parameter system, the templates of each abnormal pattern can be mapped to expected parameter offset directions. For example, for the c-th type of abnormal pattern, its expected parameter offset direction can be expressed as: ;in, Let be the generalized inverse of the aforementioned sensitivity matrix; then, compare the consistency between the offset vector obtained from real-time estimation and the expected parameter offset direction. For example, a parameter consistency index can be defined:

[0077] If the consistency index of this parameter is higher than the preset index threshold, it can be determined that the consistency condition is met, and then the candidate abnormal mode is determined as the abnormal mode corresponding to the current abnormal state.

[0078] The above process combines the residual vector calculated during anomaly pattern identification with the similarity index of each anomaly pattern in the anomaly pattern library and the parameter consistency index in attribution analysis to determine the category of the anomaly state, thus identifying the specific anomaly pattern. Only when a certain type of anomaly pattern exhibits high consistency in both the residual space and parameter space is the current anomaly state preferentially determined to belong to that type of anomaly pattern. In this way, state identification and cause attribution are no longer two unrelated tools, but rather form a mutually verifying and supporting relationship within a shared health parameter system.

[0079] Furthermore, the status assessment report generated in step S114 typically includes multiple components, such as the aforementioned abnormal status, degree of deviation, and degree of contribution. The abnormal status can further indicate whether the current industrial equipment deviates from its healthy state, the degree of deviation, and the abnormal pattern or level, etc. In addition, it can also include a comprehensive abnormal index or the statistical value of the comprehensive abnormal index within its time window; and the consistency condition determination process in the aforementioned abnormal pattern identification process, and finally output cause ranking information and maintenance recommendations for maintenance.

[0080] Furthermore, in this embodiment of the invention, the aforementioned comprehensive anomaly index, parameter offset, and contribution level can also be smoothed using a sliding window. For example, taking the degree of sharing as an example, the expression for the smoothing process can be expressed as:

[0081] in, C represents the window length. j ( k ) represents the k-th time, the first The stability of the results can be improved by smoothing the original attribution contribution of each health factor through this sliding window process.

[0082] In summary, the condition assessment method for industrial equipment provided by the embodiments of the present invention has the following technical effects: (1) The embodiments of the present invention achieve physical consistency between health reference generation and abnormal attribution by sharing a health parameter system.

[0083] While many existing methods possess both anomaly detection and causal analysis capabilities, they often rely on different models, feature spaces, or empirical rules, lacking a unified physical anchor. In this invention, the health baseline vector is used simultaneously for forward prediction of the health baseline and backward expansion of attribution sensitivity, ensuring complete consistency between the health definition upon which anomaly detection is based and the parameter benchmark upon which anomaly attribution is based. This establishes a consistent chain from health reference and residual generation to parameter offset interpretation, avoiding the fragmented problem of "one set of detection methods, one set of interpretation methods" in existing technologies.

[0084] (2) The embodiments of the present invention can achieve anomaly detection and cause attribution without relying on a large number of fault samples.

[0085] Since the core modeling object of this invention is the healthy baseline vector and its local neighborhood, rather than relying on a large number of fault samples to train the fault classification model, it is particularly suitable for application scenarios where fault samples are scarce in industrial sites. As long as data on the healthy operating status of industrial equipment can be obtained and a baseline model containing physical health parameters can be established, status assessment can be achieved by offsetting relative to the healthy baseline vector.

[0086] (3) The attribution results analysis of the embodiments of the present invention directly corresponds to physical health factors, rather than just the statistical contribution at the level of observed variables.

[0087] Traditional multivariate statistical contribution analysis typically answers "which measurement contributes more to the outlier statistic," but it doesn't directly indicate which specific industrial equipment's internal health factor is causing the problem. This invention, through a relational formula... The residual vector is directly mapped to the offset of health parameters, and the contribution of each health parameter is further calculated. Therefore, the output of this embodiment is the offset and contribution of physical health factors, such as aerodynamic performance degradation, transmission efficiency reduction, friction loss increase, or motor loss increase, which is easier for maintenance personnel to understand and use directly.

[0088] (4) The embodiments of the present invention enable mutual verification between state identification and cause attribution, thereby improving the reliability of the evaluation.

[0089] This invention preferably combines residual pattern recognition with parameter offset consistency analysis, so that the state identification result depends not only on similar patterns in the residual space, but also on the consistency constraint of the expected offset direction in the parameter space. This reduces the misjudgment that may be caused by relying solely on the observation pattern, and makes the state category determination and physical cause attribution mutually supportive, thereby improving the reliability and credibility of the evaluation results.

[0090] (5) The embodiments of the present invention have strong engineering interpretability and implementability.

[0091] Because the embodiments of this invention revolve around a set of physically meaningful health parameters, their output is closer to the information format required for maintenance decisions. Compared to simply outputting an abnormal score, a statistical measure exceeding limits, or a black-box category label, the embodiments of this invention can provide clear health factor shifts, contribution rankings, and possible causes, facilitating the development of targeted maintenance plans. Furthermore, the embodiments of this invention are applicable to both mechanistic models and semi-mechanistic or physically constrained data-driven models, thus exhibiting good engineering adaptability.

[0092] (6) The embodiments of the present invention have good versatility and scalability.

[0093] This invention is not limited to any specific type of equipment. As long as the industrial equipment can define the relationship between the input vector, output vector, and health parameters, the method of this invention can be used for condition assessment. Therefore, this invention is applicable to cooling tower ventilation equipment, fans, motors, reducers, pumps, compressors, heat exchangers, and other key industrial equipment. Different equipment only needs to define corresponding health parameter systems and health baseline models according to the characteristics of the object to share the core ideas and methodological framework of this invention.

[0094] (7) In the scenario of cooling tower ventilation equipment, the embodiments of the present invention can effectively distinguish between a variety of anomalies with similar external manifestations but different internal causes.

[0095] For example, during cooling tower operation, increased power output may stem from blade fouling, increased transmission losses, increased motor losses, or abnormal localized mechanical friction. Traditional methods often only detect "abnormal power output" or "abnormality at a certain measuring point," but struggle to directly distinguish the specific cause. In this embodiment of the invention, through residual generation and attribution analysis under a shared health parameter system, these anomalies with similar external manifestations but different internal mechanisms can be differentiated into shifts in different health factors, thereby improving the accuracy of operation and maintenance decisions.

[0096] To further facilitate understanding, the following example of a main transformer at an onshore substation will be used to illustrate the condition assessment method for industrial equipment provided in this embodiment of the invention. The main transformer is equipped with a main control unit and a high-voltage side control unit, which can continuously acquire telemetry data such as temperature, voltage, current, and tap changers during the operation of the main transformer.

[0097] In specific implementation, this embodiment of the invention uses minute-level online monitoring data, spanning approximately 46.9 hours, and obtains a total of 2815 valid sampling points. The collected data covers typical operating conditions of the main transformer equipment under different load levels, voltage states, and voltage regulation methods, and can reflect the coupling change law between the thermal and electrical states of the main transformer equipment.

[0098] Similar to wind turbines, main transformer equipment also features multiple inputs, multiple outputs, strong operating condition coupling, and scarce fault samples. On the one hand, oil temperature, winding temperature, voltage, current, and imbalance-related quantities fluctuate significantly with changes in load, voltage level, tap position, and system operating mode. On the other hand, different anomalies such as increased load loss, excitation state deviation, aggravated three-phase imbalance, and measurement chain deviation may all manifest as abnormal temperature rise or electrical quantity deviation in external observation, making it difficult for traditional monitoring methods based on a single threshold to directly distinguish the source of the anomaly.

[0099] Therefore, this embodiment of the invention applies the condition assessment method of industrial equipment to the condition assessment process of main transformer equipment, so that the health baseline generation and anomaly attribution analysis revolve around the same health parameter center, thereby demonstrating that this embodiment of the invention is not limited to wind turbine units, but can be extended to other key industrial equipment such as main transformer equipment.

[0100] Specifically, the entire condition assessment process may include the following steps: (1) Define the input vector; For the main transformer equipment, the input vector at time t under the current operating condition can be defined as:

[0101] in, Indicates active power; Indicates apparent power; This represents the average three-phase current. Indicates positive sequence voltage; This represents the average voltage of the three phases; The sliding window statistic representing the negative sequence voltage is used to characterize the level of imbalance in the current time period. The sliding window statistic representing the zero-sequence current is used to characterize the zero-sequence offset condition level during the current time period. This indicates the tap position or related operating parameters of the tap.

[0102] In this embodiment of the invention, the sliding statistics of negative-sequence voltage and zero-sequence current are included in the input vector to reflect the operating condition background; while their current observation values ​​are included in the output vector to reflect the actual response of the device or system under that operating condition. This setting avoids confusion between input and output roles and enhances the model's ability to express the thermo-electric coupling characteristics.

[0103] In specific implementations, to better reflect the gradual heating process and continuous operating conditions of the main transformer equipment, the moving average, combined, normalized, historical lag, or grouped statistical quantities of the above-mentioned quantities can be further introduced as extended input features. The specific settings can be made according to the actual usage, and the embodiments of the present invention do not impose any restrictions on this.

[0104] (2) Define the actual input vector; In this embodiment of the invention, the actual output vector of the online monitoring of the main transformer equipment is defined as:

[0105] in, Indicates oil temperature 1; Indicates oil temperature 2; Indicates the winding temperature; This represents the current negative sequence voltage observation value; This represents the zero-sequence current observation value at the current moment.

[0106] In this embodiment of the invention, both thermal state quantities and electrical imbalance quantities are selected as outputs to enable the unified health parameter space to characterize both the thermal degradation process of the main transformer equipment and the electrical state deviation process, thereby enhancing the physical interpretability and engineering maintainability of the attribution results.

[0107] (3) Definition of health parameter vector; In this embodiment of the invention, the health parameter vector of the main transformer is defined as follows:

[0108] in: It is an equivalent load loss health parameter used to characterize the changes in comprehensive load loss caused by increased copper loss, increased additional loss, or shift in heat loss. These are equivalent excitation state health parameters used to characterize the excitation operating point, voltage-side magnetic flux state, and related magnetization characteristic shifts. It is an equivalent unbalance-sensitive health parameter used to characterize the degree of response deviation of the main transformer to negative-sequence, zero-sequence, and three-phase unbalanced operating conditions; These are tap changer response health parameters used to characterize the tap changer action, tap position response characteristics, and the impact of changes in the voltage regulation link state on the output characteristics. The measurement chain offset health parameter is used to uniformly characterize the proportional offset, zero offset, or systematic drift of some measurement channels.

[0109] When the main transformer is in a healthy state, its corresponding health parameter center, i.e., the health baseline vector in this embodiment of the invention, is denoted as:

[0110] in, It serves both as the parameter center during the generation of the health reference output and as the local expansion center in the anomaly attribution analysis, thereby ensuring that state detection and cause attribution are based on the same physical reference point. In other words, in this embodiment of the invention, the health state is not defined solely by whether the observed measurements exceed limits, but rather by the central position of the main transformer equipment within a unified health parameter space. Definition. The essence of a main transformer equipment anomaly is the deviation of the current health parameters relative to the center.

[0111] (4) Construction of a health baseline model; When the main transformer is considered to be operating in a healthy or near-healthy state, historical data is collected to construct a health sample dataset. The sample data in the health sample dataset is represented as follows:

[0112] in, The number of samples in the healthy sample dataset. Using the healthy sample dataset, establish a health baseline model for the main transformer equipment:

[0113] in, Indicates the current operating condition Below, from the Health Parameter Center The generated health reference output.

[0114] In this embodiment of the invention, the health baseline model preferably adopts a modeling approach of "grouped operating condition features + unified parameter mapping". Specifically, the input features are organized into the following four groups: (a) Load loss group, used to describe the effects of operating conditions related to heat loss, such as active power, apparent power, and current; (b) Voltage excitation group, used to describe the working conditions related to the excitation state, such as positive sequence voltage and average voltage; (c) Unbalance group, used to describe the unbalanced operating conditions characterized by negative sequence voltage statistics and zero sequence current statistics; (d) Tap regulation group, used to describe the tap position and its effect on voltage regulation behavior.

[0115] In one implementation, the health baseline model in this embodiment of the invention can be represented as:

[0116] in: This is a working condition mapping matrix composed of the characteristics of each set of working conditions; This is the baseline bias term related to the current operating conditions.

[0117] Furthermore, the center of health parameters can be estimated using ridge regression or other regularized parameter estimation methods. Calibration is performed, that is:

[0118] in, This is the regularization coefficient.

[0119] Based on the current dataset, the health baseline model is established, and the coefficients of determination for fitting oil temperature 1, oil temperature 2, and winding temperature are determined. The values ​​reached 0.9207, 0.9169, and 0.9197 respectively, indicating that at the unified health parameter center... The model is now able to describe the normal thermo-electric coupling relationship of the main transformer equipment under different operating conditions.

[0120] It should be noted that the health baseline model in this embodiment of the invention is not limited to the linear or quasi-linear form described above, and may also employ a nonlinear regression model with physical constraints, a kernel mapping model, a neural network model, or a mechanism-data fusion model. This is as long as the model can explicitly or implicitly use a unified health parameter vector. Characterize the health status of equipment and center the health parameters Simultaneously serving health reference generation and abnormal attribution analysis, both fall within the protection scope of this invention's embodiments.

[0121] (5) Online residual vector generation and anomaly detection; During the online operation of the main transformer equipment, the input vector under the current operating conditions is collected in real time. and actual output vector Substituting into the health baseline model, we obtain the health reference output vector under the same working conditions:

[0122] Further construct the residual vector:

[0123] To eliminate dimensional differences and variations in health fluctuation amplitudes among different output quantities, the residuals are standardized based on the residual statistics of the health sample. Let the mean of the health sample residuals be... The covariance matrix is Then the standardized residual can be expressed as:

[0124] Based on this, a comprehensive anomaly index is constructed:

[0125] Simultaneously, a sliding window method is used to construct the average anomaly index:

[0126] in, The sliding window length is used to calculate the average anomaly index. It can reduce misjudgments caused by instantaneous fluctuations, short-term measurement spikes, and occasional disturbances.

[0127] In this embodiment of the invention, a preset threshold can be determined based on the statistical results of the healthy sample dataset. This preset threshold is also called the anomaly threshold, and is expressed as follows: For example, in this embodiment of the invention, the anomaly threshold is set to 4.80 based on the statistical results of the healthy sample dataset. It should be noted that this anomaly threshold is only an example value calculated based on specific data in this embodiment of the invention. In actual engineering applications, it can be adaptively set according to the device type, output dimension, sample distribution, expected false alarm rate, or empirical rules, and is not limited to this value.

[0128] when Persistently above the abnormal threshold At that time, it is determined that the current state has deviated from the center of health parameters. An abnormal state is defined as the presence of abnormalities or a tendency to degenerate.

[0129] Based on the actual data calculation results, multiple continuous deviation windows can also be identified during the full-time monitoring process. For example, during the monitoring process, three windows with high deviation were recorded, corresponding to: (1) 10:24 on August 27, 2025 to 10:54 on August 27, 2025; (2) 02:50 on August 29, 2025 to 03:20 on August 29, 2025; (3) 10:55 on August 27, 2025 to 11:25 on August 27, 2025.

[0130] The above results show that the embodiments of the present invention can continuously identify the state deviation of the main transformer equipment under complex operating conditions, rather than relying solely on a single oil temperature over-limit to achieve post-event alarm.

[0131] (6) Reverse attribution analysis based on the same health parameter center; The key to this invention is that after detecting an abnormal state, it does not switch to another independent diagnostic model, but continues to perform attribution analysis around the same health parameter system as the health baseline vector.

[0132] Specifically, suppose the actual health parameter vector of the chief editor's device under abnormal conditions satisfies:

[0133] The actual output vector of the main transformer can then be expressed as:

[0134] in, This refers to noise or unmodeled disturbance terms.

[0135] exist Performing a first-order Taylor expansion on the actual output vector above, we get:

[0136] The sensitivity matrix is ​​expressed as follows:

[0137] This yields an approximate relationship between the residual vector and the parameter offset:

[0138] As can be seen from the above equation, the residual vector is essentially determined by the center of the same health parameter. The deviation in health parameters is the cause, therefore anomaly detection and cause attribution are consistent at the model level.

[0139] In this embodiment of the invention, weighted regularized least squares pairs are used. Make an estimate:

[0140] in: Let be the noise covariance matrix of the residual vector; The regularization coefficient is used. It is an identity matrix.

[0141] Thus, the offset estimates of each health parameter are obtained, which is the offset vector in this embodiment of the invention, and its formula is expressed as:

[0142] Furthermore, the contribution level of each health parameter is defined:

[0143] in, Sensitivity matrix The Analysis of the abnormal windows with the largest deviations reveals that: the load loss factor contributes approximately 57.5%; the excitation state factor contributes approximately 32.0%; the tap changer factor contributes approximately 10.0%; and the imbalance factor and measurement chain offset factor contribute relatively little.

[0144] The results indicate that the current abnormal state is not merely manifested as an elevated temperature reading, but can be further explained as a composite state deviation primarily due to equivalent load loss offset and secondarily due to excitation state offset. This allows the attribution results to directly correspond to health factors of engineering significance.

[0145] (7) Verification of advantages and effects; To verify the advantages of the method in early anomaly identification and cause attribution in this invention, this invention further employs a simulation verification method based on "controlled parameter perturbation on measured healthy samples." Specifically, the following single health parameter offset scenarios are constructed near the healthy samples: load loss offset scenario, excitation state offset scenario, three-phase imbalance offset scenario, and measurement chain offset scenario.

[0146] In each of the above scenarios, the following methods and indicators are compared and analyzed: (a) The state assessment method for industrial equipment provided in this embodiment of the invention: compare the anomaly triggering times of the comprehensive anomaly index J(t) and its sliding statistics constructed based on the unified health parameter space, and calculate the Top-1 accuracy of the dominant health factor obtained by reverse attribution; (b) Single oil temperature threshold method: compare the abnormal triggering time when only whether the oil temperature exceeds the health threshold is used as the basis for alarm; (c) Single winding temperature threshold method: compare the abnormal triggering time when only the winding temperature exceeds the health threshold as the basis for alarm; (d) Observation space statistics method: compare the anomaly triggering time when constructing statistical outliers directly in the observation output space and making judgments.

[0147] For ease of explanation, Figure 2 A comparative diagram showing the average trigger time is provided, and Figure 3 A comparative diagram of average detection rates is shown, in which... Figure 2 and Figure 3 In the above embodiments of the present invention, (a) to (d) correspond to the results of the industrial equipment condition assessment method, single oil temperature threshold method, single winding temperature threshold method and observation space statistical method.

[0148] Depend on Figure 2 and Figure 3 It can be seen that the method provided by the embodiments of the present invention is superior to other methods not only in terms of average triggering time, but also in terms of average detection rate. Furthermore, experimental results show that, in the load loss offset scenario, the method of the embodiments of the present invention has an average triggering time approximately 15.62 minutes earlier than the single oil temperature threshold method, and the Top-1 attribution accuracy is approximately 70.00%; in the excitation state offset scenario, the method of the embodiments of the present invention can detect anomalies approximately 24.96 minutes earlier on average, with a Top-1 attribution accuracy of approximately 78.33%; in the three-phase imbalance offset scenario, the method of the embodiments of the present invention can detect anomalies approximately 27.29 minutes earlier on average, with a Top-1 attribution accuracy of approximately 79.17%; and in the measurement chain offset scenario, the method of the embodiments of the present invention can detect anomalies approximately 27.62 minutes earlier on average, with a Top-1 attribution accuracy of approximately 90.83%.

[0149] Therefore, the method of this invention can not only detect state deviations earlier than schemes based solely on observation thresholds, but also continue to provide explanations of causes within the same health parameter space after an anomaly occurs, thereby avoiding the problems of "detecting without attribution" or "detection and attribution models being disconnected".

[0150] (8) Format of implementation results output; In this embodiment of the invention, the final output is a status assessment report, which may include the following: Comprehensive Status Indicators: Based on the above comprehensive anomaly indicators or average abnormal indicators Calculated; Anomaly alarm levels: for example, normal, slight deviation, moderate degradation, severe anomaly; The offset of the health parameter can be extracted from the offset vector, as included in this embodiment of the invention. , , , , These are the offsets of the equivalent load loss health parameter, the equivalent excitation state health parameter, the equivalent imbalance sensitivity health parameter, the tap change response health parameter, and the measurement chain offset health parameter, respectively. Health parameter contribution ranking: can be obtained based on the degree of contribution, and the size pattern of the contribution of each health parameter will be output; Determination of dominant anomaly type: For example, load loss anomaly dominant, excitation state deviation dominant, imbalance anomaly dominant, tap changer link anomaly dominant, or measurement link anomaly dominant. Maintenance recommendations: For example, prioritize checking the source of load loss, verifying the voltage excitation status, checking the source of three-phase imbalance, checking the tap changer operation status, or troubleshooting related measurement channel offsets.

[0151] Furthermore, the key to the embodiments of the present invention does not lie in the use of a single residual analysis method, regression model, or least squares solution method, but in the unified representation of the health status of the main transformer equipment as a health parameter vector with engineering meaning. ; with health parameter center The parameters generated as a health reference output serve as a benchmark; and the health parameter center is also used as the reference. As a local expansion center for anomaly attribution analysis, it allows anomaly detection and causal attribution to share the same physical parameter system. In other words, in this embodiment of the invention, the main transformer does not first obtain residuals through one model and then interpret anomalies through another independent model. Instead, it completes a continuous closed-loop process of "health reference—anomaly offset—causal explanation" within the same unified health parameter space. Therefore, this embodiment of the invention verifies that the condition assessment method for industrial equipment can be applied not only to wind turbines but also to other industrial equipment such as main transformers that have multivariate coupling characteristics, significant changes in operating conditions, and limited fault samples, and can output condition assessment results and attribution results with clear engineering implications.

[0152] Furthermore, based on the above embodiments, this invention also provides a condition assessment device for industrial equipment, such as... Figure 4 The diagram shows a structural schematic of a condition assessment device for industrial equipment. The device includes: The acquisition module 40 is used to acquire the input vector and actual output vector of the industrial equipment under the current operating conditions during the online operation phase of the industrial equipment. Prediction module 41 is used to input the input vector into a pre-established health baseline model, and output a health reference output vector under the current operating condition based on the pre-calibrated health baseline vector through the health baseline model; wherein, the health baseline vector includes multiple physical factors of the industrial equipment in a healthy state, and the physical factors are parameters used to characterize the healthy level of the industrial equipment; Calculation module 42 is used to calculate the residual vector between the health reference output vector and the actual output vector; Anomaly module 43 is used to construct a comprehensive anomaly index based on the residual vector. When the comprehensive anomaly index is greater than a preset threshold, the industrial equipment is determined to be in an abnormal state. The mapping module 44 is used to expand the health baseline model based on the health baseline vector after determining that the industrial equipment is in an abnormal state, and establish a mapping relationship between the residual vector and the parameter offset; wherein, the parameter offset is the offset between the actual parameter vector of the industrial equipment in the current abnormal state and the health baseline vector. The quantization module 45 is used to solve the mapping relationship through an optimization algorithm to obtain the offset of each health parameter in the current abnormal state relative to the health baseline vector, and to quantify the contribution of each health parameter to the current abnormal state. The generation module 46 is used to generate a state assessment report that includes the abnormal state, the degree of offset, and the degree of contribution.

[0153] The industrial equipment condition assessment device provided in this embodiment of the invention has the same technical features as the industrial equipment condition assessment method provided in the above embodiments, so it can also solve the same technical problems and achieve the same technical effects.

[0154] Furthermore, embodiments of the present invention also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0155] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the above-described method.

[0156] Furthermore, embodiments of the present invention also provide a schematic diagram of the structure of an electronic device, such as... Figure 5 The diagram shows the structure of the electronic device, which includes a processor 51 and a memory 50. The memory 50 stores computer-executable instructions that can be executed by the processor 51, and the processor 51 executes the computer-executable instructions to implement the above-described method.

[0157] exist Figure 5 In the illustrated embodiment, the electronic device further includes a bus 52 and a communication interface 53, wherein the processor 51, the communication interface 53, and the memory 50 are connected via the bus 52.

[0158] The memory 50 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 53 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 52 may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 52 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0159] Processor 51 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 51 or by instructions in software form. Processor 51 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in the memory, and the processor 51 reads the information in the memory and uses its hardware to complete the aforementioned method.

[0160] The computer program product of the industrial equipment condition assessment method and apparatus provided in this embodiment of the invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0161] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0162] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0163] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0164] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0165] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for condition assessment of industrial equipment, characterized in that, The method includes: During the online operation phase of industrial equipment, the input vector and actual output vector of the industrial equipment under the current operating conditions are collected. The input vector is input into a pre-established health baseline model, and the health baseline model outputs a health reference output vector under the current operating condition based on the pre-calibrated health baseline vector; wherein, the health baseline vector includes multiple physical factors of the industrial equipment in a healthy state, and the physical factors are parameters used to characterize the healthy level of the industrial equipment; Calculate the residual vector between the health reference output vector and the actual output vector; A comprehensive anomaly index is constructed based on the residual vector. When the comprehensive anomaly index is greater than a preset threshold, the industrial equipment is determined to be in an abnormal state. After determining that the industrial equipment is in an abnormal state, the health baseline model is expanded based on the health baseline vector to establish a mapping relationship between the residual vector and the parameter offset; wherein, the parameter offset is the offset between the actual parameter vector of the industrial equipment in the current abnormal state and the health baseline vector. The mapping relationship is solved by optimizing the algorithm to obtain the offset of each health parameter in the current abnormal state relative to the health baseline vector, and the contribution of each health parameter to the current abnormal state is quantitatively calculated. Generate a state assessment report that includes the abnormal state, the degree of deviation, and the degree of contribution.

2. The method according to claim 1, characterized in that, The method further includes: Calculate the similarity between the residual vector and each abnormal pattern in the pre-built abnormal pattern library, and determine the abnormal patterns with similarity higher than a preset similarity threshold as candidate abnormal patterns corresponding to the current abnormal state; Compare the parameter offset with the expected parameter offset direction corresponding to the candidate anomaly pattern; If the comparison results meet the consistency condition, the candidate anomaly pattern is determined as the anomaly pattern corresponding to the current anomaly state.

3. The method according to claim 1, characterized in that, The method further includes: The historical data of the industrial equipment under healthy operating conditions are obtained, as well as the health baseline vector under healthy operating conditions; wherein, the historical data includes a historical input vector and a historical output vector corresponding to the historical input vector; A health baseline mapping relationship is established from the historical input vector to the historical output vector based on the historical data; wherein, the health baseline mapping relationship is used to characterize the health reference output vector that the industrial equipment should present when maintaining the healthy operating state corresponding to the health baseline vector under the operating conditions of the historical input vector; The health baseline function of the health baseline model is constructed based on the health baseline mapping relationship.

4. The method according to claim 3, characterized in that, The method further includes: Construct an optimization objective function, wherein the optimization objective function is used to minimize the error between the historical output vector in the historical data and the health reference output vector based on the health baseline function; The health baseline vector is calibrated based on the optimization objective function.

5. The method according to claim 1, characterized in that, The step of calculating the residual vector between the health reference output vector and the actual output vector includes: Calculate the difference between the health reference output vector and the actual output vector, and determine the difference as the residual vector.

6. The method according to claim 1, characterized in that, The steps for constructing a comprehensive anomaly index based on the residual vector include: The residual vector is standardized to obtain a standardized residual vector; An anomaly comprehensive index is constructed based on the standardized residual vector.

7. The method according to claim 1, characterized in that, The step of expanding the health baseline model based on the health baseline vector and establishing a mapping relationship between the residual vector and the parameter offset to map the residual vector to the parameter offset includes: Based on the health baseline vector, construct expressions for the true parameter vector of the industrial equipment under abnormal conditions and expressions for the actual output vector; The expression of the actual output vector is subjected to Taylor expansion based on the health baseline vector to obtain the Taylor expansion corresponding to the actual output vector, and the sensitivity matrix is ​​determined based on the Taylor expansion. Based on the sensitivity matrix, a mapping relationship is established between the residual vector and the parameter offset to map the residual vector to the parameter offset.

8. The method according to claim 1, characterized in that, The step of solving the mapping relationship using an optimization algorithm to obtain the offset of each health parameter relative to the health baseline vector under the current abnormal state includes: An optimization function for estimating the degree of offset is constructed based on the mapping relationship; Solving the optimization function yields an offset vector; where each element in the offset vector represents the degree of offset of the corresponding health parameter.

9. The method according to claim 7, characterized in that, The steps for quantifying the contribution of each of the aforementioned health parameters to the current abnormal state include: For each health parameter, the matrix component corresponding to the health parameter is extracted from the sensitivity matrix, and the contribution degree is calculated based on the offset degree corresponding to the health parameter and the matrix component; the contribution degree is used to represent the contribution of the health parameter to the current residual vector.

10. A condition assessment device for industrial equipment, characterized in that, The device includes: The data acquisition module is used to acquire the input vector and actual output vector of the industrial equipment under the current operating conditions during the online operation phase of the industrial equipment. The prediction module is used to input the input vector into a pre-established health baseline model, and output a health reference output vector under the current operating condition based on the pre-calibrated health baseline vector through the health baseline model; wherein, the health baseline vector includes multiple physical factors of the industrial equipment in a healthy state, and the physical factors are parameters used to characterize the healthy level of the industrial equipment; The calculation module is used to calculate the residual vector between the health reference output vector and the actual output vector; An anomaly module is used to construct a comprehensive anomaly index based on the residual vector. When the comprehensive anomaly index is greater than a preset threshold, the industrial equipment is determined to be in an abnormal state. The mapping module is used to expand the health baseline model based on the health baseline vector after determining that the industrial equipment is in an abnormal state, and establish a mapping relationship between the residual vector and the parameter offset; wherein, the parameter offset is the offset between the actual parameter vector of the industrial equipment in the current abnormal state and the health baseline vector. The quantization module is used to solve the mapping relationship through an optimization algorithm to obtain the offset of each health parameter in the current abnormal state relative to the health baseline vector, and to quantify the contribution of each health parameter to the current abnormal state. The generation module is used to generate a state assessment report that includes the abnormal state, the degree of offset, and the degree of contribution.

Images