A unit state maintenance decision method and system based on a multi-dimensional health score

Abstract

The application provides a kind of unit state maintenance decision method and system based on multidimensional health score, it is related to data processing technical field, method includes: the operation sensor time series data of the unit to be maintained is collected;The time-frequency transform is carried out to operation sensor time series data, obtains multimodal feature input vector;Multimodal feature input vector is input to degradation perception health representation network, and multidimensional health embedding feature matrix is output;Multidimensional health embedding feature matrix is projected to ideal health space, and standardization health score is obtained;According to the variance distribution characteristics of standardization health score, construct degradation uncertainty scenario set;Based on degradation uncertainty scenario set, construct two-stage stochastic programming model;Two-stage stochastic programming model is solved, and first stage preparation decision variable and second stage maintenance window decision variable are obtained;The state maintenance decision scheme of the unit to be maintained is determined.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for making decisions on unit condition-based maintenance based on multidimensional health scores. Background Technology

[0002] During the operation of large industrial units (such as gas turbines, steam turbines, compressor units and generator sets), the equipment is subjected to high load, complex working conditions and multi-physical field coupling environment for a long time, and its internal components will gradually undergo irreversible degradation and evolution.

[0003] With the development of industrial internet and sensor technology, multi-source time-series data such as vibration, temperature, electrical parameters, fluid state and operating conditions can be collected in real time during unit operation, thereby providing basic data support for equipment condition monitoring and health assessment.

[0004] However, traditional health assessment methods often rely on single indicators or simple weighted fusion methods, which are insufficient to fully express the nonlinear coupling relationships between multidimensional sensor information, resulting in limited health characterization capabilities. Secondly, while some machine learning or deep learning-based methods can extract high-dimensional features, they often lack physical consistency constraints, and health scores may exhibit non-monotonic fluctuations or be inconsistent with actual degradation patterns. Furthermore, existing methods lack the ability to systematically model multiple degradation scenarios and their probability distributions, making it difficult to effectively support collaborative maintenance and optimization decisions for multiple devices in complex industrial environments. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a unit condition-based maintenance decision-making method and system based on multidimensional health scoring. This method can solve the problems of traditional health assessment methods, which often rely on single indicators or simple weighted fusion methods, making it difficult to fully express the nonlinear coupling relationship between multidimensional sensor information, resulting in limited health characterization capabilities. Furthermore, while some machine learning or deep learning-based methods can extract high-dimensional features, they often lack physical consistency constraints, and the health scoring results may exhibit non-monotonic fluctuations or be inconsistent with the actual degradation patterns. In addition, they lack the ability to systematically model multiple degradation scenarios and their probability distributions, making it difficult to effectively support multi-equipment collaborative maintenance optimization decision-making in complex industrial environments.

[0006] A first aspect of this invention proposes a unit condition-based maintenance decision-making method based on multidimensional health scores, comprising: S1: Collect timing data of the operating sensors of the unit to be inspected; S2: Perform time-frequency transformation on the time-series data of the operating sensor to obtain the multimodal feature input vector; S3: Input the multimodal feature input vector into the degradation-aware health representation network and output a multidimensional health embedding feature matrix; S4: Project the multidimensional health embedding feature matrix onto the ideal health space to obtain a standardized health score; S5: Construct a set of degradation uncertainty scenarios based on the variance distribution characteristics of standardized health scores; S6: Construct a two-stage stochastic programming model based on the set of degradation uncertainty scenarios and the combination probability of each degradation uncertainty scenario in the set of degradation uncertainty scenarios; S7: Solve the two-stage stochastic programming model to obtain the first-stage preparation decision variables and the second-stage maintenance window decision variables for each component in the unit to be overhauled; S8: Based on the decision variables of the first stage preparation and the decision variables of the second stage maintenance window, determine the condition-based maintenance decision scheme for the unit to be maintained.

[0007] A second aspect of the present invention provides a unit condition-based maintenance decision system based on multidimensional health scores, comprising: a processor and a memory; The memory stores programs or instructions that can run on a processor, which, when executed by the processor, implement the steps of the unit condition-based maintenance decision-making method as described in the first aspect.

[0008] A third aspect of the present invention provides a readable storage medium storing a program or instructions, which, when executed by a processor, implement the steps of the unit condition maintenance decision-making method based on multidimensional health scores as described in the first aspect.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, by inputting multimodal features into the degradation-aware health representation network, a multidimensional health embedding feature matrix is ​​obtained, thereby achieving a deep and learnable representation of the unit's degradation state. By projecting the multidimensional health embedding onto an ideal health space, a standardized health score that satisfies the monotonic degradation law is generated, thus achieving a quantitative expression of the physical consistency of the health state. By constructing a set of degradation uncertainty scenarios based on the variance distribution of the health score, the evolutionary uncertainty of the unit under different degradation paths is characterized. By constructing a two-stage stochastic programming model, joint optimization decision-making of maintenance preparation strategy and maintenance window allocation is achieved. By solving the stochastic programming model, the maintenance strategy and maintenance window arrangement of each component are obtained, thereby forming an optimal condition maintenance decision scheme for the coordinated operation of multiple components of the unit. Attached Figure Description

[0010] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0011] Figure 1 This is a flowchart illustrating a unit condition-based maintenance decision-making method based on multidimensional health scoring, provided in an embodiment of the present invention.

[0012] Figure 2 This is an evolution diagram of a degradation uncertainty scenario based on a standardized health score, provided by an embodiment of the present invention.

[0013] Figure 3 This is a graph showing the change of failure risk cost with maintenance window under different degradation scenarios, provided by an embodiment of the present invention.

[0014] Figure 4 This is a schematic diagram of a unit condition-based maintenance decision system based on multidimensional health scoring, provided in an embodiment of the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, 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. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0016] The following description, in conjunction with the accompanying drawings, details the unit condition-based maintenance decision-making method based on multidimensional health scores provided by the embodiments of the present invention through specific examples and application scenarios.

[0017] Reference manual attached Figure 1 The diagram shows a flowchart of a unit condition-based maintenance decision-making method based on multidimensional health scores provided by an embodiment of the present invention.

[0018] This invention provides a method for unit condition-based maintenance decision-making based on multidimensional health scores, which may include the following steps: S1: Collect timing data of the operating sensors of the unit to be inspected.

[0019] Among them, the time sequence data of the operating sensors refers to the data sequence formed by various sensors (such as vibration sensors, temperature sensors, current and voltage sensors, pressure / flow sensors and operating condition monitoring devices) installed in different parts of the equipment during the actual operation of the unit, which are continuously collected and recorded in chronological order. This data can reflect the dynamic operating status changes of the unit at different points in time.

[0020] It should be noted that by collecting time-series data from operating sensors, continuous, real-time, and comprehensive monitoring of the unit's operating status can be achieved, avoiding the information lag and incompleteness problems caused by traditional manual inspections or discrete testing. At the same time, it provides a rich source of basic data for subsequent multimodal feature extraction and health assessment.

[0021] In one possible implementation, the timing data of the operating sensors includes: vibration status data, temperature status data, electrical status data, fluid status data, and operating condition data.

[0022] S2: Perform time-frequency transformation on the time-series data of the operating sensor to obtain the multimodal feature input vector.

[0023] Time-frequency transformation refers to the use of methods such as Fourier transform, wavelet transform, or short-time Fourier transform to convert a time-domain signal into a joint representation that simultaneously contains time and frequency information. Multimodal feature input vector refers to a unified high-dimensional feature representation formed by fusing features from different sensors, different physical quantities, and different transform domains (time domain / frequency domain), which is used as input to subsequent models.

[0024] It should be noted that by performing time-frequency transformation on the time-series data of the operating sensors and constructing multimodal feature input vectors, the time dynamic changes and frequency structure characteristics of the unit's operating signals can be expressed simultaneously in a unified feature space. This effectively captures transient shocks, periodic vibrations, and early weak fault information that are difficult to identify using traditional time-domain methods. At the same time, it enables the fusion expression of multi-source heterogeneous sensor data, reduces noise interference from single signals, and improves the richness and separability of feature expression.

[0025] S3: Input the multimodal feature input vector into the degradation-aware health representation network and output a multidimensional health embedding feature matrix.

[0026] Among them, the degradation-aware health representation network refers to a neural network model that can learn the degradation patterns of equipment. It performs nonlinear mapping on input features so that health status features can reflect the evolution process of equipment from health to deterioration. The multidimensional health embedding feature matrix refers to the high-dimensional structured feature representation output by the network, which is used to characterize the unit status from multiple health dimensions. Different dimensions correspond to different degradation modes or health attributes.

[0027] In one possible implementation, S3 specifically includes: S301: Construct a semi-supervised input sample set that matches the degradation-aware health representation network based on the multimodal feature input vector.

[0028] The semi-supervised input sample set refers to the training data set that includes both labeled data (healthy samples and severely degraded samples) and unlabeled running data, which is used to improve the model's learning ability when there is insufficient labeling.

[0029] The semi-supervised input sample set includes healthy samples, severely degraded samples, and unlabeled samples.

[0030] Specifically, based on the multimodal feature input vector, the labeled healthy samples and severely degraded samples are labeled and organized, while the unlabeled running data is included in the same dataset to form a semi-supervised input sample set containing three types of data, which is used for subsequent network training and feature learning.

[0031] S302: Input the semi-supervised input sample set into the degradation-aware health representation network.

[0032] S303: By using the nonlinear mapping function in the degradation-aware health representation network, each input sample in the semi-supervised input sample set is mapped to the health embedding space to obtain the health space coordinates.

[0033] The health embedding space refers to the feature representation space learned by the network, where different locations correspond to different health states or degrees of degradation. Health space coordinates refer to the specific vector representation of a sample within the health embedding space, used to characterize its health state location.

[0034] Specifically, each input sample is mapped to the health embedding space through a nonlinear mapping function within the network (such as a multilayer perceptron or a deep feature extraction structure), resulting in a health space coordinate vector corresponding to each sample.

[0035] S304: Determine the positional relationship of the input sample relative to the health hypersphere region based on the distance between the health spatial coordinates and the preset health center.

[0036] The preset health center refers to the reference center point in the health embedded space that represents the "ideal health state," and this invention does not limit this. The health hypersphere region refers to the distribution area of ​​health state with the health center as the center and a certain radius, used to distinguish between healthy and deteriorated states.

[0037] Specifically, the Euclidean distance or similarity distance between each health spatial coordinate and the preset health center is calculated, and the sample is determined to be located inside the health hypersphere, near the boundary, or in the external deterioration area based on the distance.

[0038] S305: Determine the sample weight coefficients based on the positional relationship.

[0039] Among them, the sample weight coefficient refers to the weight assigned according to the health status of the sample or the degree of deviation from the health center, which is used to strengthen the influence of key samples.

[0040] Specifically, weight coefficients are assigned to samples based on their positional relationship within the healthy hypersphere region. For example, samples in the healthy region are given lower weights, while samples at the boundary and in deterioration regions are given higher weights, in order to highlight key degradation information.

[0041] S306: Combine the sample weight coefficients to calculate the decentralized health embedding vector of each input sample relative to the preset health center.

[0042] Among them, the decentralized healthy embedding vector refers to the feature representation obtained after centering the healthy embedding vector, which is used to eliminate the influence of global offset.

[0043] Specifically, the original healthy embedding vector of the sample is subtracted from the healthy center vector to obtain the decentralized healthy embedding vector, thereby eliminating the influence of the overall offset and enhancing the ability to express relatively degraded features.

[0044] S307: Arrange the decentralized health embedding vectors in chronological order to obtain the initial health embedding matrix.

[0045] Specifically, all decentralized health embedding vectors are sorted and arranged according to the time sampling order to construct an initial health embedding matrix that evolves over time, which is used to characterize the dynamic changes in the device degradation process.

[0046] S308: Based on the initial health embedding matrix, construct the Gram matrix between the health embedding dimensions.

[0047] The Gram matrix refers to a matrix composed of vector inner products, used to describe the correlation between different health embedding dimensions.

[0048] Specifically, a dimensional correlation analysis is performed on the initial health embedding matrix, and a Gram matrix is ​​constructed by calculating the inner product relationship between each embedding dimension to represent the coupling relationship between different health feature dimensions.

[0049] S309: By combining the Gram matrix and the diversity constraint function, the health embedding dimension is decoupled to determine the multidimensional health embedding feature matrix.

[0050] Among them, the diversity constraint function refers to the mathematical function used to constrain the differences or independence between feature dimensions, and is used to prevent feature redundancy.

[0051] It should be noted that, unlike traditional methods that only employ fully supervised learning or simple feature concatenation, this method introduces a semi-supervised sample mechanism to effectively utilize unlabeled runtime data, thereby improving data utilization. Simultaneously, by constructing health centers and health hyperspheres, the health status is given a clear geometric constraint expression. Furthermore, the sample weighting mechanism and decentralization enhance the model's sensitivity to key degraded samples and reduce bias interference. Further, the Gram matrix is ​​used to characterize the relationships between embedding dimensions, and feature decoupling is achieved through diversity constraints. This significantly improves the structure, separability, and robustness of the health representation, making the output multidimensional health embedding more consistent with the device degradation mechanism and possessing stronger engineering interpretability and generalization ability.

[0052] Reference manual attached Figure 2 This diagram illustrates the evolution of a degradation uncertainty scenario based on a standardized health score, as provided in an embodiment of the present invention.

[0053] like Figure 2 As shown, the horizontal axis represents the prediction time window, indicating the prediction period extending from the current moment into the future. The vertical axis represents the standardized health score; a higher score indicates that the component's degradation is closer to the failure risk state. The figure illustrates three types of degradation uncertainty scenarios based on the standardized health score: slow degradation, expected degradation, and rapid degradation. The three scenarios have similar standardized health scores at the current prediction starting point, but as the prediction time window lengthens, the evolution speed of the health score gradually diverges under different scenarios. Specifically, the rapid degradation scenario exceeds the failure threshold around the 23rd prediction time window, indicating that under this degradation path, the component may enter a high-risk state earlier. The slow degradation scenario remains below the failure threshold, indicating a lower short-term failure risk under this path. The expected degradation scenario lies between the two, reflecting the health status evolution trend of the unit's components under normal degradation conditions.

[0054] As shown in the attached figure, this scheme can expand a single health score sequence into multiple degradation paths with probabilistic attributes, thereby characterizing the degradation uncertainty of unit components under complex operating conditions. Compared with methods that only use mean prediction or single-path prediction, this method can provide richer scenario inputs for the subsequent two-stage stochastic programming model, enabling maintenance preparation decisions and maintenance window allocation to simultaneously consider different degradation rates and their probabilities of occurrence, thus enhancing the robustness and engineering adaptability of the condition-based maintenance decision scheme.

[0055] S4: Project the multidimensional health embedding feature matrix onto the ideal health space to obtain a standardized health score.

[0056] The ideal health space refers to a predefined or learned standard health state representation space, used to constrain the health state to conform to the physical evolution law of "health → degradation", and usually includes health centers and monotonic degradation structure constraints. The standardized health score refers to mapping the projected health state to a single health index with a uniform dimension (usually in the 0-1 range), used to characterize the health level and degradation level of equipment.

[0057] In one possible implementation, S4 specifically refers to: By using alternating projection processing and order-preserving constraint processing, the multidimensional health embedding feature matrix is ​​mapped to the ideal health space, resulting in a monotonically increasing standardized health score.

[0058] In one possible implementation, S4 specifically includes: S401: Arrange the multidimensional health embedding feature matrix in time sequence to obtain the health embedding matrix to be projected.

[0059] Specifically, a multidimensional health embedding feature matrix is ​​first obtained. Each row of this matrix corresponds to a health embedding vector at a time window or sampling moment. Then, based on the timestamps of the unit's operational data collection, the timestamps corresponding to each health embedding vector are read and aligned. Next, all health embedding vectors are sorted and reorganized in chronological order (from early operating states to the current or latest operating states). During the sorting process, the data continuity of the same unit or component is maintained, and the internal dimensional structure of a single embedding vector is not changed. Only its arrangement order in the matrix is ​​adjusted. Finally, the vector sequence arranged in ascending chronological order is reconstructed into a matrix form to obtain the "health embedding matrix to be projected" reflecting the equipment degradation and evolution process, which is used for subsequent ideal health space projection processing.

[0060] S402: Construct an ideal healthy space based on the time thresholds for the healthy phase and the severe degradation phase.

[0061] The health phase time threshold is used to define the time boundary for the unit under maintenance to be in the health phase, while the severe degradation phase time threshold is used to define the time boundary for the unit under maintenance to enter the severe degradation phase. Both the health phase time threshold and the severe degradation phase time threshold can be determined based on historical operating samples, labeled degradation samples, or preset maintenance experience data.

[0062] Specifically, the ideal health space is constructed as follows: Based on the time thresholds for the healthy stage and the severe degradation stage, the health state sequence is first divided into three evolution intervals according to time. In the early health stage, the health state is constrained to be within a stable healthy or initial degradation state, representing the equipment still in normal operation or slight degradation. In the severe degradation stage, the health state is constrained to be within a failure or critical degradation state, representing the equipment having entered a stage with significant failure risk. Simultaneously, monotonically non-increasing constraints are applied throughout the entire time range to ensure that the health state evolution over time conforms to the irreversible degradation law. Through the joint definition of the above segmented constraints and monotonically increasing constraints, the ideal health space is constructed to constrain the physical consistency and degradation rationality of the health score in the time dimension.

[0063] Among them, the ideal health space is used to constrain the standardized health score to be close to the initial state in the healthy stage, close to the terminal state in the severely deteriorated stage, and monotonically increase along the time direction.

[0064] S403: Combining the health embedding matrix to be projected and the ideal health space, determine the ideal health trajectory:

[0065] in, This represents the optimal health embedding parameters. Indicates an ideal health trajectory. argmin Indicates minimization. z This indicates the state of health after projection onto an ideal healthy space. h Indicates characteristics of health status. Denotes the square of the Frobenius norm. This represents the regularization weight coefficient. Indicates the weight of health embedding. w Regular constraint terms, Y Represents the health feature mapping matrix. Represents the set of constraints for an ideal healthy space. s . t The dot (.) indicates a constraint.

[0066] Specifically, under the condition of satisfying the ideal health space constraint, the formula enables the observed health state to achieve optimal projection and consistent reconstruction under structural constraints by jointly optimizing the health embedding mapping parameters and the health trajectory representation.

[0067] S404: Under the constraint of the ideal health trajectory, the health embedding weights of the degenerate-perceived health representation network are updated to obtain the first health trajectory in the current iteration:

[0068]

[0069] in, Indicates the first n Healthy embedding weights updated after +1 iteration. Indicates the first n The ideal health trajectory obtained from the next iteration. Indicates the first n The health status feature representation after +1 iteration update.

[0070] Specifically, the formula takes the health score sequence as the constraint target and iteratively optimizes the health embedding parameters to make the health state generation model gradually approach the ideal health evolution trajectory that satisfies the physical constraints, thereby improving the reliability and consistency of the health score.

[0071] S405: Under the time evolution constraint of the first health trajectory, the health embedding matrix to be projected is projected onto the ideal health space to generate a second health trajectory that satisfies the order-preserving constraint and is monotonically increasing.

[0072] Specifically, under the temporal evolution constraint of the health trajectory, the health embedding matrix to be projected is first used as the initial representation of the current health state. In each iteration, the distance between the current health state and all feasible health states in the ideal health space is measured, and the target point that minimizes the squared error between the two is selected as the projection result, thereby achieving optimal approximation to the ideal health space. Simultaneously, a time-order-preserving constraint is introduced during the projection update process, applying a monotonic relationship that progresses with time to the health state sequence, ensuring that the health state at subsequent time points is not less than the health state at the previous time point, thus avoiding non-physical rebounds in the health score. Through the alternating update mechanism of "minimum distance projection" and "order-preserving constraint correction," the health embedding matrix to be projected is gradually mapped to the ideal health space, ultimately generating a health trajectory that satisfies temporal consistency, order preservation constraints, and monotonically increasing characteristics.

[0073] S406: Perform boundary normalization on the second health trajectory to obtain a standardized health score.

[0074] Specifically, based on the health trajectory, the health status at each time point is first subjected to interval constraints, limiting it to a preset standard interval. That is, when the health status is below the lower bound, it is truncated to the lower bound value; when the health status is above the upper bound, it is truncated to the upper bound value, thus eliminating non-physical values ​​caused by abnormal fluctuations. Subsequently, a temporal consistency constraint is applied to the processed health status sequence to ensure that the health score at subsequent time points is not less than the health score at the previous time point, thereby maintaining the monotonic evolution characteristics of the health degradation process. Through the combined effect of the above "interval truncation normalization" and "temporal monotonic constraint," the original health trajectory is mapped to a standardized health score sequence located within the standard interval and monotonically increasing over time, which is used for subsequent health status assessment and degradation modeling.

[0075] It should be noted that by projecting the multidimensional health embedding feature matrix onto the ideal health space, complex high-dimensional health representations can be structurally compressed within a unified physical constraint framework. This transforms health information from different sources and dimensions into comparable standardized health scores, effectively eliminating scale inconsistencies and redundant coupling between multidimensional features. Furthermore, introducing the ideal health space constraint ensures that health scores evolve according to a monotonic degradation law over time, avoiding non-physical fluctuations in health indicators found in traditional methods. This improves the stability, interpretability, and engineering consistency of health assessment results and provides a unified quantitative basis for subsequent degradation scenario modeling and maintenance decisions.

[0076] S5: Construct a set of degradation uncertainty scenarios based on the variance distribution characteristics of standardized health scores.

[0077] Among them, variance distribution characteristics refer to the statistical properties of the dispersion of standardized health scores in a time series or sample set, used to characterize the volatility and uncertainty of health status. The set of degradation uncertainty scenarios refers to the set of multiple possible degradation paths obtained based on the volatility of health scores.

[0078] Specifically, firstly, standardized health scores are processed into time series data to form a continuous health change curve. Then, the variance of this health score series is calculated across different time windows or the overall sample to measure the degree of fluctuation in health status. Next, the trend of health score changes is segmented based on the magnitude of variance; for example, low variance corresponds to stable degradation, medium variance to normal fluctuation degradation, and high variance to drastic fluctuation degradation. Subsequently, corresponding degradation pattern categories are set according to different variance intervals, and time segments or samples with similar variance characteristics are grouped into the same degradation scenario. Finally, all degradation scenario sets are summarized to form a set of degradation uncertainty scenarios containing multiple degradation rates and evolution paths, which is used as input for subsequent stochastic programming models.

[0079] In one possible implementation, the set of degradation uncertainty scenarios includes: slow degradation scenarios, expected degradation scenarios, and fast degradation scenarios.

[0080] It should be noted that by constructing a set of degradation uncertainty scenarios based on the variance distribution characteristics of standardized health scores, the originally singular and deterministic health assessment results can be expanded into multiple possible degradation paths, thereby effectively characterizing the evolutionary uncertainty of the unit under different operational disturbances and degradation rates. Compared with traditional methods that rely solely on mean or single trend modeling, this method can utilize variance information to reflect the intensity of health status fluctuations, making scenario division more statistically grounded and adaptive, improving the realism and robustness of degradation modeling, and providing a more reasonable and diversified decision input basis for subsequent two-stage stochastic programming.

[0081] Reference manual attached Figure 3 The figure shows a graph illustrating the variation of failure risk cost with maintenance window under different degradation scenarios, as provided by an embodiment of the present invention.

[0082] like Figure 3 As shown in the figure, the horizontal axis represents the candidate maintenance window, indicating that maintenance arrangements are gradually postponed from earlier windows to later windows. The vertical axis represents the failure risk cost index, which characterizes the failure risk, downtime losses, and uncertainty fluctuation costs incurred by the continued operation of unit components under different maintenance windows. The figure shows the change of the failure risk cost index as the candidate maintenance window is postponed under three degradation scenarios: slow degradation, expected degradation, and rapid degradation. As the candidate maintenance window is postponed from window 1 to window 6, the failure risk cost increases in all degradation scenarios, with the rapid degradation scenario showing a larger increase in risk cost. This change indicates that in scenarios with rapid degradation, further postponement of the maintenance window will lead to a simultaneous increase in failure probability fluctuation risk and expected downtime losses.

[0083] As shown in the attached figure, this scheme can transform degradation uncertainty scenarios into quantifiable risk cost inputs, thus providing a basis for the selection of maintenance windows in the subsequent two-stage stochastic programming model. Compared with methods that rely solely on fixed maintenance cycles or a single average remaining life for decision-making, this approach can more reasonably reflect the risk differences caused by delayed maintenance under different degradation paths, thereby enhancing the engineering adaptability and realism of the maintenance decision scheme.

[0084] S6: Construct a two-stage stochastic programming model based on the set of degradation uncertainty scenarios and the combination probability of each degradation uncertainty scenario in the set of degradation uncertainty scenarios.

[0085] Here, the combination probability refers to the likelihood of each degradation scenario occurring in actual operation, expressed as a probability value, and used to describe the uncertainty weights of different degradation paths. The two-stage stochastic programming model refers to a phased decision optimization model, where the first stage makes an initial decision before uncertainty occurs, and the second stage makes supplementary or adjusted decisions after the specific scenario is implemented, aiming for overall optimization with the goal of optimizing the expected cost.

[0086] In one possible implementation, S6 specifically includes: S601: Based on the set of degradation uncertainty scenarios, determine the combination of degradation scenarios corresponding to each component in the unit to be overhauled.

[0087] Among them, the degradation scenario combination refers to the joint degradation situation formed by combining the states of multiple components under different degradation uncertainty scenarios, which is used to describe the overall degradation evolution path of the unit.

[0088] Specifically, based on the set of degradation uncertainty scenarios, the states of each component under different degradation scenarios are combined and matched. For example, the slow degradation of component 1 and the rapid degradation of component 2 are arranged and combined to form a "degradation scenario combination set" that describes the overall degradation state of the unit, which is used to characterize the joint degradation of multiple components.

[0089] S602: Calculate the combination probability of each degradation scenario combination.

[0090] Among them, the combination probability refers to the probability of a certain degradation scenario combination occurring in all possible combinations, and is used to quantify the uncertainty weight of multi-component joint degradation.

[0091] Specifically, for each combination of degraded scenarios, the probability of each individual scenario is multiplied or the joint probability is calculated (e.g., based on the independent assumption or related correction model) to obtain the combination probability of each combination of degraded scenarios. All combination probabilities are then normalized to ensure that the sum of probabilities is 1.

[0092] S603: Define the decision variables for the first stage preparation and the second stage maintenance window respectively.

[0093] The first-stage preparation decision variables are used to characterize whether the component to be inspected has entered the nearest maintenance window, whether it has entered the maintenance preparation state, or whether monitoring should continue.

[0094] The second-stage maintenance window decision variable is used to characterize the state of the component to be inspected being assigned to the corresponding maintenance window under any combination of degradation scenarios.

[0095] It should be noted that the decision variables for the first stage preparation and the second stage maintenance window are decision variables in the optimization model, and their optimal values ​​are obtained by solving the two-stage stochastic programming model in S7.

[0096] S604: Calculate the failure risk cost of the component under maintenance by combining its degradation state and maintenance window under the corresponding degradation scenario combination.

[0097] in, m Indicates the optional repair window number. Indicates the first j The current degradation state of each component. Indicates the first j Individual components combined in degradation scenarios The future state of degradation, Indicates the first j Failure threshold of each component Indicates the first j The overall failure risk cost of each component Indicates the first j The failure risk cost weighting coefficient for each component, Var j ( ) indicates the first j Measurement of the uncertainty of the failure probability of each component. Indicates the first j Each component is in the maintenance window. m The probability of failure occurring is as follows. Indicates the first j Each component is in the maintenance window. m The expected downtime during maintenance.

[0098] Specifically, unlike traditional methods that rely solely on a single metric (such as failure probability or mean remaining lifetime) for evaluation, this method introduces a failure probability fluctuation term driven by degradation scenarios. Compared with expected downtime losses E j This allows risk assessment to not only reflect the "average failure probability" but also characterize the "uncertainty fluctuations under different degradation paths." Therefore, driven by multidimensional health scores, this formula can map embedded health information into risk costs that can be used for optimization decisions, making it more suitable for maintenance optimization problems under multi-scenario stochastic degradation conditions.

[0099] S605: Combining the decision variables of the first stage preparation and the decision variables of the second stage maintenance window, determine the planning objective function with the goal of minimizing the expected total maintenance cost:

[0100] in,Z ( ) represents the objective function of the program. x k,j Indicates the first j The first phase of maintenance strategy selection for each component includes... x 1,j Indicates the first j Each component must be inspected and repaired immediately. x 2,j Indicates the first j Each component is ready for maintenance. x 3,j Indicates the first j These components will not be inspected at this time. Indicates the combination of degraded scenarios Next j Are the components arranged in the first...? m j One maintenance window, min Indicates minimization. Indicates the first j The weighting parameters for the first-stage maintenance strategy selection of each component Indicates the first j Basic maintenance preparation costs for each component. Indicates a combination of degenerate scenarios The probability of occurrence, Indicates the combination of degraded scenarios Maintenance window m j The next j The overall maintenance cost of a single component. Indicates the first j The basic maintenance preparation cost for each component under different maintenance strategies Cost adjustment coefficient, S This indicates shared entry costs. max This indicates taking the maximum value.

[0101] when =1 indicates a combination of degenerate scenarios. Next, the j The components are arranged in the first... m j Each maintenance window performs repairs. =0 indicates a degenerate scenario combination Next, the j The component was not arranged in the first m j Each maintenance window performs repairs.

[0102] Specifically, a two-stage stochastic optimization model driven by degradation scenarios is constructed based on multidimensional health scores. By jointly optimizing maintenance preparation decisions, maintenance window allocation, and multi-equipment collaborative maintenance costs, the global cost-optimal decision of the unit condition-based maintenance strategy under uncertain degradation conditions is achieved.

[0103] S606: Under the constraints of the first-stage preparation decision variables and the second-stage maintenance window decision variables, a two-stage stochastic programming model is constructed by minimizing the programming objective function.

[0104] The first-stage preparation decision variables refer to the decision variables regarding whether a component should enter maintenance preparation, undergo immediate maintenance, or postpone maintenance before the future degradation scenario is clearly defined. The second-stage maintenance window decision variables refer to the decision variables regarding whether each component will be scheduled for maintenance in different maintenance windows after the specific degradation scenario is realized.

[0105] It should be noted that by constructing combinations of degradation scenarios and their probability distributions, the uncertainty of degradation across multiple components is uniformly incorporated into a stochastic optimization framework. Compared to traditional methods based solely on single predicted values ​​or deterministic remaining life models, this method can simultaneously characterize the joint degradation relationships and scenario coupling effects among multiple devices. Furthermore, it introduces the variance of failure probability and the expected loss from downtime into the risk cost function, achieving dual modeling of "average risk + volatility risk." Simultaneously, through a two-stage decision-making structure, maintenance preparation decisions and maintenance window allocation are decoupled and optimized, making the decisions both forward-looking and scenario-adaptive. This significantly improves the unit's global cost optimization capability and robustness under complex and uncertain degradation environments.

[0106] S7: Solve the two-stage stochastic programming model to obtain the first-stage preparation decision variables and the second-stage maintenance window decision variables for each component in the unit to be overhauled.

[0107] The first-stage preparation decision variables refer to the decision outcome variables regarding whether each component should enter maintenance preparation, undergo immediate maintenance, or continue monitoring when the degradation scenario is unknown. The second-stage maintenance window decision variables refer to the decision outcome variables regarding which each component is assigned to a specific maintenance window after the specific degradation scenario has been determined.

[0108] In one possible implementation, S7 specifically includes: S701: Based on the decision preparation variables of the first stage, generate a set of decision preparation combinations that satisfy the mutual exclusion constraints of the first stage.

[0109] In this context, the first-stage maintenance decision combination represents a specific maintenance strategy scheme that satisfies the mutual exclusion constraint, while the first-stage mutual exclusion constraint preparation decision combination set represents the set of all feasible maintenance strategy schemes that satisfy the mutual exclusion constraint, which is used to constitute the feasible region of the first-stage optimization problem.

[0110] Specifically, firstly, based on the first-stage decision variables (immediate maintenance, preparation for maintenance, and no maintenance) of each component in the unit, all possible combinations of first-stage maintenance strategies are generated according to mutual exclusion rules. Then, combinations that satisfy "each component corresponds to only one strategy" are selected to form the first-stage preparation decision combination set, which is used as the feasible domain input for subsequent optimization.

[0111] S702: Based on the prepared decision combination set, determine the set of optional maintenance windows for each component to be inspected in the second stage:

[0112] in, Indicates the first phase of maintenance decision combination Next, the j A collection of optional maintenance windows for each component. This indicates the total number of available repair windows. Indicates the first phase of maintenance decision combination Next, the j Whether the first maintenance strategy, i.e., the immediate maintenance strategy, is selected for each component. Indicates the first phase of maintenance decision combination Next, the j Whether the component selects the second maintenance strategy, i.e., enters the ready state. Indicates the first phase of maintenance decision combination Next, the j Whether to select the third maintenance strategy for each component, i.e., not to maintain it for the time being.

[0113] It should be noted that, , as well as The value can be either 0 or 1. When it is 1, it means that the corresponding strategy is selected, and when it is 0, it means that no strategy is selected.

[0114] This formula is used to map the maintenance strategies of each component of the unit to a corresponding set of optional maintenance windows based on the maintenance decision in the first stage, thereby realizing the constraint coupling between maintenance decision and maintenance sequence.

[0115] Specifically, the first-stage maintenance decision variables are used to characterize the maintenance status selection of a single component, while the first-stage maintenance decision combination is used to represent the overall configuration of the first-stage maintenance status variables of all components, and is used to characterize the complete unit maintenance strategy scheme.

[0116] S703: Calculate the minimum scenario cost of the second-stage maintenance window decision variables within the set of available maintenance windows:

[0117] in, Indicates the first phase of maintenance decision combination Combination with degraded scenarios The minimum scenario cost of the second-stage maintenance window decision variables. Indicates the combination of degraded scenarios Next, the j Are the components arranged in the first...? m One maintenance window, Indicates the combination of degraded scenarios Maintenance window m The next j The overall maintenance cost of a single component. J This indicates the total number of components in the unit. Indicates the first m Is the maintenance window enabled, i.e., is any device in the... m One maintenance window was inspected.

[0118] Specifically, this formula is used to minimize the total maintenance cost of the unit in a single scenario by optimizing the allocation scheme of each component in different maintenance windows, given a first-stage maintenance strategy and a degradation uncertainty scenario.

[0119] S704: Combining the combination probabilities of each degradation scenario combination, the minimum scenario cost is weighted to obtain the expected total maintenance cost corresponding to each preparation decision combination in the preparation decision combination set.

[0120] The expected total maintenance cost refers to the average cost after weighting the costs of all scenarios according to probability.

[0121] S705: Compare the various expected total maintenance costs to determine the optimal preparation decision combination.

[0122] Among them, the optimal preparation decision combination refers to the first-stage maintenance strategy combination with the lowest expected cost.

[0123] Specifically, the expected total maintenance cost is calculated for each candidate scheme in the first-stage preparation decision combination set, and a horizontal comparison is made. The combination with the smallest expected cost is selected as the optimal first-stage maintenance strategy scheme for the unit.

[0124] S706: Combining the optimal preparation decision combination and the minimum scenario cost, determine the first-stage preparation decision variables and the second-stage maintenance window decision variables for each component in the unit to be overhauled.

[0125] Specifically, after determining the optimal first-stage combination, it is substituted into the second-stage optimization model. Combined with the optimal maintenance window allocation results for each degradation scenario, the maintenance preparation status of each component in the first stage and the maintenance window arrangement in each scenario are finally determined, forming a complete two-stage optimal maintenance execution plan.

[0126] It should be noted that by constructing a first-stage decision combination set and a second-stage maintenance window optimization process, the unit maintenance problem is decomposed into a hierarchical structure of "strategy generation—window mapping—scenario optimization—expectation assessment." Compared with traditional methods based solely on a single rule or static optimization, this scheme can explicitly characterize the constraint impact of the first-stage maintenance strategy on the feasible region of subsequent maintenance windows, achieving strong coupling modeling between decision variables. Simultaneously, through scenario cost decomposition and probability weighting mechanisms, maintenance costs under different degradation scenarios can be uniformly incorporated into the evaluation system, thereby significantly improving the global optimality, constraint consistency, and engineering feasibility of maintenance decisions in multi-scenario uncertain environments.

[0127] S8: Based on the decision variables of the first stage preparation and the decision variables of the second stage maintenance window, determine the condition-based maintenance decision scheme for the unit to be maintained.

[0128] Among them, the condition-based maintenance decision plan refers to a complete execution plan consisting of the first-stage maintenance strategy and the second-stage maintenance window arrangement, including the maintenance status of each component, maintenance time arrangement and collaborative maintenance organization method, which is used to guide the actual maintenance execution of the unit.

[0129] It should be noted that by integrating the decision variables of the first stage preparation with the decision variables of the second stage maintenance window, the results obtained from the phased optimization can be transformed into a unified and structured condition-based maintenance execution plan. Compared with the traditional method that only outputs a single maintenance suggestion or static maintenance plan, this step can fully express the end-to-end decision results of "strategy selection - time arrangement", so that the maintenance plan has both decision executability and scenario consistency, and can be directly used for on-site scheduling and resource allocation, thereby significantly improving the engineering implementation capability and execution efficiency of the unit maintenance plan.

[0130] For example, firstly, the first-stage preparation decision variables and the second-stage maintenance window decision variables obtained from the solution are collected and aligned to ensure that the decision results of each component under different degradation scenarios are complete and traceable. Then, based on the first-stage decisions, the maintenance status category of each component is determined, such as immediate maintenance, preparation for maintenance, or no maintenance for the time being. Next, based on the second-stage maintenance window decision variables, each component is mapped to a specific maintenance time window or maintenance batch. Subsequently, the maintenance arrangements between different components are integrated to form a collaborative maintenance structure that considers resource constraints and time sequence. Finally, the maintenance status and maintenance window information of all components are summarized and output to form a complete condition-based maintenance decision scheme to guide the actual maintenance execution of the unit.

[0131] In this embodiment of the invention, by inputting multimodal features into the degradation-aware health representation network, a multidimensional health embedding feature matrix is ​​obtained, thereby achieving a deep and learnable representation of the unit's degradation state. By projecting the multidimensional health embedding onto an ideal health space, a standardized health score that satisfies the monotonic degradation law is generated, thus achieving a quantitative expression of the physical consistency of the health state. By constructing a set of degradation uncertainty scenarios based on the variance distribution of the health score, the evolutionary uncertainty of the unit under different degradation paths is characterized. By constructing a two-stage stochastic programming model, joint optimization decision-making of maintenance preparation strategy and maintenance window allocation is achieved. By solving the stochastic programming model, the maintenance strategy and maintenance window arrangement of each component are obtained, thereby forming an optimal condition maintenance decision scheme for the coordinated operation of multiple components of the unit.

[0132] Reference manual attached Figure 4 The diagram shows a structural schematic of a unit condition-based maintenance decision system based on multidimensional health scoring provided by an embodiment of the present invention.

[0133] This invention provides a unit condition-based maintenance decision system 20 based on multidimensional health scores, comprising: a processor 201 and a memory 202; The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-mentioned unit condition maintenance decision-making method based on multidimensional health scores and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0134] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0135] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM).

[0136] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means; wherein the wireless means include infrared, wireless, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0137] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0138] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0139] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0140] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0141] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0143] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part 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 of 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.

[0144] This invention provides a readable storage medium that stores a program or instructions. When the program or instructions are executed by a processor, they implement the steps of the above-described unit condition maintenance decision-making method based on multidimensional health scores and achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications 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. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A unit condition-based maintenance decision-making method based on multidimensional health scores, characterized in that, include: S1: Collect timing data of the operating sensors of the unit to be inspected; S2: Perform time-frequency transformation on the time-series data of the operating sensor to obtain a multimodal feature input vector; S3: Input the multimodal feature input vector into the degradation-aware health representation network and output a multidimensional health embedding feature matrix; S4: Project the multidimensional health embedding feature matrix onto the ideal health space to obtain a standardized health score; S5: Based on the variance distribution characteristics of the standardized health scores, construct a set of degradation uncertainty scenarios; S6: Based on the set of degradation uncertainty scenarios and the combination probability of each degradation uncertainty scenario in the set of degradation uncertainty scenarios, construct a two-stage stochastic programming model; S7: Solve the two-stage stochastic programming model to obtain the first-stage preparation decision variables and the second-stage maintenance window decision variables for each component in the unit to be overhauled; S8: Determine the condition-based maintenance decision scheme for the unit to be maintained based on the decision variables of the first stage preparation and the decision variables of the second stage maintenance window.

2. The unit condition-based maintenance decision-making method based on multidimensional health scoring according to claim 1, characterized in that, The timing data from the operating sensors includes: vibration status data, temperature status data, electrical status data, fluid status data, and operating condition data.

3. The unit condition-based maintenance decision-making method based on multidimensional health scores according to claim 1, characterized in that, S3 specifically includes: S301: Based on the multimodal feature input vector, construct a semi-supervised input sample set that matches the degradation-aware health representation network; S302: Input the semi-supervised input sample set into the degradation-sensing health representation network; S303: Using the nonlinear mapping function in the degradation-aware health representation network, each input sample in the semi-supervised input sample set is mapped to the health embedding space to obtain health space coordinates; S304: Determine the positional relationship of the input sample relative to the health hypersphere region based on the distance between the health spatial coordinates and the preset health center; S305: Determine the sample weight coefficients based on the aforementioned positional relationships; S306: Calculate the decentralized health embedding vector of each input sample relative to the preset health center by combining the sample weight coefficients; S307: Arrange the decentralized health embedding vectors in chronological order to obtain the initial health embedding matrix; S308: Construct a Gram matrix between the health embedding dimensions based on the initial health embedding matrix; S309: Combine the Gram matrix and the diversity constraint function to decouple the health embedding dimension and determine the multidimensional health embedding feature matrix.

4. The unit condition-based maintenance decision-making method based on multidimensional health scores according to claim 1, characterized in that, Specifically, S4 is: By alternating projection processing and order-preserving constraint processing, the multidimensional health embedding feature matrix is ​​mapped to the ideal health space to obtain a monotonically increasing standardized health score.

5. The unit condition-based maintenance decision-making method based on multidimensional health scores according to claim 4, characterized in that, S4 specifically includes: S401: Arrange the multidimensional health embedding feature matrix in a temporal sequence to obtain the health embedding matrix to be projected; S402: Construct the ideal health space based on the time threshold of the healthy stage and the time threshold of the severe degradation stage; S403: Combine the health embedding matrix to be projected and the ideal health space to determine the ideal health trajectory; S404: Under the constraint of the ideal health trajectory, update the health embedding weights of the degenerate perception health representation network to obtain the first health trajectory in the current iteration; S405: Under the time evolution constraint of the first health trajectory, the health embedding matrix to be projected is projected into the ideal health space to generate a second health trajectory that satisfies the order-preserving constraint and is monotonically increasing. S406: Perform boundary normalization processing on the second health trajectory to obtain the standardized health score.

6. The unit condition-based maintenance decision-making method based on multidimensional health scoring according to claim 1, characterized in that, The set of degradation uncertainty scenarios includes: slow degradation scenarios, expected degradation scenarios, and fast degradation scenarios.

7. The unit condition-based maintenance decision-making method based on multidimensional health scores according to claim 1, characterized in that, S6 specifically includes: S601: Based on the set of degradation uncertainty scenarios, determine the combination of degradation scenarios corresponding to each component in the unit to be repaired; S602: Calculate the combination probability of each of the aforementioned combinations of degradation scenarios; S603: Define the decision variables for the first stage preparation and the decision variables for the second stage maintenance window, respectively; S604: Calculate the failure risk cost of the component under maintenance by combining the degradation state and maintenance window of the component under maintenance in the corresponding degradation scenario combination; S605: Combining the decision variables of the first stage preparation and the decision variables of the second stage maintenance window, determine the planning objective function with the goal of minimizing the expected total maintenance cost; S606: Under the constraints of the first-stage preparation decision variables and the second-stage maintenance window decision variables, the two-stage stochastic programming model is constructed by minimizing the planning objective function.

8. The unit condition-based maintenance decision-making method based on multidimensional health scores according to claim 7, characterized in that, Specifically, S7 includes: S701: Based on the decision preparation variables of the first stage, generate a set of decision preparation combinations that satisfy the mutual exclusion constraints of the first stage; S702: Based on the prepared decision combination set, determine the set of optional maintenance windows for each of the components to be inspected in the second stage; S703: Calculate the minimum scenario cost of the second-stage maintenance window decision variables within the set of optional maintenance windows; S704: Combining the combination probabilities of each of the aforementioned degradation scenario combinations, the minimum scenario cost is weighted to obtain the expected total maintenance cost corresponding to each preparation decision combination in the preparation decision combination set. S705: Compare the expected total maintenance costs of each item to determine the optimal preparation decision combination; S706: Combining the optimal preparation decision combination and the minimum scenario cost, determine the first-stage preparation decision variables and the second-stage maintenance window decision variables for each component in the unit to be repaired.

9. A unit condition-based maintenance decision-making system based on multidimensional health scoring, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the unit condition maintenance decision-making method based on multidimensional health scores as described in any one of claims 1 to 8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the unit condition-based maintenance decision-making method as described in any one of claims 1 to 8.

Images