Decision-making method for maintenance of complex electromechanical systems under over-temperature conditions

By constructing an over-temperature cumulative index and a thermo-mechanical coupling term, and combining the covariance matrix adaptive evolution strategy to optimize the model parameters, the problem of the difficulty in characterizing the nonlinear coupling relationship between temperature and load in the existing technology is solved, and accurate prediction of maintenance timing and improvement of system reliability under over-temperature conditions are achieved.

CN123070937APending Publication Date: 2026-11-10ROCKET FORCE UNIV OF ENG
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611549126.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-08
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing maintenance timing decision-making methods based on confidence rule bases are unable to accurately characterize the nonlinear coupling relationship between temperature and load under temperature exceeding limits, and cannot reflect the nonlinear acceleration characteristics of device degradation. Furthermore, the fixed rule weights cannot be dynamically adjusted, resulting in insufficient evaluation accuracy and limited generalization ability.

Method used

A super-temperature cumulative index is constructed, a state-dependent thermo-mechanical coupling term is introduced, and the model parameters are optimized through an adaptive evolution strategy of the covariance matrix. An exponential mapping function is established, and the rule weights are dynamically modulated to realize the expression of the coupling relationship between temperature and load and the characterization of nonlinear degradation law.

Benefits of technology

It improves the accuracy of maintenance timing prediction and the safety and reliability of system operation under ultra-high temperature conditions, can accurately reflect the nonlinear coupling effect of temperature and load, and enhances the model's adaptability and evaluation accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070937A_ABST
    Figure CN123070937A_ABST
Patent Text Reader

Abstract

This invention discloses a method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions, belonging to the field of health management of complex electromechanical systems. The method collects ambient temperature, monitoring device surface temperature, and load values ​​to construct and normalize an extreme temperature accumulation index. Using normalized ambient temperature, temperature difference, load, and the normalized extreme temperature index as input attributes, a confidence rule base performance evaluation model is constructed. Rule weights are dynamically modulated based on the normalized extreme temperature index to construct state-related thermo-mechanical coupling terms and calculate activation weights. An adaptive evolutionary strategy using the covariance matrix is ​​employed to jointly optimize model parameters, and an evidence-based reasoning algorithm is used to fuse rule outputs and calculate the output utility value. The output utility value is normalized to obtain a health score, and an exponential maintenance timing function is constructed based on the health score to output the optimal maintenance timing. This invention can accurately characterize the temperature accumulation effect and the temperature-load nonlinear coupling effect, achieving accurate prediction of maintenance timing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of health management of complex electromechanical systems, and more specifically to a method for determining the timing of maintenance of complex electromechanical systems under extreme temperature conditions. Background Technology

[0002] Complex electromechanical systems frequently encounter operating conditions where temperatures exceed design thresholds. These include prolonged high-intensity operation leading to increased component temperatures, high-temperature overload during sudden tasks, and thermal stress caused by rising ambient temperatures. Exceeding temperature limits significantly accelerates the performance degradation of electronic components, making the system more susceptible to phenomena such as zero-bias drift, noise amplification, material aging, and structural fatigue, thereby increasing the risk of failure. For complex electromechanical systems in fields such as aerospace, rail transportation, and energy equipment, which consist of highly coupled subsystems including mechanical, electronic, and control systems, health status assessments and maintenance timing decisions under temperature-exceeding conditions directly impact the safety and reliability of system operation.

[0003] Traditional methods for maintenance decision-making in complex electromechanical systems primarily employ scheduled maintenance or reactive repair. Scheduled maintenance, performed on a fixed cycle, fails to reflect the dynamic changes in the system's health status; reactive repair, implemented only after a fault occurs, carries significant downtime risks and safety hazards. To overcome the shortcomings of traditional methods, condition monitoring-based maintenance timing decision-making methods have gained increasing attention. Among these, confidence rule base models, capable of integrating expert knowledge with monitoring data and handling uncertain information, have been applied in the field of health assessment for complex electromechanical systems. Introducing confidence rule bases into maintenance timing decision-making aims to achieve dynamic assessment of the system's health status by integrating multi-source monitoring information and expert knowledge, thereby providing a quantitative basis for determining maintenance timing.

[0004] However, existing maintenance timing decision-making methods based on confidence rule bases still face several technical challenges under conditions of excessive temperature.

[0005] First, existing methods generally lack a quantitative description of the cumulative effect of temperature exceeding limits. The impact of temperature on device degradation depends not only on the instantaneous temperature but also on factors such as the duration and amplitude of high-temperature exposure. The contributions of short-term high temperatures and long-term low temperatures to the degradation process differ significantly. Existing methods typically focus only on instantaneous temperature changes, neglecting the nonlinear cumulative relationship between high-temperature exposure time and temperature amplitude, making it difficult to accurately characterize the accelerated degradation process under temperature exceeding limits.

[0006] Secondly, there is a mutually reinforcing coupling relationship between temperature and load, which existing models struggle to reflect. High temperatures alter the mechanical properties of materials, increasing the stress response under the same load level; conversely, high loads cause internal heating within the device, further raising the temperature. This creates a positive feedback loop, resulting in a significant nonlinear acceleration of the degradation rate. This nonlinear synergistic effect cannot be described by simply superimposing the effects of temperature and load. Traditional confidence rule base models have fixed rule weights, which cannot be dynamically adjusted with temperature changes and fail to reflect the coupling relationship between temperature and load at the rule activation level, leading to significantly insufficient accuracy in performance evaluation under high-temperature conditions.

[0007] Furthermore, parameters such as rule confidence, rule weight, and attribute weight in confidence rule base models are usually set based on human experience, which is highly subjective and lacks adaptive optimization capabilities, making it difficult to adapt to different operating conditions. In addition, existing maintenance timing inference methods mostly use empirical thresholds or simple linear mappings, which cannot reflect the nonlinear characteristics of degradation rate changes with temperature, and it is also difficult to establish a continuous quantitative relationship that matches the degree of degradation. This results in a large bias in maintenance timing prediction, which limits the generalization ability and robustness of the model.

[0008] To address the impact of temperature on maintenance timing decisions, existing technologies attempt to incorporate temperature, along with other environmental stresses, as an overload factor. For example, Chinese invention patent CN121836691A discloses a method for determining the maintenance timing of complex electromechanical systems under overload conditions. This method improves the model's sensitivity to overload conditions by constructing an overload cumulative index and embedding its matching degree into the activation weights of a confidence rule base. This method treats environmental stresses such as temperature, pressure, vibration, and current as unified overload factors, calculates the cumulative index using a linear weighted summation method, and introduces the matching degree of each overload factor as an independent factor into the activation weights. In the activation weight calculation of this method, the temperature matching degree and the load matching degree are multiplied as independent factors, with no interaction term between them. Therefore, the correlation between the cumulative characteristics of temperature overload and the load state is not included in the modeling scope of this method.

[0009] Therefore, how to independently characterize the cumulative effect of temperature overload at the activation weight level of the confidence rule base, establish the dynamic coupling relationship between temperature and load, and construct a maintenance timing mapping mechanism that can reflect the nonlinear degradation law are technical problems that urgently need to be solved in this field. Summary of the Invention

[0010] To address the shortcomings of existing technologies, this invention proposes a method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions. By constructing an extreme temperature cumulative index, introducing state-related thermo-mechanical coupling terms into the activation weights of the confidence rule base, employing an adaptive evolutionary strategy based on the covariance matrix to jointly optimize model parameters, and utilizing an exponential mapping function to establish a continuous correspondence between health status and maintenance timing, this method enables accurate prediction of maintenance timing for complex electromechanical systems under extreme temperature conditions, thereby improving the safety and reliability of system operation.

[0011] To achieve the above objectives, the present invention provides the following technical solution:

[0012] A method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions includes the following steps:

[0013] Step 1: Collect multiple sets of historical operating data for the complex electromechanical system. Each set of historical operating data includes the ambient temperature value, the surface temperature value of the monitoring device in the complex electromechanical system, and the load value of the monitoring device. Construct an instantaneous overtemperature quantity based on the ambient temperature value and the surface temperature value of the monitoring device. Construct an overtemperature accumulation index based on the instantaneous overtemperature quantity and the power penalty factor. Normalize the overtemperature accumulation index to obtain a normalized overtemperature index value. Normalize the ambient temperature values ​​in the multiple sets of historical operating data to obtain normalized ambient temperature values.

[0014] The normalized ambient temperature, temperature difference, monitoring device load, and normalized over-temperature index of each set of historical data constitute an input sample. The temperature difference is the difference between the surface temperature of the monitoring device and the ambient temperature. All input samples are divided into training set and validation set.

[0015] Step 2: Using normalized ambient temperature, temperature difference, monitoring device load, and normalized over-temperature index as input attributes, construct a confidence rule base performance evaluation model:

[0016] The performance status of the monitoring device is divided into: The confidence rule base performance evaluation model has several levels, and the rule consequents are the confidence distributions at each performance state level.

[0017] The matching degree of each input attribute in each rule is calculated based on the reference level and membership function of the input attributes; the rule weights are dynamically modulated based on the normalized hypertemperature index to obtain the modulated rule weights.

[0018] A state-related thermo-mechanical coupling term is constructed based on the matching degree of the monitoring device load and the matching degree of the normalized overtemperature index. The activation weight of each rule is calculated based on the modulated rule weight, the matching degree of the input attribute, and the thermo-mechanical coupling term.

[0019] Step 3: Input the training samples in the training set into the confidence rule base performance evaluation model, multiply the rule confidence of each rule by the activation weight, and obtain the rule output with activation weight;

[0020] An evidence-based reasoning algorithm is used to fuse the outputs of all rules with activation weights to obtain the performance state level confidence distribution of the training samples. The output utility value of the training samples is then calculated based on the performance state level confidence distribution and the performance level utility vector.

[0021] Using the minimization of mean square error as the objective function, an adaptive evolution strategy based on the covariance matrix is ​​employed to jointly optimize the rule confidence, rule weight, attribute weight, over-temperature modulation coefficient, and thermo-mechanical coupling parameters of the confidence rule base performance evaluation model, resulting in an optimized confidence rule base performance evaluation model.

[0022] Step 4: Input the sample to be tested into the optimized confidence rule base performance evaluation model, calculate the predicted utility value of the sample to be tested; normalize the predicted utility value to obtain the health score.

[0023] Step 5: Construct an exponential maintenance timing function based on the health status, and output the optimal maintenance timing based on the exponential maintenance timing function.

[0024] Further, in step 2, the expression for the thermo-mechanical coupling term is:

[0025]

[0026] In the formula, For the first Thermo-mechanical coupling terms of the rule, For the first The monitoring device load matching degree according to the rules, For the first The normalized overtemperature exponent matching degree of the rule, To monitor the threshold of device load matching, The threshold for the normalized overtemperature exponential matching degree. The load power factor, It is the power factor of the overtemperature exponent. and All are greater than 1. For the normalized supertemperature index, This is the state-dependent coupling coefficient function.

[0027] Furthermore, the state-related coupling coefficient function The expression is:

[0028]

[0029] In the formula, These are the basic values ​​for the coupling coefficient; This is the cumulative effect coefficient of temperature. This is the load influence coefficient; , and It is obtained through joint optimization using an adaptive evolutionary strategy for the covariance matrix.

[0030] Furthermore, in step 2, the formula for dynamically modulating the rule weights using the normalized supertemperature exponent is:

[0031]

[0032] In the formula, For the first Modulated rule weights of the rules For the first The rule weight of each rule For the first The overtemperature modulation coefficients of the rule are obtained by optimization using an adaptive evolutionary strategy for the covariance matrix. The normalized supertemperature index.

[0033] Furthermore, in step 2, the formula for calculating the activation weight of each rule is as follows:

[0034]

[0035] In the formula, For the first The activation weight of the rule, For the first Modulated rule weights of the rules For the first Thermo-mechanical coupling terms of the rule, For the first The input attribute in the first The degree of matching in the rule, For the first The attribute weights of each input attribute. To input the number of attributes, This represents the total number of rules.

[0036] Furthermore, in step 3, the covariance matrix adaptive evolution strategy adopts a population parallel evolution strategy to generate multiple candidate solutions in the parameter space. Based on the mean square error value, the multiple candidate solutions are selected, reorganized, and the covariance matrix is ​​updated. The rule confidence, rule weight, attribute weight, over-temperature modulation coefficient, and thermo-mechanical coupling parameter are iteratively optimized. During the optimization process, a normalization constraint is applied to the rule confidence so that the sum of the rule confidence of each rule is 1.

[0037] Furthermore, in step 5, the expression for the exponential maintenance timing function is:

[0038]

[0039] In the formula, For health, The maintenance timing corresponding to the sample to be tested. To minimize maintenance time, This is the longest possible maintenance window. To adjust the parameters of the curve shape, .

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0041] (1) The effect of temperature on device degradation depends not only on the instantaneous temperature but also on the duration and amplitude of high-temperature exposure. The contributions of short-term high temperature and long-term low temperature to the degradation process differ significantly. This invention introduces a power-law penalty factor to sum the instantaneous overtemperature quantity exponentially, making the contribution of short-term high temperature to system degradation significantly higher than that of long-term low temperature conditions, thereby accurately characterizing the accelerated degradation process under temperature over-limit conditions. The normalized over-temperature index is used as an independent input attribute in the confidence rule base inference, enabling the model to directly perceive the impact of temperature accumulation levels on complex electromechanical systems.

[0042] (2) High temperature alters the mechanical properties of materials, increasing the stress response under the same load level; high load, in turn, causes internal heating of the device, further raising the temperature. This creates a positive feedback loop, resulting in a significant nonlinear acceleration of the degradation rate. This invention uses a threshold-triggered structure to amplify the thermo-mechanical coupling term only when both the load matching degree and the over-temperature index matching degree exceed the corresponding thresholds. A power-law amplification structure characterizes the nonlinear accelerated degradation under high temperature and high load conditions. A state-dependent coupling coefficient function dynamically changes the coupling strength with the over-temperature index and the load matching degree, thus achieving a joint coupling expression between the temperature accumulation effect and the load state. The thermo-mechanical coupling term significantly amplifies the activation intensity of degradation-related rules in the high temperature and high load range, while maintaining stable activation weights under normal operating conditions, enabling the model to accurately reflect the accelerated degradation trend caused by thermo-mechanical coupling.

[0043] (3) The rule weights in traditional confidence rule bases are fixed and cannot be dynamically adjusted with temperature changes, making it difficult to reflect the accelerating effect of high temperature on the degradation process. This invention dynamically modulates the rule weights based on the normalized hypertemperature index, so that the rule weights are adaptively adjusted with the temperature accumulation level. When the gyroscope is in a high-temperature state, the weights of degradation-related rules are amplified, and the model's sensitivity to thermal degradation is enhanced; when the gyroscope is in a normal temperature state, the rule weights remain stable and do not produce additional amplification effects. The hypertemperature modulation coefficient, rule confidence, rule weights, attribute weights, and thermo-mechanical coupling parameters are jointly optimized by the covariance matrix adaptive evolution strategy, overcoming the subjectivity of manually set parameters, enabling the model parameters to be adaptively adjusted according to the training data, thereby improving the evaluation accuracy and generalization ability of the model under hypertemperature conditions.

[0044] (4) Traditional methods for inferring maintenance timing often employ empirical thresholds or simple linear mappings, which fail to reflect the nonlinear characteristics of degradation rate changes with temperature and are prone to sudden decision-making changes near the threshold. This invention normalizes the utility value to a health level and continuously maps the health level to maintenance timings through an exponential maintenance timing function, making the maintenance timings monotonically increase with the health level. The shortest maintenance timing is output when the health level is zero, and the longest maintenance timing is output when the health level is highest. Compared to linear mapping, exponential mapping makes the maintenance timings change faster in the lower health level range. When the system performance slightly declines, the maintenance timings can be significantly shortened. In a healthy state, the maintenance timings remain at a high level, avoiding sudden decision-making changes caused by traditional piecewise threshold methods. Attached Figure Description

[0045] Figure 1 This is a flowchart of the method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions according to the present invention;

[0046] Figure 2 A scatter plot of normalized ambient temperature values ​​for each group of historical time data;

[0047] Figure 3 A scatter plot of normalized surface temperature values ​​for each group of historical time data;

[0048] Figure 4 A scatter plot of load current for each group of historical running data;

[0049] Figure 5 This is a comparison chart of the predicted utility values ​​and actual utility values ​​of the model of this invention, comparative model 1, and comparative model 2 on the validation set;

[0050] Figure 6 This is a diagram showing the optimal maintenance timing on the validation set obtained using the model of this invention. Detailed Implementation

[0051] The technical solutions of the embodiments 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, and 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.

[0052] Example

[0053] This embodiment uses the gyroscope in an aircraft inertial navigation system as the specific experimental object. The aircraft inertial navigation system is a typical complex electromechanical system. As the core sensitive device in the aircraft inertial navigation system, the gyroscope's output performance is highly sensitive to temperature changes. Under excessive temperature conditions, it is prone to thermal degradation characteristics such as increased zero-bias drift, increased noise level, and decreased output stability.

[0054] refer to Figure 1 This embodiment proposes a method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions, which is achieved through the following steps:

[0055] Step 1: Collect multiple sets of historical operating data for the complex electromechanical system. Each set of historical operating data includes the ambient temperature value, the surface temperature value of the monitoring device in the complex electromechanical system, and the load value of the monitoring device. Construct an instantaneous overtemperature quantity based on the ambient temperature value and the surface temperature value of the monitoring device. Construct an overtemperature cumulative index based on the instantaneous overtemperature quantity and the power penalty factor. Normalize the overtemperature cumulative index to obtain a normalized overtemperature index value. Normalize the ambient temperature values ​​in the multiple sets of historical data to obtain normalized ambient temperature values. The normalized ambient temperature value, temperature difference value, monitoring device load value, and normalized overtemperature index value of each set of historical data constitute an input sample, where the temperature difference value is the difference between the surface temperature value of the monitoring device and the ambient temperature value. Divide all input samples into a training set and a validation set.

[0056] In this embodiment, multiple sets of historical operational data from the gyroscope in the aircraft's inertial navigation system are collected. Each set of historical operational data includes the ambient temperature, the gyroscope surface temperature, and the gyroscope load value, where the load is the load current. This embodiment applies temperature stress to the gyroscope using a staged thermal cycling method. Ambient temperature, gyroscope surface temperature, and gyroscope load current are collected synchronously only within the stable operating range of each stage. No transitional samples are collected during heating, cooling, or operating condition switching. The first stage collects operational data after the ambient temperature rises; the second stage collects operational data after the gyroscope surface temperature rises; and the third and fourth stages collect operational data after cooling and recovery. 200 sets of historical operational data are collected in each stage, resulting in a total of 800 sets of historical operational data.

[0057] Construct the instantaneous overtemperature quantity based on the ambient temperature and the gyroscope surface temperature:

[0058]

[0059] In the formula, For the first Instantaneous temperature exceedance data from historical moments in operation. For the first Ambient temperature values ​​from historical data points. For the first Surface temperature values ​​from historical data points. This definition ensures that an over-temperature condition is only considered to exist when the surface temperature is higher than the ambient temperature, thus avoiding an overestimation of normal temperature differences.

[0060] Based on the instantaneous overtemperature quantity and the power-law penalty factor, an overtemperature accumulation index is constructed:

[0061]

[0062] in, For the first The cumulative temperature index of the historical data at different times. For the first Instantaneous temperature exceedance data from historical moments in operation. It is a power-order penalty factor and In this embodiment, The power-law penalty factor is used to amplify the effects of high-temperature conditions, making the contribution of short-term high temperatures to system degradation significantly higher than that of long-term low-temperature conditions.

[0063] To facilitate its integration with other attributes in confidence rule base inference, the supertemperature cumulative index is normalized to obtain the normalized supertemperature index value:

[0064]

[0065] in, For the first Normalized overtemperature index values ​​of historical runtime data. This is the minimum value of the cumulative temperature index among 800 sets of historical data. The normalized overtemperature index is the maximum value of the cumulative overtemperature index from 800 sets of historical data. The normalized overtemperature index ranges from [0,1] and can be directly used as the fourth input attribute of the confidence rule base performance evaluation model to characterize the degree of thermal degradation of the system throughout the entire operation process.

[0066] The ambient temperature values ​​from 800 sets of historical data were normalized to obtain normalized ambient temperature values:

[0067]

[0068]

[0069] In the formula, For the first Normalized ambient temperature values ​​from historical runtime data. This represents the minimum ambient temperature among 800 sets of historical data points. This represents the maximum ambient temperature from 800 sets of historical data. For the first Normalized surface temperature values ​​from historical data points. This represents the minimum surface temperature among 800 sets of historical data points. This represents the maximum surface temperature from 800 sets of historical data.

[0070] Figure 2 The normalized ambient temperature values ​​of the historical data for each group are displayed. Figure 3 The normalized surface temperature values ​​of the historical data for each set of time points are displayed. Figure 4 The load current values ​​of the historical operating data for each group are displayed.

[0071] For each set of historical data, the normalized ambient temperature, temperature difference, gyroscope load, and normalized overtemperature index constitute an input sample, where the temperature difference is... .

[0072] The 800 input samples were randomly divided into training and validation sets in a 1:1 ratio.

[0073] Step 2: Using normalized ambient temperature, temperature difference, monitoring device load, and normalized over-temperature index as input attributes, construct a confidence rule base performance evaluation model.

[0074] A confidence rule base performance evaluation model is constructed using normalized ambient temperature, temperature difference, gyroscope load, and normalized overtemperature index as input attributes.

[0075] In this embodiment, the performance status of the gyroscope is divided into four levels: ,in For slight thermal drift, Corresponding to mild degradation, Corresponding to moderate degradation, The corresponding degradation is significant.

[0076] Let the normalized ambient temperature be... The temperature difference is The gyroscope load is The normalized supertemperature index is The reference levels and quantification values ​​for each input attribute are shown in Table 1.

[0077] Table 1 Reference levels and quantification values ​​of input attributes

[0078]

[0079] The reference level quantification value for each input attribute is determined based on the range and distribution characteristics of the collected data for that attribute, so that each reference level can cover the typical range of the collected data.

[0080] Calculate the degree of matching between the input attribute and each reference level based on the reference level and quantification value of the input attribute. Assume... If the value is 0.6, then:

[0081] The normalized ambient temperature is a "high" level of fit for the reference. for:

[0082]

[0083] The normalized ambient temperature matches the reference level of "medium-high". for:

[0084]

[0085] The normalized ambient temperature is a good match for the reference level "low". .

[0086] For input samples The degree of matching for each input attribute in each rule is calculated based on the membership function:

[0087]

[0088] In the formula, For the first input sample The input attribute in the first In this embodiment, the matching degree in the rule is... , For the first The input attribute in the first In this embodiment, the membership function in the rule is a trigonometric function. For the first in the input sample Each input attribute, namely for , for , for , for .

[0089] Construct a system based on the reference level combination relationship of the input attributes, containing... A rule base of rules. Each rule consists of input conditions and output performance status level confidence, generally in the following form:

[0090]

[0091] In the formula, For the first Rule 1 For the first The input attribute in the first The reference level in the rule, The first gyroscope performance status Each level For the first The rule is Confidence level on ; .

[0092] The rule weights are dynamically modulated based on the normalized overtemperature exponent to obtain the modulated rule weights:

[0093]

[0094] In the formula, For the first Modulated rule weights of the rules For the first The rule weight of each rule For the first The overtemperature modulation coefficients are obtained by an adaptive evolutionary strategy to optimize the covariance matrix. The parameters to be optimized are initially set to 0.01, with a range of [0,1]. They are jointly optimized along with the confidence level, rule weights, attribute weights, and thermal-mechanical coupling parameters using an adaptive evolutionary strategy for the covariance matrix. The rule weights are dynamically modulated based on the normalized hypertemperature index, so that when the gyroscope is in a high-temperature state, the weights of degradation-related rules are amplified, thereby enhancing the model's sensitivity to thermal degradation.

[0095] State-dependent thermo-mechanical coupling terms are constructed based on the matching degree of the gyroscope load and the matching degree of the normalized overtemperature index in each rule:

[0096]

[0097] In the formula, For the first Thermo-mechanical coupling terms of the rule, For the first The gyroscope load matching degree according to the rules, For the first The normalized overtemperature exponent matching degree of the rule, The threshold for gyroscope load matching. The threshold for the normalized overtemperature exponential matching degree. The load power factor, It is the power factor of the overtemperature exponent. and All are greater than 1. This is the state-dependent coupling coefficient function.

[0098] The state-dependent coupling coefficient function is:

[0099]

[0100] In the formula, This is the base value for the coupling coefficient, used to set the baseline level for the coupling coefficient; This is the temperature cumulative effect coefficient, used to adjust the degree of influence of the normalized overtemperature exponent on the coupling strength; This is the load influence coefficient, used to adjust the degree of influence of load attribute matching on coupling strength; , and All parameters are to be optimized and are obtained through joint optimization using an adaptive evolution strategy for the covariance matrix.

[0101] The state-dependent coupling coefficient function allows the coupling strength to vary with... and Dynamic changes, thereby achieving The temperature accumulation effect characterized and The joint coupling expression between the load states is represented. The constant 1 in the thermo-mechanical coupling term ensures that the activation weight remains stable under normal operating conditions, while the coupling term will significantly amplify the activation intensity of the degradation correlation rule in the high temperature and high load range.

[0102] The activation weight of each rule is calculated based on the modulated rule weights, the matching degree of the input attributes, and the state-related thermo-mechanical coupling term:

[0103]

[0104] In the formula, For the first The activation weight of the rule, For the first The attribute weights of each input attribute. Enter the number of attributes.

[0105] Step 3: Using the minimum mean square error as the objective function, the confidence level, rule weight, attribute weight, over-temperature modulation coefficient, and thermo-mechanical coupling parameters of the confidence rule base performance evaluation model are jointly optimized using the covariance matrix adaptive evolution strategy to obtain the optimized confidence rule base performance evaluation model.

[0106] In this embodiment, the training set partitioned in step 1 is used for parameter optimization of the confidence rule base performance evaluation model. The parameters to be optimized include rule confidence. Rule weight Attribute weight Over-temperature modulation coefficient And thermo-mechanical coupling parameters. Among these, the thermo-mechanical coupling parameters include the load matching threshold. Over-temperature index matching threshold Load power factor Supertemperature power factor State-related coupling coefficient parameters , and .

[0107] Rule confidence For the first The rule following the first clause corresponds to the first clause. There are 81 rules for the confidence level of each performance status level, with each rule corresponding to 4 performance status levels. Therefore, there are 81 × 4 parameters to be optimized for rule confidence. The initial value of the rule confidence is set according to the frequency of occurrence of each performance level and satisfies the normalization constraint that the sum of the output confidence of each rule is 1. For example, for a certain rule, if the frequencies of occurrence of "Excellent", "Good", "Medium", and "Poor" in its corresponding samples are 0.4, 0.3, 0.2, and 0.1, respectively, then the initial confidence of the consequent of this rule is set to... , , , .

[0108] Rule weight For the first The rule weights of the rules are 81 parameters to be optimized, and the initial values ​​of all of them are set to 1.

[0109] Attribute weight For the first The attribute weights of the input attributes are 4 parameters to be optimized, and the initial value of each is set to 1.

[0110] Over-temperature modulation coefficient For the first The overtemperature modulation coefficients for each rule consist of 81 parameters to be optimized, all initially set to 0.01. These overtemperature modulation coefficients are used to control the amplification effect of the normalized overtemperature exponent on the rule weights.

[0111] Load matching threshold and temperature matching threshold The initial values ​​are all set to 0.3. The load matching threshold is used to control the critical value at which load attribute matching triggers coupling amplification, and the over-temperature matching threshold is used to control the critical value at which over-temperature attribute matching triggers coupling amplification.

[0112] Load power factor Supertemperature power factor The initial values ​​are all set to 2. The load power factor is used to characterize the degree of nonlinear accelerated degradation after the load matching degree exceeds the load matching degree threshold, and the overtemperature power factor is used to characterize the degree of nonlinear accelerated degradation after the overtemperature matching degree exceeds the overtemperature matching degree threshold.

[0113] , , The initial values ​​are all set to 0.1.

[0114] During parameter optimization, training samples from the training set are input into the confidence rule base performance evaluation model, following step 2. The calculation method calculates the first The training sample corresponding to the th training sample Activation weight of the rule In this embodiment, Confidence level With activation weight By combining these, we obtain the rule output with activation weights. .

[0115] An evidence-based reasoning algorithm is used to fuse the rule outputs with activation weights corresponding to all rules to obtain the first rule. The training sample at the th ... Performance status levels Fusion confidence The fusion process satisfies the normalization constraint. In this embodiment, the total number of performance levels =4.

[0116] In obtaining the first Confidence distribution of performance state levels of training samples Then, the first [level] is calculated based on the confidence distribution of the performance state levels and the utility value of each performance state level. Output utility value of each training sample :

[0117]

[0118] in, The performance level utility vector is composed of the utility values ​​of each performance state level. for The transpose of. In this embodiment... That is, the utility value corresponding to the performance level "Excellent" is 4, the utility value corresponding to the performance level "Good" is 3, the utility value corresponding to the performance level "Medium" is 2, and the utility value corresponding to the performance level "Poor" is 1.

[0119] Using the minimization of mean squared error as the objective function, the formula for calculating mean squared error is:

[0120]

[0121] In the formula, Mean square error, The number of training samples. For the first The true utility value corresponding to each training sample is determined based on the true performance state level and performance level utility vector of the training sample. In this embodiment, the true performance state level is given by experts based on the actual degradation degree of the gyroscope.

[0122] The covariance matrix adaptive evolution strategy employs a population-parallel evolution approach, generating multiple candidate solutions in the parameter space. Based on the objective function value, it selects, reorganizes, and updates the covariance matrix of these candidate solutions, iteratively optimizing rule confidence, rule weights, attribute weights, hypertemperature modulation coefficients, and thermo-mechanical coupling parameters. During optimization, a normalization constraint is applied to the rule confidence, ensuring that the sum of the output confidence of each rule is 1. .

[0123] In this embodiment, the population size of the covariance matrix adaptive evolution strategy is set to 50, the maximum number of iterations is set to 300, the initial step size is set to 0.5, and the covariance matrix is ​​initially set to the identity matrix. After 300 iterations, the optimization process converges, yielding the optimal parameter vector. Based on the optimal parameter vector, the optimized confidence rule base performance evaluation model is obtained.

[0124] Step 4: Input the sample to be tested into the optimized confidence rule base performance evaluation model, and obtain the predicted utility value corresponding to the sample to be tested according to the calculation method of the output utility value in Step 3.

[0125] The normalization benchmarks for the ambient temperature value and the cumulative over-temperature index in the test sample are consistent with those of the training sample. That is, the ambient temperature value of the test sample is normalized by the minimum and maximum values ​​of the ambient temperature in 800 sets of historical data. The cumulative over-temperature index of the test sample is normalized by the minimum and maximum values ​​of the cumulative over-temperature index in 800 sets of historical data.

[0126] Health level is calculated based on predicted utility value. :

[0127]

[0128] In the formula, The predicted utility value corresponding to the sample to be tested. This is the minimum reference value for utility. This is the maximum reference value for utility. In this embodiment, The value is the utility value corresponding to the performance level "poor". The value corresponds to the utility value of the performance level "Excellent". The higher the health level, the better the gyroscope performance.

[0129] Step 5: Construct an exponential maintenance timing function based on health status:

[0130]

[0131] In the formula, This refers to the maintenance timing corresponding to the sample under test, i.e., the time interval from the sampling time of the sample to the next maintenance. To minimize maintenance time, This is the longest possible maintenance window. To adjust the parameters of the curve shape and >0. The exponential maintenance timing function monotonically increases with health, and outputs the shortest maintenance timing when health is zero.

[0132] In this embodiment, the shortest maintenance time is set according to the actual maintenance requirements of the gyroscope. =40 days, longest maintenance period =120 days, adjust the parameters of the curve shape =3. Apply the exponential maintenance timing function to the test samples to obtain the optimal maintenance timing output for each test sample. The test samples can be obtained from online testing or reserved samples.

[0133] To better illustrate the beneficial effects of the present invention, a performance comparison was conducted on the confidence rule base performance evaluation model (the model of the present invention), comparison model 1, and comparison model 2 provided in this embodiment.

[0134] The difference between Comparative Model 1 and the model of this invention is that Comparative Model 1 is a confidence rule base model that does not consider the over-temperature condition. Comparative Model 1 does not introduce the normalized over-temperature index as an input attribute, does not perform dynamic modulation of rule weights, and does not introduce thermo-mechanical coupling terms. It only uses normalized ambient temperature, temperature difference, and gyroscope load as input attributes and uses a confidence rule base to evaluate the performance of the gyroscope.

[0135] The difference between Comparative Model 2 and the model of this invention is that Comparative Model 2 is an improved confidence rule base model that only introduces the supertemperature cumulative index. Comparative Model 2 introduces the normalized supertemperature index as an input attribute and performs dynamic modulation of rule weights, but does not introduce thermo-mechanical coupling terms. That is, the activation weight calculation of Comparative Model 2 does not include state-related thermo-mechanical coupling terms.

[0136] Under the training set partitioned in step 1 and the same covariance matrix adaptive evolution strategy optimization conditions, the model of this invention, the comparative model 1 and the comparative model 2 were trained respectively, and the prediction utility values ​​of the three models were compared on the validation set partitioned in step 1. Figure 5 The predicted utility values ​​and actual utility values ​​of the model of this invention, comparative model 1, and comparative model 2 are shown; from Figure 5 It can be seen that the predicted utility value of the model in this invention is closest to the actual utility value. The mean squared error of the model in this invention is calculated to be 0.0193, compared to 0.0944 for comparative model 1 and 0.0259 for comparative model 2. This indicates that the over-temperature modulation mechanism and thermo-mechanical coupling structure proposed in this invention can significantly improve the model's ability to characterize degradation under over-temperature conditions, giving the model higher sensitivity and accuracy under high temperature and high load conditions.

[0137] Figure 6 This demonstrates the optimal maintenance timing on the validation set obtained using the model of this invention, from... Figure 6 As can be seen, there are significant differences in the timing of maintenance under different temperature conditions. This indicates that the maintenance timing obtained using the model of this invention can reflect the differences in the degree of degradation of the gyroscope under different temperature conditions.

[0138] The specific embodiments of the present invention enable those skilled in the art to understand or implement the invention. Various modifications to the embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention.

[0139] It should be understood that the present invention is not limited to the content already described above, and various modifications and changes can be made without departing from its scope. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for determining the timing of maintenance for complex electromechanical systems under extreme temperature conditions, characterized in that, Includes the following steps: Step 1: Collect multiple sets of historical operating data for the complex electromechanical system. Each set of historical operating data includes the ambient temperature value, the surface temperature value of the monitoring device in the complex electromechanical system, and the load value of the monitoring device. Construct an instantaneous overtemperature quantity based on the ambient temperature value and the surface temperature value of the monitoring device. Construct an overtemperature accumulation index based on the instantaneous overtemperature quantity and the power penalty factor. Normalize the overtemperature accumulation index to obtain a normalized overtemperature index value. Normalize the ambient temperature values ​​in the multiple sets of historical operating data to obtain normalized ambient temperature values. The normalized ambient temperature, temperature difference, monitoring device load, and normalized over-temperature index of each set of historical data constitute an input sample. The temperature difference is the difference between the surface temperature of the monitoring device and the ambient temperature. All input samples are divided into training set and validation set. Step 2: Using normalized ambient temperature, temperature difference, monitoring device load, and normalized over-temperature index as input attributes, construct a confidence rule base performance evaluation model: The performance status of the monitoring device is divided into: The confidence rule base performance evaluation model has several levels, and the rule consequents are the confidence distributions at each performance state level. The matching degree of each input attribute in each rule is calculated based on the reference level and membership function of the input attributes; the rule weights are dynamically modulated based on the normalized hypertemperature index to obtain the modulated rule weights. A state-related thermo-mechanical coupling term is constructed based on the matching degree of the monitoring device load and the matching degree of the normalized overtemperature index. The activation weight of each rule is calculated based on the modulated rule weight, the matching degree of the input attribute, and the thermo-mechanical coupling term. Step 3: Input the training samples in the training set into the confidence rule base performance evaluation model, multiply the rule confidence of each rule by the activation weight, and obtain the rule output with activation weight; An evidence-based reasoning algorithm is used to fuse the outputs of all rules with activation weights to obtain the performance state level confidence distribution of the training samples. The output utility value of the training samples is then calculated based on the performance state level confidence distribution and the performance level utility vector. Using the minimization of mean square error as the objective function, an adaptive evolution strategy based on the covariance matrix is ​​employed to jointly optimize the rule confidence, rule weight, attribute weight, over-temperature modulation coefficient, and thermo-mechanical coupling parameters of the confidence rule base performance evaluation model, resulting in an optimized confidence rule base performance evaluation model. Step 4: Input the sample to be tested into the optimized confidence rule base performance evaluation model, calculate the predicted utility value of the sample to be tested; normalize the predicted utility value to obtain the health score. Step 5: Construct an exponential maintenance timing function based on the health status, and output the optimal maintenance timing based on the exponential maintenance timing function.

2. The method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions according to claim 1, characterized in that, In step 2, the expression for the thermo-mechanical coupling term is: In the formula, For the first Thermo-mechanical coupling terms of the rule, For the first The monitoring device load matching degree according to the rules, For the first The normalized overtemperature exponent matching degree of the rule, To monitor the threshold of device load matching, The threshold for the normalized overtemperature exponential matching degree. The load power factor, It is the power factor of the supertemperature exponent. and All are greater than 1. For the normalized supertemperature index, This is the state-dependent coupling coefficient function.

3. The method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions according to claim 2, characterized in that, The state-related coupling coefficient function The expression is: In the formula, These are the basic values ​​for the coupling coefficient; This is the cumulative effect coefficient of temperature. This is the load influence coefficient; , and It is obtained through joint optimization using an adaptive evolutionary strategy for the covariance matrix.

4. The method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions according to claim 1, characterized in that, In step 2, the formula for dynamically modulating the rule weights using the normalized supertemperature exponent is: In the formula, For the first Modulated rule weights of the rules For the first The rule weight of each rule For the first The overtemperature modulation coefficients of the rule are obtained by optimization using an adaptive evolutionary strategy for the covariance matrix. The normalized supertemperature index.

5. The method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions according to claim 1, characterized in that, In step 2, the formula for calculating the activation weight of each rule is as follows: In the formula, For the first The activation weight of the rule, For the first Modulated rule weights of the rules For the first Thermo-mechanical coupling terms of the rule, For the first The input attribute in the first The degree of matching in the rule, For the first The attribute weights of each input attribute. To input the number of attributes, This represents the total number of rules.

6. The method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions according to claim 1, characterized in that, In step 3, the covariance matrix adaptive evolution strategy adopts a population parallel evolution strategy to generate multiple candidate solutions in the parameter space. Based on the mean square error value, the multiple candidate solutions are selected, reorganized and the covariance matrix is ​​updated. The rule confidence, rule weight, attribute weight, over-temperature modulation coefficient and thermo-mechanical coupling parameter are iteratively optimized. During the optimization process, a normalization constraint is applied to the rule confidence, so that the sum of the rule confidence of each rule is 1.

7. The method for determining the maintenance timing of complex electromechanical systems under extreme temperature conditions according to claim 1, characterized in that, In step 5, the expression for the exponential maintenance timing function is: In the formula, For health, The maintenance timing corresponding to the sample to be tested. To minimize maintenance time, This is the longest possible maintenance window. To adjust the parameters of the curve shape, .

Citation Information

Patent Citations

  • Complex electromechanical system maintenance opportunity decision-making method under overload operation

    CN121836691A