Reliability assessment methods, equipment and media for key components of turbine governors

Abstract

The present invention belongs to the technical field of power system equipment reliability assessment and provides a reliability assessment method, device, and medium for key components of a turbine governor. The method comprises: applying a hierarchical analysis method to screen key components of the turbine governor; processing original fault operation and maintenance records of the key components to obtain a data set containing censored data; calculating the censoring rate based on the data set; and dynamically selecting the optimal method based on different censoring rates to complete the reliability assessment of the key components. The present invention achieves adaptive selection of parameter estimation methods by partitioning censoring rates into intervals, improving adaptability to data with varying censoring levels while also balancing accuracy and robustness under different censoring rates.

Description

Technical Field

[0001] The present invention belongs to the technical field of reliability assessment of power system equipment. Specifically, it relates to a reliability assessment method, equipment and medium for key components of a turbine governor. The method is suitable for life modeling, reliability analysis and maintenance decision optimization of key components such as complex electromechanical systems and turbine generator sets. Background Art

[0002] As the core control device in hydropower generation systems, turbine speed governors achieve unit speed control, power regulation, and grid frequency stability by adjusting the guide vane opening and blade angle in real time. Key components, including electro-hydraulic converters, PID (Proportional Integral Derivative) controllers, and servomotors, are subject to the combined effects of hydraulic pulsation, mechanical wear, and electrical aging. Failure of key governor components can directly lead to increased unit oscillation, decreased grid connection stability, and even unplanned downtime, resulting in significant economic losses. Current failure data records primarily include the time of failure, location, cause, severity, and handling. Due to the long service life and high reliability of governors, some components have experienced no functional failures from commissioning to the end of the observation period. Consequently, failure data records exhibit the characteristics of a timed right-truncation test and are generally highly censored.

[0003] The so-called reliability of a product refers to the ability of the product to complete the specified functions under specified conditions and within a specified time. Reliability analysis includes quantitative and qualitative analysis. Qualitative analysis mainly analyzes changes in system reliability from the perspectives of failure mechanism, failure mode, failure consequences, etc., and is generally related to the specific research object. Quantitative analysis mainly discusses the reliability of the system from quantitative indicators such as statistical analysis of failure data, estimation of reliability, and failure rate function. The reliability analysis process can provide necessary guidance for reliability modeling, including modeling methods and model selection. For the study of actual cases, establishing an accurate reliability model is the prerequisite for formulating a reasonable maintenance strategy. This section will apply reliability theory for quantitative analysis, study and determine the theoretical optimal regular maintenance cycles of various components of the unit, meet the personalized needs of users, and assist in making maintenance timing decisions.

[0004] In the reliability assessment of complex systems, Weibull distribution is widely used in life modeling due to its flexibility. Since the different key components of the turbine governor during operation have different operating characteristics, service life and reliability levels, the level of deletion of the obtained data set is also quite different. It is difficult to perform reliability assessment on all key components using a single assessment method. Therefore, there is a need for a complete reliability assessment solution that can dynamically select the optimal parameter method according to the deletion rate level. In addition, field data often presents the characteristics of "large sample, high deletion". The traditional parameter estimation method has the following problems: the maximum likelihood estimation method (MLM) has high deletion rate data, and the shape parameter ( β ) have significant estimation biases, and the mean time to failure (MTTF) is overestimated; the least squares method (LSM) relies on fitting empirical distribution functions, is sensitive to the distribution of censoring time, and has large error fluctuations; the Bayesian method requires prior information, the heuristic algorithm is highly complex, and none of them systematically quantifies the impact of censoring features. Therefore, there is an urgent need for a reliability assessment method that can use multiple hierarchical strategies, integrate censoring features, and effectively utilize censoring information to improve the estimation accuracy and robustness under highly censored data. After searching the patents of the existing technologies, it was found that the common methods are mainly the following:

[0005] Method 1: Chinese Patent Publication Number: CN112733281A, Patent Name: A Machine Tool Reliability Assessment Method Considering Censored Data. This patent discloses a machine tool reliability assessment method considering censored data to improve the accuracy of reliability assessment parameters. First, the normal data is screened based on the IQR (Interquartile Range) method. Secondly, the total failure time processing rule is used to reduce the censored data in the data and expand the sample size of failure data. Then, the average rank method is used to preliminarily obtain the initial value of the Weil parameter. Then, based on the new empirical distribution function, the least squares method is used to complete the new Weil parameter estimation. Finally, the KS method is used for convergence test. Method 1 focuses on the processing of censored data and the expansion of sample size, and improves the accuracy of machine tool reliability assessment through multi-stage statistical methods.

[0006] Method 2: Chinese Patent Publication Number: CN112507487A, Patent Name: A Reliability Assessment Method and System for a Turbine Governor Relay. This patent provides a reliability assessment method and system for a turbine governor relay, comprising: learning the mapping relationship between the state index of the turbine governor relay and the relay action rate based on a neural network and historical turbine data; determining the relay action rate at any time based on the state index of the turbine governor relay and the mapping relationship; determining the relay function value based on the relay action rate at any time and a preset relay minimum rate limit; determining the relay reliability based on the distribution of the relay function value in a standard normal coordinate system; the reliability reflects whether the relay's operating status meets the requirements. The present invention effectively fills a gap in the field of turbine governor servo reliability assessment, can obtain more accurate action rate values, and makes the calculated reliability closer to the actual system reliability. Method 2 focuses on dynamic mapping modeling and performance function analysis based on neural networks, filling the gap in the field of turbine servo reliability assessment.

[0007] Method 3: Chinese Patent Publication Number: CN111291486A, Patent Name: A Method for Reliability Assessment of CNC Machine Tool System Components. This patent relates to a reliability assessment method for CNC machine tool components, including the following steps: 1. Dividing the system components, collecting on-site CNC machine tool fault information, and performing fault analysis; 2. Calculating the equivalent failure interval time and equivalent test truncation time of the components and machine tool system; 3. Applying the Johnson method to perform rank correction on the equivalent failure interval time and constructing component and machine tool system reliability models; 4. Constructing a system series reliability model and applying the correlation index method to verify the rationality of component reliability modeling based on the equivalent sample method. Under the assumption of repair as new, the present invention applies the equivalent sample method to calculate the component failure interval time, which conforms to the definition of system component life. The method uses the total failure time method and the equivalent sample method to correct the component failure interval time, expands the sample size, and conforms to the sampling principle. Compared with traditional component reliability modeling methods based on system information, this method is more in line with engineering practice. Method 3 focuses on component-level reliability modeling and system-level verification, and expands the sample size through the equivalent sample method to improve the engineering applicability of the assessment. Summary of the Invention

[0008] The present invention provides a reliability assessment method, device and medium for key components of a turbine governor. A phased parameter estimation strategy is proposed for scenarios with high missing data, aiming to solve the problems of large parameter estimation deviation and poor robustness in traditional methods under high missing data conditions.

[0009] In a first aspect, the present invention provides a reliability assessment method for key components of a turbine governor, comprising:

[0010] The key components of turbine governor are selected by using analytic hierarchy process;

[0011] Processing the original fault operation and maintenance records of the key components to obtain a data set containing censored data;

[0012] Calculate the censoring rate based on the data set;

[0013] The optimal method is dynamically selected according to different deletion rates to complete the reliability assessment of key components.

[0014] In some embodiments, the step of applying the analytic hierarchy process to screen out key components of the turbine governor includes:

[0015] Establishing a hierarchical structure model according to the system structure of the turbine governor, and constructing a judgment matrix according to the hierarchical structure model;

[0016] Using consistency index to test the consistency of the judgment matrix;

[0017] After consistency is met, key components are screened by calculating component weights.

[0018] In some embodiments, processing the original fault operation and maintenance records of the key components to obtain a data set containing censored data includes:

[0019] Filter out the original fault operation and maintenance records of the key components according to the component names from the provided original fault operation and maintenance records and process them;

[0020] In the original failure operation and maintenance records of the key components, after normalizing the component commissioning date, the component failure data obtained by the end of the observation time conforms to the random truncation test, thereby obtaining the number of failed components and the component failure time. The operating time of the components without failure is right-truncated to obtain the missing time; since the life data of the speed regulator components obeys the Weibull distribution, after eliminating the outliers in the failure time and the missing time, finally a data set containing missing data is obtained.

[0021] In some embodiments, calculating the censoring rate based on the data set includes:

[0022]

[0023] in, c is the censoring rate, n is the number of failed parts, m is the number of parts that have not failed.

[0024] In some embodiments, dynamically selecting the optimal method according to different levels of deletion rates to complete the reliability assessment of key components includes:

[0025] For low deletion rates, the maximum likelihood estimation method is used to complete the reliability assessment of key components;

[0026] For high deletion rates, a step-by-step parameter estimation method based on integrated deletion characteristic information is used to complete the reliability assessment of key components;

[0027] For extremely high deletion rates, the reliability assessment of key components is completed by combining semi-supervised data conversion method and maximum likelihood estimation method;

[0028] The low censoring rate, high censoring rate and extremely high censoring rate are determined according to a preset censoring rate range.

[0029] In some embodiments, the reliability assessment of key components using a maximum likelihood estimation method includes:

[0030]

[0031] in, L is the likelihood function, n is the number of failed parts, m is the number of components that have not failed, t i Indicates the i The failure time of a fault sample, For the j The censoring time of the censored samples;

[0032] F ( t ) is the shape parameter of the Weibull distribution β and scale parameters η The failure distribution function is expressed as:

[0033]

[0034] f ( t ) is the shape parameter of the Weibull distribution β and scale parameters η The density distribution function is expressed as:

[0035]

[0036] in, , t is a random variable, namely the failure time or deletion time, and exp() is an exponential function with the natural constant e as the base.

[0037] In some embodiments, the step-by-step parameter estimation method using integrated censored feature information to complete the reliability assessment of key components includes:

[0038] Calculate the censoring feature statistics, including the censoring time coefficient of variation and the censoring time skewness; the censoring time coefficient of variation is the ratio of the standard deviation of the censoring time to the mean; the censoring time is obtained based on the standard deviation and mean of the censoring time;

[0039] Constructing a correction function based on the censoring rate and the censoring time skewness, and using the correction function to correct the mean life MTTF; wherein the correction function is a function based on the censoring rate and the censoring time skewness;

[0040] The corrected mean life MTTF is iteratively optimized based on the maximum likelihood estimation method:

[0041] Step 1: Fix the estimated value of the mean life MTTF corrected by the correction function, substitute it into the calculation formula of the mean life MTTF, substitute the calculated scale parameter into the maximum likelihood estimation method to estimate the shape parameter;

[0042] Step 2: fixing the estimated shape parameter, estimating the scale parameter using the maximum likelihood estimation method, and updating the estimated value of the mean time to failure (MTTF) using the estimated scale parameter;

[0043] Step 3: fix the updated estimated value of the mean time MTTF, re-estimate the shape parameters according to step 1, and determine whether the re-estimated shape parameters meet the convergence condition. If not, repeat steps 1 to 2 until the convergence condition is met.

[0044] In some embodiments, the reliability assessment of key components is completed by combining the semi-supervised data conversion method and the maximum likelihood estimation method, including:

[0045] Introducing operating condition parameters of turbine governor components into a data set and using them as feature vectors of data set samples; the data set includes fault samples and censored samples;

[0046] Combining feature similarity and survival curve similarity to construct a weight matrix;

[0047] Calculate the transfer matrix based on the weight matrix;

[0048] Construct an initial label matrix based on the failure time of the faulty samples and the censoring time of the censored samples;

[0049] Iteratively update the pseudo survival time based on the transition matrix and the initial label matrix to obtain the optimal pseudo label of the censored sample;

[0050] Replacing the censored samples in the data set with the optimal pseudo labels and treating them as faulty samples, thereby updating the data set;

[0051] Based on the updated data set, the reliability assessment of key components is completed using the maximum likelihood estimation method.

[0052] In a second aspect, the present invention provides an electronic device, comprising:

[0053] at least one processor; and a memory communicatively coupled to the at least one processor;

[0054] The memory stores instructions that can be executed by the at least one processor, and the at least one processor executes the above method by executing the instructions stored in the memory.

[0055] In a third aspect, the present invention provides a computer-readable storage medium for storing instructions, which implement the above method when the instructions are executed.

[0056] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0057] 1. Due to the differences in operating characteristics, service life and reliability levels of different key components of the turbine speed governor during operation, the acquired data set deletion levels also vary greatly. It is difficult to perform reliability evaluation on all key components using a single evaluation method. In response to this problem, the present invention can dynamically select the optimal method according to the deletion rate level. For example, when the deletion rate is <40%, the maximum likelihood estimation method (MLE) is used; when the deletion rate is 40% to 80%, the step-by-step parameter estimation method based on comprehensive deletion feature information is used; when the deletion rate is >80%, the semi-supervised data conversion method and the maximum likelihood estimation method are combined. The present invention realizes the adaptive selection of parameter estimation methods through deletion rate interval division, improves the adaptability to data with different deletion degrees, and can also take into account the accuracy and robustness under different deletion rates. Through the above-mentioned hierarchical and multi-strategy solution, the statistical rigor and method feasibility can be effectively balanced in engineering practice to ensure the accuracy of reliability evaluation of turbine speed governors.

[0058] 2. The step-by-step parameter estimation method proposed in this invention, which integrates censoring characteristic information, is suitable for scenarios with high censoring rates, such as those with censoring rates between 40% and 80%. This method comprehensively reflects the censoring characteristics of the data by combining multiple statistical features (such as the censoring rate, the coefficient of variation of the censoring time, and the skewness). It also uses the statistical information of the censored data to correct the initial estimates, effectively utilizing the implicit characteristic information in the censored data. Through staged optimization, it avoids the coupling problems caused by the simultaneous estimation of multiple parameters. This method effectively addresses the bias issues of traditional methods under high censoring rates, significantly improving the estimation accuracy and robustness under these conditions.

[0059] 3. For extremely high censoring data, such as censoring rate > 80%, traditional parameter estimation methods fail. The present invention introduces a new method of survival curve difference measurement and dynamic weight combination in the framework of semi-supervised learning, and combines feature similarity and survival time distribution similarity through weight coefficients when constructing the weight matrix. A semi-supervised data conversion method that can be used for survival analysis is proposed. This method generates multiple pseudo candidate times, i.e. pseudo labels, for the survival time of censored samples through iterative search, and then optimizes and selects the candidate pseudo label that minimizes the target loss function as the optimal pseudo label for the censored sample, so that it is as close to the actual failure time as possible. The censored samples are gradually converted into fault samples, and finally the maximum likelihood estimation method is used to perform parameter estimation on the data set after the conversion data and the original fault data are merged. This method can adjust the similarity weight ratio of the feature and survival time distribution according to the data distribution to adapt to different high censoring scenarios, and the model has strong adaptability. The method of semi-supervised data conversion combined with maximum likelihood estimation significantly overcomes the limitations of traditional methods in high censoring data. Its core advantages are:

[0060] (1) By being data-driven, it can flexibly capture the characteristics of survival distribution without making strong model assumptions.

[0061] (2) Maximizing information utilization: Integrating the characteristics and survival time information of censored data, effectively using the information of fault samples to infer the survival time of censored samples, thereby improving the prediction performance of the model under high censored data.

[0062] (2) Robustness: The similarity metric and iterative update rules are reasonably defined to ensure that the generated pseudo-labels conform to both the data distribution and the physical constraints of the survival time. This approach is particularly suitable for scenarios where fault data is scarce but censored data is abundant (e.g., high-censored data for turbine speed regulators). By intelligently utilizing the potential information in censored data, it significantly improves model performance, making it an ideal choice for fields such as industrial predictive maintenance and medical diagnosis, and providing a new solution for modeling high-censored data in fields such as industrial reliability analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flow chart of a reliability assessment method for key components of a turbine governor provided by an embodiment of the present invention.

[0064] Figure 2 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0066] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0067] Example

[0068] In order to solve the problem of large parameter estimation deviation and poor robustness of traditional methods under high deletion data. Figure 1 As shown, an embodiment of the present invention provides a reliability assessment method for key components of a turbine governor, comprising the following steps:

[0069] S100, using the analytic hierarchy process to screen out key components of the turbine governor.

[0070] In some embodiments, step S100 includes the following sub-steps:

[0071] S101, a hierarchical structure model is established according to the system structure of the turbine governor, and a judgment matrix is ​​constructed according to the hierarchical structure model. The size of the judgment matrix is n × n ,in n is the number of devices at the corresponding level.

[0072] S102, using consistency index to test the consistency of the judgment matrix. In this embodiment, the consistency ratio CR As a consistency indicator, when the consistency ratio CR < CR TH ( CR TH A threshold value is preset for the consistency ratio, which is set according to needs and actual conditions. For example, when the consistency ratio is 0.1, it is considered that the consistency of the judgment matrix is ​​acceptable; when the consistency ratio is CR ≥ CR TH , then make appropriate corrections to the judgment matrix.

[0073] S103, after the consistency is met, the key components are screened by calculating the component weights. Specifically, the weight vector corresponding to each level component is calculated.w , and thus derive the weight corresponding to each component. The higher the component weight, the more likely it is to fail and the more worthy of attention when making maintenance decisions. The higher the weight (relatively high weight or exceeding the preset weight threshold) w TH ) are considered as key components for subsequent component reliability evaluation.

[0074] S200: Processing original fault operation and maintenance records of the key components to obtain a data set containing censored data.

[0075] The turbine governor is used as the analysis object, and the failure of key components under the turbine governor is mainly studied. Among the original fault operation and maintenance records provided, the original fault operation and maintenance records of the key components are filtered out according to the component name and processed.

[0076] In the original fault operation and maintenance records of the key components, due to the different commissioning dates of the components, the commissioning dates of the components are normalized to t = 0, and by the end of the observation time, the component failure data obtained conforms to the random truncation test, and the number of failed components is obtained. n and component failure time t i (the time when the component fails during actual operation), m The running time of the components is right-censored, and the censored time (i.e., right-censored time, for the convenience of description, it will be uniformly referred to as censored time in the following) is recorded as Since the life data of the speed regulator components follow the Weibull distribution, after removing the outliers in the failure time and censoring time, we finally obtain a data set containing censored data, which is expressed as:

[0077] (1)

[0078] in, t i Indicates the i The failure time of each failed component (i.e., the failure sample in the data set), Indicates the j The censoring time is the running time of the components that have not failed (that is, the censored samples in the data set).

[0079] S300: Calculate a deletion rate based on the data set.

[0080] In this embodiment, the deletion rate is calculated based on the number of failed components, the number of components that have not failed, and the total number of key components. c The formula is as follows:

[0081] (2)

[0082] in, n is the number of failed parts, m is the number of components that have not failed, n + m = N , N is the total number of key components.

[0083] The censoring rate level can be determined based on a preset censoring rate range. The preset censoring rate range is set according to needs and actual conditions. In this embodiment, the setting is as follows:

[0084] Low censoring rate: censoring rate <40%;

[0085] High censoring rate: censoring rate is between 40% and 80%;

[0086] Very high censoring rate: censoring rate >80%.

[0087] S400 dynamically selects the optimal method based on different deletion rate levels to complete the reliability assessment of key components.

[0088] S410, for low deletion rate (<40%), the reliability assessment of key components is completed using the maximum likelihood estimation method. The details are as follows:

[0089] Shape parameter of the Weibull distribution β (shape parameter) and scale parameter η The maximum likelihood estimate of (scale parameter) is obtained by maximizing the log-likelihood function of Equation (5).

[0090] Failure distribution function of the two-parameter Weibull distribution F ( t )for:

[0091] (3)

[0092] The Probability Density Function (PDF) of the two-parameter Weibull distribution is:

[0093] (4)

[0094] in, , is the scale parameter (or characteristic life) of the Weibull distribution, and exp() is the exponential function with the natural constant e as the base; is the shape parameter of the Weibull distribution; t is a random variable, namely the failure time or censoring time.

[0095] (5)

[0096] S420, for high deletion rates between 40% and 80%, a step-by-step parameter estimation method based on comprehensive deletion feature information is used to complete the reliability assessment of key components.

[0097] By combining multiple statistical features (such as the censoring rate, the coefficient of variation over time, and the censoring time skewness), the censoring characteristics of the data are fully reflected, and the statistical information of the censored data is used to correct the initial estimate. Through staged optimization, the coupling problem caused by estimating multiple parameters simultaneously is avoided. The specific steps of the step-by-step parameter estimation method are as follows:

[0098] S421, calculate the censoring characteristic statistics, including the censoring time coefficient of variation and the censoring time skewness.

[0099] (1) Calculate the coefficient of variation of the censoring time ρ :

[0100] The coefficient of variation of the censoring time is the ratio of the standard deviation of the censoring time to the mean, reflecting the degree of dispersion of the censoring time. Its calculation formula is as follows:

[0101] (6)

[0102] in, σ c is the standard deviation of the censoring time, which reflects the discrete degree of the censoring time distribution. The calculation formula is:

[0103] (7)

[0104] μ c is the mean of the censored time, and the calculation formula is:

[0105] (8)

[0106] (2) Calculate the censoring time skewness s :

[0107] The censoring time skewness reflects the asymmetry of the censoring time distribution and is obtained based on the censoring time and the standard deviation and mean of the censoring time. Its calculation formula is:

[0108] (9)

[0109] S422, based on censoring rate c and censoring time skewness s A correction function is constructed, and the mean time to failure (MTTF) is corrected using the correction function.

[0110] Assuming an estimate of the initial mean life MTTF μ 0≈2 μ c , by introducing the correction function g ( T , K ) adjusts the initial MTTF estimate, and for the correction function g ( T , K ), whose parameter cuts off the time T is the average cutoff time of the observation interval, which means the average of the cutoff time points of all observation windows. It can reflect the overall length of the observation window. For example, if the components start running at different time points, the end time of each observation window is different. T It is the average of the end time of each observation window, that is, the average of the running time of the entire component in this stage, which affects the distribution characteristics of the missing data and indirectly affects the missing rate. c and censoring time skewness s Another parameter K is the number of time points at which data is truncated, that is, the number of different batches or different observation intervals in the governor system, that is, the number of observation windows, which reflects the complexity of data truncation. K The larger the value, the more observation windows the data is truncated in, the more complex the deletion pattern is, and the more indirectly affects the deletion rate. c If the parts are divided n batches are put into operation, and each batch has an independent observation window. K = n From this we can find the correction function g ( T , K ) can be converted into the censoring rate c and censoring time skewness s Function:

[0111] (10)

[0112] Among them, the coefficient a 1, a 2 The coefficients were determined by multivariate linear regression to quantify the impact of missing features on the estimation error.

[0113] The estimated value of the mean life MTTF corrected by the correction function is expressed as:

[0114] (11)

[0115] in, μ c is the mean of the censored time.

[0116] S423, iteratively optimize the corrected mean life MTTF:

[0117] Mean time to failure (MTTF) refers to the average working time of a component before it fails. The calculation formula is:

[0118] (12)

[0119] in, Γ ( x ) is the Gamma function, η is the scale parameter, β is the shape parameter, t is the component failure time.

[0120] Step 1: Fix μ 0, the estimated value of the mean life MTTF corrected by the correction function μ Substituting 0 into formula (12) yields , substitute it into formula (5) and use the maximum likelihood estimation method to estimate the shape parameters β 0;

[0121] Step 2: Fix β 0, use formula (5) to estimate the scale parameter using the maximum likelihood estimation method η* ,get , the updated MTTF estimate is μ 1= η* Γ(1+1 / β 0);

[0122] Step 3: Fix μ 1. Re-estimate β 1. Determine whether the re-estimated shape parameters meet the convergence conditions of formula (13). If not, repeat steps 1 to 2 until the convergence conditions are met.

[0123] (13)

[0124] in, k is the number of iterations, is the shape parameter iteration threshold, which is set according to needs and actual conditions. In this embodiment, it is set to 0.01.

[0125] S430, for extremely high deletion rates (>80%), combines semi-supervised data conversion methods and maximum likelihood estimation methods to complete reliability assessment of key components.

[0126] This method introduces the operating parameters of turbine governor components such as oil pressure, oil temperature, vibration, etc. based on the existing data set and uses them as the feature vectors of each sample in the data set to establish a data conversion function in the form of formula (14). By comparing i Fault samples and j Censored samples j Features x i and x j , effectively use the information of the fault sample to infer the failure time of the censored sample and the pseudo failure time of the censored sample , using the iteration to generate multiple pseudo candidate times, i.e. pseudo labels, for the censored time of the censored sample, and then select the candidate pseudo label that minimizes the target loss function as the optimal pseudo label for the censored sample through the target loss function as shown in formula (15), so that it is as close as possible to the true failure time T j ; Thus, the censored samples are gradually converted into fault samples, and finally the maximum likelihood estimation method is used to estimate the parameters of the data set after the conversion data is combined with the original fault data.

[0127] (14)

[0128] in: For the i The pseudo candidate failure time of fault samples, x i For the i The feature vector of the fault sample, x j For the j The feature vector of the censored samples, C j For the j The censoring time of the censored samples, T i For the i The actual failure time of the fault sample, δ j For the j The missing index of the censored samples, δ j =1 indicates a fault (failure event occurs), δ j =0 indicates censoring (failure event did not occur).

[0129] (15)

[0130] Formally, the supervised loss function for faulty data is given by and an unsupervised loss function for pairs of faulty or censored data points ,in, y i For the i Fault samples, For the i Pseudo labels for fault samples, y j For the j censored samples, For the j Pseudo labels for censored samples, W ij Contains the edge weights of all node pairs, λ Controls the relative importance of the supervision term. The purpose of this function is to select the candidate pseudo-label that minimizes the target loss function. As the optimal pseudo label for the censored samples.

[0131] The specific steps are as follows:

[0132] S431, data preparation:

[0133] Based on the existing data set, the operating parameters of the turbine governor components such as oil pressure, oil temperature, vibration, etc. are introduced and used as the feature vectors of the data set samples. x The data set is divided into two categories, namely fault data and censored data ,in, x i For the i The feature vector of the fault sample, T i For the i The actual failure time of the fault sample, x j For the j The feature vector of the censored samples, C j For the observed j The censoring time of the censored samples.

[0134] S432, combining feature similarity and survival curve similarity to construct a weight matrix W ij

[0135] The feature similarity can be obtained by using the Gaussian kernel function to calculate the distance in the feature space using formula (16). The survival curve similarity can be obtained by using formula (17) based on the difference of Kaplan-Meier survival curves. For the failure samples, the survival curve can be directly calculated using the Kaplan-Meier estimator formula (18). For the missing samples, the true failure time is unknown (only the Tj ≥ C j ), the survival curve needs to be estimated with the help of global information formula (19). The global survival curve is the population average survival function estimated by all samples (including censored samples and fault samples). Then the comprehensive weight of these two similarities is calculated by formula (18).

[0136] (16)

[0137] in, is the bandwidth parameter that controls how quickly the similarity decays. A larger value will make the similarity distribution smoother, while a smaller value will emphasize local structures. is the eigenvector x i and x j The Euclidean distance of .

[0138] (17)

[0139] in, control KS The strength of the difference's effect on the weights, KS is the Kolmogorov-Smirnov statistic, S i ( t ) is the i The survival function of the failure sample is estimated.

[0140] (18)

[0141] in, d i It's time t i The number of failures that occurred, n i is the size of the risk set, expressed at time point t i The number of failures that did not occur before.

[0142] Assume that the survival curve of the censored sample is at the censoring time C j The result is consistent with the global survival curve, that is:

[0143] (19)

[0144] (20)

[0145] in, a is the weight coefficient, , used to control the weight of feature similarity.

[0146] S433, calculate the transfer matrix based on the weight matrix P :

[0147] By normalizing the rows in formula (19), we can ensure that the sum of the transition probabilities of each node is 1, and use this to construct the probability transfer matrix P .

[0148] (19)

[0149] S434, construct an initial label matrix based on the failure time of the faulty samples and the censoring time of the censored samples:

[0150] The initial label of the fault sample is the failure time , the initial label of the censored sample is set to the censoring time , merge the two to form the initial label matrix .

[0151] S435, based on the transfer matrix P and label matrix Y , iteratively update the pseudo survival time:

[0152] In the typical right-censored case, the actual failure time of the censored sample is unknown, but greater than the observed failure time. Based on this fact, the iteration result is required to meet the constraint after each iteration. In each iteration, the labels of the data points of the censored samples are updated according to the weighted average of the labels of their neighbor nodes, while the labels of the faulty data are fixed. The mathematical expression is formula (20), where the transfer matrix P and label matrix Y The blocks are marked (subscript l ) and unmarked (subscript u ) part is as in formula (21), and the iterative formula can be simplified to update only the unlabeled part as formula (22). Repeat the above steps until the iteration converges to a convex solution such as formula (23). The final result is the optimal pseudo label of the censored sample. Y u .

[0153] (twenty two)

[0154] in, is the damping factor, , the weight used to control the weight of information propagation.

[0155] (twenty three)

[0156] (twenty four)

[0157] (25)

[0158] S436, Update Dataset

[0159] After the above steps, the censoring time C j The optimal pseudo label will be obtained through the semi-supervised data conversion method Replaced by, it is considered as a fault sample. Update the original data set:

[0160] (26)

[0161] in, D For the updated dataset.

[0162] S437, based on the updated data set, the reliability assessment of key components is completed using the maximum likelihood estimation method. Specifically, assuming that the event time data in the updated data set follows a Weibull distribution, maximum likelihood estimation is performed using Equation (5) to obtain the final parameter estimation results.

[0163] Based on the same technical concept, an embodiment of the present invention also provides an electronic device that can implement the reliability assessment method flow of the key components of the turbine governor provided in the above embodiment of the present invention. In one embodiment, the electronic device can be a server, or a terminal device or other electronic device. Figure 2 As shown, the electronic device may include:

[0164] At least one processor, and a memory connected to the at least one processor. The embodiment of the present invention does not limit the specific connection medium between the processor and the memory. Figure 2 The example in this article is that the processor and memory are connected via a bus. Figure 2 The connections between the other components are shown in bold lines, which are only for illustration and not intended to be limiting. The bus can be divided into address bus, data bus, control bus, etc. Figure 2 The processor is represented by a single thick line, but this does not mean that there is only one bus or only one type of bus. Alternatively, the processor can also be called a controller, without any limitation on the name.

[0165] In an embodiment of the present invention, the memory stores instructions that can be executed by at least one processor. The at least one processor can execute the reliability assessment method of the key components of the turbine governor discussed above by executing the instructions stored in the memory. The processor can implement Figure 2 The functions of each module in the device shown.

[0166] Among them, the processor is the control center of the device, which can use various interfaces and lines to connect the various parts of the entire control device, and monitor the device as a whole by running or executing instructions stored in the memory and calling data stored in the memory, the various functions of the device and processing data.

[0167] In an optional design, the processor may include one or more processing units, and the processor may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interface, and application programs, and the modem processor primarily processes wireless communications. It is understood that the modem processor may not be integrated into the processor. In some embodiments, the processor and memory may be implemented on the same chip, or in some embodiments, they may be implemented on separate chips.

[0168] The processor can be a general-purpose processor, such as a CPU, a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component, and can implement or execute the methods, steps, and logic diagrams disclosed in the embodiments of the present invention. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the reliability assessment method for key components of a turbine governor disclosed in the embodiments of the present invention can be directly executed by a hardware processor, or by a combination of hardware and software modules within the processor.

[0169] As a non-volatile computer-readable storage medium, memory can be used to store non-volatile software programs, non-volatile computer executable programs and modules. Memory can include at least one type of storage medium, for example, can include flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), programmable read-only memory (Programmable Read Only Memory, PROM), read-only memory (Read Only Memory, ROM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), magnetic memory, disk, optical disk, etc. Memory is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory in the embodiment of the present invention can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0170] By designing and programming a processor, the code corresponding to the reliability assessment method for a key component of a hydroturbine governor described in the aforementioned embodiment can be embedded in the chip, enabling the chip to execute the steps of the method in the aforementioned embodiment when running. Designing and programming a processor is well known to those skilled in the art and will not be further described here.

[0171] Based on the same inventive concept, an embodiment of the present invention further provides a storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer executes the reliability assessment method for key components of a turbine governor discussed above.

[0172] In some optional embodiments, the present invention also provides various aspects of a method for reliability assessment of key components of a turbine speed governor, which can also be implemented in the form of a program product, which includes program code. When the program product is run on the device, the program code is used to enable the control device to execute the steps of a method for reliability assessment of key components of a turbine speed governor according to various exemplary embodiments of the present invention described above in this specification.

[0173] It should be noted that although several units or subunits of the device are mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of a unit described above can be further divided into multiple units to be embodied. In addition, although the operations of the method of the present invention are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be decomposed into multiple steps.

[0174] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0175] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a server, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0176] Program code for performing the operations of the present invention may be written using any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computing device, partially on the user's device, as a stand-alone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0177] Where a remote computing device is involved, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0178] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0179] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0180] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A reliability assessment method for key components of a turbine governor, characterized in that: include: The key components of turbine governor are selected by using analytic hierarchy process; Processing the original fault operation and maintenance records of the key components to obtain a data set containing censored data; Calculate the censoring rate based on the data set; Dynamically select the optimal method based on different deletion rates to complete the reliability assessment of key components; The calculating the censoring rate based on the data set includes: in, c is the censoring rate, n is the number of failed parts, m is the number of components that have not failed; The method of dynamically selecting the optimal method according to different deletion rate levels to complete the reliability assessment of key components includes: For low deletion rates, the maximum likelihood estimation method is used to complete the reliability assessment of key components; For high deletion rates, a step-by-step parameter estimation method based on integrated deletion characteristic information is used to complete the reliability assessment of key components; For extremely high deletion rates, the reliability assessment of key components is completed by combining semi-supervised data conversion method and maximum likelihood estimation method; Wherein, the low censoring rate, high censoring rate and extremely high censoring rate are determined according to a preset censoring rate range; The step-by-step parameter estimation method using comprehensive deleted feature information to complete the reliability assessment of key components includes: Calculate the censoring feature statistics, including the censoring time coefficient of variation and the censoring time skewness; the censoring time coefficient of variation is the ratio of the standard deviation of the censoring time to the mean; the censoring time is obtained based on the standard deviation and mean of the censoring time; Constructing a correction function based on the censoring rate and the censoring time skewness, and using the correction function to correct the mean life MTTF; wherein the correction function is a function based on the censoring rate and the censoring time skewness; The corrected mean life MTTF is iteratively optimized based on the maximum likelihood estimation method: Step 1: Fix the estimated value of the mean life MTTF corrected by the correction function, substitute it into the calculation formula of the mean life MTTF, substitute the calculated scale parameter into the maximum likelihood estimation method to estimate the shape parameter; Step 2: fixing the estimated shape parameter, estimating the scale parameter using the maximum likelihood estimation method, and updating the estimated value of the mean time to failure (MTTF) using the estimated scale parameter; Step 3: fix the updated estimated value of the mean time to failure (MTTF), re-estimate the shape parameters according to step 1, and determine whether the re-estimated shape parameters meet the convergence condition. If not, repeat steps 1 to 2 until the convergence condition is met. The reliability assessment of key components is completed by combining the semi-supervised data conversion method and the maximum likelihood estimation method, including: Introducing operating condition parameters of turbine governor components into a data set and using them as feature vectors of data set samples; the data set includes fault samples and censored samples; Combining feature similarity and survival curve similarity to construct a weight matrix; Calculate the transfer matrix based on the weight matrix; Construct an initial label matrix based on the failure time of the faulty samples and the censoring time of the censored samples; Iteratively update the pseudo survival time based on the transition matrix and the initial label matrix to obtain the optimal pseudo label of the censored sample; Replacing the censored samples in the data set with the optimal pseudo labels and treating them as faulty samples, thereby updating the data set; Based on the updated data set, the reliability assessment of key components is completed using the maximum likelihood estimation method.

2. The reliability assessment method for key components of a turbine governor according to claim 1, characterized in that: The key components of the turbine governor are screened out using the analytic hierarchy process, including: Establishing a hierarchical structure model according to the system structure of the turbine governor, and constructing a judgment matrix according to the hierarchical structure model; Using consistency index to test the consistency of the judgment matrix; After consistency is met, key components are screened by calculating component weights.

3. The reliability assessment method for key components of a turbine governor according to claim 1, characterized in that: The processing of the original fault operation and maintenance records of the key components to obtain a data set containing censored data includes: Filter out the original fault operation and maintenance records of the key components according to the component names from the provided original fault operation and maintenance records and process them; In the original failure operation and maintenance records of the key components, after normalizing the component commissioning date, the component failure data obtained by the end of the observation time conforms to the random truncation test, thereby obtaining the number of failed components and the component failure time. The operating time of the components without failure is right-truncated to obtain the missing time; since the life data of the speed regulator components obeys the Weibull distribution, after eliminating the outliers in the failure time and the missing time, finally a data set containing missing data is obtained.

4. The reliability assessment method for key components of a turbine governor according to claim 1, characterized in that: The reliability assessment of key components is completed by using the maximum likelihood estimation method, including: in, L is the likelihood function, n is the number of failed parts, m is the number of components that have not failed, t i Indicates the i The failure time of a fault sample, For the j The censoring time of the censored samples; F ( t ) is the shape parameter of the Weibull distribution β and scale parameters η The failure distribution function is expressed as: f ( t ) is the shape parameter of the Weibull distribution β and scale parameters η The density distribution function is expressed as: in, , t is a random variable, namely the failure time or deletion time, and exp() is an exponential function with the natural constant e as the base.

5. An electronic device, characterized in that: include: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the at least one processor executes the method according to any one of claims 1 to 4 by executing the instructions stored in the memory.

6. A computer-readable storage medium, characterized in that The computer-readable storage medium is used to store instructions, and when the instructions are executed, the method according to any one of claims 1 to 4 is implemented.

Images