System reliability modeling method based on multi-source data fusion

Abstract

The invention provides a system reliability modeling method based on multi-source data fusion. The method comprises the following steps: collecting data in a ship complex system; preprocessing the collected data to obtain multi-source data; performing reliability modeling based on the multi-source data to obtain a basic model; performing simulation on the basis of the multi-source data and the data of the basic model to evaluate the reliability level of the system, and performing statistical analysis on the data obtained by system failure simulation to obtain a system reliability evaluation index; the reliability evaluation indexes comprise the reliability and the average life of the system. According to the method, data processing in each stage is developed and developed according to reliability test identification requirements of a complex system, on this basis, a system reliability model is constructed and fault and failure characteristics of the system are analyzed by combining the structure and functional characteristics of the system, and reliability simulation evaluation of typical application scenarios is developed.

Description

Technical Field

[0001] The invention belongs to the technical field of modeling, and in particular relates to a system reliability modeling method based on multi-source data fusion. Background Art

[0002] At present, the data analysis process mainly involves people collecting, screening, summarizing and analyzing the disorganized and all-encompassing data and information to form meaningful data, files or records (i.e. information), and focuses on data collection, organization and analysis. Reliability data analysis is a comprehensive process from data generation and collection to analysis of reliability parameters and life indicators. There are two models for reliability data analysis, one is the physical property model and the other is the probability model. The physical property model studies where the failure occurs in the product and in what form. It analyzes from the failure mechanism. This is a microscopic analysis and a root-seeking approach. The probability model studies the relationship between failure and time, and uses mathematical statistics to find the probability distribution of the failure time. This is a macroscopic analysis method.

[0003] The purpose and tasks of reliability data analysis are closely related to the needs of reliability work. In the development stage, the parameters should be evaluated according to the test results, the causes of product failures should be analyzed, the weak links should be found, and improvement measures should be proposed to provide the possibility for the gradual increase of product reliability. From the end of development to the period before production, the reliability level can be evaluated based on the results of the reliability identification test to see whether it meets the design requirements, thereby providing management information for production decisions. In the batch production stage, the reliability can be evaluated based on the data of the acceptance test to check whether its various levels can meet the reliability required by the product. In the early stage of product use, sufficient attention should be paid to the collection of reliability data on the use site, and it should be analyzed and evaluated in time to find out the early failures of the product and their main causes, improve or strengthen quality management, and strengthen reliability screening, which can greatly reduce the early failure rate of the product and improve the reliability of the product. During use, reliability analysis and evaluation need to be carried out regularly, and products with low reliability should be improved to meet the indicators required by the design.

[0004] Reliability data has the characteristics of timeliness, randomness, value, timeliness, and plasticity. Among them, timeliness means that reliability data is mostly expressed in time; randomness means that product failures occur randomly; value is reflected in the fact that data collection requires a lot of material and financial resources, and at the same time, reliability data that has been analyzed and processed has great value for reliability work; timeliness means that the generation and use of reliability data are closely related to each stage of the product life cycle, and the data generated at each stage reflects the reliability level of the product at that stage; plasticity means that as time goes by, reliability data reflects the trend and process of product reliability development.

[0005] Complex systems, especially those with small samples, are limited by system development time, usage environment, data collection difficulties and scientific research funds. The sample size of field tests is extremely limited, and the data distribution is diverse and discrete. It is difficult to give an objective evaluation of such systems using traditional mathematical statistics methods based on large samples, and this becomes a key problem in system reliability assessment. In the entire life cycle of complex systems, there are many sources of information that are beneficial to reliability assessment. Scientific reliability assessment requires fully tapping the potential of this information. At the same time, the cross-cutting, multi-source, and even heterogeneous nature of reliability information increases the difficulty of processing and utilizing reliability information. Therefore, the research on reliability information fusion technology has become the key to improving the utilization rate of reliability information and conducting scientific reliability assessment, and is a research hotspot in the field of reliability assessment. Common reliability information mainly includes expert experience information, component and subsystem information, and reliability growth information.

[0006] The data information available in the process of reliability growth of complex equipment presents the characteristics of multi-source, heterogeneous, and the coexistence of big data and small samples. In the process of research and development of complex equipment, on the one hand, due to the large number of knowledge fields, industrial categories, professional types and research and development cooperation departments involved, the long research and development supply chain, the wide range involved, and the extremely complex collaborative research and development relationship, the data information available in the process of research and development and reliability growth has the characteristics of multi-source, heterogeneous and big data; on the other hand, due to the fact that the research and development of complex equipment is almost multi-variety and small batch, and its structure is extremely complex, the types and quantities of systems, components and parts involved in the equipment are huge, so the data information involved in the process of research and development and reliability growth presents a pyramid-shaped feature with a large amount of data information at the bottom of the system and a small amount of data information at the top of the system.

[0007] The task of data processing is to provide accurate and effective data for operational reliability analysis and ensure that the quality of input data meets the standards of operational reliability analysis. Common processing methods include data cleaning, data transformation, data reduction, etc. The data types of complex systems in different application fields may be different. At this time, it is necessary to select relevant data processing methods based on the characteristics of the data.

[0008] Commonly used data processing methods are as follows:

[0009] (1) Data cleaning

[0010] Data cleaning refers to the process of re-examining and verifying data, with the aim of deleting duplicate information, correcting existing errors, and ensuring data consistency. The main technical means include abnormal data processing, missing data processing, and duplicate data processing. Abnormal data refers to data in a data sequence that is far away from the general level of the sequence. Compared with other data in the data set, abnormal data does not conform to the general rules of the data set. The mainstream abnormal data detection technologies include abnormal data detection methods based on statistics, density, distance, and clustering. In the process of running data collection, noise is inevitably present due to natural or human factors. Noise in data refers to the random error or variance of a certain attribute value in the data set. The main methods of smoothing noise include binning method, regression method, and clustering method. Missing data processing is an important step in data cleaning. There are three types of missing data: completely random missing, random missing, and non-random missing. The processing of missing data can be roughly divided into three categories: deleting incomplete data tuples, ignoring missing data, and filling missing values. Duplicate data refers to data that appears twice or even multiple times in the data. The repeated appearance of data makes the weight of the repeated sample greater than that of other data, resulting in a biased result in the subsequent running reliability analysis. For controllable amounts of duplicate data, comparison algorithms are often used to eliminate them. For complex data, their weights are often reduced proportionally, and the data are rearranged to form a probability distribution.

[0011] (2) Data transformation

[0012] When conducting operational reliability analysis, since the algorithm has its own specific restrictions and requirements on the format and scope of data, this requires the conversion of the data format and scope of these data sets with different formats to unify the format and scope of all data.

[0013] (3) Data Specification

[0014] Data reduction mainly includes data dimension reduction and data sample reduction. Data dimension reduction is to reduce the number of data attributes by screening data features or using spatial changes. Data sample reduction is to reduce the amount of data by finding a representative subset in the original data or reconstructing smaller or shorter data and discretizing the data. The core of data dimension reduction is to select or fuse features to obtain the attributes that can describe the key features. The purpose is to reduce redundant features and irrelevant features in the data set to improve the accuracy and quality of subsequent operational reliability analysis. The feature selection strategy should meet the requirements of low computational cost and finding the best or near-optimal feature subset. In order to meet the above conditions, it is necessary to weigh whether the selected feature set is optimal for the subsequent operational reliability analysis task. The test standard is that the results of the operational reliability analysis method applied to the feature subset should be the same as or relatively better than the results applied to the entire feature set. Among them, feature selection is mainly divided into two types: linear selection method and nonlinear selection method. Linear selection methods include factor analysis and principal component analysis, and nonlinear selection methods include local linear embedding, equidistant feature mapping and its derivative methods. The core of data sample reduction is to reduce the data sample size or select smaller data representations from the data set and convert continuous data into discrete values.

[0015] The small sample data processing method is as follows:

[0016] The basis of reliability analysis is reliability test data. Due to the limitations of many factors, the test sample size is often small, and it is difficult to obtain accurate and reasonable results using classical reliability evaluation methods. Many experts have proposed different methods to solve this problem, such as the best linear unbiased overall estimation method, support vector machine technology, reliability evaluation method under fuzzy prior information, and empirical Bayesian estimation method.

[0017] (1) Bayesian method

[0018] The Bayesian method solves problems from a different perspective than the classical probability method. Its main feature is that it can make full use of various information (including historical data, expert information and other prior information and field test data) to make reasonable statistical inferences on the problem and obtain posterior information. The (posterior) probability density function of the Bayesian method is expressed as:

[0019]

[0020] Among them, π(θ|x) is the conditional distribution of θ given the sample x, which is called the posterior distribution of θ; the likelihood function f(x|θ) represents the field experimental information provided by the observation information X; p(θ) is the prior probability density, that is, the prior distribution of the prior information; θ is the distribution parameter, and the Bayesian method assumes that θ is a random variable. The Bayesian method concentrates all information about θ from various information such as the population, sample, and prior information, and excludes all information irrelevant to θ. Therefore, it is effective and more reasonable to make statistical inferences about θ based on the posterior distribution π(θ|x).

[0021] (2) Information fusion related methods

[0022] Many people have conducted research on information fusion. For example, Liu Han proposed a Bayesian evaluation method for success-failure unit reliability that integrates multi-source pre-verification information; Fang Genhai conducted in-depth research on the application of Bayesian fusion method in product reliability evaluation, the fusion of multi-source fuzzy reliability information, and the application of evidence theory in the fusion of multi-source reliability information; Feng Jing et al. studied the application of fuzzy logic operator methods in fusion; and Man Jun studied the multi-source information fusion method based on credibility.

[0023] On the basis of previous research, considering the compatibility test and weighted processing of historical data information from different sources, the support vector machine (SVM) theory is introduced to deal with the fusion problem of multi-source prior information. First, a distribution pattern recognition model based on support vector machine technology is constructed, and then this model is used to perform compatibility test and recognition on field data and historical data. Based on field data as the main basis, the relevance of historical data to field samples is examined, that is, historical information is used as the weight of prior distribution. The main process of distribution pattern recognition using support vector machine technology is as follows:

[0024] 1) Construct SVM multi-classifier

[0025] Construct a multi-classifier from the SVM binary classifier. There are two main types of SVM multi-classification algorithms based on binary classification: one-to-one and one-to-many. Use the one-to-one multi-classification algorithm to construct a multi-classifier. The algorithm is as follows: combine N types of training samples, so that a total of N(N-1) / 2 binary classifiers can be constructed, and the algorithm of each binary classifier is known. In the classification of test samples, the voting method is used. After all N(N-1) / 2 two-class classifiers classify the test sample X, the class with the most votes among the N classes determines which class the test sample belongs to.

[0026] 2) Selection of data distribution characteristic parameters

[0027] The selection of distribution characteristic parameters has a great influence on the results of data distribution pattern recognition. Through a large number of experiments, six characteristic parameters including three quartiles of data samples (lower quartile, median, upper quartile), dispersion coefficient, skewness and kurtosis are selected as sample parameters of the recognition model.

[0028] 3) Generate a random data sequence of any distribution in 400 subdivided distribution patterns of four distribution types by computer simulation, calculate its six input feature parameter values, and randomly select them into two parts, one of which is the training sample set and the other constitutes the test sample set.

[0029] 4) Multivariate classification SVM is used to learn the classification of the training sample set, and the kernel function uses the radial basis kernel function to obtain the memory weights for automatic recognition of distribution patterns, that is, to obtain the optimal classification hyperplane.

[0030] 5) Use the obtained multivariate classification SVM optimal classification hyperplane to perform distribution pattern recognition on the test sample set.

[0031] Since field data is key information in the field of test analysis and identification, people often judge pre-test information based on field test results. If the pre-test information passes the compatibility test, then in a statistical sense, the reliability of the pre-test information is confirmed to some extent. However, due to the small sample size, the pre-test information is extremely valuable. In order not to lose any useful information, pattern recognition technology is applied to compatibility testing and credibility determination.

[0032] (3) Parameter estimation method

[0033] The maximum likelihood estimation method (MLE) is a parameter estimation method based on statistical theory. It is a statistical method based on a large amount of data. There are theoretical defects in applying it to small sample parameter estimation. Therefore, the estimation results of the first two methods are unreliable. Although the second method uses a new algorithm, it is only a correction on the wrong results. The SVM method is a parameter estimation method based on statistical learning theory. Although it can fully mine the implicit information of field data compared with the MLE method, it abandons important prior information and historical data information, and is also lacking in theory. The Bayes method can comprehensively consider historical information and field data and is an ideal parameter estimation method. The estimated values ​​of the Weibull shape parameters of the latter two methods are not much different. It can be considered that the sources of field data and historical prior information are basically the same, which can show that the assumptions of the Bayes method are valid. At the same time, the estimated values ​​of the life of the two methods are relatively close, and tend to converge to the actual use value, indicating that the Bayes method can maximize the use of prior information and reduce estimation errors.

[0034] The Bayes method can effectively use empirical information to make the estimation results as close to the actual value as possible. Support vector machine technology has its own advantages in dealing with small sample reliability analysis problems, that is, it can effectively mine data information. Here, in order to obtain a more realistic and effective prior distribution, a new information fusion method based on SVM technology is proposed, and the traditional Bayes method is improved to conduct secondary mining of the implicit information in the field data and historical data to make full use of it. If the method can be improved by using multiple field data, it is believed that a more reasonable estimation method can be obtained, which needs further exploration.

[0035] Data characteristics and processing methods in various fields

[0036] The following analyzes the relevant reliability data characteristics and reliability data processing methods in combination with different fields.

[0037] (1) Aerospace

[0038] Reliability assessment in the field of complex aerospace systems has always been one of the problems that plague the engineering community. This is because the reliability tests of complex systems such as launch vehicles, manned spacecraft, and communication satellites are expensive and have long cycles, and there is little field test data, which makes the reliability assessment method based on traditional statistical large sample theory difficult to apply. In fact, in all aspects of product design, development, production, and use, there is useful information for product reliability assessment, such as simulation data, component and subsystem test data, expert experience, historical test data, and similar product test data. Therefore, in reliability assessment, only by fully tapping the potential of this information can accurate and reliable assessment conclusions be drawn. Research by domestic and foreign scholars has shown that the Bayes method is a feasible way to reasonably fuse multi-source reliability information. The key to determining the success or failure of the Bayes method is how to determine a reasonable fusion pre-test distribution based on multi-source pre-test information. Therefore, the multi-source pre-test information fusion method has become a hot topic for domestic and foreign scholars. Existing fusion methods include weighted fusion, environmental factor fusion, and multi-stage variable population fusion. In actual working conditions, the uncertainty of reliability analysis of complex polymorphic systems caused by factors such as complex structure and insufficient reliability data has attracted widespread attention from scholars. The system reliability analysis methods that have been developed include: fuzzy fault tree, Takagi-Sugeno fuzzy fault tree, fuzzy Petri net, generating function method and Bayesian network (BN) method, among which BN can not only clearly and effectively express the complex logical relationship of polymorphic systems, but also has a unique two-way reasoning mechanism, which is particularly suitable for reliability analysis of high-reliability, small-sample complex systems, and has been applied to reliability analysis, risk assessment, fault diagnosis and other fields.

[0039] (2) Nuclear power equipment field

[0040] A nuclear power plant is a very complex and huge system. It is difficult to imagine how many types of data there are and how large the amount of data is. At present, China has accumulated a large amount of reliability-related data for various components of nuclear power plants, but it is far from enough to systematically count and analyze them. Therefore, it is very necessary to organize and count these reliability data and find out the rules. Reliability data analysis is becoming more and more important in such a complex system. The use of Bayesian methods to process the probabilistic safety assessment (PSA) data of nuclear power equipment is a case of no information prior. When processing the reliability data of nuclear power equipment, the "two-step" method combined with the general database of nuclear power plants is used to transform the nuclear power equipment data processing that belongs to no information prior into data processing with information prior.

[0041] (3) Chemical equipment field

[0042] During the production and operation of chemical enterprises, with the continuous development of science and technology, chemical equipment is updated at a fast speed. Many equipment do not fail at all in short-cycle operation, which makes the failure data of the same model equipment less. The failure life data obtained from the maintenance survey of in-service chemical equipment is very likely to show a relatively typical small sample feature. In order to accurately obtain the failure rate model of a certain equipment, a Weibull distribution model is established based on the failure sample data generated during the operation of the equipment. The specific implementation method is to calculate it using the average rank method or the median rank method, and use the correlation coefficient optimization method to calculate the location parameters of the three-parameter Weibull distribution, and then linearize the Weibull distribution model, and then use the fitting method to identify the distribution parameters. The current parameter estimation methods include gray estimation method, least squares method, maximum likelihood estimation method and other methods.

[0043] (4) Ship equipment field

[0044] For large, complex and expensive ships, the reliability test data is extremely limited and far from meeting the reliability data volume requirements of reliability assessment. Collecting test data from ship equipment development, identification and acceptance tests for ship reliability assessment is a feasible method.

[0045] Reliability assessment is generally divided into unit reliability assessment and system reliability assessment. From the perspective of the current mathematical model of reliability assessment, reliability assessment at home and abroad is generally divided into three categories: the confidence lower limit formula method of the classical method, the Bayesian formula method and the trust method. These three methods can be used for both unit reliability assessment and system reliability assessment. The theory of the classical method is relatively mature, and the results obtained are relatively reliable, but how to determine the sample size in the classical method is a key issue. The smaller the sample size, the less accurate the result obtained. To be more precise, the more conservative the confidence lower limit of the obtained reliability parameter is. Therefore, there are certain technical difficulties in describing large and complex systems by the classical method. The Bayesian method can use prior information to save test time and money, but the Bayesian method depends on prior information. How to choose the prior distribution is the key, which is also the difficulty of using the Bayesian method. This difficulty limits its application in engineering. The trust method has not yet established a clear rule for determining the trust distribution for the density family of multiple parameters. It has only obtained some results on certain issues, and there are also some inherent difficulties. It cannot be widely used at present.

[0046] In view of the fact that the classical confidence lower limit formula method, Bayesian formula method and trust method are unable to provide a comprehensive and systematic evaluation method for large and complex systems, it is necessary to explore a suitable reliability evaluation method as a reliability modeling and evaluation method for small sample ships.

[0047] Therefore, how to provide a system reliability modeling method based on multi-source data fusion has become a technical problem that needs to be solved urgently in this field. Summary of the invention

[0048] The purpose of the present invention is to provide a system reliability modeling method based on multi-source data fusion.

[0049] The present invention provides a system reliability modeling method based on multi-source data fusion, comprising:

[0050] Step S1, collecting data from the complex system of the ship;

[0051] Step S2, preprocessing the collected data to obtain multi-source data;

[0052] Step S3, performing reliability modeling based on the multi-source data to obtain a basic model;

[0053] Step S4: Reliability modeling is performed based on the basic model under a time-varying uncertain environment to obtain an environmental dynamic characteristic model and complete the modeling.

[0054] Preferably, in step S1, the data includes parameters in various devices and control systems;

[0055] The device includes a frequency converter and a motor;

[0056] The parameters of the equipment include physical dimensions of size, weight, flatness, coaxiality and moment of inertia;

[0057] The parameters of the control system include thermal parameters of temperature, flow and pressure; and electrical components of voltage, current, resistance, capacitance, reactance, harmonics, phase angle, active power and reactive power.

[0058] Preferably, in step S2, the preprocessing method comprises:

[0059] Step 21: Format the data and obtain primary data;

[0060] Step 22: normalizing the primary data to obtain normalized data;

[0061] Step 23: Further process the normalized data using the lower confidence limit formula method of the classical method to obtain multi-source data.

[0062] Preferably, in step S3, during the reliability modeling process, the method includes:

[0063] According to the obtained multi-source data, data fusion is performed to obtain a fusion mathematical model;

[0064] Let f 1 (t) represents the life distribution function of the system during the sea trial phase, f 2 (t) represents the life distribution function of the system in the design stage, f 3 (t) represents the lifetime distribution function of the system in the identification test, f 4 (t) represents the life distribution function based on real ship data;

[0065] A multi-source data weighted fusion model is adopted, the number of data sources is set to m, and the fusion mathematical model is:

[0066]

[0067] Among them, w i is the weighted value of each data source, f i (t) is the life distribution density function obtained from the reliability data at different development stages.

[0068] Preferably, in step S3, since the data sources of the multi-source data are different, it is necessary to determine the weights of the multi-source data, and the specific method is as follows:

[0069] Assume that the product life distributions of each source data support each other, and establish support vectors for m life distributions:

[0070] S=(S 11 ,S 12 ,…,S1m )

[0071] in,

[0072]

[0073] f 1 (t) represents the life distribution function of the system during the trial stage, f i (t), i=2,3,...,m are the life distribution functions of the system based on system reliability design, test and actual ship data; S represents f i (t) for f 1 The higher the support level, the higher the S 1i The smaller the value;

[0074] The weights of each multi-source data are:

[0075]

[0076] Under small sample test conditions, the life distribution obtained from field test data is often not completely the same as the actual life distribution. 1 (t) Relative credibility of actual life distribution ρ = w 1 As f 1 The weight of (t), w 1 The larger it is, the greater the weight of the life distribution of the field data in the fusion; the larger the amount of field data, the closer its life distribution is to reality and the higher its credibility.

[0077] Preferably, in step S4, the time-varying uncertain environment includes wind speed and wave height, and the method for modeling the specific environmental dynamic characteristics includes:

[0078] Discretize the wind speed and wave height states.

[0079] The wind speed is divided into "light wind, small effect on component failure", "moderate wind, large effect on component failure" and "strong wind, large effect on component failure", which are represented by "1", "2" and "3" respectively;

[0080] The wave height is divided into "light waves, with less failure effect on components", "medium waves, with greater failure effect on components" and "strong waves, with greater failure effect on components", which are also represented by "1", "2" and "3" respectively;

[0081] The continuous-time Markov chain is used to model the environmental factors: the above two-dimensional environmental factors are transformed into a one-dimensional environment, the minimum generator matrix is ​​Q, and the corresponding transition probability is p ij (t),i,j=1,2,...,9;

[0082] Reliability R(t) refers to the probability that a product can complete a specified function under specified conditions and within a specified time interval. When the test data is of life type, R(t) is a function of time. Let H represent the diagonal matrix of the failure rate function of the system, where the (k, k) element of H is:

[0083]

[0084] That is, the sum of the failure rates of all components under environmental state k, k = 1, 2, ..., 9;

[0085] Let α represent the probability vector of each environment at the initial moment of the system, where α k represents the probability that the initial state of the equipment is k; let e represent a column vector whose elements are all 1, then the reliability function R(t) of the system can be expressed as:

[0086] R(t)=αexp((QH)t)e

[0087] Where exp((QH)t) is the matrix exponential;

[0088] The mean time between failures (MTBF) is further expressed as:

[0089]

[0090] Then the expression of Q is as follows:

[0091]

[0092] Preferably, in step S4, the specific method of converting the two-dimensional environmental factors into a one-dimensional environment is:

[0093] The two-dimensional environment (1,1) is transformed into the corresponding environment 1;

[0094] The two-dimensional environment (1,2) is transformed into the corresponding environment 2;

[0095] The two-dimensional environment (1,3) is transformed into the corresponding environment 3;

[0096] The two-dimensional environment (2,1) is transformed into the corresponding environment 4;

[0097] The two-dimensional environment (2,2) is transformed into the corresponding environment 5;

[0098] The two-dimensional environment (2,3) is transformed into the corresponding environment 6;

[0099] The two-dimensional environment (3,1) is transformed into the corresponding environment 7;

[0100] The two-dimensional environment (3,2) is transformed into the corresponding environment 8;

[0101] The two-dimensional environment (3,3) is transformed into the corresponding environment 9.

[0102] Preferably, in step S4, in order to obtain the environmental dynamic characteristic model, it is also necessary to determine the environmental factors, and the specific method includes:

[0103] If F i (t) and F j (t) respectively represent the product under stress S i and S j The cumulative failure rate under the action of i (t i )=F j (t j ), then the stress S i Stress S j The environmental factors are:

[0104]

[0105] For electronic products that obey exponential distribution, the failure rate is constant. If the failure rates in two environments are λ i and λ j , then the environmental factor of environment 1 relative to environment 2 is

[0106] Preferably, the method further comprises performing simulation based on the multi-source data and the data of the basic model to evaluate the reliability level of the system, and obtaining a system reliability evaluation index by performing statistical analysis on the data obtained from the system failure simulation.

[0107] Preferably, the reliability evaluation index includes the reliability and life expectancy of the system.

[0108] It can be seen from the above scheme that the embodiment of the present invention provides a system reliability modeling method based on multi-source data fusion, which has the following beneficial effects:

[0109] Aiming at the needs of reliability test and identification of complex systems, this application carries out data analysis, deduplication, screening, cleaning, merging, classification and other processing at various stages of development. Based on this, combined with the system structure and functional characteristics, the application uses probability, Monte Carlo simulation, parameter estimation and other technical methods to build a system reliability model, analyze the fault and failure characteristics of the system, and carry out reliability simulation evaluation of typical application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0110] Figure 1 A flowchart of a system reliability modeling method based on multi-source data fusion provided according to an embodiment;

[0111] Figure 2Reliability simulation flow chart of complex electromechanical systems;

[0112] Figure 3 Comparison chart of simulated and theoretical values ​​of average life of a single component under different simulation times;

[0113] Figure 4 Subsystem propulsion inverter typical composition block diagram;

[0114] Figure 5 Block diagram of typical components of the propulsion motor subsystem;

[0115] Figure 6 Comparison chart of simulation results and theoretical values ​​of subsystem reliability function when the number of simulations is 2000;

[0116] Figure 7 Comparison chart of simulated and theoretical average life of the system under different simulation times;

[0117] Figure 8 Comparison chart of simulation results and theoretical values ​​of system reliability function when the number of simulations is 2000. DETAILED DESCRIPTION

[0118] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0119] Embodiment 1:

[0120] The invention discloses a system reliability modeling method based on multi-source data fusion. Figure 1 FIG. 1 is a flow chart of a system reliability modeling method based on multi-source data fusion according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0121] Step S1, collecting data from the complex system of the ship;

[0122] Step S2, preprocessing the collected data to obtain multi-source data;

[0123] Step S3, performing reliability modeling based on the multi-source data to obtain a basic model;

[0124] Step S4: Reliability modeling is performed based on the basic model under a time-varying uncertain environment to obtain an environmental dynamic characteristic model and complete the modeling.

[0125] In the step S1, the data includes parameters in each device and control system;

[0126] The equipment includes a frequency converter and a motor, such as Figure 4 and Figure 5 As shown;

[0127] The parameters of the equipment include physical dimensions of size, weight, flatness, coaxiality and moment of inertia;

[0128] The parameters of the control system include thermal parameters of temperature, flow and pressure; and electrical components of voltage, current, resistance, capacitance, reactance, harmonics, phase angle, active power and reactive power.

[0129] In step S2, the preprocessing method includes:

[0130] Step 21: Format the data and obtain primary data;

[0131] Step 22: normalizing the primary data to obtain normalized data;

[0132] Step 23: Further process the normalized data using the lower confidence limit formula method of the classical method to obtain multi-source data.

[0133] In step S3, during the reliability modeling process, the method includes:

[0134] According to the obtained multi-source data, data fusion is performed to obtain a fusion mathematical model;

[0135] Let f 1 (t) represents the life distribution function of the system during the sea trial phase, f 2 (t) represents the life distribution function of the system in the design stage, f 3 (t) represents the lifetime distribution function of the system in the identification test, f 4 (t) represents the life distribution function based on real ship data;

[0136] A multi-source data weighted fusion model is adopted, the number of data sources is set to m, and the fusion mathematical model is:

[0137]

[0138] Among them, w i is the weighted value of each data source, f i (t) is the life distribution density function obtained from the reliability data at different development stages.

[0139] In step S3, since the data sources of the multi-source data are different, it is necessary to determine the weights of the multi-source data. The specific method is as follows:

[0140] Assume that the product life distributions of each source data support each other, and establish support vectors for m life distributions:

[0141] S=(S 11 ,S 12 ,…,S 1m )

[0142] in,

[0143]

[0144] f 1 (t) represents the life distribution function of the system during the trial stage, f i (t), i=2,3,...,m are the life distribution functions of the system based on system reliability design, test and actual ship data; S represents f i (t) for f 1 The higher the support level, the higher the S 1i The smaller the value;

[0145] The weights of each multi-source data are:

[0146]

[0147] Under small sample test conditions, the life distribution obtained from field test data is often not completely the same as the actual life distribution. 1 (t) Relative credibility of actual life distribution ρ = w 1 As f 1 The weight of (t), w 1 The larger it is, the greater the weight of the life distribution of the field data in the fusion; the larger the amount of field data, the closer its life distribution is to reality and the higher its credibility.

[0148] In step S4, the time-varying uncertain environment includes wind speed and wave height, and the specific method of modeling the dynamic characteristics of the environment includes:

[0149] Discretize the wind speed and wave height states.

[0150] The wind speed is divided into "light wind, small effect on component failure", "moderate wind, large effect on component failure" and "strong wind, large effect on component failure", which are represented by "1", "2" and "3" respectively;

[0151] The wave height is divided into "light waves, with less failure effect on components", "medium waves, with greater failure effect on components" and "strong waves, with greater failure effect on components", which are also represented by "1", "2" and "3" respectively;

[0152] The continuous-time Markov chain is used to model the environmental factors: the above two-dimensional environmental factors are transformed into a one-dimensional environment, the minimum generator matrix is ​​Q, and the corresponding transition probability is p ij (t),i,j=1,2,...,9;

[0153] Reliability R(t) refers to the probability that a product can complete a specified function under specified conditions and within a specified time interval. When the test data is of life type, R(t) is a function of time. Let H represent the diagonal matrix of the failure rate function of the system, where the (k, k) element of H is:

[0154]

[0155] That is, the sum of the failure rates of all components under environmental state k, k = 1, 2, ..., 9;

[0156] Let α represent the probability vector of each environment at the initial moment of the system, where α k represents the probability that the initial state of the equipment is k; let e represent a column vector whose elements are all 1, then the reliability function R(t) of the system can be expressed as:

[0157] R(t)=αexp((QH)t)e

[0158] Where exp((QH)t) is the matrix exponential;

[0159] The mean time between failures (MTBF) is further expressed as:

[0160]

[0161] Then the expression of Q is as follows:

[0162]

[0163] In step S4, the specific method of converting the two-dimensional environmental factors into a one-dimensional environment is:

[0164] The two-dimensional environment (1,1) is transformed into the corresponding environment 1;

[0165] The two-dimensional environment (1,2) is transformed into the corresponding environment 2;

[0166] The two-dimensional environment (1,3) is transformed into the corresponding environment 3;

[0167] The two-dimensional environment (2,1) is transformed into the corresponding environment 4;

[0168] The two-dimensional environment (2,2) is transformed into the corresponding environment 5;

[0169] The two-dimensional environment (2,3) is transformed into the corresponding environment 6;

[0170] The two-dimensional environment (3,1) is transformed into the corresponding environment 7;

[0171] The two-dimensional environment (3,2) is transformed into the corresponding environment 8;

[0172] The two-dimensional environment (3,3) is transformed into the corresponding environment 9.

[0173] In step S4, in order to obtain the environmental dynamic characteristic model, it is also necessary to determine the environmental factors, and the specific method includes:

[0174] If F i (t) and F j (t) respectively represent the product under stress S i and S j The cumulative failure rate under the action of i (t i )=F j (t j ), then the stress S i Stress S j The environmental factors are:

[0175]

[0176] For electronic products that obey exponential distribution, the failure rate is constant. If the failure rates in two environments are λ i and λ j , then the environmental factor of environment 1 relative to environment 2 is

[0177] The method further includes performing simulation based on the multi-source data and the data of the basic model to evaluate the reliability level of the system, and obtaining system reliability evaluation indicators by performing statistical analysis on the data obtained from the system failure simulation, wherein the reliability evaluation indicators include the reliability and average life of the system.

[0178] Embodiment 2:

[0179] In order to achieve the above purpose, the specific implementation scheme of the present invention is:

[0180] Complex system reliability data processing, system reliability modeling based on multi-source data fusion, system reliability modeling based on multi-source fusion data in time-varying uncertain environments, complex system reliability simulation, and complex system reliability evaluation.

[0181] Step S1, collecting data from the complex system of the ship;

[0182] Step S2, preprocessing the collected data to obtain multi-source data;

[0183] 1) In view of the problems of complicated data records, various formats, and inconsistent record formats, basic sorting and preprocessing were mainly carried out;

[0184] 2) Through the confidence lower limit formula method of the classical method, the theory of the classical method is relatively mature and the results obtained are relatively reliable;

[0185] 3) Weighted data fusion method: For data at different stages, the weight of data at each stage is scientifically determined using the weighted data fusion method, and then the data at each stage are integrated for reliability modeling and evaluation.

[0186] Step S3, performing reliability modeling based on the multi-source data to obtain a basic model;

[0187] Let f 1 (t) represents the life distribution function of the system during the sea trial phase, f 2 (t) represents the life distribution function of the system in the design stage, f 3 (t) represents the lifetime distribution function of the system in the identification test, f 4 (t) represents the life distribution function based on real ship data.

[0188] 1) Data fusion model

[0189] Considering the above system, a series of quality data are generated during the production, manufacturing and identification process. Due to the different sources of fusion data, the following multi-source data weighted fusion model will be used to fuse various reliability data obtained during the development process of the system. Assuming the number of data sources is m, the fusion mathematical model is:

[0190]

[0191] Among them, w i is the weighted value of each data source, f i (t) is the life distribution density function obtained from the reliability data at different development stages.

[0192] 2) Data weight determination

[0193] Since the fusion data sources are different, the product life distributions of each source data are assumed to support each other, and support vectors are established for m life distributions:

[0194] S=(S 11 ,S 12 ,…,S 1m )

[0195] in,

[0196]

[0197] Here f 1 (t) represents the life distribution function of the system during the trial stage, f i (t), i = 2, 3, ..., m are life distribution functions based on system reliability design, test and actual ship data. S represents f i (t) for f 1 The higher the support level, the higher the S 1i The smaller the value.

[0198] The following will focus on the weight determination of each data source, assuming that:

[0199]

[0200] Under small sample test conditions, the life distribution obtained from field test data is often not completely the same as the actual life distribution. 1 (t) Relative credibility of actual life distribution ρ = w 1 As f 1 The weight of (t), w 1 The larger the value, the greater the weight of the life distribution of the field data in the fusion. The larger the amount of field data, the closer its life distribution is to reality and the higher its credibility. The weights of other data sources are:

[0201]

[0202] Step S4: Reliability modeling is performed based on the basic model under a time-varying uncertain environment to obtain an environmental dynamic characteristic model and complete the modeling.

[0203] 1) Modeling of environmental dynamic characteristics

[0204] The main factors affecting the system operation are wind speed and wave height. Let e(t) = (e 1 (t),e 2 (t)) represents the environment of the system at time t, where e 1 (t) and e 2 (t) represent the wind speed and wave height at time t. The wind speed and wave height are discretized, and the wind speed is divided into "small, small effect on component failure", "medium, large effect on component failure" and "strong wind, large effect on component failure", represented by "1", "2" and "3" respectively; similarly, the wave height can be divided into "small, small effect on component failure", "medium, large effect on component failure" and "strong wave, large effect on component failure", also represented by "1", "2" and "3" respectively.

[0205] The following mainly uses the continuous time Markov chain to model the environmental factors: First, the above two-dimensional environmental factors are converted into a one-dimensional environment. The specific conversion rules are shown in Table 5 below. The system has a total of 9 possible operating states, and its minimum generator matrix is ​​Q, and the corresponding transition probability is p ij (t), i, j=1, 2,...,9.

[0206] Table 2 Two-dimensional environment conversion table

[0207]

[0208] Reliability R(t) refers to the probability that a product can complete a specified function under specified conditions and within a specified time interval. It is mainly for success-failure test data and is often represented by R(t). When the test data is of life type, R(t) is a function of time. Let H represent the diagonal matrix of the failure rate function of the system, where the (k,k) element of H is

[0209]

[0210] That is, the sum of the failure rates of all components in the environment state k, k = 1, 2, ..., 9. Let α represent the probability vector of each environment in which the system is located at the initial moment, where α k represents the probability that the initial state of the device is k. Let e ​​represent a column vector whose elements are all 1, then the reliability function R(t) of the system can be expressed as:

[0211] R(t)=αexp((QH)t)e

[0212] Where exp((QH)t) is the matrix exponential.

[0213] The mean time between failures (MTBF) can be further expressed as:

[0214]

[0215] Consider the following expression for Q:

[0216]

[0217] 2) Multi-source data fusion model

[0218] a) Determination of environmental factors

[0219] The definition of environmental factors is: i (t) and F j (t) respectively represent the product under stress S i and S j The cumulative failure rate under the action ofi (t i )=F j (t j ), then the stress S i Stress S j The environmental factors are:

[0220]

[0221] For electronic products that obey exponential distribution, the failure rate is constant. If the failure rates in two environments are λ i and λ j , then the environmental factor of environment 1 relative to environment 2 is The application of environmental factors must meet the following three prerequisites.

[0222] Premise 1: Consistency of failure mechanism. The failure mechanism of the product remains unchanged under different environmental stress levels. Only when the failure mechanism remains consistent can the reliability information under different stress levels be converted and integrated, and the environmental factor becomes meaningful.

[0223] Premise 2: Distribution homogeneity, the life distribution of products under different environmental factors obeys the same form of distribution. This condition indicates that the distribution form of life data under different stress levels should be the same, but there are differences in distribution parameters.

[0224] Premise 3: Nelson hypothesis: the residual life of a product depends only on the accumulated failures and current environmental stresses, and has nothing to do with the accumulation method. This hypothesis was proposed by Nelson, who actually takes the cumulative failure probability as an external manifestation of the environmental damage to the product, and believes that even in different environments, as long as the cumulative failure probability of the product is the same, the cumulative damage in the product is also the same.

[0225] The life of equipment in operation at sea generally follows an exponential distribution. According to the definition of environmental factors, for electronic products that follow an exponential distribution, the environmental factor K in environment 1 relative to environment 2 is equal to the ratio of the failure rate of the product in these two environments, that is, If the test information is abundant, the failure rate of the product in two environments is known and can be calculated directly. When the information is insufficient, a reliability prediction-based method can be used to predict the failure rate of the product in different environments. The system reliability modeling is integrated with the information in the design stage, the experimental identification stage, the actual ship information and the sea trial information, and the failure rate of the whole product is further predicted. The environmental factor conversion method based on the failure rate is complex to calculate and requires the collection of a large amount of product information, but it is relatively accurate. It comprehensively considers the basic failure rates of various components and parts of the product, quality control levels, application environment categories, performance ratings and types, structures and other factors affecting the failure rate. In summary, the following will choose to determine the environmental factors of the water tank test relative to the actual sea trial based on the failure rate prediction method. Assumptions

[0226] Then, the failure rate of component i at time t can be expressed as:

[0227] λ i (t,e(t))=λ i / C(e 1 (t),e 2 (t))

[0228] That is, the failure rate of each component of the system is related to its inherent design and manufacturing level, as well as its operating environment (e 1 (t),e 2 (t)) is relevant.

[0229] b) Determination of system failure rate

[0230] In the design phase, environmental factors are used in combination with formulas

[0231] λ i (t,e(t))=λ i C(e 1 (t),e 2 (t))

[0232] The failure rates of various components in environments 1 to 9 can be determined. Furthermore, let λ design (i) represents the failure rate of the system under environmental state i during the design phase. The following table gives the specific values ​​of the failure rate.

[0233] Table 2 Failure rate of system design stage under different environments (10 -4 / Hour)

[0234] Impact arrival rate Environment 1 Environment 2 Environment 3 Environment 4 Environment 5 Environment 6 Environment 7 Environment 8 Environment 9 <![CDATA[λ design (i)]]> 3.81 3.93 4.01 4.08 4.21 4.27 4.35 4.50 4.59

[0235] In the experimental stage, let h shock (i) represents the impact arrival rate of the system when the environmental state is i. The following table gives the specific values ​​of the impact arrival rate under different environments.

[0236] Table 3 System impact rate under different environments (10 -5 / Hour)

[0237] Impact arrival rate Environment 1 Environment 2 Environment 3 Environment 4 Environment 5 Environment 6 Environment 7 Environment 8 Environment 9 <![CDATA[h shock (i)]]> 1 1.2 1.3 1.4 1.5 1.7 1.8 1.9 2

[0238] Let λ experiment (i) represents the failure rate of the system under environmental state i during the design phase. The following table gives the specific values ​​of the failure rate.

[0239] Table 4 System impact rate under different environments (10 -4 / Hour)

[0240] System failure rate Environment 1 Environment 2 Environment 3 Environment 4 Environment 5 Environment 6 Environment 7 Environment 8 Environment 9 <![CDATA[λ experiment (i)]]> 3.91 4.05 4.14 4.22 4.36 4.44 4.53 4.69 4.79

[0241] Let λ boat (i) represents the failure rate based on actual ship data under system environmental state i. The following table gives the specific values ​​of the failure rate.

[0242] Table 5 System impact rate under different environments (10 -4 / Hour)

[0243] System failure rate Environment 1 Environment 2 Environment 3 Environment 4 Environment 5 Environment 6 Environment 7 Environment 8 Environment 9 <![CDATA[λ boat (i)]]> 3.826 3.826 3.826 3.829 3.829 3.829 3.832 3.829 3.829

[0244] During the sea trial phase, since the system provides fault information under actual operation scenarios, the impact of environmental factors is no longer considered.

[0245] 3) System reliability modeling

[0246] a) Multi-source data fusion

[0247] Multi-source data fusion still assumes that w 1 =0.3, the weights of the remaining stages can be calculated as:

[0248] w 2 =0.2415,w 3 =0.2678,w 4 =0.1907

[0249] Therefore, the comprehensive failure rate of the system under environment i can be obtained as:

[0250] λ overall (i) = w 1 λ 1 (i)+w 2 λ design (i)+w 3 λ experiment (i)+w 4 λ boat (i)

[0251] The system failure rate matrix H is as follows:

[0252]

[0253] Step S5: Perform reliability simulation on the system;

[0254] System reliability analysis regards the product as a system containing basic components and hierarchical functional structures, and quantitatively evaluates the reliability level of the product through the reliability data of the basic components and the hierarchical reliability model. Monte Carlo simulation (MC simulation) is a simulation method that treats the evaluation object as a random process, randomly generates the data required for evaluation, and achieves the evaluation purpose through numerical calculation. It is widely used in complex problems that are difficult to solve with traditional mathematical methods or physical experiments. System reliability modeling and analysis is a typical random problem. The MC simulation method can be used to simulate and generate unit failure data samples that meet the requirements, and system-level virtual experiments can be carried out based on the logical relationship between unit failure and system failure to obtain system failure simulation data samples. System reliability indicators such as system reliability and average life can be obtained by statistically analyzing the system failure simulation data.

[0255] In recent years, the dynamic reliability of complex systems has attracted people's attention. At present, the main methods of reliability assessment include Markov analysis, Petri net theory, dynamic fault tree and Monte Carlo simulation. This patent mainly takes the ship integrated electric propulsion system as the evaluation object, integrates multi-stage data such as design, test, and failure, adopts the Monte Carlo simulation method, and uses MATLAB software for simulation programming to evaluate the dynamic reliability of key components, subsystems and systems of the system when the component life follows different distributions, including the average life simulation when the life of each stage follows the exponential distribution, the reliability function simulation when the life of each stage follows the exponential distribution, the average life simulation when the life distribution of each stage is a general distribution, and the reliability function simulation when the life distribution of each stage is a general distribution, such as Figure 2 shown.

[0256] 1) Single component reliability simulation

[0257] The secondary circuit refers to an electrical circuit in which secondary devices are interconnected in an electrical system to monitor, control, regulate and protect the primary equipment. These devices include the secondary winding of the mutual inductor, measuring and monitoring instruments, relays, automatic devices, etc., which are connected to form a circuit through a control cable. The main function of the secondary circuit is to convert the current, voltage and other information in the primary circuit into corresponding electrical signals, and transmit them to the protection device, automation or metering instrument to achieve protection, control, monitoring and metering functions. In addition, the secondary circuit can also be classified according to different classification standards, such as by power supply properties (AC current circuit and DC circuit), by function and by wiring method (measurement circuit, relay protection circuit, switch control and signal circuit, operation power circuit, and electrical locking circuit of circuit breaker and disconnector, etc.). The secondary circuit is an important part of the inverter, which is of great significance to the stable operation and safe power supply of the system.

[0258] Assume that the secondary circuit related fault data includes four stages, and the data weights of each stage are w i ,

[0259] That is, the probability density function of the secondary circuit life is:

[0260]

[0261] The reliability function and average life of the system can then be calculated. The specific simulation process is as follows: Figure 1 Typical simulation results are shown in Figure 2 shown.

[0262] 2) Subsystem reliability simulation

[0263] The propulsion inverter provides optimal dynamic control and protection for load changes under different propulsion conditions, thereby ensuring the optimal performance and safe operation of the propulsion equipment. The inverter will control the speed of the propulsion motor, so that the system can continuously adjust the speed of the propulsion motor according to the instructions, drive the propeller, and make the ship sail smoothly. The propulsion inverter consists of a rectifier, a brake unit, an inverter unit, a cooling device, an output filter, a secondary circuit, a control system, etc. Figure 3 shown.

[0264] The propulsion motor provides shaft power that meets the ship's needs at rated speed, can achieve variable frequency speed regulation, can be continuously adjusted within the speed regulation range, and can operate for a long time. The permanent magnet propulsion motor mainly includes the motor body, cooling system, mounting base and electrical signal system, such as Figure 4 shown.

[0265] The reliability simulation analysis is carried out with the propulsion inverter as the subsystem. Typical simulation results are as follows: Figure 5 shown.

[0266] 3) System reliability simulation

[0267] A typical integrated electric propulsion system transforms the traditional mechanical transmission mode of the main engine with a gearbox shaft system and a propeller into a mode in which the power station generates electricity to supply the inverter, and the inverter drives the electric motor to drive the propeller. Through integrated control, the inverter, propulsion motor and other core equipment are combined into a system with the characteristics of high efficiency, high reliability, high automation and low maintenance, which will become the development trend of future ship power systems.

[0268] The operating principle of a typical integrated electric propulsion system is as follows: the three-phase AC power is connected to the distribution board as the input power of the inverter, and after passing through the incoming line unit and the PWM rectifier unit, it is shaped into a DC power supply and then output to the inverter unit. Under the control of the control system, its main power device switches according to the rules required for closed-loop control, converting the DC power into PWM pulse voltages of different phases, fundamental voltage amplitude and frequency required by the motor winding, and driving the three-phase permanent magnet synchronous motor to rotate according to the command speed. When the motor is braked, the control system monitors the change of the DC side voltage and controls the action of the brake unit in real time. While ensuring that the DC bus voltage is maintained at the required upper and lower thresholds, part of the motor kinetic energy is consumed in the brake resistor to achieve rapid braking of the propulsion motor. Among them, the propulsion inverter performs optimal dynamic control and protection of load changes under different propulsion conditions, thereby ensuring the optimal performance and safe operation of the propulsion equipment. The inverter will control the speed of the propulsion motor, so that the system can continuously adjust the speed of the propulsion motor according to the command, drive the propeller, and make the ship sail smoothly. The propulsion inverter consists of a rectifier, a brake unit, an inverter unit, a cooling device, an output filter, a secondary circuit, a control system, etc. The propulsion motor provides shaft power that meets the ship's needs at rated speed, can achieve variable frequency speed regulation, can be continuously adjusted within the speed regulation range, and can operate for a long time. The permanent magnet propulsion motor mainly includes the motor body, cooling system, mounting base and electrical signal system.

[0269] Typical simulation results are as follows: Figure 6 and Figure 7 shown.

[0270] Step S6: Evaluate the reliability of the system.

[0271] At the component level, the data of four stages are integrated respectively, and the two situations that the life of each stage follows exponential distribution and general distribution are considered. Under the exponential distribution situation, the reliability function of the secondary circuit under the two situations is given; at the subsystem level, the propulsion inverter subsystem containing multiple components is investigated, and the situations that the life of each component follows exponential distribution and general distribution are also given. Under the exponential distribution situation, the reliability function of the inverter under the two situations is given respectively; at the system level, the integrated electric propulsion system is taken as the investigation object, and the system reliability of the two subsystems including the inverter and the propulsion motor is evaluated. The average life of the system is obtained when the life of each component follows the exponential distribution, and the average life of the system is obtained under the general distribution situation, and the reliability function of the system under the two situations is given respectively. Through specific simulation and comparison with theoretical values, such as Figure 3 , Figure 6 , Figure 7 and Figure 8 As mentioned above, the effectiveness of the simulation and the correctness of the model solution are verified, providing technical support and reference for the effective evaluation of the reliability of complex systems.

[0272] In summary, the solution proposed in the present invention mainly targets the needs of reliability test and identification of complex systems, and carries out data analysis, deduplication, screening, cleaning, merging, classification and other processing at various stages of development. On this basis, combined with the system structure and functional characteristics, the system reliability model is constructed by using probability, Monte Carlo simulation, parameter estimation and other technical methods, the fault and failure characteristics of the system are analyzed, and reliability simulation evaluation of typical application scenarios is carried out.

[0273] 1) Reliability processing of complex systems: processing and analyzing the system’s design phase data, test phase data, actual ship data, and sea trial data, and obtaining the failure rate of each component in different stages and scenarios, providing a basis for subsequent reliability modeling and analysis.

[0274] 2) Complex system reliability modeling and analysis. On the basis of reliability data analysis, a system reliability model is constructed and analyzed. Two reliability assessment models are proposed: Model 1 is the basic model, which proposes the basic ideas and methods of data fusion. Model 2 considers the dynamic uncertainty of the equipment operating environment, and uses Markov chain to describe the environmental evolution characteristics. It considers the reliability modeling and assessment of the system under different operating environment factors. The data fusion method is the same as Model 1. The analysis shows that the complex and changeable environment has a certain impact on the system reliability.

[0275] 3) Reliability simulation research on complex systems under typical scenarios, integrating data from four stages at the component level, subsystem level, and system level, and obtaining the average lifespan when the lifespan of each component follows an exponential distribution, as well as the average lifespan under a general distribution. Reliability functions for the two situations are given to support quantitative evaluation.

[0276] 4) The above provides technical support for the effective reliability evaluation of complex electromechanical systems of ships, and also provides a reference for similar products to integrate multiple types of data for reliability evaluation and verification.

[0277] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the particular order or sequential order shown to achieve the desired results. In some implementations, multitasking and parallel processing may be advantageous.

[0278] The above are preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A system reliability modeling method based on multi-source data fusion, characterized in that: include: Step S1, collecting data from the complex system of the ship; Step S2, preprocessing the collected data to obtain multi-source data; Step S3, performing reliability modeling based on the multi-source data to obtain a basic model; Step S4: Reliability modeling is performed based on the basic model under a time-varying uncertain environment to obtain an environmental dynamic characteristic model and complete the modeling.

2. The system reliability modeling method based on multi-source data fusion according to claim 1 is characterized in that: In the step S1, the data includes parameters in each device and control system; The device includes a frequency converter and a motor; The parameters of the equipment include physical dimensions of size, weight, flatness, coaxiality and moment of inertia; The parameters of the control system include thermal parameters of temperature, flow and pressure; and electrical components of voltage, current, resistance, capacitance, reactance, harmonics, phase angle, active power and reactive power.

3. The system reliability modeling method based on multi-source data fusion according to claim 2 is characterized in that: In step S2, the preprocessing method includes: Step 21: Format the data and obtain primary data; Step 22: normalizing the primary data to obtain normalized data; Step 23: Further process the normalized data using the lower confidence limit formula method of the classical method to obtain multi-source data.

4. The system reliability modeling method based on multi-source data fusion according to claim 3 is characterized in that: In step S3, during the reliability modeling process, the method includes: According to the obtained multi-source data, data fusion is performed to obtain a fusion mathematical model; Let f1(t) represent the life distribution function of the system in the sea trial test stage, f2(t) represent the life distribution function of the system in the design stage, f3(t) represent the life distribution function of the system in the identification test, and f4(t) represent the life distribution function based on the actual ship data; A multi-source data weighted fusion model is adopted, the number of data sources is set to m, and the fusion mathematical model is: Among them, w i is the weighted value of each data source, f i (t) is the life distribution density function obtained from the reliability data at different development stages.

5. The system reliability modeling method based on multi-source data fusion according to claim 4 is characterized in that: In step S3, since the data sources of the multi-source data are different, it is necessary to determine the weights of the multi-source data. The specific method is as follows: Assume that the product life distributions of each source data support each other, and establish support vectors for m life distributions: S=(S 11 ,S 12 ,…,S 1m ) in, f1(t) represents the life distribution function of the system during the trial phase, f i (t), i=2,3,...,m are the life distribution functions of the system based on system reliability design, test and actual ship data; S represents f i The higher the support of S 1i The smaller the value; The weights of each multi-source data are: Under small sample test conditions, the life distribution obtained from field test data is often not exactly the same as the actual life distribution. The credibility of f1(t) relative to the actual life distribution ρ=w1 is used as the weight of f1(t). The larger w1 is, the greater the weight of the life distribution of field data in the fusion; the larger the amount of field data, the closer its life distribution is to reality and the higher its credibility is.

6. The system reliability modeling method based on multi-source data fusion according to claim 5 is characterized in that: In step S4, the time-varying uncertain environment includes wind speed and wave height, and the specific method of modeling the dynamic characteristics of the environment includes: Discretize the wind speed and wave height states. The wind speed is divided into "light wind, small effect on component failure", "moderate wind, large effect on component failure" and "strong wind, large effect on component failure", which are represented by "1", "2" and "3" respectively; The wave height is divided into "light waves, with less effect on component failure", "medium waves, with greater effect on component failure" and "strong waves, with greater effect on component failure", which are also represented by "1", "2" and "3" respectively; The continuous-time Markov chain is used to model the environmental factors: the two-dimensional environmental factors of wind speed and wave height are transformed into a one-dimensional environment, the minimum generator matrix is ​​Q, and the corresponding transition probability is p ij (t),i,j=1,2,...,9; Reliability R(t) refers to the probability that a product can complete a specified function under specified conditions and within a specified time interval. When the test data is of life type, R(t) is a function of time. Let H represent the diagonal matrix of the failure rate function of the system, where the (k, k) element of H is: That is, the sum of the failure rates of all components under environmental state k, k = 1, 2, ..., 9; Let α represent the probability vector of each environment at the initial moment of the system, where α k represents the probability that the initial state of the equipment is k; let e represent a column vector whose elements are all 1, then the reliability function R(t) of the system can be expressed as: R(t)=αexp((QH)t)e Where exp((QH)t) is the matrix exponential; The mean time between failures (MTBF) is further expressed as: Then the expression of Q is as follows:

7. The system reliability modeling method based on multi-source data fusion according to claim 6 is characterized in that: In step S4, the specific method of converting the two-dimensional environmental factors into a one-dimensional environment is: The two-dimensional environment (1,1) is transformed into the corresponding environment 1; The two-dimensional environment (1,2) is transformed into the corresponding environment 2; The two-dimensional environment (1,3) is transformed into the corresponding environment 3; The two-dimensional environment (2,1) is transformed into the corresponding environment 4; The two-dimensional environment (2,2) is transformed into the corresponding environment 5; The two-dimensional environment (2,3) is transformed into the corresponding environment 6; The two-dimensional environment (3,1) is transformed into the corresponding environment 7; The two-dimensional environment (3,2) is transformed into the corresponding environment 8; The two-dimensional environment (3,3) is transformed into the corresponding environment 9.

8. The system reliability modeling method based on multi-source data fusion according to claim 7 is characterized in that: In step S4, in order to obtain the environmental dynamic characteristic model, it is also necessary to determine the environmental factors, and the specific method includes: If F i (t) and F j (t) respectively represent the product under stress S i and S j The cumulative failure rate under the action of i (t i )=F j (t j ), then the stress S i Stress S j The environmental factors are: For electronic products that obey exponential distribution, the failure rate is constant. If the failure rates in two environments are λ i and λ j , then the environmental factor of environment 1 relative to environment 2 is 9. The system reliability modeling method based on multi-source data fusion according to claim 8 is characterized in that: The method further includes performing simulation based on the multi-source data and the data of the basic model to evaluate the reliability level of the system, and obtaining a system reliability evaluation index by performing statistical analysis on the data obtained from the system failure simulation.

10. The system reliability modeling method based on multi-source data fusion according to claim 9 is characterized in that: The reliability evaluation index includes the reliability and average life of the system.

Images