Method for predicting and evaluating vibration signal state of engine in ascending stage of carrier rocket

Abstract

The invention discloses a multi-layer and multi-factor prediction and evaluation method for a vibration signal state of an engine in a rising stage of a carrier rocket. Comprising the following steps: firstly, analyzing and predicting sampled historical data in a short time by using a more effective feature extraction method of an LSTM deep network, and obtaining higher prediction precision at the same time; a method based on comparative statistics of predicted waveforms and ideal waveforms is adopted, and higher evaluation precision is obtained; obtaining weight factors of various vibration signals in combination with an analytic hierarchy process; and finally, obtaining a health value of the research target from bottom to top by adopting a fuzzy comprehensive evaluation method. The method is suitable for complex fuzzy systems such as carrier rockets, health scores are obtained through historical data interception, data processing, data prediction and comprehensive evaluation and are transmitted back to a ground test launch control platform, ground monitoring personnel can conveniently and accurately master the flight state of the rocket after launch, and the method is used for making rocket launch and flight strategies.

Description

Technical Field

[0001] The present invention belongs to the field of health status assessment of rocket engine vibration signals, and specifically relates to a multi-layer and multi-factor launch vehicle ascent stage engine vibration signal status prediction and assessment method based on deep learning and fuzzy comprehensive evaluation. Background Art

[0002] During the launch and flight of a launch vehicle, the intense thrust pulses and aerodynamic mutations will cause various types of vibrations. During the flight of a rocket, the vibration signals cover low, medium and high frequencies, which seriously affect the normal operation of the equipment structure and electrical system of multiple systems of the rocket. At the same time, low-frequency vibrations will cause astronauts to have difficulty in physical movement and make decisions in a confused manner, which can easily lead to the failure of the rocket launch mission and casualties. According to data from the China Academy of Launch Vehicle Technology, more than 50% of rocket failures are caused by vibrations encountered during the period. The state of the rocket under various vibration conditions is the focus of attention of ground test and launch control personnel. Therefore, fully automatic state assessment can effectively reduce the monitoring pressure of ground personnel and is of great significance in the rocket launch mission that races against time. At present, the vibration signals for launch vehicle engines mainly include the following six categories: POGO vibration, low-frequency vibration, high-frequency vibration, pulse pressure, impact signal and noise. POGO vibration is the pressure generated in multiple engines of the launch vehicle during the flight of the rocket and the periodic vibration generated by the external structure of the rocket, which can easily lead to pressure loss of internal pipelines and damage to instruments. At the same time, when the rocket flies at a high Mach number, the strong collision between the rocket structure boundary and the external gas environment will generate strong pulsating pressure, which is easy to cause further equipment vibration. When multiple rocket engines are ignited and launched, the rocket acceleration increases sharply, and low- and high-frequency signals and impact signals will appear in the thrust system. At the same time, there is widespread noise when a large number of mechanical equipment are working. For the ground, as the load scale and thrust of my country's future launch vehicles gradually increase, the harm caused by vibration will also increase at the same time. Solving the vibration problem is the key to improving the level of my country's aerospace science and technology.

[0003] At present, there are two methods for rocket vibration failure: pre-prevention and post-security. Pre-prevention mainly includes summarizing the failure modes caused by vibration, suppressing the possibility of vibration, and predicting and evaluating the rocket's health status; post-security mainly provides redundant means to maintain the rocket's flight when a failure occurs, with the main purpose of reducing the loss and harm after the accident. Pre-prevention is the key weak link in current research, which is limited by the rapid iteration of rocket failure modes, the technological level of rocket hardware materials, and the difficulty of developing current evaluation technologies. The current mainstream evaluation method is based on historical data and expert experience. According to the trend of telemetry data, the specific state is extrapolated to propose a method to suppress vibration signals. At the same time, the hierarchy of modern equipment is complex, and the correlation between systems and components is strong. Predicting and evaluating a single signal or a signal of the same level cannot fully reflect the state of the object under study. There will be problems of false alarms and false alarms when the actual project is implemented. Therefore, in order to cope with multiple data categories, it is necessary to conduct a comprehensive and integrated evaluation of the multi-level structure system. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting and evaluating the vibration signal state of an engine in the ascent phase of a launch vehicle to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a method for predicting and evaluating the state of vibration signals of an engine in the ascending stage of a launch vehicle, comprising the following steps:

[0006] Step 1: Data preprocessing is the process of cleaning outliers and correcting data formats in the original data, which is a pre-processing step for deep data prediction. Compared with other algorithms, this method is based on the LSTM deep network, which extracts the local dependencies between the previous and next parts of the time series signal with the powerful feature learning ability of the neural network, and completes the prediction of long sequences.

[0007] Step 2: The evaluation set is used as the evaluation index of health assessment, which is obtained by combining subjective experience and objective data analysis. The statistical probability of each evaluation value in the real and predicted data is divided based on multiple thresholds and used as the input of fuzzy comprehensive evaluation;

[0008] Step 3: Weight analysis is based on the analytic hierarchy process (AHP), which decomposes the decision-making system into the overall goal, sub-goals at each level, and evaluation criteria, and fuzzy quantifies the weights of the complex system at multiple levels and multiple indicators;

[0009] Step 4: Weight set generation Based on the weight analysis module, the weight matrix required for fuzzy comprehensive evaluation is generated; the launch vehicle engine system is a typical fuzzy environment, and the details inside the system are difficult to fit and mathematically characterize;

[0010] Step 5: Input the evaluation set and weight set of the prediction results, consider the influence of multiple factors, and obtain the final health score through bottom-up evaluation.

[0011] Preferably, the data prediction uses LSTM deep network for feature extraction, which can more effectively analyze and predict the sampled historical data. LSTM network is a recursive neural network with the characteristics of memory unit, which can better process time series data and extract relevant features in the data to improve prediction accuracy.

[0012] Preferably, the LSTM method is developed from the recurrent neural network RNN ​​method, and an input gate, a forget gate and an output gate are added to the RNN method. The setting of the gate control unit in the network structure improves the applicability of the RNN method and solves the problems existing in the RNN algorithm network. Since the launch vehicle is tested under various vibration conditions, the collected data is always in a dynamic change process. The correlation information before and after the data is the key to predicting the data trend. Compared with the RNN method, the LSTM algorithm improves the prediction ability of long-term data and solves the problems of gradient explosion and gradient disappearance in the data training process of the RNN method.

[0013] Preferably, the evaluation set generation, the process of determining the evaluation set based on the membership function in the traditional fuzzy comprehensive evaluation method is not applicable to the current research object. The evaluation set generation method in the present invention is based on the predicted data rather than the prediction error, and each predicted point of the predicted waveform is included in the consideration range of the evaluation generation, so as to more accurately and objectively evaluate the state of the vibration signal and judge the health of the engine. Compared with the traditional single threshold division method, this method explores the original characteristics of the data and is more objective.

[0014] Preferably, the weight analysis adopts the analytic hierarchy process (AHP) method to obtain the weight factors of various vibration signals. The AHP method is a multi-criteria decision analysis method, which obtains the relative importance of each factor to the research target by comparing and weighting different factors. This method generates a multi-layer multi-factor weight matrix inside the launch vehicle engine based on the analytic hierarchy process, optimizes the general analytic hierarchy process in the evaluation level division method, and makes the final evaluation process more comprehensive.

[0015] Preferably, the analytic hierarchy process subdivides the weight influence relationship into the divided small levels, and performs a weight calculation of the same level on each level, which is conducive to intuitive judgment of complex problems and decision-making. Since the analytic hierarchy process is simple and easy to understand, it can decompose the fuzzy complex system into small systems that are easy to inject expert knowledge, making up for the cognitive limitations of the human brain for fuzzy complex systems, enhancing the objectivity and logic of judgment and decision-making between multiple factors, and more scientifically assigning weights to each indicator factor in the process of evaluating complex systems.

[0016] Preferably, the weight set is generated based on the established evaluation hierarchy, and the weight of each factor in each layer of the evaluation hierarchy is obtained by the hierarchical analysis method. The weight of the sub-target layer can be obtained according to the calculation to form a weight set. Considering the hierarchical complexity of the system equipment, a multi-layer multi-factor weight matrix can be generated in combination with the hierarchical analysis method.

[0017] Preferably, the fuzzy comprehensive evaluation method integrates the weight factors and evaluation results of various vibration signals. This method is applicable to the predictive evaluation process of the fuzzy system. In view of the problems of strong subjectivity, poor applicability, and insufficient risk prevention value for current missions in the current aerospace health assessment system, the autonomy, objectivity, versatility, comprehensiveness, and timeliness of the evaluation are optimized on the basis of meeting the actual needs of aerospace health assessment.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. The vibration signal state prediction and evaluation method of the launch vehicle ascending stage engine is designed to address the current situation where current research focuses on real-time evaluation but ignores future trends, and the health assessment system is not universal. A fuzzy prediction and evaluation system based on vibration data is designed. The evaluation process is formed by combining prediction algorithms, hierarchical analysis methods, box plot methods, and fuzzy comprehensive evaluation methods. The overall process is based on data prediction technology, and the evaluation is performed on high-precision vibration trends, which enhances the reference significance of risk prevention. The evaluation process is applicable to the fuzzy system of launch vehicle engines, with a modular system architecture and a hierarchical evaluation structure, which is easy for technology migration and system expansion.

[0020] 2. The prediction and evaluation method of the vibration signal state of the launch vehicle's ascending stage engine has designed a space division method for the healthy state of the launch vehicle engine based on mathematical statistics to address the problem that the current evaluation method is too subjective. Compared with the traditional threshold judgment method, it reduces the introduction of human experience. The improved scheme explores the distribution characteristics of vibration data, calculates the upper and lower limits of the amplitude distribution of the ideal data of the ground test, and uses this as the basis for division. The method is more objective and avoids the evaluation process from being too dependent on human experience.

[0021] 3. The vibration signal state prediction and evaluation method for the launch vehicle ascent stage engine is based on the characteristics of the launch vehicle engine's complex internal hierarchy, fuzzy system mechanism, and difficulty in comprehensively stratifying the engine's systems and components. The launch vehicle engine evaluation hierarchy is abstracted based on vibration data and weighted. Compared with the traditional hardware stratification method that monitors key components and completely ignores other components, the data-driven evaluation hierarchy division method makes the evaluation results more comprehensive. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings constituting a part of the present application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention:

[0023] Figure 1 is a diagram of an overall evaluation architecture according to an embodiment of the present invention;

[0024] Figure 2 is a weight generation hierarchical graph according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0026] This embodiment provides a method for predicting and evaluating the state of a launch vehicle ascent phase engine vibration signal, comprising the following steps:

[0027] Step 1: Data preprocessing is the pre-processing step for data deep prediction, which involves cleaning outliers and correcting data formats in the original data. Compared with other algorithms, this method is based on the LSTM deep network, which extracts the local dependencies between the previous and next parts of the time series signal and completes the prediction of long sequences by relying on the powerful feature learning ability of the neural network.

[0028] The LSTM method is developed from the recurrent neural network (RNN) method. Since the launch vehicle is tested under various vibration conditions, the collected data is in dynamic change. The correlation information before and after the data is the key to predicting data trends. Compared with the RNN method, the long short-term memory network improves the prediction ability of long-term data and solves the gradient explosion and gradient disappearance problems in the data training process of the RNN method. For actual engineering implementation, LSTM is more stable.

[0029] The LSTM-based prediction method mainly includes the following three steps:

[0030] (1) Data preprocessing: Clean the original vibration data of the launch vehicle;

[0031] (2) Model training and error calculation: Use the LSTM model to train the data;

[0032] (3) LSTM model prediction execution: predict the data waveform, save the data, and calculate the root mean square error (RMSE) of the prediction error value;

[0033] Among them, the error formula is as follows:

[0034] Y i represents each acquisition value in the test data, f(x i ) represents each acquisition point of the predicted data, N represents the total number of acquisition points, and the prediction accuracy is compared by the root mean square error algorithm.

[0035] Step 2: The evaluation set is used as the evaluation index for health assessment, which is obtained by combining subjective experience and objective data analysis. Based on the multi-threshold threshold, the statistical probability of each evaluation value in the real and predicted data is divided, and used as the input of the fuzzy comprehensive evaluation.

[0036] When facing the actual results of the vibration signal prediction of the launch vehicle engine, a method for generating an evaluation matrix of mathematical statistics is proposed, that is, compared with the ideal waveform, and the probability value of the predicted value in each deviation distribution interval is calculated as the value of each evaluation index. The evaluation set generation method of the present invention mainly combines the actual experience of the historical launches of the rocket, and takes the ideal signal state under normal conditions as the benchmark. In the ideal state, the collector is a stable signal with small amplitude fluctuations. For the sake of simplifying the calculation, the fluctuation range can be ignored compared with the set threshold.

[0037] According to the expert experience, the comment set is set as: M = {excellent, good, normal, bad, deteriorating}, corresponding to each color block respectively.

[0038] The average waveform amplitude of the signal before the vibration condition occurs is A, the total number of sampling points of the test data is N, and the upper and lower bound thresholds of the reference waveform are set as U i , D i (0 < i <= 4), and the number of points in each threshold interval is n i .

[0039] Each state interval is set as represents the excellent state, represents the good state, and so on:

[0040]

[0041] Each single evaluation P in the evaluation set matrix is statistically analyzed for the state interval by each sampling point of the predicted data, and a percentage score is given according to the interval probability:

[0042] Predict the six types of vibration types in the element layer and the data of multiple groups of collectors, and compare the real test data, predicted data with the ideal reference state. According to expert experience, the upper and lower bound thresholds of the four segments are set as 0.3, 0.6, 0.8, and 0.9 times the mean value of the first 20% of the real test data, respectively, and the deterioration index is the upper and lower limits greater than 0.9 times the mean value benchmark.

[0043] Step 3: Weight analysis is based on the analytic hierarchy process (AHP). The decision-making system is decomposed into the overall goal, sub-goals at each level, and evaluation criteria. The weights of the complex system are fuzzily quantified at multiple levels and with multiple indicators.

[0044] For the health value assessment of the launch vehicle engine system, there are multiple conditions and quantitative analysis of multiple factors. The electrical environment inside the launch vehicle engine is a typical giant system and fuzzy environment. Therefore, it is impossible to try to globally grasp the quantitative value of the weight of each factor in the system under the same weight standard by relying solely on manual experience. Based on the summary of historical data and past failure modes, experts can obtain the weight relationship between each factor, that is, the quantitative impact of factor A on the entire system relative to factor B, but when there are many factors involved, it is impossible for experts to directly obtain the impact weight relationship of multiple levels and multiple factors. Therefore, the present invention uses the analytic hierarchy process (AHP) to generate a multi-layer multi-factor weight matrix. The analytic hierarchy process was first proposed by American scientist TLSaaty in the 1970s and applied to network system theory and multi-objective comprehensive evaluation. It mainly solves the decision-making problems of complex systems that are interrelated and mutually constrained, and is widely used in aerospace, electrical, operations research and other fields.

[0045] When using the hierarchical analysis method to construct a rocket engine vibration signal model and solve the weight of each type of factor, it is generally divided into the following steps:

[0046] (1) Establish a hierarchical model based on the types and batches of launch vehicle engine vibration signals;

[0047] (2) Based on historical statistics and expert experience, a judgment matrix is ​​constructed according to the weight relationship between factors of the same level, also known as a positive reciprocal matrix:

[0048]

[0049] (3) According to criterion A ij ω=λωCalculate the maximum eigenvalue λ of the positive reciprocal matrix max and eigenvector ω, which is convenient for subsequent calculations;

[0050] (4) Normalize the feature vector and calculate the weight values ​​of all elements contained in each layer:

[0051]

[0052] (5) Perform consistency check on the matrix, in which the credibility of the manually judged matrix is ​​checked. If there is a contradiction in the weight relationship between the elements, the consistency check fails. The credibility of the result of introducing the manually weighted positive reciprocal matrix is ​​increased. The consistency index is:

[0053]

[0054] The corresponding random average consistency index is RI = 0.52, so the consistency ratio is:

[0055] When CR<0.1, it is considered to have passed the test, where n is the order of the established system hierarchy.

[0056] (6) Generate a subjective weight vector.

[0057] Step 4: Weight set generation Based on the weight analysis module, the weight matrix required for fuzzy comprehensive evaluation is generated; the launch vehicle engine system is a typical fuzzy environment, and the details inside the system are difficult to fit and mathematically characterize.

[0058] The weight distribution structure of this system is shown in the attached Figure 2 As shown in the figure, E = {E1, E2, E3, E4, E5, E6}, where E1, E2, E3, E4, E5, E6 represent the six sub-goals of POGO vibration, low-frequency vibration, high-frequency vibration, pulsating pressure, impact, and noise respectively. The overall system is divided into three levels, so the consistency test order n is 3.

[0059] The overall hierarchical structure of the vibration signal of the launch vehicle engine in the ascending phase is divided into three levels, in which the sub-target layer contains a weight vector W(E1, E2, E3, E4, E5, E6); the factor layer includes W1(E 11 ,E 12 ,E 13 ), W2(E 21 ,E 22 ,E 23 ), W3(E 31 ,E 32 ), W4(E 41 ,E 42 ,E 43 ), W5(E 51 ,E 52 ), W6(E 61 ,E 62 ) A total of six weight vectors.

[0060] Step 5: Input the evaluation set and weight set of the prediction results, consider the influence of multiple factors, and obtain the final health score through bottom-up evaluation.

[0061] The fuzzy comprehensive evaluation method has the advantages of clear results and strong systematicity. It is suitable for more comprehensive and in-depth evaluation of multi-level and multi-factor objects in a fuzzy environment. The expert systems currently used in a large number of spacecraft health management use historical experience to establish a knowledge base, and make system evaluation more intelligent through extrapolation, mapping and other methods. But in fact, the operating conditions of launch vehicle engines are unstable, which is closely related to the launch environment and flight orbit. At the same time, the frequent iterations of launch vehicle engine processes and technologies have led to the inability to quickly supplement the expert system knowledge base. Therefore, the evaluation method based on the expert system is not universal, accurate and comprehensive.

[0062] The evaluation steps of the fuzzy comprehensive evaluation method are as follows:

[0063] (1) Establish an overall framework of research objectives and create a multi-level structure to facilitate subsequent bottom-up evaluation;

[0064] (2) Establish a comment set M = {m1, m2...m n};

[0065] (3) Obtain the weight moments W of the factors at each level and the evaluation matrix V of the underlying signal;

[0066] (4) Perform fuzzy evaluation operation to obtain the fuzzy evaluation vector of the target layer: B = V·W, and normalize the vector;

[0067] (5) According to the project requirements, assign the score that can be obtained to each type of evaluation to generate a score vector;

[0068] (6) Calculate the health score to represent the health level of the system.

[0069] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting and evaluating the vibration signal state of an engine in the ascent phase of a launch vehicle, characterized in that: The following steps are involved: Step 1: Data preprocessing: After cleaning the original vibration data of the launch vehicle, use the LSTM model to train the data and calculate the error, predict the data waveform and save the data, and calculate the root mean square error of the predicted error value; Step 2: Analyze the data to obtain the evaluation set, and divide the statistical probability of each evaluation value in the real and predicted data based on the multi-threshold threshold as the input of the fuzzy comprehensive evaluation; Step 3: Based on the analytic hierarchy process (AHP), a rocket engine vibration signal model is constructed and the weight of each factor is solved. The decision-making system is decomposed into the overall goal, sub-goals at each level, and evaluation criteria, and the weight of the complex system is fuzzy quantified. Step 4: Based on the weight analysis module, generate the weight matrix required for fuzzy comprehensive evaluation; Step 5: Based on the evaluation set and weight set of the prediction results, the final health score is obtained through bottom-up evaluation.

2. A method for predicting and evaluating the vibration signal state of a launch vehicle ascending stage engine according to claim 1, characterized in that: The LSTM method is developed from the recurrent neural network (RNN) method. On the basis of the RNN method, an input gate, a forget gate, and an output gate are added, and a gate control unit is set in the network structure.

3. The method for predicting and evaluating the vibration signal state of a launch vehicle ascending stage engine according to claim 1, characterized in that: The evaluation set generation method is based on the prediction data, and each prediction point of the prediction waveform is taken into consideration for evaluation generation, so as to accurately and objectively evaluate the state of the vibration signal.

4. The method for predicting and evaluating the vibration signal state of a launch vehicle ascending stage engine according to claim 1, characterized in that: The hierarchical analysis method subdivides the weight influence relationship into small levels and calculates the weight of each level at the same level.

5. The method for predicting and evaluating the vibration signal state of a launch vehicle ascending stage engine according to claim 4, characterized in that: Based on the established evaluation hierarchy, the weight of each factor in each layer of the evaluation hierarchy is obtained through the hierarchical analysis method. According to the calculation, the weight of the sub-target layer can be obtained to form a weight set.

6. The method for predicting and evaluating the vibration signal state of a launch vehicle ascending stage engine according to claim 1, characterized in that: The fuzzy comprehensive evaluation method is used to comprehensively evaluate the weight factors and evaluation results of various vibration signals.

7. A method for predicting and evaluating the vibration signal state of a launch vehicle ascending stage engine according to claim 6, characterized in that: The fuzzy comprehensive evaluation method comprises the following steps: (1) Establish an overall framework of research objectives and create a multi-level structure to facilitate subsequent bottom-up evaluation; (2) Establish a review set based on the actual object; (3) Obtain the weight matrix W of the factors in each level and the evaluation matrix V of the underlying signal; (4) Perform fuzzy evaluation operation to obtain the fuzzy evaluation vector of the target layer: B = V·W, and normalize the vector; (5) According to the project requirements, assign the score that can be obtained to each type of evaluation to generate a score vector F; (6) Calculate the health score to evaluate the health of the system.

Images