A Time Tracking Feature-Level Fusion Method for Slow-Varying and Fast-Varying Signals

Abstract

The present invention discloses a time-tracking feature-level fusion method for slow-varying and fast-varying signals, including: obtaining the mean and variance of a preset characteristic frequency on a given time duration for the downsampling result; 5) For the fast-varying signal, according to the mean and variance obtained in step 4), constructing a fast-varying fusion function composed of the statistical characteristics of normal samples and test samples; 6) For the slow-varying signal, according to the mean and variance obtained in step 4), constructing a slow-varying fusion function composed of the statistical characteristics of normal samples and test samples; 7) Obtaining the objective function to be optimized for the fast-varying signal to obtain the fusion parameters of the fast-varying signal; 8) Obtaining the objective function to be optimized for the slow-varying signal to obtain the fusion parameters of the slow-varying signal; 9) Constructing a full-parameter slow-varying signal fusion function; 10) Calculating the final fusion result of the slow-varying and fast-varying signals according to the full-parameter slow-varying signal fusion function. This method can achieve the time-tracking feature-level fusion of slow-varying and fast-varying signals.

Description

Technical Field

[0001] The present invention belongs to the field of health assessment of liquid rocket engines, and relates to a time-tracking feature-level fusion method for slow-varying and fast-varying signals. Background Art

[0002] Liquid rocket engines are the core power systems of China's space launch vehicles, and their health is related to the success or failure of space launches and the development of future reusable rocket technologies. During their service stage, it is necessary to analyze the actual signals to judge their real working processes, understand their mechanical, thermal, and acoustic environments, and timely and accurately evaluate their health states. However, the turbine machinery measurement systems in various countries are relatively large. For example, the number of measurement parameters for one test run of the US F-1 engine is 575, and the test run parameters of China's high-pressure staged combustion liquid oxygen-kerosene rocket engine are about 200. These parameters include static parameters and dynamic parameters, and the dynamic parameters can be further divided into slow-varying parameters and fast-varying parameters. Slow-varying parameters such as pressure, flow rate, temperature, turbine speed, and thrust can basically reflect the working state of turbine machinery. Slow-varying parameters are limited by the resolution of sensors and have a low upper limit of measurement frequency; fast-varying parameters include fluid pressure pulsation and mechanical vibration, etc., which have a high sampling frequency and diverse change characteristics. During the start-up, shutdown, working condition adjustment, and long-term operation stages of liquid rocket engines, the slow-varying signals will have obvious fluctuations or slow degradation, and the fast-varying signals always exhibit non-stationarity and transience. Currently, it is usually difficult to fuse slow-varying and fast-varying signals in fault analysis and health identification, resulting in the difficulty of existing analysis methods to truly characterize the real health state of turbine machinery. How to conduct research on information fusion methods based on actual signals with different physical information, change rates, and sampling frequencies, so as to accurately and effectively evaluate the health state of engines is of great importance.

[0003] Looking at the current theoretical research field of multi-source sensor information fusion, the main fusion theory methods include Bayesian theory, Dempster-Shafer evidence theory, Jeffrey-like rules, random sets, and the combination of the above methods and neural networks in recent years. One of the main difficulties faced in the multi-source fusion field is the lack of a unified mathematical tool and method for accurately describing physical phenomena. At the same time, the diversity and complexity of multi-source signals further restrict the theoretical development and engineering application of information fusion technology. At the same time, there are obvious differences in the physical characteristics and change rates of vibration, fluid pulsation signals, pressure, speed, temperature, and different signals, which bring difficulties to their feature-level information fusion. More importantly, the physical signals of liquid rocket engines during the start-up stage, working condition adjustment stage, and long-term service stage are usually a process that changes with time, and there are certain differences in the change laws. It is more difficult to obtain the information fusion of their dynamic evolution laws over time. Therefore, it is of great significance to develop a time-tracking feature-level fusion method for slow-varying and fast-varying signals. Summary of the Invention

[0004] The object of the present invention is to overcome the above-mentioned disadvantages of the prior art, and provides a time-tracking feature-level fusion method for slow-varying and fast-varying signals, which can realize the time-tracking feature-level fusion of slow-varying and fast-varying signals.

[0005] To achieve the above object, the time-tracking feature-level fusion method for slow-varying and fast-varying signals according to the present invention includes:

[0006] 1) Perform a bilateral second-order synchrosqueezing transform on the original fast-varying signal x(t) to obtain the three-dimensional time-frequency feature of the original fast-varying signal x(t)

[0007] 2) Calculate the instantaneous frequency feature M(u, ζ) of the fast-varying signal according to the three-dimensional time-frequency feature of the original fast-varying signal x(t)

[0008] 3) Downsample the instantaneous frequency feature M(u, ζ) of the fast-varying signal to obtain the downsampling result M(kΔT, ζ);

[0009] 4) Obtain the mean and variance of the preset characteristic frequency on the given time length of the downsampling result M(kΔT, ζ);

[0010] 5) For the fast-varying signal, construct a fast-varying fusion function composed of the statistical features of normal samples and test samples according to the mean and variance obtained in step 4);

[0011] 6) For the slow-varying signal, construct a slow-varying fusion function composed of the statistical features of normal samples and test samples according to the mean and variance obtained in step 4);

[0012] 7) Construct the boundary conditions corresponding to the fast-varying signal, obtain the objective function to be optimized of the fast-varying signal, and obtain the fusion parameter γ of the fast-varying signal;

[0013] 8) Construct the boundary conditions corresponding to the slow-varying signal, obtain the objective function to be optimized of the slow-varying signal, and obtain the fusion parameters λ, k, γ of the slow-varying signal;

[0014] 9) Substitute the fusion parameter γ of the fast-varying signal into the fast-varying fusion function obtained in step 5), substitute the fusion parameters λ, k, γ of the slow-varying signal into the slow-varying fusion function obtained in step 6), and then construct a full-parameter slow-varying signal fusion function according to the fast-varying fusion function and the slow-varying fusion function;

[0015] 10) Calculate the final fusion result of the slow-varying and fast-varying signals according to the full-parameter slow-varying signal fusion function obtained in step 9).

[0016] The specific operation of step 2) is:

[0017] Let the three-dimensional time-frequency feature​ There are n characteristic frequency components, and the center frequency of the i-th characteristic frequency component is ξ 0i , and the frequency modulation bandwidth is Δξ 0i , the signal duration is T, and the quasi-steady signal is compressed to its center frequency ξ 0i . The instantaneous frequency characteristic M(u, ζ) of the rapidly varying signal is as follows:

[0018]

[0019] Among them,

[0020] The downsampling result M(kΔT, ζ) in step 3) is:

[0021]

[0022] In step 5), the rapidly varying fusion function composed of the statistical characteristics of normal samples and test samples is:

[0023]

[0024] Among them, γ is the function to be optimized, and μ i (t) is the time dynamic mean of the normal training samples of the i-th sensor, and X i (t) is the normal test sample value of the i-th sensor.

[0025] In step 6), the slowly varying fusion function composed of the statistical characteristics of normal samples and test samples is:

[0026]

[0027]

[0028] In step 7), the objective function to be optimized for the rapidly varying signal and the optimization results of its parameters are respectively:

[0029]

[0030]

[0031] In step 8), the objective function to be optimized for the slowly varying signal and the optimization results of its parameters are respectively:

[0032]

[0033]

[0034]

[0035] In step 9), the fusion functions of the fully parameterized slowly varying and rapidly varying signals are respectively:

[0036]

[0037]

[0038] The final fusion result of the slow-varying signal is:

[0039]

[0040] The present invention has the following beneficial effects:

[0041] When the time-tracking feature-level fusion method of slow-varying and fast-varying signals according to the present invention is specifically operated, the instantaneous frequency of the fast-varying signal is obtained by using the bilateral second-order synchrosqueezing transform, so that the time-domain index of the fast-varying signal and the original features of the slow-varying signal are used as the overall features of the signal, and the instantaneous frequency of the fast-varying signal is used as the detailed features of the signal, so that it can not only describe the overall operating state of the device, but also focus on the key features and core components of the device, overcoming the defect that the traditional method can only focus on details or only characterize macroscopic features; in addition, the present invention adjusts the fast-varying signal and the slow-varying signal into digital expressions with the same sampling frequency and similar mathematical features by performing time-frequency domain statistical feature transformation on the fast-varying signal and the slow-varying signal, changing the defect that the physical laws of multi-source information in the traditional method are different and there is no unified mathematical tool, so as to realize the time-tracking feature-level fusion of slow-varying and fast-varying signals.

[0042] Further, the present invention constructs a dynamic time-tracking function of the slow-varying signal and the fast-varying signal composed of the hyperbolic tangent function and the modified arctangent function to obtain the time dynamic confidence upper limit function of the fast-varying signal and the confidence interval function of the slow-varying signal; then, by performing time product operations on the characteristic indexes of the fast-varying signal and the true values of the slow-varying signal with their corresponding dynamic time-tracking functions respectively, the normalized time dynamic tracking features of each index are obtained, improving the defect that the time-tracking effect in the current information fusion field is not ideal. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of the present invention;

[0044] Figure 2a is a schematic diagram of y = softplus(Xi(t));

[0045] Figure 2b is for different e r corresponding and the distribution diagram of φ2;

[0046] Figure 2c is for the distribution diagram of φ2 and ψ0;

[0047] Figure 2dFor different u i (t) corresponding ψ0 function graph;

[0048] Figure 2e For different e r Function graph corresponding to ψ0;

[0049] Figure 2f For different σ i (t) corresponding function graph of ψ0;

[0050] Figure 3a Normalized health index graph of the gas generator flow signal;

[0051] Figure 3b Normalized health index graph of the gas generator speed signal;

[0052] Figure 3c Normalized health index graph of the gas generator time-frequency characteristics;

[0053] Figure 3d Normalized health index graph of the gas generator peak-to-peak value;

[0054] Figure 3e Normalized health index graph of the gas generator RMS;

[0055] Figure 3f Normalized health index graph of the gas generator kurtosis;

[0056] Figure 4a Schematic diagram of the rapid-varying signal fusion;

[0057] Figure 4b Schematic diagram of the slow-varying signal fusion;

[0058] Figure 5 Graph of the starting health degree change law after the signal fusion of the gas generator of the liquid rocket engine in the present invention;

[0059] Figure 6a Normalized health index graph of the flow rate of the slow-varying signal of the turbopump;

[0060] Figure 6b Normalized health index graph of the speed of the slow-varying signal of the turbopump;

[0061] Figure 6c Normalized health index graph of the rotor radial displacement of the slow-varying signal of the turbopump;

[0062] Figure 6d Normalized health index graph of the peak-to-peak value of the slow-varying signal of the turbopump;

[0063] Figure 6e Normalized health index graph of the RMS of the slow-varying signal of the turbopump;

[0064] Figure 6f It is the kurtosis-normalized health index diagram of the slow-changing signal of the turbopump;

[0065] Figure 7a It is the schematic diagram of the fast-changing signal fusion of the dynamic health index of the turbopump;

[0066] Figure 7b It is the schematic diagram of the slow-changing signal fusion of the dynamic health index of the turbopump;

[0067] Figure 7c It is the schematic diagram of the slow-fast-changing signal fusion of the dynamic health index of the turbopump. Detailed implementation manners

[0068] In order to enable those skilled in the art of the present technology to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments, and are not intended to limit the scope of the present invention disclosure. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts disclosed in the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0069] The structural schematic diagrams according to the disclosed embodiments of the present invention are shown in the accompanying drawings. These figures are not drawn to scale, and for the purpose of clear expression, some details are enlarged and some details may be omitted. The shapes of various regions and layers shown in the figures, as well as their relative sizes and positional relationships, are only exemplary. In practice, there may be deviations due to manufacturing tolerances or technical limitations, and those skilled in the art can design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0070] Reference Figure 1 , the time-tracking feature-level fusion method for slow-changing and fast-changing signals described in the present invention is used for the health assessment of liquid rocket engines, and includes the following steps:

[0071] 1) Perform a bilateral second-order synchrosqueezing transform on the original fast-changing signal x(t) to obtain the three-dimensional time-frequency characteristics of the original fast-changing signal x(t)

[0072] 2) Assume that the three-dimensional time-frequency characteristics has n characteristic frequency components, the center frequency of the i-th characteristic frequency component is ξ 0i , the frequency modulation bandwidth is Δξ 0i , the signal duration is T, and the quasi-steady signal is compressed to its center frequency ξ0i At this point, the instantaneous frequency characteristic M(u, ζ) of the rapidly varying signal is:

[0073]

[0074] where ζ = [ξ1, ξ2, …… ξ n ,

[0075] 3) Perform downsampling calculation on the instantaneous frequency characteristic M(u, ζ) of the rapidly varying signal obtained in step 2), and the downsampling result M(kΔT, ζ) is:

[0076]

[0077] 4) Obtain the mean and variance of the preset characteristic frequency on the given time duration of the downsampling result M(kΔT, ζ), so that the rapidly varying signal and the slowly varying signal have similar mathematical expression laws, that is

[0078]

[0079] 5) To avoid the phenomenon that the mathematical expectation of the signal is zero, resulting in infinite health indicators or confidence intervals in subsequent transformations, the system signal state matrix S(t) composed of n sensors and its test sample signal S mea (t) are approximated and non-zero transformed using the softplus function. As shown in Figure 2, theoretical proof shows that the softplus function approaches zero on the negative side of the independent variable and approaches the true value of the independent variable on the non-negative side, that is:

[0080]

[0081] S(t) = [S1(t), S2(t), …… S n (t)]

[0082]

[0083] where S i (t) and are the test values of the i-th sensor signal matrix corresponding to the test system and the test sample respectively.

[0084] For the test sample composed of multi-source signals, the statistical indicators representing the health state of the mechanical equipment are within the confidence interval, that is, the statistical value S mea (X i (t)) of each sensor measurement signal all satisfy:

[0085]

[0086] where is the normalized health index of the i-th sensor.

[0087] The normalized indexes of the time-varying characteristics of all signals are:

[0088]

[0089] where q is a preset non-negative integer;

[0090] 6) For rapidly varying signals, construct a rapidly varying fusion function composed of the statistical characteristics of normal samples and test samples, that is:

[0091]

[0092] where γ is the function to be optimized, μ i (t) is the time dynamic mean of the normal training samples of the i-th sensor, and X i (t) is the normal test sample value of the i-th sensor.

[0093] 7) For slowly varying signals, construct a slowly varying fusion function composed of the statistical characteristics of normal samples and test samples, that is:

[0094]

[0095]

[0096] where λ, k, γ are the target parameters to be optimized.

[0097] 8) Construct the boundary conditions corresponding to the rapidly varying signals. The test values of the rapidly varying signals are within the normal sample confidence interval [0, μ i (t) + 3σ i (t)], and the health index is close to or equal to 1 (assumed to be 1 - e r ). Outside the interval, it approaches zero; at the same time, the greater the deviation of the test sample from the statistical interval, the closer its health index is to zero; to obtain the boundary equation with time tracking effect, and obtain the target function to be optimized and its parameter solution results of the rapidly varying signals, that is:

[0098]

[0099] Obtain the boundary equation with time tracking effect, and the parameters of the fusion function of the rapidly varying signals are:

[0100]

[0101] 9) Construct the boundary conditions corresponding to the slowly varying signals. The test values of the slowly varying signals are within the normal sample confidence interval [μ i (t) - 3σ i (t), μ i (t) + 3σi (t)], the health index is close to or equal to 1, assumed to be 1 - e r , and approaches zero outside the interval; at the same time, the greater the deviation of the test sample from the statistical interval, the closer its health index should be to zero, that is:

[0102]

[0103] Obtain the boundary equation with time tracking effect, and the fusion function parameters of the slow - varying signal are:

[0104]

[0105]

[0106] 10) Substitute the fusion parameter γ of the fast - varying signal into the fast - varying fusion function to obtain the fast - varying signal fusion function with all parameters; substitute the fusion parameters λ, k, γ of the slow - varying signal into the slow - varying fusion function, and the fusion functions of the fast - varying and slow - varying signals with all parameters are respectively:

[0107]

[0108]

[0109] 11) Perform time - dynamic normalization on the fusion result of the slow - varying and fast - varying signals, and the final fusion result of the slow - and fast - varying signals is:

[0110]

[0111] 12) For a mechanical system with n measurement sensors, the health degree characterized by each sensor is within the interval. At the same time, the health index weights of each fast - varying signal and slow - varying signal are obtained respectively, that is:

[0112]

[0113]

[0114] Among them, w i-slow is the weight of the i - th slow - varying signal, and w i-rapid is the weight of the i - th fast - varying signal.

[0115] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above - mentioned are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A time-tracking feature-level fusion method for slow-varying and fast-varying signals, characterized in that Including: 1) Performing a bilateral second-order synchrosqueezing transform on the original rapidly varying signal to obtain a three-dimensional time-frequency feature expression of the original rapidly varying signal; 2) Calculating the instantaneous frequency feature of the rapidly varying signal according to the three-dimensional time-frequency feature expression of the original rapidly varying signal; 3) Downsampling the instantaneous frequency feature of the rapidly varying signal to obtain a downsampling result; 4) Obtaining the mean and variance of a preset characteristic frequency on a given time duration for the downsampling result; 5) For the rapidly varying signal, according to the mean and variance obtained in step 4), constructing a rapidly varying fusion function composed of statistical characteristics of normal samples and test samples; 6) For the slowly varying signal, according to the mean and variance obtained in step 4), constructing a slowly varying fusion function composed of statistical characteristics of normal samples and test samples; 7) Constructing boundary conditions corresponding to the rapidly varying signal, obtaining the objective function to be optimized for the rapidly varying signal, and obtaining the fusion parameters of the rapidly varying signal; 8) Constructing boundary conditions corresponding to the slowly varying signal, obtaining the objective function to be optimized for the slowly varying signal, and obtaining the fusion parameters of the slowly varying signal; 9) Substituting the fusion parameters of the rapidly varying signal into the rapidly varying fusion function obtained in step 5), substituting the fusion parameters of the slowly varying signal into the slowly varying fusion function obtained in step 6), and then constructing a fusion function of the slowly varying and rapidly varying signals with full parameters according to the rapidly varying fusion function and the slowly varying fusion function; 10) Calculating the final fusion result of the slowly varying and rapidly varying signals according to the full-parameter slowly varying signal fusion function obtained in step 9).

2. The time tracking feature-level fusion method for slow-varying and fast-varying signals according to claim 1, wherein The specific operation of step 2) is: Let the three-dimensional time-frequency feature expression There are n characteristic frequency components, and the center frequency of the i-th characteristic frequency component is ξ 0i , the frequency modulation bandwidth is △ξ 0i , the signal duration is T, and the quasi-steady signal is compressed to its center frequency ξ 0i . The instantaneous frequency feature M(u, ζ) of the rapidly varying signal is as follows: Among them, ζ = [ξ1, ξ2, …… ξ n , 3. The time-tracking feature-level fusion method for slow-varying and fast-varying signals according to claim 2, characterized in that The downsampling result M(kΔT, ζ) in step 3) is: where k is the target parameter to be optimized.

4. The time tracking feature-level fusion method for slow-varying and fast-varying signals according to claim 3, wherein In step 5), the rapidly varying fusion function composed of statistical characteristics of normal samples and test samples is: where γ is the function to be optimized, and μ i (t) is the time dynamic mean of the normal training samples of the i-th sensor, and X i (t) is the normal test sample value of the i-th sensor.

5. The time tracking feature-level fusion method for slow-varying and fast-varying signals according to claim 4, wherein In step 6), the slowly varying fusion function composed of statistical characteristics of normal samples and test samples is: where λ, k, γ are target parameters to be optimized, and e is the natural constant.

6. The time-tracking feature-level fusion method for slow-varying and fast-varying signals according to claim 5, wherein The objective function to be optimized for the rapidly varying signal and the solution result of the parameters in step 7) are respectively:

7. The time-tracking feature-level fusion method for slow-varying and fast-varying signals according to claim 6, wherein The objective function to be optimized for the slowly varying signal in step 8) is:

8. The time-tracking feature-level fusion method for slow-varying and fast-varying signals according to claim 7, wherein The rapidly varying signal fusion function and the slowly varying signal fusion function with full parameters in step 9) are respectively:

9. The method for time-tracking feature-level fusion of slow-varying and fast-varying signals according to claim 8, wherein The final fusion result of the slowly varying and rapidly varying signals is: Among them, G(X i (t), μ i (t)) is the final fusion result function of the slow-varying signal.

Images