Rotating machinery fault prediction method and system based on health index and particle filtering

Abstract

The present invention discloses a method and system for predicting rotating machinery faults based on health index and particle filtering. First, a suitable health index is extracted based on multi-feature evaluation; then, based on the abnormal change of the amplitude of the health index, abnormal samples are mined; and based on the periodic characteristics of rotating machinery faults, fast spectral kurtosis-multi-point optimal minimum entropy deconvolution is used to enhance early fault features, approach the early fault occurrence node, determine the starting point of the remaining service life prediction, and provide more samples with fault information for subsequent remaining service life prediction; finally, based on the particle filtering method, the probability prediction of the remaining service life of the rotating machinery is realized. The fault prediction method provided by the present invention has the advantages of simple theory, strong feasibility, high confidence in results, etc., and can be applied to fault prediction under the confidence interval of rotating machinery.

Description

Technical Field

[0001] The present invention relates to the field of rotating machinery fault prediction, and in particular to a rotating machinery fault prediction method and system based on health index and particle filtering. Background Art

[0002] Scientific remaining useful life (RUL) prediction can effectively improve the safety performance of mechanical equipment and maximize the equipment capacity. Rotating machinery is widely used in mechanical equipment. Rotating machinery RUL prediction is a systematic project contained in the field of prognostics and health management (PHM). It needs to rely on a series of preliminary technical work (such as data perception, health index extraction, fault prediction, etc.) to ultimately serve the intelligent operation and maintenance of equipment. Around how to achieve RUL prediction of rotating machinery, researchers have carried out a lot of work and proposed a series of RUL prediction methods. Specifically, the prediction work is based on extracting a suitable health index (Health Indicator, HI), determining a scientific prediction starting point, and constructing a suitable prediction degradation model. Existing RUL prediction work rarely systematically considers the above process.

[0003] In response to the problem of feature extraction, researchers have carried out a series of research work. Traditional feature index types include typical time domain features and time-frequency domain features, among which time domain features are more common than frequency domain features. Typical time domain feature indicators include: waveform index, root mean square value, kurtosis index, etc., and time-frequency features include various entropy values. In addition to traditional time domain feature extraction methods, quantitative description methods based on nonlinear signals have also attracted more and more attention from researchers, such as Lyapunov exponents, etc., but the current algorithms have low computational efficiency and take a long time to calculate. With the rise of deep learning methods, state indicator extraction methods based on deep learning have sprung up like mushrooms after rain, but there is still a lack of appropriate mechanism explanations. Scientific HI can accurately characterize the degradation trend of the predicted object and has good monotonicity, robustness and trend. Overall, this research work is relatively weak.

[0004] Based on the different complexity of the research object, the existing RUL prediction technology can be divided into prediction methods based on physical models, prediction methods based on data-driven, and prediction methods that are a mixture of the two. Among them, the prediction method based on physical models needs to be based on a profound study of the failure mechanism of the prediction object. For simple prediction objects (such as flat plates, simply supported beams, etc.), the prediction method based on physical models can obtain higher RUL prediction accuracy. However, for complex prediction objects, their failure mechanisms are relatively complex, and it is difficult to establish a suitable failure model. Therefore, the prediction method based on physical models is not competent for RUL prediction of complex equipment. Thanks to the advancement of data perception methods, more complex equipment degradation state data can be obtained. Subsequently, the RUL prediction method based on data-driven was born and developed. However, the prediction results of the data-driven prediction method are directly affected by the data. When the data is missing or not ideal, it will be difficult to guarantee the prediction accuracy. Then, RUL prediction based on hybrid methods is proposed. Combining the advantages of the prediction method based on physical models and the prediction method based on data-driven prediction, the hybrid prediction method has been widely used in the field of RUL prediction. However, most of the above methods obtain prediction accuracy rather than prediction results. For engineering practice, the existing prediction methods lack overall consideration, are time-consuming, and have complex theories. How to obtain scientific prediction results has become a problem that needs to be solved. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a rotating machinery fault prediction method and system based on health index and particle filtering, which can solve the problem that the existing prediction technology cannot obtain the prediction results scientifically.

[0006] According to the first aspect of the present invention, a method for predicting rotating machinery faults based on health index and particle filtering comprises the following steps:

[0007] S100, acquiring degradation vibration time series data x based on prediction object perception signal , and then extract multiple degradation feature indices F i ;

[0008] S200, based on the multi-evaluation index Eva fusion decision strategy, screen and obtain the optimal degradation feature index F optimal , construct the health index HI;

[0009] S300, based on health index HI amplitude anomaly detection, mining original abnormal samples x abnormal ,Then early fault diagnosis is achieved through the fault enhancement method and the ,starting point of fault prediction is determined;

[0010] S400, based on the early fault diagnosis results, the fault sample information is converted into a priori information of the fault prediction degradation model, and combined with the particle filter probability statistics, the remaining service life prediction result of the rotating machinery is obtained.

[0011] The rotating machinery fault prediction method based on health index and particle filtering according to the first aspect of the present invention has at least the following technical effects:

[0012] (1) The implementation method of the present invention takes into consideration the extraction of health index, determination of prediction starting point, construction of prediction degradation model and uncertainty assessment of prediction result as a whole, and has clear theory and is simple and easy to implement.

[0013] (2) The embodiment of the present invention uses a variety of degradation characteristic indicators as screening and evaluation objects, comprehensively evaluates and obtains a relatively better HI, and then mines abnormal samples based on HI amplitude anomaly detection, thereby providing more faulty samples for subsequent RUL prediction. This process is less time-consuming and has a clear direction.

[0014] (3) The embodiment of the present invention combines HI extraction with a particle filtering method to obtain a prediction result with high accuracy and higher confidence.

[0015] According to some embodiments of the present invention, the degradation characteristic index F in step S100 i Including mean square value, root mean square value, waveform index, peak index, margin index, pulse index, kurtosis index, Shannon entropy and vibration severity.

[0016] According to some embodiments of the present invention, the expression of the mean square value is:

[0017] The expression of the RMS value is:

[0018] The expression of the waveform indicator is:

[0019] The expression of the peak index is:

[0020] The expression of Shannon entropy is

[0021] The expression of the vibration severity is:

[0022] Among them, x i represents the original time series, T is the length of the measured signal, and v(t) represents the vibration speed of the object.

[0023] According to some embodiments of the present invention, the multiple evaluation indicators in step S200 include monotonicity, robustness and trend.

[0024] According to some embodiments of the present invention, the calculation formula of monotonicity is:

[0025]

[0026] Where X = {x k} k＝1:K Represents the health indicator sequence at time t k The corresponding health indicator x k ; K represents the number of health indicators; d / dx=x k+1 -x k Characterizes the difference of the health indicator sequence; No.ofd / dx>0 and No.ofd / dx<0 represent the positive and negative differences respectively; Mon1(X) represents the sum of the derivatives of all health indicators X.

[0027] According to some embodiments of the present invention, the calculation formula of the robustness is:

[0028]

[0029] in Indicates t k The average trend of health indicators at all times, K represents the number of health indicators, X = {x k} k＝1:K Represents the health indicator sequence at time t k The corresponding health indicator x k .

[0030] According to some embodiments of the present invention, the calculation formula of the trend is:

[0031]

[0032] Among them, Tre1(X,T) varies from -1 to 1, K represents the number of health indicators, X = {x k} k＝1:K Represents the health indicator sequence at time t k The corresponding health indicator x k .

[0033] According to some embodiments of the present invention, the fault enhancement method in step S300 adopts the MOMEDA method.

[0034] According to some embodiments of the present invention, the specific steps of step S400 are:

[0035] S401. Construct a prediction degradation model based on historical prior knowledge or historical statistical data regression fitting;

[0036] S402, establish state transfer equation and measurement equation:

[0037] The state transfer equation is

[0038]

[0039] The measurement equation is

[0040] h k =f(θ k ,k)+η,η~N(0,σ0)

[0041] Among them, θ k =[a k ,b k ,…] represents the parameter vector of the degradation model at time k, h k represents the observed value obtained at time k, ω a ,ω b and η represent random variables that obey Gaussian distribution, σ a , σ b and σ0 are the Gaussian distribution variances of the three variables;

[0042] S403, performing degradation model parameter estimation based on the PF method, and updating parameters of the predicted degradation model;

[0043] S404, performing RUL extrapolation prediction of the research object through the parameter-updated prediction degradation model, and obtaining the time when multiple particles reach the set threshold and the number of particles;

[0044] S405: Evaluate the uncertainty of the prediction results to obtain the prediction results of the remaining service life of the rotating machinery.

[0045] According to the second aspect of the present invention, a rotating machinery fault prediction system based on health index and particle filtering includes:

[0046] The memory is used to store computer programs; the processor is used to implement the above-mentioned rotating machinery fault prediction method based on health index and particle filtering when executing the computer program.

[0047] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0049] Figure 1 This is a flow chart of a rotating machinery fault prediction method based on health index and particle filtering in an embodiment of the present invention.

[0050] Figure 2 It is a time series waveform diagram of the full life degradation acceleration of the bearing in the embodiment of the present invention.

[0051] Figure 3 1 is a graph of various degradation characteristic indicators in an embodiment of the present invention.

[0052] Figure 4 It is a diagram of multi-index evaluation results in an embodiment of the present invention.

[0053] Figure 5 1 is a health index diagram in an embodiment of the present invention.

[0054] Figure 6 1 is a diagram of HI amplitude abnormality detection results in an embodiment of the present invention.

[0055] Figure 7 This is the optimal demodulation frequency band diagram in the embodiment of the present invention.

[0056] Figure 8 This is a diagram of early weak fault detection results in an embodiment of the present invention.

[0057] Fig. 9 It is a particle filter probability prediction diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0059] In the description of the invention, it should be understood that descriptions involving orientation, such as up, down, front, back, left, right, etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the invention.

[0060] In the description of the invention, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed" etc. are understood to exclude the number itself, and "above", "below", "within" etc. are understood to include the number itself. If there is a description of the first or the second, it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0061] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention in combination with the specific content of the technical solution.

[0062] refer to Figure 1 , a rotating machinery fault prediction method based on health index and particle filtering, comprising the following steps:

[0063] S100, acquiring degradation vibration time series data x based on prediction object perception signal , and then extract multiple degradation feature indices F i ;

[0064] Among them, the degradation characteristic index F i Including mean square value, root mean square value, waveform index, peak index, margin index, pulse index, kurtosis index, Shannon entropy and vibration severity. The expression of mean square value is The expression for the root mean square value is The expression of the waveform indicator is The expression of the peak index is The expression of Shannon entropy is The expression of vibration severity is: And other indicators, etc.; Among them, x i represents the original time series, T is the length of the measured signal, v(t) represents the vibration speed of the object, and N represents the number of sampling points of a single sample.

[0065]

[0066] I n,m I represents the nth data sample of the m-layer network, and there are n×m such samples in total. n,m It can also be expressed as Where L ker represents the convolution kernel size, j represents the jth point in the sample. * represents convolution, w k,n,m represents the convolution kernel connecting the nth sample in layer m and the kth sample in layer m+1, b n,m For bias. represents the intermediate output of the convolutional layer, and f(·) represents the activation function.

[0067] Then the above degradation characteristic indicators are normalized to facilitate the subsequent degradation characteristic indicator evaluation and optimization.

[0068] S200, based on the multi-evaluation index Eva fusion decision strategy, screen and obtain the optimal degradation feature index F optimal , construct the health index HI;

[0069] Monotonicity, robustness and trend are selected as evaluation indicators. The specific calculation expressions are as follows:

[0070] Monotonicity:

[0071]

[0072] Where X = {x k} k＝1:K Represents the health indicator sequence at time t k The corresponding health indicator x k ; K represents the number of health indicators; d / dx=x k+1 -x k Characterizes the difference of the health indicator sequence; No.ofd / dx>0 and No.ofd / dx<0 represent the positive and negative differences respectively; Mon1(X) represents the sum of the derivatives of all health indicators X. The monotonicity index evaluation results range from 0 to 1. The larger the score, the better the monotonicity.

[0073] robustness:

[0074]

[0075] in, Indicates t k The mean trend of the health index at each moment is generally obtained through smoothing. Similar to the monotonicity evaluation result, the larger the score, the better the robustness.

[0076] Trend:

[0077]

[0078] Among them, the range of Tre1(X,T) is from -1 to 1. When it approaches -1 or 1, the surface health index has a strong correlation with time.

[0079] Then, the monotonicity, robustness and trend evaluation values ​​corresponding to the multiple degradation characteristic indicators extracted in step S100 are respectively obtained, and the performance of the above degradation characteristic indicators is comprehensively evaluated, and finally a relatively better degradation characteristic indicator is obtained as a health index for subsequent RUL prediction.

[0080] S300, based on health index HI amplitude anomaly detection, mining original abnormal samples x abnormal , and then the early fault diagnosis is realized through the fault enhancement method to determine the starting point of fault prediction, as follows:

[0081] Firstly, the constructed HI is subjected to amplitude abnormal change detection to obtain abnormal original samples;

[0082] Then, the early fault feature enhancement is realized through the multi-point optimal minimum entropy deconvolution (MOMEDA) method, and then the fault prediction starting point is determined. The implementation process of the MOMEDA method is as follows:

[0083] The MOMEDA method uses the multi-point D norm as the target cost function to construct an optimal deconvolution filter f = [f1, f2, ... f L ] (where L is the filter length) maximizes the target cost function. The specific expression of the target cost function is:

[0084]

[0085] Among them, y is the target signal, t is the target vector of the same length as the target signal, and t determines the position and weight of the periodic pulse to be convolved.

[0086] In order to find the maximum MDN, the above expression (1) is derived with respect to the filter coefficient f, and the process is as follows:

[0087]

[0088] Where N is the length of the original signal y, let get:

[0089]

[0090] based on And assume If exists, the optimal convolution filter coefficients can be obtained:

[0091]

[0092]

[0093] In order to obtain the pulse period, a series of assumed periods T are discretely obtained within a period range. m , and construct a series of target vectors t m ,Right now:

[0094]

[0095] Among them, δ n represents a pulse at position n in the time series, round(T m ) indicates rounding the period. Then, a series of deconvolution signals y can be obtained. m ,Right now:

[0096]

[0097] Get each group of convolution signals y separately m Spectral kurt:

[0098]

[0099] Then, based on the maximum spectral kurtosis criterion, the fault period is obtained to determine the optimal deconvolution signal.

[0100] In summary, the implementation process of the MOMEDA method is as follows:

[0101] a. Set the range vector r = [r1…r n ] and window function w = [1…1];

[0102] b. Calculate vector t i (1+m*r i )=1,m=0,1,2... Store these vectors in the matrix T=[t1…t n ]middle;

[0103] c. Calculate the improved matrix T * =filter(w,1,T);

[0104] d. Calculate F = (X0X0 T ) -1 XOt * and Y = X0 T F;

[0105] e. From Y = [y1…y n ] Select Max Kurtosis y i , then this vector y is the best filtering result.

[0106] In addition to the MOMEDA method, the fault enhancement method can also use the MED method (Minimum Entropy Deconvolution) or the MCKD method (Maximum Correlated Kurtosis Deconvolution). Compared with the MED and MCKD methods, the MOMEDA method is better.

[0107] S400, based on the early fault diagnosis results, the fault sample information is converted into the prior information of the fault prediction degradation model, and combined with the particle filter probability statistics, the remaining service life prediction result of the rotating machinery is obtained, as follows:

[0108] S401. Establish a scientific degradation model.

[0109] Based on historical prior knowledge or regression fitting of historical statistical data, a suitable prediction degradation model is constructed.

[0110] x=a×exp(b×t)+c×exp(d×t)+δ

[0111] S402 establishes a state transfer equation and a measurement equation.

[0112]

[0113] h k =f(θ k ,k)+η,η~N(0,σ0)

[0114] Among them, θ k =[a k ,b k ,…] represents the parameter vector of the degradation model at time k, h k Represents the observation value obtained at time k. ω a ,ω b and η represent random variables that obey Gaussian distribution, σ a , σ b And σ0 are the Gaussian distribution variances of the above three variables.

[0115] S403, performing degradation model parameter estimation based on PF method (Particle Filter method, Particle Filter).

[0116] Assume that the system has a total of K observation samples. At each k moment, k = 1, 2, ... K, the prior probability can be updated based on the observation samples.

[0117] Prediction: At time k, based on the importance distribution, N particles are sampled: Used to represent the prior distribution of system states;

[0118] Update: Calculate and obtain the weight of each particle.

[0119]

[0120]

[0121] The minimum mean square estimate of the system state x at time k is obtained as:

[0122]

[0123] Resampling: Resample the particles based on the weights to obtain a new set of particles

[0124] S404: RUL extrapolation prediction based on the predicted degradation model.

[0125] The RUL extrapolation prediction of the research object is carried out based on the parameter-updated predictive degradation model to obtain the time and number of particles reaching the set threshold.

[0126] S405: Evaluate the uncertainty of the prediction results to obtain the prediction results of the remaining service life of the rotating machinery.

[0127] The following is an example of a typical representative of rotating machinery, a rolling bearing, to illustrate the specific implementation of the present invention: As a key power transmission component in rotating machinery, the health status of the rolling bearing has a significant impact on the safety of the equipment, and the prediction of the remaining service life of the rolling bearing is of great significance. The method of the present invention is used to predict the remaining service life of the rolling bearing, and the specific implementation steps are as follows:

[0128] S100, obtaining degradation vibration time series data x based on rolling bearing perception signal , and obtain the result as Figure 2 As shown; then extract multiple degradation feature indicators F i , respectively, and the mean square values ​​were extracted RMS value Waveform indicators Peak Index Margin Index Pulse Indicator Kurtosis index Shannon Entropy Vibration severity And the degradation characteristic indicators obtained based on deep learning, etc., the results are as follows Figure 3 shown.

[0129] S200, based on the Eva fusion decision strategy of multiple evaluation indicators, monotonicity, robustness and trend are selected as evaluation indicators, and the evaluation results are obtained as follows Figure 4 As shown; Screening to obtain the optimal degradation characteristic index F optimal , and obtain the result as Figure 5 As shown, RMS and vibration severity are used as health index HI to characterize the degradation state of rolling bearings, wherein amplitude anomaly detection is performed based on RMS, and RUL prediction is performed based on vibration severity.

[0130] S300, based on the abnormal detection of the health index HI amplitude, the results are obtained as follows Figure 6 As shown, the original sample corresponding to the 21st second is taken as the abnormal sample x abnormal Then, based on MOMEDA technology, early fault diagnosis is realized and the results are obtained as follows: Figure 7 and Figure 8 As shown, the fault prediction starting point is determined to be the 21st second.

[0131] S400, based on the HI extracted from the fault samples, i.e. the vibration intensity and the particle filtering method, obtain the prior knowledge of the fault prediction degradation model, and obtain the prediction results of the remaining service life of the rotating machinery under the probability statistical characteristics, such as Fig. 9 shown.

[0132] The present invention also relates to a rotating machinery fault prediction system based on health index and particle filtering, comprising a memory for storing a computer program; and a processor for implementing the rotating machinery fault prediction method based on health index and particle filtering as described above when executing the computer program.

[0133] In summary, the present invention first extracts a suitable health index based on multi-feature evaluation; then, based on the abnormal change of the amplitude of the health index, it mines abnormal samples; and based on the periodic characteristics of rotating machinery faults, it uses fast spectral kurtosis-multi-point optimal minimum entropy deconvolution to achieve early fault feature enhancement, approach the early fault occurrence node, determine the remaining service life prediction starting point, and provide more samples with fault information for subsequent remaining service life prediction; finally, based on the particle filtering method, it realizes the probability prediction of the remaining service life of rotating machinery. The fault prediction method provided by the present invention has the advantages of simple theory, strong feasibility, high confidence in results, etc., and can be applied to fault prediction under the confidence interval of rotating machinery.

[0134] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the knowledge scope of ordinary technicians in the technical field without departing from the purpose of the present invention.

Claims

1. A method for predicting rotating machinery faults based on health index and particle filtering, characterized in that: The following steps are involved: S100, acquiring degradation vibration time series data x based on prediction object perception signal , and then extract multiple degradation feature indices F i ; S200, based on the multi-evaluation index Eva fusion decision strategy, screen and obtain the optimal degradation feature index F optimal , construct the health index HI; S300, based on health index HI amplitude anomaly detection, mining original abnormal samples x abnormal ,Then early fault diagnosis is achieved through the fault enhancement method and the ,starting point of fault prediction is determined; S400, based on the results of early fault diagnosis, converting the fault sample information into prior information of the fault prediction degradation model, combining with particle filter probability statistics, to obtain the prediction result of the remaining service life of the rotating machinery; In step S100, the degradation characteristic index F i Including mean square value, root mean square value, waveform index, peak index, margin index, pulse index, kurtosis index, Shannon entropy and vibration severity; The expression of the mean square value is: The expression of the RMS value is: The expression of the waveform indicator is: The expression of the peak index is: The expression of Shannon entropy is The expression of the vibration severity is: Among them, x i represents the original time series, T is the length of the measured signal, v(t) represents the vibration speed of the object, and N represents the number of sampling points of a single sample; The specific steps of step S400 are: S401. Construct a prediction degradation model based on historical prior knowledge or historical statistical data regression fitting; S402, establish state transfer equation and measurement equation: The state transfer equation is The measurement equation is h k =f(θ k ,k)+η,η:N(0,σ0) Among them, θ k =[a k ,b k ,…] represents the parameter vector of the degradation model at time k, h k represents the observed value obtained at time k, ω a ,ω b and η represent three random variables that obey Gaussian distribution, σ a , σ b and σ0 are the Gaussian distribution variances of the three variables; S403, performing degradation model parameter estimation based on the PF method, and updating parameters of the predicted degradation model; S404, performing RUL extrapolation prediction of the research object through the parameter-updated prediction degradation model, and obtaining the time when multiple particles reach the set threshold and the number of particles; S405: Evaluate the uncertainty of the prediction results to obtain the prediction results of the remaining service life of the rotating machinery.

2. The method for predicting rotating machinery faults based on health index and particle filtering according to claim 1 is characterized in that: The multiple evaluation indicators in step S200 include monotonicity, robustness and trend.

3. The method for predicting rotating machinery faults based on health index and particle filtering according to claim 2 is characterized in that: The calculation formula of monotonicity is: Where X = {x k } k＝1:K Represents the health indicator sequence at time t k The corresponding health indicator x k ; K represents the number of health indicators; d / dx=x k+1 -x k Characterizes the difference of the health indicator sequence; No.ofd / dx>0 and No.ofd / dx<0 represent the positive and negative differences respectively; Mon1(X) represents the sum of the derivatives of all health indicators X.

4. The method for predicting rotating machinery faults based on health index and particle filtering according to claim 2 is characterized in that: The calculation formula of the robustness is: in Indicates t k The average trend of health indicators at all times, K represents the number of health indicators, X = {x k } k＝1:K Represents the health indicator x corresponding to the health indicator sequence at time tk k .

5. The method for predicting rotating machinery faults based on health index and particle filtering according to claim 2, characterized in that: The calculation formula of the trend is: Among them, Tre1(X,T) varies from -1 to 1, K represents the number of health indicators, X = {x k } k＝1:K Represents the health indicator sequence at time t k The corresponding health indicator x k .

6. The method for predicting rotating machinery faults based on health index and particle filtering according to claim 1, characterized in that: The fault enhancement method in step S300 adopts the MOMEDA method.

7. A rotating machinery fault prediction system based on health index and particle filtering, characterized in that: include Memory for storing computer programs; A processor is used to implement the rotating machinery fault prediction method based on health index and particle filtering as described in any one of claims 1 to 6 when executing the computer program.

Images