CN115859777A - Method for predicting service life of product system in multiple fault modes - Google Patents
Method for predicting service life of product system in multiple fault modes Download PDFInfo
- Publication number
- CN115859777A CN115859777A CN202211395606.0A CN202211395606A CN115859777A CN 115859777 A CN115859777 A CN 115859777A CN 202211395606 A CN202211395606 A CN 202211395606A CN 115859777 A CN115859777 A CN 115859777A
- Authority
- CN
- China
- Prior art keywords
- performance degradation
- life prediction
- tested
- clustering
- cluster
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for predicting the service life of a product system in a multi-fault mode, which comprises the following steps: the method is characterized in that a performance degradation path identification model and a residual service life prediction model are integrated comprehensively, systems with different performance degradation paths are distinguished to improve the prediction precision of the residual service life, organic fusion of the two prediction models is realized, in the key technology, a performance degradation feature extraction model based on data dimension reduction, a performance degradation path clustering model based on track clustering and a residual service life prediction model based on a deep neural network are integrated in sequence, and the residual service life prediction with high consistency, high accuracy and high reliability is realized for subsystems and parts with service life degradation features of a product system.
Description
Technical Field
The invention relates to the technical field of service life prediction, in particular to a service life prediction method of a product system in a multi-fault mode.
Background
The fault prediction and health management technology can track and predict the running health state of the equipment and the system, and can give timely and accurate maintenance suggestions before faults occur, thereby effectively ensuring the long-term safe and reliable running of the equipment and the system.
The existing research mainly aims at the individual difference and degradation characteristics of equipment or a system to carry out life prediction, and the research on the influence of different failure modes on the prediction result is less. For example, the state parameters of the rocket control system with typical time series characteristics reflect changes in the state parameters due to system degradation under the influence of time stress, and the remaining service life of the system can be predicted through massive historical data by adopting a data-driven method. However, due to the general unstable situations that the state parameters of the rocket control system are complex, the system has a plurality of failure modes, and the like, the life prediction method based on data driving has three problems: (1) rocket control system state parameters are complex: the rocket control system has more parts, more state parameters needing to be recorded, and a large amount of noise interference exists in the work, so that the service life prediction is influenced; (2) control system multiple failure modes: the rocket control system has a plurality of fault modes, the rocket control system reflects that a plurality of performance degradation paths exist in a system which is not in service life, and the influence of different performance degradation paths on the residual service life is different, so that the service life prediction is influenced; (3) the control system long-time sequence: the time sequence of the state parameter data of the rocket control system is long, and degradation information in the sequence is difficult to capture and learn, so that the life prediction is influenced. Due to the three problems, accurate life prediction cannot be carried out on the rocket control system at present.
Disclosure of Invention
The invention provides a method for predicting the service life of a product system in a multi-fault mode, which aims to solve the technical problem that the accurate service life prediction cannot be carried out on the product system, such as a rocket control system.
The invention provides a method for predicting the service life of a product system in a multi-fault mode, which comprises the following steps:
analyzing and processing all time sequence data of a plurality of product systems under a multi-fault mode as a training set, and determining a plurality of cluster categories of performance degradation paths of the product systems of the training set;
according to the training set data of the product system of each clustering class, a service life prediction model corresponding to each clustering class is constructed and trained to obtain a well-trained service life prediction model of each clustering class;
analyzing and processing time series data before the product system in the multi-fault mode is in fault as a test set, determining a plurality of cluster categories of a performance degradation path of the product system of the test set, and testing the trained life prediction model of each cluster category by using the test set data of the product system of each cluster category to obtain a tested life prediction model of each cluster category;
the method comprises the steps of analyzing and processing time sequence data of a product system to be tested, determining a clustering category of a performance degradation path of the time sequence data, and performing life prediction processing by using a tested life prediction model corresponding to the clustering category to obtain the residual service life of the product system to be tested.
Preferably, the determining the clustering category of the time-series data by analyzing the time-series data of the product system to be tested includes:
performing data smoothing noise reduction processing on the time sequence data of the product system to be detected to obtain time sequence data with random noise filtered;
performing Principal Component Analysis (PCA) processing on the time sequence data with the random noise filtered to obtain performance degradation characteristics related to a performance degradation path of the product system to be tested;
and clustering the time sequence data of the performance degradation characteristics to obtain the cluster category of the performance degradation path of the product system to be detected.
Preferably, the obtaining of the time-series data of the performance degradation feature related to the performance degradation path of the product system to be tested by performing Principal Component Analysis (PCA) on the time-series data of the filtered random noise includes:
and carrying out Principal Component Analysis (PCA) processing on the time sequence data with the random noise filtered to obtain a plurality of PCA characteristics of the time sequence data with the random noise filtered, and taking the PCA characteristics with the variance contribution rate of the PCA characteristics larger than a set threshold value as the performance degradation characteristics related to the performance degradation path of the product system to be tested.
Preferably, the obtaining of the cluster type of the system performance degradation path of the product to be tested by clustering the time series data of the performance degradation features comprises:
calculating the similarity between the performance degradation characteristic time sequence data by using a Dynamic Time Warping (DTW) algorithm;
and performing K-Medoids clustering on the similarity between the performance degradation characteristic time sequence data to obtain the clustering category of the performance degradation path of the product system to be tested.
Preferably, the step of performing life prediction processing by using the tested life prediction model corresponding to the cluster category to obtain the remaining service life of the product system to be tested includes:
determining a tested life prediction model corresponding to the cluster type according to the cluster type of the performance degradation path of the product system to be tested;
and carrying out life prediction processing on the product system to be tested by using the tested life prediction model corresponding to the cluster category to obtain the residual service life of the product system to be tested.
Preferably, the analyzing and processing all time series data of the plurality of product systems in the multiple failure modes as a training set, and determining a plurality of cluster categories of the performance degradation paths of the product systems of the training set includes:
sequentially carrying out data smoothing noise reduction processing and Principal Component Analysis (PCA) processing on the training set data to obtain performance degradation characteristics related to performance degradation paths of the multiple product systems;
and clustering the time series data of the performance degradation characteristics related to the performance degradation paths of the multiple product systems by using a Dynamic Time Warping (DTW) algorithm and a K-Medoids clustering method to obtain multiple clustering categories of the performance degradation paths of the product systems of the training set.
Preferably, the constructing and training a life prediction model corresponding to each cluster category according to the training set data of the product system of each cluster category to obtain the trained life prediction model of each cluster category comprises:
according to each cluster category, constructing a life prediction model of each cluster category;
and training the life prediction model of each cluster type according to the training set data of the product system of each cluster type to obtain the trained life prediction model of each cluster type.
Preferably, the life prediction model is a product system residual service life prediction model under multiple failure modes based on a Transformer deep neural network.
The invention has the beneficial effects that: by carrying out recession characteristic extraction on complex state parameter data of a product system and carrying out performance degradation path identification, residual service life prediction is carried out on the basis, and therefore the aim of improving the accuracy of service life prediction is achieved.
Drawings
FIG. 1 is a flow chart of a method for rocket control system life prediction in multiple failure modes provided by the present invention;
FIG. 2 is a detailed flow chart of a method for predicting the service life of a rocket control system in a multi-fault mode according to the invention;
FIG. 3 is a schematic diagram of a simulation model of a power supply module of the ultra-wideband wireless communication system provided by the invention;
FIG. 4 is a time domain characteristic diagram of output end ripples of the average value of peak-to-peak value, variance, root mean square, minimum value and absolute value of No. 1 system sample in two failure modes of inductance L1 degradation and capacitance C3 degradation provided by the invention;
fig. 5 is an output end ripple time domain characteristic diagram of an average value of peak-to-peak value, variance, root mean square, minimum value and absolute value of output end ripple time sequence characteristics of a system 1 sample after being processed in two fault modes of inductance L1 degradation and capacitance C3 degradation provided by the invention;
FIG. 6 is a summary diagram of the ripple time domain characteristics of the output terminal of the average of the peak-to-peak value, variance, root mean square, minimum, and absolute values of the ripple time sequence characteristics of the output terminals of all the system samples provided by the present invention;
FIG. 7 is a bar chart of the PCA contribution to the ripple time domain characteristics at the output end of the simulation model 1 of the power supply module of the ultra-wideband wireless communication system provided by the present invention;
FIG. 8 is a schematic diagram showing a summary of results of all sample characteristics 1 of an electrical module simulation model 1 of an UWB wireless communication system according to the present invention;
FIG. 9 is a schematic diagram showing a summary of similarity results of all samples of an electrical module simulation model 1 of an UWB wireless communication system according to the present invention;
FIG. 10 is a schematic diagram of a performance degradation path identification result of a power supply module simulation model 1 of the ultra-wideband wireless communication system provided by the invention;
FIG. 11 is a diagram illustrating the predicted result of the transform deep neural network model provided in the present invention.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention. In the following description, suffixes such as "module", "part", or "unit" used to denote elements are used only for facilitating the explanation of the present invention, and have no peculiar meaning in itself. Thus, "module", "component" or "unit" may be used mixedly.
The invention integrates the performance degradation path identification model and the residual service life prediction model, and distinguishes systems with different performance degradation paths to improve the prediction precision of the residual service life and realize the organic fusion of the two prediction models. In the key technology, a performance degradation feature extraction model based on data dimension reduction, a performance degradation path clustering model based on track clustering and a residual service life prediction model based on a deep neural network are integrated in sequence, and residual service life prediction with high consistency, high accuracy and high credibility is realized for subsystems and parts with service life degradation features of a control system. The invention takes a rocket control system as an example.
Fig. 1 is a flowchart of a method for predicting the life of a rocket control system in multiple failure modes, as shown in fig. 1, including:
step S101: analyzing and processing all time sequence data of a plurality of rocket control systems in a multi-fault mode as a training set, and determining a plurality of clustering categories of performance degradation paths of the rocket control systems of the training set;
step S102: establishing and training a life prediction model corresponding to each clustering class according to the training set data of the rocket control system of each clustering class to obtain a trained life prediction model of each clustering class;
step S103: analyzing and processing time series data before the rocket control system in a multi-fault mode fails as a test set, determining a plurality of cluster types of performance degradation paths of the rocket control system of the test set, and testing the trained life prediction model of each cluster type by using the test set data of the rocket control system of each cluster type to obtain a tested life prediction model of each cluster type;
step S104: the method comprises the steps of analyzing and processing time sequence data of a rocket control system to be tested, determining a clustering class of a performance degradation path of the time sequence data, and performing life prediction processing by using a tested life prediction model corresponding to the clustering class to obtain the remaining service life of the rocket control system to be tested.
Further, the determining the cluster type of the time sequence data by analyzing and processing the time sequence data of the rocket control system to be tested comprises: performing data smoothing noise reduction processing on the time sequence data of the rocket control system to be tested to obtain time sequence data with random noise filtered; performing Principal Component Analysis (PCA) processing on the time sequence data with the random noise filtered to obtain performance degradation characteristics related to a performance degradation path of the rocket control system to be tested; and clustering the time sequence data of the performance degradation characteristics to obtain the clustering category of the performance degradation path of the rocket control system to be tested. Performing Principal Component Analysis (PCA) processing on the time sequence data with the random noise filtered out to obtain performance degradation characteristic time sequence data related to a performance degradation path of the rocket control system to be tested comprises the following steps: and carrying out Principal Component Analysis (PCA) processing on the time sequence data with the random noise filtered to obtain a plurality of PCA characteristics of the time sequence data with the random noise filtered, and taking the PCA characteristics with the variance contribution rate of the PCA characteristics larger than a set threshold value as performance degradation characteristics related to a performance degradation path of the rocket control system to be tested. The clustering of the time sequence data of the performance degradation characteristics to obtain the clustering category of the performance degradation path of the rocket control system to be tested comprises the following steps: calculating the similarity between the performance degradation characteristic time sequence data by using a Dynamic Time Warping (DTW) algorithm; and performing K-Medoids clustering on the similarity between the performance degradation characteristic time sequence data to obtain the clustering category of the performance degradation path of the rocket control system to be tested.
Specifically, the obtaining of the remaining service life of the rocket control system to be tested by using the tested service life prediction model corresponding to the cluster category includes: determining a tested life prediction model corresponding to the cluster type according to the cluster type of the performance degradation path of the rocket control system to be tested; and carrying out life prediction processing on the rocket control system to be tested by using the tested life prediction model corresponding to the cluster category to obtain the residual service life of the rocket control system to be tested.
Further, the analyzing and processing all time series data of the plurality of rocket control systems in the multi-fault mode as a training set, and the determining a plurality of cluster categories of the performance degradation paths of the rocket control systems of the training set includes: sequentially carrying out data smoothing noise reduction processing and Principal Component Analysis (PCA) processing on the training set data to obtain performance degradation characteristics related to performance degradation paths of the plurality of rocket control systems; and clustering the time sequence data of the performance degradation characteristics related to the performance degradation paths of the plurality of rocket control systems by using a Dynamic Time Warping (DTW) algorithm and a K-Medoids clustering method to obtain a plurality of clustering categories of the performance degradation paths of the rocket control systems of the training set.
Further, the constructing and training a life prediction model corresponding to each cluster category according to the training set data of the rocket control system of each cluster category to obtain the trained life prediction model of each cluster category comprises: according to each cluster category, constructing a life prediction model of each cluster category; and training the service life prediction model of each cluster type according to the training set data of the rocket control system of each cluster type to obtain the trained service life prediction model of each cluster type.
Preferably, the service life prediction model is a rocket control system residual service life prediction model under multiple failure modes based on a Transformer deep neural network.
Fig. 2 is a detailed flowchart of a method for predicting the service life of a rocket control system in multiple failure modes, as shown in fig. 2, including:
the method comprises the following steps: performance degradation feature extraction
Firstly, sequentially obtaining data of all time sequences to be processed, analyzing the state parameter time sequences of s rocket control systems, and recording as X 1 ,X 2 ,…,X s For the control system i, a time and b state parameter data points are collected together, and then the time sequence X is obtained i Can be expressed as
Random noise in original data is filtered through a data smoothing noise reduction technology based on a local weighted regression method, degradation information of state parameter data is reserved, and therefore the degradation information in the original data is purified, and the method specifically comprises the following steps:
(1) And determining a window range, wherein the window range is used for controlling the smooth scale of the local weighted regression data.
(2) Within a certain window range n, for all points q therein k K =1,2, …, n, by a weighting function ω k (q i ) To q is i And developing d-order polynomial fitting.
(3) Q is obtained through calculation i Fitting value p of i Used in place of q i 。
Function omega k (q i ) The distribution of the weight values is determined, and a more common weight function is selected, and the formula is as follows.
For different independent variable values, parameters in the model change along with the change of the independent variable values, and in the range of an independent variable space, the model automatically gives an estimated value of a function subjected to local weighted regression, so that the noise reduction processing work of the state parameter data of the control system is realized.
And (3) realizing the dimensionality reduction of the state parameters of the control system by adopting a Principal Component Analysis (PCA) algorithm, reconstructing a feature space by utilizing variance importance contribution, and mining and extracting features with higher importance contribution to realize data dimensionality reduction. Recording the state degradation matrix of the rocket control system as X, wherein each row represents different monitoring parameters X k ＝(x k1 ,x k2 ,...,x kn ) Each column represents a degeneration process of the parameter. A certain performance state of the drive train may be defined by a random variable x k Description, then x k The covariance matrix of (a) is:
in the formula, N is the number of sampling points of the degradation state,
mean values of the monitored parameters:
solving for R x All eigenvalues λ of i (i =1,2, · n) and a feature vector v i The characteristic value λ i Arranging in the order from big to small: lambda [ alpha ] 1 ＞λ 2 ＞...＞λ n Corresponding feature vector is v i (i =1,2,. N). Sample x j Projection onto a feature vector v i The principal components corresponding to the direction are obtained as follows:
all the feature vectors are expanded into an n-dimensional orthogonal space, and x is projected to the orthogonal space to obtain corresponding n-dimensional principal components. The larger the eigenvalue corresponding to the eigenvector is, the larger its contribution at the time of reconstruction is, and the smaller the eigenvalue is, the smaller the contribution of the eigenvector at the time of reconstruction is. Is provided withThe first m principal components in orthogonal space are y 1 ,y 2 ,...,y m The cumulative variance contribution rate is:
when the cumulative variance contribution rate of the current several principal components is large enough, for example, h (m) > 95%, that is, more than 95% of original data information is retained in the several principal components, taking the first m (m < n) principal components to represent the original information, and obtaining a system performance degradation characteristic matrix Y (s is a control system number) after dimension reduction:
Y＝{Y 1 ,Y 2 ,…,Y s }。
step two: performance degradation path clustering
Because the rocket control system has a plurality of failure modes, the rocket control system is reflected to have a plurality of performance degradation paths on the system which is not in service life, and the influence of different performance degradation paths on the residual service life has difference, the performance degradation paths of the rocket control system need to be clustered from the degradation characteristics extracted in the first step, and the clustering process can be divided into two sub-steps: and calculating the path similarity and the path clustering.
Firstly, calculating the similarity of characteristic time sequence tracks among different rocket control system samples by a Dynamic Time Warping (DTW) method, and extracting the sample data vector Y of the rocket control system after the physical sign is extracted a ＝[y a1 ,y a2 ,…,y am ] T ，Y b ＝[y b1 ,y b2 ,…,y bn ] T (a, b are system sample numbers, m, n are corresponding operating cycle numbers), the specific steps are as follows:
(1) Constructing an m x n matrix grid with matrix elements (i, j) representing x ai And x bj Distance d (x) between two points ai ,x bj ) Each matrix element (i, j) represents x ai And x bj And (4) aligning.
(2) Find a regular path through this trellis and represent by W: w = [ W = 1 ,w 2 ,…,w k ],max{m,n}≤k＜m+n-1
(3) Path boundary condition constraints, i.e. w 1 ＝(1,1)，w k = m, n, ensuring that the selected path must start from the common start point of the sequence and end at the common end point of the sequence.
(4) Path continuity constraint if w p-1 = (r ', s'), then for the next point w of the path p-1 = (r, s) needs to meet the requirements that (r-r ') is less than or equal to 1 and (s-s') is less than or equal to 1, and X is ensured a And X b Each coordinate in W appears.
(4) Path monotonicity constraint if w p-1 = (r ', s'), then for the next point w of the path p-1 = (r, s) needs to satisfy (r-r ') > 0 and (s-s') > 0, the points above the constraint W must be monotonic over time.
(5) Repeating (2), (3) and (4) to obtain the path with the minimum regular cost
K in the denominator is used to compensate for the different length of the warping path.
Obtaining a distance matrix D (s is the number of control systems) after DTW:
and then automatically classifying similar samples into the same category by adopting a K-Medoids method for the distance matrix D, and finally, the data points in the same group should have similar attributes and/or characteristics, and the data points in different groups should have highly different attributes and/or characteristics, and the specific steps are as follows:
(1) Randomly selecting k (k =2 in the case according to the number of specific case fault modes) samples as initial clustering centers a = [ a ] 1 ,a 2 ,…,a k ]。
(2) For each sample x i And respectively calculating the distances of the k clustering centers, and marking the distances to the clustering center with the minimum distance.
(3) Recalculating cluster mean points
Cluster center a j ＝min(a,a j '), i.e., the new cluster center is the closest sample point to the cluster mean point.
(4) And (4) repeating the steps (2) and (3) circularly until the conditions such as the iteration times, the minimum error change and the like meet the requirements.
Clustering system samples through a performance degradation path similarity matrix of the rocket control system to obtain clustering results (performance degradation path 1 or performance degradation path 2) of different system samples, thereby realizing identification of the rocket control system according to a performance degradation track time sequence.
Step three: remaining useful life prediction
The Transformer deep neural network obtains the correlation relation and the attention matrix between the model input and the model output completely through an attention mechanism, so that the Transformer deep neural network can allow more parallelization, and therefore higher processing efficiency and prediction quality are obtained. The model main body framework is divided into three parts of position coding, coder and full connection layer, and the specific steps of the model training process are as follows:
(1) Two transform life prediction models are respectively established, and correspond to two performance degradation paths (a performance degradation path 1 and a performance degradation path 2). Dividing all aviation control systems into two types according to performance degradation paths, and respectively dividing all time sequence data X of the control systems after preprocessing and smooth noise reduction processing i As input to the model (for control system i, a total of a time and b state parameter data points were collected):
(2) Converting the model input after position coding into the input of the encoder, transmitting the input into the encoder, attaching position information to the input, and comparing the position information with the input X i Position encoding PE is performed so that the input of the encoder is X i ′＝{X i,1 +PE 1 ,X i,2 +PE 2 ,…,X i,a +PE n And (4) calculating the formula of PE as follows.
Wherein j is X i,j In the sequence X i K is a parameter x i,j,k In X i,j Position of (d) model The dimension for coding w is 5.
(3) The encoder part comprises 3 encoder layers, each encoder layer comprises four steps of a multi-head self-Attention mechanism, residual connection and normalization, a feed-forward network, residual connection and normalization, and finally the self-Attention moment array Attention is transmitted into the full-connection layer, the Attention calculation formula is as follows, d k Is the square root of the K dimension:
where K and Q must satisfy the preconditions (otherwise no computation is possible) in the same high-dimensional space, and V does not necessarily need to be in the same high-dimensional space as K, Q, but only needs to satisfy the output of the final model and V in the same high-dimensional space.
(4) And taking the output of the encoder as the input of the full connection layer, and finally converting the output into the single-step output of the final model through the operation of the full connection layer, namely the predicted value of the RUL.
(5) And calculating loss by taking the error between the RUL predicted value and the RUL actual value as a loss function, updating relevant parameters in the network through back propagation, and repeating training until the loss meets the requirement.
The specific steps of the model testing process are as follows:
(1) Sampling the entire time-series data X of this control system i The control system is obtained after performance degradation feature extraction and performance degradation path identificationThe performance degradation path of (2) is to control the whole time series data X of the system after the preprocessing and the smoothing noise reduction processing i And putting the life prediction model corresponding to the performance degradation path as the input of the model encoder.
(2) To X i Position encoding PE is performed so that the input of the encoder is X i ′＝{X i,1 +PE 1 ,X i,2 +PE 2 ,…,X i,a +PE n }。
(3) The encoder part comprises 3 encoder layers, each encoder layer comprises four steps of a multi-head self-Attention mechanism, residual connection and normalization, a feed-forward network, residual connection and normalization, and finally the self-Attention moment array Attention is transmitted into a full-connection layer.
(4) And taking the output of the encoder as the input of the full connection layer, and finally converting the output into the single-step output of the final model, namely the predicted value of the RUL, through the operation of the full connection layer.
(5) And calculating the error between the RUL predicted value and the RUL actual value, and evaluating the prediction result.
The following describes a method flow by taking two rocket control system state parameters of an ultra-wideband wireless communication system power supply module circuit simulation model 1 and an ultra-wideband wireless communication system power supply module circuit simulation model 2 as embodiments of specific implementation modes.
Ultra-wideband wireless communication system power supply module circuit simulation model 1
Step 11: system state parameter acquisition
And performing circuit simulation on the power supply module of the ultra-wideband wireless communication system by using Multisim, wherein a simulation model is shown in fig. 3. For the power supply module simulation model 1 of the ultra-wideband wireless communication system, two fault modes of inductor L1 degradation and capacitor C3 degradation exist.
For the power supply module simulation model 1 of the ultra-wideband wireless communication system, two fault modes of inductor L1 degradation and capacitor C3 degradation exist, and an output end ripple time domain characteristic diagram of an average value of a peak-to-peak value, a variance, a root mean square, a minimum value and an absolute value of a system sample No. 1 in the two fault modes is respectively drawn as shown in fig. 4.
Step 12: system performance degradation feature extraction
The data are smoothed and normalized, and output end ripple time domain characteristic graphs of the peak-to-peak value, the variance, the root mean square, the minimum value and the average value of the absolute value of the output end ripple time sequence characteristics after the processing of the system sample No. 1 in the two fault modes are respectively drawn as shown in fig. 5.
A summary diagram of output end ripple time domain characteristics of the average values of the peak-to-peak value, the variance, the root-mean-square, the minimum value and the absolute value of the processed ripple time sequence characteristics at the output ends of all the system samples is shown in fig. 6.
The PCA technology is adopted to reduce the dimension of data, and a histogram of the PCA contribution to the ripple time domain characteristics of the output end of the simulation model 1 of the power supply module of the ultra-wideband wireless communication system is plotted as shown in FIG. 7.
After PCA processing, the contribution degree of the feature 1 of the simulation model 1 of the power supply module of the ultra-wideband wireless communication system reaches 99.4%, and the feature 1 is considered to contain most of information of the original time domain feature, so that the original time domain feature is reduced to 1 dimension. Fig. 8 is a summary of results of all sample characteristics 1 of the module simulation model 1 of the uwb wireless communication system.
Step 13: system performance degradation path identification
DTW is carried out on all sample characteristics 1 of a power supply module simulation model 1 of the ultra-wideband wireless communication system, the path distance between samples is calculated and is used as the similarity between the samples, the smaller the path distance (the brighter the color) is, the higher the similarity between the two samples is, namely, the samples belong to the same fault mode, and the larger the path distance (the darker the color) is, the lower the similarity between the two samples is, namely, the samples belong to different fault modes. Fig. 9 is a summary of similarity results of all samples of the simulation model 1 of the uwb wireless communication system.
K-Medoids clustering is performed on similarity results of all samples of the simulation model 1 of the power supply module of the ultra-wideband wireless communication system, so that a performance degradation path identification result of the simulation model 1 of the power supply module of the ultra-wideband wireless communication system is obtained, and a drawing is shown in FIG. 10.
Step 14: system remaining useful life prediction
The method comprises the steps of identifying performance degradation paths of system samples of a simulation model 1 of a power supply module of the ultra-wideband wireless communication system, dividing the system samples into two types, establishing two transform deep neural network models for training corresponding to a fault mode represented by each type of performance degradation path, bringing system samples of a test set into the models for verification after the training is finished, and obtaining model prediction results as shown in figure 11, wherein the abscissa is a system sample number, the ordinate is a period, the solid line represents an RUL real value, and scattered points are transform deep neural network prediction results.
The accuracy index calculation formula is as follows:
the accuracy of calculation is 98.15% by substituting the prediction result of the transform deep neural network model of the electric module simulation model 1 of the ultra-wideband wireless communication system into a formula, and the transform deep neural network model meets the requirement that the accuracy is more than or equal to 98%.
Power supply module circuit simulation model 2 of ultra-wideband wireless communication system
Step 21: system state parameter acquisition
The circuit simulation is carried out on the ultra-wideband wireless communication system power supply module by using Multisim, and two fault modes of inductance L2 degradation and capacitance C5 degradation exist in the ultra-wideband wireless communication system power supply module simulation model 2.
For the power supply module simulation model 2 of the ultra-wideband wireless communication system, two fault modes of inductor L2 degradation and capacitor C5 degradation exist, and output end ripple time domain characteristic diagrams of average values of peak-to-peak values, variances, root-mean-square, minimum values and absolute values of No. 1 system samples in the two fault modes are obtained respectively.
Step 22: system performance degradation feature extraction
And smoothing and normalizing the data, and respectively acquiring output end ripple time domain characteristic diagrams of average values of peak-to-peak value, variance, root mean square, minimum value and absolute value of output end ripple time sequence characteristics of the processed No. 1 system sample in two fault modes.
And acquiring an output end ripple time domain characteristic summary chart of average values of peak-to-peak values, variances, root-mean-square, minimum values and absolute values of the ripple time sequence characteristics of the output ends of all the processed system samples.
And (3) reducing the dimension of the data by adopting a PCA technology, and acquiring a PCA contribution histogram of ripple time domain characteristics at the output end of the power supply module simulation model 2 of the ultra-wideband wireless communication system.
After PCA processing, the contribution degree of the feature 1 of the simulation model 2 of the power supply module of the ultra-wideband wireless communication system reaches 94.9%, and the feature 1 is considered to contain most of information of the original time domain feature, so that the original time domain feature is reduced to 1 dimension. And summarizing results of all sample characteristics 1 of the electric module simulation model 2 of the ultra-wideband wireless communication system.
Step 23: system performance degradation path identification
DTW is carried out on all sample characteristics 1 of a power supply module simulation model 2 of the ultra-wideband wireless communication system, the path distance between the samples is calculated and used as the similarity between the samples, the smaller the path distance is, the higher the similarity between the two samples is, namely, the samples belong to the same fault mode, and the larger the path distance is, the lower the similarity between the two samples is, namely, the samples belong to different fault modes. And acquiring a summary of similarity results of all samples of the electric module simulation model 1 of the ultra-wideband wireless communication system.
And performing K-Medoids clustering on similarity results of all samples of the simulation model 2 of the power supply module of the ultra-wideband wireless communication system to obtain a performance degradation path identification result of the simulation model 2 of the power supply module of the ultra-wideband wireless communication system.
And step 24: system remaining useful life prediction
The method comprises the steps of identifying performance degradation paths of system samples of a power supply module simulation model 2 of the ultra-wideband wireless communication system, dividing the system samples into two types, establishing two transform deep neural network models for training corresponding to fault modes represented by each type of performance degradation path, and bringing test set system samples into the model for verification after the training is finished to obtain model prediction results.
The accuracy index calculation formula is as follows:
the accuracy of the transform deep neural network model prediction result of the electric module simulation model 2 of the ultra-wideband wireless communication system is substituted into a formula to be calculated to be 98.08%, and the transform deep neural network model meets the requirement that the accuracy is more than or equal to 98%.
In summary, the invention has the following advantages:
(1) According to the method, the path similarity among a plurality of rocket control systems with different decay rates and different fault initiation degrees is evaluated through a DTW (delay tolerant shift) method, and a K-Medoids performance degradation path clustering method is introduced on the basis, so that a large number of rocket control system degradation paths can be rapidly and accurately classified;
(2) The invention provides a Ttansformer deep neural network-based rocket control system residual service life prediction model, which realizes high-precision service life prediction by comprehensively considering the influence of performance degradation paths on rocket control system residual service life prediction.
(3) The invention takes the failure mode as a consideration factor, identifies the performance degradation path of the rocket control system, brings the rocket control system with the same performance degradation path into the same model for training and prediction, and is beneficial to improving the residual life prediction precision of the rocket control system.
The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, and the scope of the invention is not limited thereby. Any modification, equivalent replacement, and improvement made by those skilled in the art without departing from the scope and spirit of the present invention should be within the scope of the claims of the present invention.
Claims (8)
1. A method for product system life prediction in multiple failure modes, comprising:
analyzing and processing all time sequence data of a plurality of product systems under a multi-fault mode as a training set, and determining a plurality of cluster categories of performance degradation paths of the product systems of the training set;
according to the training set data of the product system of each clustering class, a service life prediction model corresponding to each clustering class is constructed and trained to obtain a well-trained service life prediction model of each clustering class;
analyzing and processing time series data before the product system in the multi-fault mode is in fault as a test set, determining a plurality of cluster categories of a performance degradation path of the product system of the test set, and testing the trained life prediction model of each cluster category by using the test set data of the product system of each cluster category to obtain a tested life prediction model of each cluster category;
the method comprises the steps of analyzing and processing time sequence data of a product system to be tested, determining a clustering category of a performance degradation path of the time sequence data, and performing life prediction processing by using a tested life prediction model corresponding to the clustering category to obtain the residual service life of the product system to be tested.
2. The method according to claim 1, wherein the determining the cluster category of the time-series data by analyzing the time-series data of the product system to be tested comprises:
performing data smoothing noise reduction processing on the time sequence data of the product system to be detected to obtain time sequence data with random noise filtered;
performing Principal Component Analysis (PCA) processing on the time sequence data with the random noise filtered out to obtain performance degradation characteristics related to a performance degradation path of the product system to be detected;
and clustering the time sequence data of the performance degradation characteristics to obtain the clustering category of the system performance degradation path of the product to be tested.
3. The method of claim 2, wherein the obtaining the time-series data of the performance degradation feature related to the performance degradation path of the product system to be tested by performing Principal Component Analysis (PCA) on the time-series data of the random noise filtered out comprises:
and performing Principal Component Analysis (PCA) processing on the time sequence data with the random noise filtered to obtain a plurality of PCA characteristics of the time sequence data with the random noise filtered, and taking the PCA characteristics with the PCA characteristic variance contribution rate larger than a set threshold value as the performance degradation characteristics related to the performance degradation path of the product system to be tested.
4. The method of claim 3, wherein the obtaining of the cluster category of the system performance degradation path of the product to be tested by clustering the time-series data of the performance degradation feature comprises:
calculating the similarity between the performance degradation characteristic time sequence data by using a Dynamic Time Warping (DTW) algorithm;
and performing K-Medoids clustering on the similarity between the performance degradation characteristic time sequence data to obtain the clustering category of the performance degradation path of the product system to be tested.
5. The method according to claim 4, wherein the obtaining the remaining service life of the product system to be tested by performing the service life prediction processing by using the tested service life prediction model corresponding to the cluster type comprises:
determining a tested life prediction model corresponding to the cluster type according to the cluster type of the performance degradation path of the product system to be tested;
and carrying out life prediction processing on the product system to be tested by using the tested life prediction model corresponding to the cluster category to obtain the residual service life of the product system to be tested.
6. The method of claim 1, wherein analyzing all time series data of the plurality of product systems in multiple failure modes as a training set, wherein determining a plurality of cluster categories of performance degradation paths of the product systems of the training set comprises:
sequentially carrying out data smoothing noise reduction processing and Principal Component Analysis (PCA) processing on the training set data to obtain performance degradation characteristics related to performance degradation paths of the multiple product systems;
and clustering the time series data of the performance degradation characteristics related to the performance degradation paths of the multiple product systems by using a Dynamic Time Warping (DTW) algorithm and a K-Medoids clustering method to obtain multiple clustering categories of the performance degradation paths of the product systems of the training set.
7. The method of claim 6, wherein the constructing and training the life prediction model corresponding to each cluster category according to the training set data of the product system of each cluster category to obtain the trained life prediction model of each cluster category comprises:
according to each cluster category, constructing a life prediction model of each cluster category;
and training the life prediction model of each cluster type according to the training set data of the product system of each cluster type to obtain the trained life prediction model of each cluster type.
8. The method according to any one of claims 1 to 7, wherein the life prediction model is a product system residual service life prediction model in a multiple failure mode based on a Transformer neural network.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2022106505023 | 2022-06-09 | ||
| CN202210650502 | 2022-06-09 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115859777A true CN115859777A (en) | 2023-03-28 |
Family
ID=84067164
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211022629.7A Pending CN115375026A (en) | 2022-06-09 | 2022-08-25 | Method for predicting service life of aircraft engine in multiple fault modes |
| CN202211395606.0A Pending CN115859777A (en) | 2022-06-09 | 2022-11-09 | Method for predicting service life of product system in multiple fault modes |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211022629.7A Pending CN115375026A (en) | 2022-06-09 | 2022-08-25 | Method for predicting service life of aircraft engine in multiple fault modes |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN115375026A (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116203929B (en) * | 2023-03-01 | 2024-01-05 | 中国矿业大学 | Industrial process fault diagnosis method for long tail distribution data |
| CN117454671B (en) * | 2023-12-22 | 2024-04-12 | 广东力宏微电子有限公司 | Artificial intelligence-based field effect transistor life assessment method |
-
2022
- 2022-08-25 CN CN202211022629.7A patent/CN115375026A/en active Pending
- 2022-11-09 CN CN202211395606.0A patent/CN115859777A/en active Pending
Also Published As
| Publication number | Publication date |
|---|---|
| CN115375026A (en) | 2022-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113496262B (en) | Data-driven active power distribution network abnormal state sensing method and system | |
| Soualhi et al. | Prognosis of bearing failures using hidden Markov models and the adaptive neuro-fuzzy inference system | |
| CN111914873A (en) | Two-stage cloud server unsupervised anomaly prediction method | |
| CN115859777A (en) | Method for predicting service life of product system in multiple fault modes | |
| CN110610035A (en) | Rolling bearing residual life prediction method based on GRU neural network | |
| CN112115004B (en) | Hard disk service life prediction method based on back propagation Bayes deep learning | |
| CN109917777B (en) | Fault detection method based on mixed multi-sampling rate probability principal component analysis model | |
| CN115293280A (en) | Power equipment system anomaly detection method based on space-time feature segmentation reconstruction | |
| CN112070322B (en) | High-voltage cable line running state prediction method based on long-short term memory network | |
| CN112257263B (en) | Equipment residual life prediction system based on self-attention mechanism | |
| CN113342597B (en) | System fault prediction method based on Gaussian mixture hidden Markov model | |
| CN116300691A (en) | State monitoring method and system for multi-axis linkage numerical control machining | |
| CN114297918A (en) | Aero-engine residual life prediction method based on full-attention depth network and dynamic ensemble learning | |
| CN112082639A (en) | Engine state diagnosis method and diagnosis modeling method thereof | |
| CN115017826B (en) | Method for predicting residual service life of equipment | |
| CN113673346A (en) | Motor vibration data processing and state recognition method based on multi-scale SE-Resnet | |
| CN114580291A (en) | LSTM-VAE-based mobile equipment health condition assessment method | |
| CN114266289A (en) | Complex equipment health state assessment method | |
| CN116384224A (en) | Aero-engine life prediction method based on conditional parameter dynamic convolutional neural network | |
| CN111241629B (en) | Intelligent prediction method for performance change trend of hydraulic pump of airplane based on data driving | |
| CN117435997A (en) | Elevator early warning and on-demand maintenance method based on transducer and time sequence compression mechanism | |
| CN116840658A (en) | Fault prediction and diagnosis method for chip firmware | |
| CN114167837B (en) | Intelligent fault diagnosis method and system for railway signal system | |
| CN113722833B (en) | Turbofan engine residual service life prediction method based on double-channel long-short-term memory network | |
| CN115828744A (en) | White light LED fault on-line diagnosis and service life prediction method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |