CN116561927A - Digital twin-driven small sample rotary machine residual life prediction method and system - Google Patents

Abstract

The invention discloses a method and a system for predicting the residual life of a small sample rotary machine driven by digital twinning, which use a rotary machine early degradation signal training convolution self-encoder to learn an early degradation signal mode; then reconstructing the test signal by using a convolution self-encoder, calculating a reconstruction error, mapping the reconstruction error to a [0,1] interval as a health factor, fitting a Weibull reliability function according to the health factor, and predicting the residual service life of the rotary machine; and repeating the steps based on the real-time data in the continuous running process of the rotary machine to realize the real-time updating of the residual life of the rotary machine. The prediction algorithm for the residual life of the rotary machine constructed by the invention can predict the residual life of the bearing on the premise of not needing an end degradation signal, can adaptively update a model along with the increase of data quantity so as to improve the prediction precision in real time, and has the feasibility of predicting the life of the rotary machine with a small sample in an actual industrial scene.

Description

Method and system for predicting remaining life of small-sample rotating machinery driven by digital twins

technical field

The invention belongs to the technical field of life prediction of rotating machinery, and in particular relates to a method and system for predicting the remaining life of small sample rotating machinery driven by digital twins.

Background technique

The remaining service life prediction methods of rotating machinery mainly include two types: prediction methods based on mechanism models and data-driven prediction methods. The former uses failure mechanism, probability statistics and other methods to analyze the degradation process of rotating machinery, and builds a degradation mechanism model and predicts the remaining life through theoretical analysis and experimental verification. Data-driven remaining life prediction is to build an end-to-end model between sensor signals and the remaining service life of rotating machinery based on various prediction algorithms. With the support of big data, the remaining life of rotating machinery can be predicted without expert experience and domain knowledge.

Affected by factors such as application scenarios, operating conditions, and working hours, the remaining life of rotating machinery usually evolves dynamically. The above methods can only predict the static average life of rotating machinery at a specific moment in the design stage or during use, and cannot To describe the dynamic degradation process of rotating machinery, it is difficult to achieve real-time online prediction of remaining life. In addition, when using traditional mechanism models or data-driven models, a large amount of degraded data is required for model construction. When the data used for model construction is insufficient, the problem of model prediction accuracy will decrease. However, due to the limitation of data collection capabilities and the needs of safe production, the actual In the process, it is often difficult to collect a large amount of complete degradation data for building a prediction model.

Contents of the invention

The technical problem to be solved by the present invention is to provide a method and system for predicting the remaining life of small-sample rotating machinery driven by digital twins in view of the deficiencies in the above-mentioned prior art, and to construct a data-driven rotating machinery by fusing convolutional autoencoder and Weibull distribution. The mechanical degradation behavior model is used to solve the technical problem that it is difficult to accurately predict the remaining service life of rotating machinery under small sample data, and realize accurate real-time prediction of the remaining service life of rotating machinery under small sample data.

The present invention adopts following technical scheme:

A small-sample rotating machinery residual life prediction method driven by digital twins, including the following steps:

S1. Signal preprocessing of rotating mechanical equipment;

S2. Establish a training convolutional autoencoder to evaluate the unhealthy degree of the signal obtained in step S1;

S3. Establish an index mapping function, and map the unhealthy degree obtained in step S2 into a health factor that can directly reflect the health status of the equipment;

S4. Based on the Weibull reliability function and the health factor obtained in step S3, combined with the gradient descent method, the real-time iterative prediction of the remaining service life of the rotating machinery is performed.

Specifically, step S1 is specifically:

S101. For a piece of measured equipment signal z(t), use a one-dimensional Kalman filter algorithm to establish a state equation of the equipment signal, and estimate the optimal estimated value obtained by superimposing the measured value of the signal with error and the observed value with error;

S102. Divide the filtered signal x(t) obtained in step S101 into several signal segments whose length is N, and identify the signal in the first 30% of the time as the early degradation signal of the rotating machinery for the training of the subsequent algorithm model, The rest are test data.

Further, in step S101, z(t) is used as the observed value z _k and the initial estimated value at k=0, given the state transition matrix A, the process excitation noise covariance Q, the measurement noise covariance R, and the transition matrix H The initial value of z(t) is calculated iteratively to obtain the signal x(t) after the denoising of z(t) by Kalman filter.

Further, in step S102, the division length N of the device signal is:

Among them, k is an artificially given scaling factor, f _rotate is the rotation frequency of the rotating machine, and f _collect is the signal collection frequency.

Specifically, step S2 is specifically:

S201. Construct a convolutional autoencoder neural network model;

S202. The early degraded signal obtained after preprocessing in step S1 is x _early (t), and the input convolutional autoencoder is encoded and decoded to obtain a reconstructed signal Calculate the reconstruction error using the squared difference loss function;

S203. Input the test signal into the trained convolutional autoencoder to perform signal reconstruction, calculate the reconstruction error between the test signal and the reconstructed signal, and use the reconstruction error as the unhealthy degree of the equipment corresponding to the test signal.

Specifically, step S3 is specifically:

S301. Construct an exponential mapping function based on the definition of the unhealthy degree and the health factor HI and determine hyperparameters of the exponential mapping function;

S302. Sort the M test signal segments x={x _1N ,x _2N ,...,x _MN } obtained after preprocessing in chronological order, and then input them into the convolutional autoencoder and the exponential mapping function to obtain the health of the rotating machinery Factor HI set HI={HI _1N ,HI _2N ,...,HI _MN }.

Further, in step S301, the index mapping function data is specifically:

Among them, k and b are the shape coefficient and bias constant of the exponential mapping function, respectively, is the reconstruction error.

Specifically, step S4 is specifically:

S401, initializing the Weibull reliability function;

S402. Calculate the remaining service life of the rotating machinery based on the Weibull reliability function;

S403. According to the HI set obtained from the real-time measurement data in step S3, use the gradient descent method to update the parameters of the Weibull reliability function, repeat steps S402 and S403, and realize the real-time iterative prediction of the remaining service life of the rotating machinery under the condition of small samples.

Specifically, in step S402, the remaining service life RUL _t of the rotating machine at time t is:

RUL _t =T _f -T _t =η[(-ln R _f ) ^1/β -(-ln R _t ) ^1/β ]

Among them, T _f is the theoretical failure working time of the rotating machine, T _t is the theoretical working time of the rotating machine, η is the scale parameter, R _f is the failure threshold reliability, R _t is the reliability, and β is the shape parameter.

In the second aspect, an embodiment of the present invention provides a small-sample rotating machinery residual life prediction system driven by digital twins, including:

Preprocessing module, preprocessing the signal of rotating machinery equipment;

The evaluation module establishes a training convolutional autoencoder to evaluate the unhealthy degree of the signal obtained by the preprocessing module;

The mapping module establishes an index mapping function to map the unhealthy degree obtained by the evaluation module into a health factor that can directly reflect the health status of the equipment;

The prediction module is based on the Weibull reliability function and the health factor obtained by the mapping module, combined with the gradient descent method to perform real-time iterative prediction of the remaining service life of the rotating machinery.

Compared with the prior art, the present invention has at least the following beneficial effects:

A method for predicting the remaining life of small-sample rotating machinery driven by digital twins. Considering that rotating machinery usually has the characteristics of long degradation period, high data density, and high acquisition cost, it is difficult to collect a large number of full-life cycle degradation signals in practice. Construction and training mechanism The prediction model or data-driven prediction model makes it possible to accurately predict the remaining service life of rotating machinery in a small sample scenario. A method for predicting the remaining life of rotating machinery combining Weibull reliability function and convolutional autoencoder is established. In the training phase, it is only necessary to use the early degradation signal of the rotating machinery to train the convolutional autoencoder (Convolutional Autoencoder, CAE), and evaluate the real-time degradation degree of the equipment and construct the health factor (Health Index, HI). The Weibull reliability function is introduced as a degradation model to describe the degradation process of rotating machinery, the parameters of the Weibull reliability function are updated in real time according to the health factors, and the remaining service life is predicted based on the Weibull reliability function. The method for predicting the remaining life of rotating machinery constructed in the present invention can well solve the dynamic description of the degradation process of rotating machinery and the real-time prediction of the remaining service life in the case of small samples, and provides a basis for the application of digital twin technology in the field of monitoring, operation and maintenance of rotating machinery. new ideas.

Further, the preprocessing based on the one-dimensional Kalman filter algorithm and signal segment interception for the original measurement signal of the rotating machinery can remove the noise and abnormal signals in the original measurement signal, and convert the original signal into a signal segment with the same data format , which is convenient for the unified calculation and processing of the subsequent steps of the algorithm.

Furthermore, the one-dimensional Kalman filter algorithm uses the estimated value of the rotating machinery signal at the previous moment and the signal measurement value at the current moment to obtain the optimal estimate of the rotating machinery signal at the current moment, remove a large amount of noise in the rotating machinery measurement signal, and avoid Due to noise, errors are generated in subsequent reconstruction errors and health factor construction steps. At the same time, the one-dimensional Kalman filter algorithm has the characteristics of a small amount of calculation, which can quickly filter and reduce the noise of the original equipment signal, and improve the real-time performance of the remaining life prediction of the rotating machinery.

Furthermore, the traditional method of intercepting signal fragments is to artificially set a length. If the length is too small, the signal fragment cannot reflect the process of one revolution of the rotating machine; In addition, the complexity and calculation amount of the prediction algorithm will be increased. In this patent, the rotation frequency of the rotating machinery and the measurement frequency of the sensor signal are considered to calculate the signal length corresponding to one revolution of the rotating machinery, and the scaling factor is combined to ensure the rationality of the length setting of the signal segment, so that the signal segment not only contains the data of at least one complete rotation cycle, but also Does not contain excessive redundant data.

Further, the convolutional neural network can effectively extract the signal characteristics of the degraded signal of the rotating machinery, learn the data pattern of the degraded signal, and use the convolutional autoencoder trained only with the early degraded signal to have different reconstruction errors for the signal of the rotating machinery in different degradation periods. The characteristics of , can realize the quantitative estimation of the unhealthy degree of the equipment corresponding to the degradation signal of each period of the rotating machinery in the case of only the early degradation signal.

Further, due to the differences in the operating conditions of rotating machinery and equipment types, the unhealthy distribution ranges of different test signals evaluated by the convolutional autoencoder are different, and the unhealthy degree of the test signal is mapped to [0,1] using an exponential mapping function The health factor on the interval is convenient for subsequent parameter update and life prediction of the Weibull reliability function.

Furthermore, the exponential mapping function has the characteristics of continuous derivation, monotonically decreasing, and saturation, and can flexibly set the corresponding shape coefficient and bias constant according to the working conditions and equipment types of the rotating machinery, so as to improve the interpretability and accuracy of the health factor mapping. Computational efficiency.

Further, using the Weibull reliability function as the degradation behavior model of the rotating machinery, on the one hand, it can directly reflect the degradation trend of the rotating machinery and improve the interpretability of the remaining life prediction of the rotating machinery; After the function parameters are updated, the remaining life is predicted to ensure the accuracy and real-time performance of the remaining life prediction process of rotating machinery.

Furthermore, the failure threshold reliability can be flexibly set according to the requirements of life prediction of rotating machinery and the needs of actual application scenarios of rotating machinery, so as to predict the remaining service life of rotating machinery more accurately and efficiently.

It can be understood that, for the beneficial effects of the second aspect above, reference may be made to the relevant description in the first aspect above, and details are not repeated here.

In summary, the present invention uses the early degradation signal of the rotating machinery to train the convolutional autoencoder, evaluates the degree of unhealthiness corresponding to the degradation signal of the rotating machinery, and uses an exponential mapping function to map it into a health factor. By introducing the Weibull reliability function to describe the degradation trend of the rotating machinery, the parameters of the Weibull reliability function are updated and the remaining life prediction is performed based on real-time data, so that the remaining life of the rotating machinery can be predicted more accurately and in real time.

The technical solutions of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

Description of drawings

Fig. 1 is the overall frame diagram of the present invention;

Fig. 2 is the overall flow of remaining life prediction of the present invention;

Figure 3 is a comparison diagram between the original signal of bearing vibration and the vibration signal after Kalman filtering;

Figure 4 is a comparison diagram of the bearing vibration signal and the reconstruction signal of the convolutional self-encoder;

Fig. 5 is a curve diagram of the reconstruction error curve of the vibration signal of the bearing sample in the training set;

Fig. 6 is a curve diagram of the health factor HI and Weibull reliability function of the bearing samples in the training set;

Fig. 7 is a real-time iterative prediction result diagram of the remaining life of the test set bearing sample 1_3;

Fig. 8 is a diagram of the remaining life prediction results of bearing sample 1_3 in the test set.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

In the description of the present invention, it should be understood that the terms "comprising" and "comprising" indicate the presence of described features, integers, steps, operations, elements and/or components, but do not exclude one or more other features, Presence or addition of wholes, steps, operations, elements, components and/or collections thereof.

It should also be understood that the terminology used in the description of the present invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the present invention. As used in this specification and the appended claims, the singular forms "a", "an" and "the" are intended to include plural referents unless the context clearly dictates otherwise.

It should also be further understood that the term "and/or" used in the description of the present invention and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations , for example, A and/or B, may mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" in the present invention generally indicates that the contextual objects are an "or" relationship.

It should be understood that although the terms first, second, third, etc. may be used in the embodiments of the present invention to describe preset ranges, etc., these preset ranges should not be limited to these terms. These terms are only used to distinguish preset ranges from one another. For example, without departing from the scope of the embodiments of the present invention, the first preset range may also be called the second preset range, and similarly, the second preset range may also be called the first preset range.

Depending on the context, the word "if" as used herein may be interpreted as "at" or "when" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrases "if determined" or "if detected (the stated condition or event)" could be interpreted as "when determined" or "in response to the determination" or "when detected (the stated condition or event) )" or "in response to detection of (a stated condition or event)".

Various structural schematic diagrams according to the disclosed embodiments of the present invention are shown in the accompanying drawings. The figures are not drawn to scale, with certain details exaggerated and possibly omitted for clarity of presentation. The shapes of various regions and layers shown in the figure and their relative sizes and positional relationships are only exemplary, and may deviate due to manufacturing tolerances or technical limitations in practice, and those skilled in the art may Regions/layers with different shapes, sizes, and relative positions can be additionally designed as needed.

The invention provides a small-sample rotating machinery residual life prediction method driven by digital twins. Aiming at the problem that the remaining service life of rotating machinery is difficult to accurately predict under small sample data, a data-driven rotary Mechanical degradation behavior model. First, the convolutional autoencoder is trained using the early degradation signal of the rotating machinery to learn the early degradation signal pattern; then the convolutional autoencoder is used to reconstruct the test signal and calculate the reconstruction error, and the reconstruction error is mapped to [0 ,1] interval is used as the health factor, and the Weibull reliability function is fitted according to the health factor to predict the remaining service life of the rotating machinery; during the continuous operation of the rotating machinery, the above steps are repeated based on real-time data to realize the real-time update of the remaining life of the rotating machinery. The remaining life prediction algorithm of rotating machinery constructed in the present invention can predict the remaining life of bearings without the need for terminal degradation signals, and can update the model adaptively as the amount of data increases to improve the prediction accuracy in real time. Feasibility of small-sample rotating machinery life prediction.

Please refer to Fig. 2, a method for predicting the remaining life of a small-sample rotating machinery driven by a digital twin of the present invention, including the following steps:

S1. Carry out signal preprocessing of rotating machinery equipment

The original equipment signals of rotating machinery are usually high-frequency and low-density continuous time-domain signals, which need to be de-noised and filtered before they can be used for subsequent model training and prediction. First, the one-dimensional Kalman filter algorithm is used to filter and reduce the noise of the original signal; then, the length of the time window is determined by comparing the rotational speed of the rotating machinery with the signal acquisition frequency, and the equipment signal is divided into equal-length signal segments by using a fixed time window.

S101. For a piece of measured equipment signal z(t), use a one-dimensional Kalman filter algorithm to establish the state equation of the equipment signal, and estimate the optimal estimated value obtained by superimposing the measured value of the signal with error and the observed value with error, In this way, the filtering and noise reduction of the equipment signal can be realized.

The one-dimensional Kalman filter prediction formula used is as follows:

in, The formula is the posterior state estimation value of the signal at time k-1, which is the filtering result of the Kalman filter algorithm, also known as the optimal estimation value; A is the state transition matrix; /> is the prior estimated state value at time k; /> is the prior estimate covariance at time k; P _k-1 is the posterior estimate covariance at k-1 time; Q is the process excitation noise covariance, which is used to characterize the error between the state transition matrix and the actual process.

The one-dimensional Kalman filter update formula used is as follows:

Among them, K _k is the filter gain matrix at time k; H is the transformation matrix from state variables to observed values; R is the measurement noise covariance; z _k is the signal measurement value at time k.

Taking z(t) as the initial estimated value of observed value z _k and k=0, given the initial values of state transition matrix A, process excitation noise covariance Q, measurement noise covariance R, and transition matrix H, the above The five formulas are iteratively calculated to obtain the signal x(t) after z(t) is denoised by Kalman filter.

S102. Perform interception and division of rotating mechanical equipment signals;

If the rotation frequency of the rotating machinery is f _rotate and the signal collection frequency is f _collect , then the division length N of the equipment signal is:

Among them, k is an artificially given scaling factor, which is used to adjust the division length N according to the actual characteristics of the device signal.

After the division length N is obtained, the filtered signal x(t) obtained in step S101 is divided into several signal segments of length N, and the signal in the first 30% of the time is identified as the early degradation signal of the rotating machinery for subsequent use. Algorithm model training, and the rest are test data.

S2. Realize the unhealthy evaluation of the input signal by establishing a training convolutional autoencoder

The health status of rotating machinery shows a gradual deterioration trend with the intensification of the degradation process, which is reflected in the equipment signal, that is, the equipment signal in different degradation stages has different data patterns, so it can be measured by measuring the relationship between the test signal and the early degradation signal. The degree of difference reflects the unhealthy degree of the equipment corresponding to the test signal. In order to achieve this goal, first construct a convolutional autoencoder neural network model; then use the early degraded signal to train the convolutional autoencoder, so that it can learn the early degraded signal pattern, and can achieve high-precision reconstruction of the early degraded signal; finally Input the test signal into the trained convolutional autoencoder for signal reconstruction, and calculate the reconstruction error between the test signal and the reconstructed signal. Since the convolutional autoencoder only learns the data pattern of the early degraded signal, it cannot accurately restore other The test signal in the degradation stage, and the higher the degree of degradation, the greater the difference with the early degradation signal, that is, the greater the reconstruction error, so the reconstruction error can be directly used as the unhealthy degree of the equipment corresponding to the test signal.

S201. Construct a convolutional autoencoder model;

The autoencoder consists of two parts: an encoder and a decoder. The encoder reduces the dimensionality of the input signal to the feature space, and the decoder reconstructs the signal encoded in the feature space and outputs it. The convolutional self-encoder is the self-encoder that uses the convolutional neural network as the encoder and decoder. For the convolutional self-encoder with k convolution kernels, the formulas for the encoding and decoding stages are as follows:

h ^k ＝σ(x*W ^k +b ^k ) (7)

Among them, h ^k is the feature code corresponding to the kth convolution kernel; W ^k is the kth convolution kernel; b ^k is the bias constant corresponding to the kth convolution kernel; σ() is the activation function; U ^k is the kth deconvolution kernel; c ^k is the bias constant corresponding to the kth deconvolution kernel.

S202. Training a convolutional autoencoder based on an early degraded signal

The early degradation signal obtained after step S1 preprocessing is x _early (t), and the input convolutional autoencoder is encoded and decoded to obtain the reconstructed signal Compute the reconstruction error using the squared difference loss function:

The reconstruction error of the early degraded signal is backpropagated in the convolutional autoencoder, and the parameters in the autoencoder are updated to reduce the reconstruction error of the early degraded signal. When the reconstruction error is small after multiple parameter updates, it can be considered that the convolutional self-encoder has learned the mapping relationship between the input signal and the feature signal, and the data pattern feature extraction of the input signal can be realized without additional information. .

S203. Calculation of the reconstruction error of the test signal.

The test signal obtained after preprocessing in step S1 is x(t), and the input convolutional autoencoder is encoded and decoded to obtain a reconstructed test signal Compute the reconstruction error using the squared difference loss function:

At this time, the unhealthy degree of the equipment corresponding to the test signal is

S3. Establish an index mapping function to map the unhealthy degree into a health factor that can directly reflect the health status of the equipment

According to the definition and calculation process of the unhealthy degree of equipment, the value range of the unhealthy degree is [0, +∞), and the lower the degree of equipment degradation, the closer the unhealthy degree is to 0. Rotating machinery has multiple degradation failure modes, and different failure situations correspond to different CAE unhealthy ranges. In order to unify different failure situations, a mapping function is used to map the unhealthy degree on the [0,+∞) interval to the health factor HI on the [0,1] interval. HI can quantitatively describe the health status of rotating machinery. HI=1 means that the equipment has not started to degrade, and HI=0 means that the equipment has completely degraded and failed.

S301. Construct an exponential mapping function based on the definition of the unhealthy degree and the health factor HI and determine hyperparameters of the exponential mapping function;

According to the degree of unhealthiness, the definition of health factors and the degradation law of rotating machinery, the mapping function should meet the following requirements:

(1) The definition domain is [0,+∞), and the value range is [0,1].

(2) It is monotonously decreasing in the domain of definition and can be derived smoothly, which is convenient for calculation.

(3) It is saturated, that is, the function value tends to 1 when the independent variable approaches 0, and the function value approaches 0 when the independent variable approaches +∞.

Improve the sigmoid function to obtain an exponential mapping function that meets the above requirements:

Among them, k and b are the shape coefficient and bias constant of the exponential mapping function, respectively.

The bias constant b is used to adjust the health factor HI of the equipment at the initial moment, and the shape coefficient k is used to control the deceleration rate of the health factor HI as the unhealthy degree increases. The above two parameters need to be given artificially according to the change law of the initial healthy state and unhealthy degree of the rotating machinery.

S302. Calculation of rotating machinery health factors

The M test signal segments obtained after preprocessing x={x _1N ,x _2N ,...,x _MN } are sorted in chronological order and then input into the convolutional autoencoder and exponential mapping function to obtain the health factor HI of the rotating machinery Set HI={HI _1N , HI _2N , . . . , HI _MN }.

S4. Real-time iterative prediction of the remaining service life of rotating machinery based on the Weibull reliability function combined with the gradient descent method.

The main steps of real-time iterative prediction of the remaining life of rotating machinery using the Weibull reliability function are as follows: Initialize the Weibull reliability function; calculate the remaining service life of the rotating machinery based on the Weibull reliability function; according to the HI set obtained from the real-time measurement data in step S3, use gradient The descending method updates the parameters of the Weibull reliability function; repeat the above two steps. The specific process is as follows:

S401. Initialize the Weibull reliability function

The Weibull distribution is a commonly used distribution in reliability theory, and the expression of the two-parameter Weibull reliability function is:

Among them, R(t) is the Weibull reliability, which represents the reliability of mechanical equipment, and β and η are the shape parameters and scale parameters of the Weibull reliability function, respectively.

In the initial stage, the initial values of shape parameters and scale parameters can be artificially given according to the specific working conditions and historical degradation laws of the rotating machinery, or the shape parameters and scale parameters can be initialized randomly.

S402. Given a failure threshold reliability R _f , substituting the formula (12) in S401 to obtain the theoretical failure working time T _f of the rotating machinery corresponding to R _f ;

T _f ＝η(-ln R _f ) ^1/β (13)

According to step S3, the health factor HI _t corresponding to the rotating mechanical equipment at time t is obtained. The distribution intervals of Weibull reliability and health factors are both [0,1], and both health factors HI and reliability can be used to characterize the health status of rotating machinery, so HI _t can be used as reliability R _t , and HI _t can be substituted into In formula (12) of S401, the theoretical working time T _t of the rotating machinery corresponding to HI _t is obtained as:

T _t =η(-ln R _t ) ^1/β (14)

Combining Equation (13) and Equation (14), the remaining service life RUL _t of the rotating machinery at time t is obtained as:

RUL _t =T _f -T _t =η[(-ln R _f ) ^1/β -(-ln R _t ) ^1/β ] (15)

S403. Calculate the health factor set HI={HI _1N ,HI _2N ,...,HI _MN } from the degeneration of the rotating machinery to time t according to S3, and substitute the health factor and its corresponding time into formula (12), select The square difference loss function calculates the fitting error of the Weibull reliability function at time t as follows:

The partial derivatives of the fitting error with respect to the shape parameter and the scale parameter are:

Given a learning rate a, the shape parameters and scale parameters can be updated according to the learning rate and partial derivatives:

S402 and S403 are repeated every time a new HI _t is calculated according to S3, so as to realize real-time iterative prediction of the remaining service life of the rotating machinery under the condition of a small sample.

In yet another embodiment of the present invention, a system for predicting the remaining life of a small-sample rotating machinery driven by a digital twin is provided, which can be used to implement the method for predicting the remaining life of a small-sample rotating machinery driven by a digital twin. Specifically, the digital twin The driven small sample rotating machinery remaining life prediction system includes a preprocessing module, an evaluation module, a mapping module and a prediction module.

Among them, the preprocessing module preprocesses the signals of rotating mechanical equipment;

The evaluation module establishes a training convolutional autoencoder to evaluate the unhealthy degree of the signal obtained by the preprocessing module;

The mapping module establishes an index mapping function to map the unhealthy degree obtained by the evaluation module into a health factor that can directly reflect the health status of the equipment;

The prediction module is based on the Weibull reliability function and the health factor obtained by the mapping module, combined with the gradient descent method to perform real-time iterative prediction of the remaining service life of the rotating machinery.

In yet another embodiment of the present invention, a terminal device is provided, the terminal device includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the computer The program instructions stored in the storage medium. The processor may be a central processing unit (Central Processing Unit, CPU), or other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gates Array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, are suitable for implementing one or more instructions, specifically It is suitable for loading and executing one or more instructions to realize the corresponding method flow or corresponding functions; the processor described in the embodiment of the present invention can be used for the operation of the method for predicting the remaining life of small sample rotating machinery driven by digital twins, including:

Preprocess the signal of rotating mechanical equipment; establish a training convolutional autoencoder to evaluate the unhealthy degree of the signal; establish an exponential mapping function to map the unhealthy degree into a health factor that can directly reflect the health status of the equipment; based on Weibull reliability function and health Factor, combined with the gradient descent method for real-time iterative prediction of the remaining service life of rotating machinery.

In yet another embodiment of the present invention, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a terminal device for storing programs and data. . It can be understood that the computer-readable storage medium here may include a built-in storage medium in the terminal device, and certainly may include an extended storage medium supported by the terminal device. The computer-readable storage medium provides storage space, and the storage space stores the operating system of the terminal. Moreover, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions may be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here may be a high-speed RAM memory, or a non-volatile memory (Non-Volatile Memory), such as at least one magnetic disk memory.

One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor, so as to realize the corresponding steps in the method for predicting the remaining life of a small-sample rotating machinery driven by a digital twin in the above-mentioned embodiment; the instructions in the computer-readable storage medium One or more instructions are loaded by the processor and executed as follows:

Preprocess the signal of rotating mechanical equipment; establish a training convolutional autoencoder to evaluate the unhealthy degree of the signal; establish an exponential mapping function to map the unhealthy degree into a health factor that can directly reflect the health status of the equipment; based on Weibull reliability function and health Factor, combined with the gradient descent method for real-time iterative prediction of the remaining service life of rotating machinery.

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments It is a part of embodiments of the present invention, but not all embodiments. The components of the embodiments of the invention generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Accordingly, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Example

The method for predicting the remaining life of small-sample rotating machinery driven by digital twins disclosed in the present invention will be further described below by substituting specific bearing vibration signals.

The method for predicting the remaining life of small-sample rotating machinery driven by digital twins disclosed in the present invention includes the following steps:

S1. Carry out signal preprocessing of rotating mechanical equipment:

Collect the vibration signal of the whole life cycle of the bearing, and use the Kalman filter algorithm to filter the vibration signal to obtain the signal after noise reduction. The signal of one of the bearings is shown in Figure 3. The rotational speed of the bearing is 1800r/min, the sampling frequency of the vibration sensor is 25600Hz, the scaling factor is selected as 1.5, and the segment length of the vibration signal of the bearing is determined to be 1280. Observing the vibration signal of the bearing after noise reduction in Figure 1, it can be found that the amplitude of the vibration signal of the bearing is small and has no obvious change before 200 seconds, and the amplitude increases slowly after 200 seconds and increases sharply around 280 seconds, indicating that this When the bearing has completely failed, based on this, it can be considered that the first 40% of the time, that is, the bearing is in the early degradation stage within the first 100 seconds, and the bearing signal segment during this period is selected as the training data set. Use the same method and parameters to process the other collected bearing life cycle degradation signals to obtain a training data set composed of early degradation signals.

S2. Realize the unhealthy evaluation of the input signal by establishing a training convolutional autoencoder:

In the convolutional autoencoder model, the convolution kernel size of the convolutional layer and the transposed convolutional layer is set to 5, and the step size is set to 2; the pooling size and step size of the pooling layer and the unpooling layer are both set to 4 , when training the network, the batch size is 64, the number of training rounds is 50, and the learning rate is set to 0.001. Use the early degraded data training dataset in S1 to train the convolutional autoencoder, and input the early degraded signal and the late degraded signal into the convolutional autoencoder for reconstruction. The comparison between the original signal and the reconstructed signal is shown in Figure 4 . It can be seen from the comparison figure that for the early degraded signal, the convolutional autoencoder has learned the data pattern of the early degraded data well, and can reconstruct the original signal with high precision, but it cannot restore and reconstruct the signal at the end of the degraded stage.

Using the trained convolutional autoencoder to calculate the reconstruction error of the vibration signal of the bearing's full life cycle, the calculation results of the reconstruction error are shown in Figure 5. It can be seen from the figure that the reconstruction errors of the bearing signals are close to 0 in the early stage of degradation, which is relatively stable, but the reconstruction errors increase sharply when approaching failure, as shown in Figure 5, the reconstruction errors of the three bearing signals are at Both are greater than 1 at the time of failure, and the reconstruction error of bearing sample 1_1 exceeds 5.

S3. Establish an index mapping function to map the unhealthy degree to a health factor that can directly reflect the health status of the equipment:

According to the reconstruction error corresponding to the vibration signal of the bearing's full life cycle shown in Figure 3, it can be seen that the reconstruction error of the bearing at the initial moment is close to 0, that is, no degradation occurs, and the health factor at this time can be considered as 0. When the bearing is in the period of rapid degradation or failure, its reconstruction error is close to or much greater than 1, and it can be determined that the HI of the bearing at this time is less than 0.8, that is, HI(1)<0.8. According to the above observations, when b=-4, -5, -6, HI(0)=0.982, 0.993, 0.997, considering the effect of the bias constant, set the bias constant to -5; according to HI(1)< 0.8, the solution is k>3.61, so the shape parameter is set to 4. Taking k=4, b=-6 as the parameters of the exponential mapping function, and substituting the reconstruction error calculated according to the convolutional autoencoder in S2, the health factor curves corresponding to the vibration signals of the three bearing samples can be obtained, as shown in Figure 6.

S4. Real-time iterative prediction of the remaining service life of rotating machinery based on the Weibull reliability function combined with the gradient descent method:

The Weibull failure reliability threshold is set to 0.05, the bearing health factor HI is determined to be less than 0.05 and the bearing fails, and the shape parameter β and size parameter η of the Weibull reliability function are initialized to 2 and 15000, respectively. Taking the test bearing sample 1_3 as an example, the test data includes the first 18010 seconds of the bearing’s life cycle signal, and the first 2700 seconds of the signal is selected as the CAE training data, and then the Weibull reliability function is updated every 1000 seconds and the remaining life of the current bearing is predicted. The iterative prediction results are shown in Figure 7. Figure 7 shows that as the working time increases and the parameters are updated iteratively, the error between the prediction result of the bearing remaining life and the real remaining life of the Weibull reliability function gradually decreases, indicating that the Weibull reliability function has learned the relationship between the bearing vibration signal and the remaining life. mapping relationship between them.

Fig. 8 is the remaining life prediction result of Weibull reliability function of bearing sample 1_3 in the test set at 18000 seconds. Fig. 8 shows that the remaining life prediction result is 6876 seconds, the actual remaining life is 5730 seconds, and the prediction percentage error is 20%. Using the same method and steps to predict the remaining life of the remaining test bearing samples, the prediction results of this method and the comparison with other commonly used traditional algorithms are shown in Table 1.

Table 1

The comparison shows that this method can achieve high-precision prediction of the remaining service life of rotating machinery using only early degradation data.

In summary, the present invention is a method and system for predicting the remaining life of small-sample rotating machinery driven by digital twins, which only uses early degradation signals to train convolutional autoencoders, and uses convolutional autoencoders and exponential mapping functions to construct health factors. The Weibull reliability function is introduced as a degradation behavior model to describe the degradation trend of the rotating machinery. Based on the health factor, the parameters of the Weibull reliability function are updated and the remaining life of the rotating machinery is calculated. The accurate and real-time prediction of the remaining life of the rotating machinery is realized in the case of small samples.

Those skilled in the art can clearly understand that for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used for illustration. In practical applications, the above-mentioned functions can be assigned to different functional units, Completion of modules means that the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment may be integrated into one processing unit, or each unit may exist separately physically, or two or more units may be integrated into one unit, and the above-mentioned integrated units may adopt hardware It can also be implemented in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the protection scope of the present application. For the specific working processes of the units and modules in the above system, reference may be made to the corresponding processes in the aforementioned method embodiments, and details will not be repeated here.

In the above-mentioned embodiments, the descriptions of each embodiment have their own emphases, and for parts that are not detailed or recorded in a certain embodiment, refer to the relevant descriptions of other embodiments.

Those skilled in the art can appreciate that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed in the present invention can be realized by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the present invention.

In the embodiments provided in the present invention, it should be understood that the disclosed device/terminal and method may be implemented in other ways. For example, the device/terminal embodiments described above are only illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or Components may be combined or integrated into another system, or some features may be omitted, or not implemented. In another point, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

If the integrated module/unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention realizes all or part of the processes in the methods of the above embodiments, and can also be completed by instructing related hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer When the program is executed by the processor, the steps in the above-mentioned various method embodiments can be realized. Wherein, the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a removable hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (Read-Only Memory, ROM) , random access memory (Random Access Memory, RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable media can be based on the requirements of legislation and patent practice in the jurisdiction Appropriate additions and subtractions, for example, in some jurisdictions, by virtue of legislation and patent practice, computer readable media exclude electrical carrier signals and telecommunication signals.

The present application is described with reference to flowcharts and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

The above content is only to illustrate the technical ideas of the present invention, and cannot limit the protection scope of the present invention. Any changes made on the basis of the technical solutions according to the technical ideas proposed in the present invention shall fall within the scope of the claims of the present invention. within the scope of protection.

Claims (10)

1. The method for predicting the remaining life of small sample rotating machinery driven by digital twins, characterized in that it comprises the following steps: S1. Signal preprocessing of rotating machinery equipment; S2. Establish a training convolutional autoencoder to evaluate the unhealthy degree of the signal obtained in step S1; S3. Establish an index mapping function, and map the unhealthy degree obtained in step S2 into a health factor that can directly reflect the health status of the equipment; S4. Based on the Weibull reliability function and the health factor obtained in step S3, combined with the gradient descent method, the real-time iterative prediction of the remaining service life of the rotating machinery is performed. 2. The method for predicting the remaining life of small-sample rotating machinery driven by digital twins according to claim 1, wherein step S1 is specifically: S101. For a piece of measured equipment signal z(t), use a one-dimensional Kalman filter algorithm to establish a state equation of the equipment signal, and estimate the optimal estimated value obtained by superimposing the measured value of the signal with error and the observed value with error; S102. Divide the filtered signal x(t) obtained in step S101 into several signal segments whose length is N, and identify the signal in the first 30% of the time as the early degradation signal of the rotating machinery for the training of the subsequent algorithm model, The rest are test data. 3. The method for predicting the remaining life of small-sample rotating machinery driven by digital twins according to claim 2, characterized in that in step S101, z(t) is used as the observed value z _k and the initial estimated value at the moment k=0, Given the state transition matrix A, the process excitation noise covariance Q, the measurement noise covariance R, and the initial value of the transition matrix H, iteratively calculate the signal x(t) after z(t) has been denoised by Kalman filtering. 4. The method for predicting the remaining life of small-sample rotating machinery driven by digital twins according to claim 2, characterized in that, in step S102, the division length N of the equipment signal is: Among them, k is an artificially given scaling factor, f _rotate is the rotation frequency of the rotating machine, and f _collect is the signal collection frequency. 5. The method for predicting the remaining life of small-sample rotating machinery driven by digital twins according to claim 1, wherein step S2 is specifically: S201. Construct a convolutional autoencoder neural network model; S202. The early degraded signal obtained after preprocessing in step S1 is x _early (t), and the input convolutional autoencoder is encoded and decoded to obtain a reconstructed signal Calculate the reconstruction error using the squared difference loss function; S203. Input the test signal into the trained convolutional autoencoder to perform signal reconstruction, calculate the reconstruction error between the test signal and the reconstructed signal, and use the reconstruction error as the unhealthy degree of the equipment corresponding to the test signal. 6. The method for predicting the remaining life of small-sample rotating machinery driven by digital twins according to claim 1, wherein step S3 is specifically: S301. Construct an exponential mapping function based on the definition of the unhealthy degree and the health factor HI and determine hyperparameters of the exponential mapping function; S302. Sort the M test signal segments x={x _1N ,x _2N ,...,x _MN } obtained after preprocessing in chronological order, and then input them into the convolutional autoencoder and the exponential mapping function to obtain the health of the rotating machinery Factor HI set HI={HI _1N ,HI _2N ,...,HI _MN }. 7. The method for predicting the remaining life of small-sample rotating machinery driven by digital twins according to claim 6, characterized in that, in step S301, the index mapping function data is specifically: Among them, k and b are the shape coefficient and bias constant of the exponential mapping function, respectively, is the reconstruction error. 8. The method for predicting the remaining life of small-sample rotating machinery driven by digital twins according to claim 1, wherein step S4 is specifically: S401, initialize the Weibull reliability function; S402. Calculate the remaining service life of the rotating machinery based on the Weibull reliability function; S403. According to the HI set obtained from the real-time measurement data in step S3, use the gradient descent method to update the parameters of the Weibull reliability function, repeat steps S402 and S403, and realize the real-time iterative prediction of the remaining service life of the rotating machinery under the condition of small samples. 9. The small-sample remaining life prediction method for rotating machinery driven by digital twins according to claim 1, wherein in step S402, the remaining service life RUL _t of the rotating machinery at time t is: RUL _t =T _f -T _t =η[(-lnR _f ) ^1/β -(-lnR _t ) ^1/β ] Among them, T _f is the theoretical failure working time of the rotating machine, T _t is the theoretical working time of the rotating machine, η is the scale parameter, R _f is the failure threshold reliability, R _t is the reliability, and β is the shape parameter. 10. A small-sample rotating machinery residual life prediction system driven by digital twins, characterized in that it includes: Preprocessing module, preprocessing the signal of rotating machinery equipment; The evaluation module establishes a training convolutional autoencoder to evaluate the unhealthy degree of the signal obtained by the preprocessing module; The mapping module establishes an index mapping function to map the unhealthy degree obtained by the evaluation module into a health factor that can directly reflect the health status of the equipment; The prediction module is based on the Weibull reliability function and the health factor obtained by the mapping module, combined with the gradient descent method to perform real-time iterative prediction of the remaining service life of the rotating machinery.