CN116304849A - Two-dimensional piston pump fault diagnosis method based on local cut space arrangement and gating circulation network - Google Patents

Abstract

The invention discloses a piston pump fault diagnosis method based on local cut space arrangement and a gate control circulation network, which comprises the following steps: collecting vibration signals reflecting the normal state and the fault state of the piston pump; constructing a total fault sample set based on the original vibration data after noise reduction and reconstruction by using a time-frequency analysis method; dividing the total fault sample set into a training set and a testing set according to the proportion; constructing a fault diagnosis model for diagnosing faults of the piston pump, taking a training sample set as input of the fault diagnosis model, taking a state label as output of the fault diagnosis model, training the fault diagnosis model, and carrying out optimization training on the fault diagnosis model through fusion characteristics of a local tangent space arrangement method to obtain an optimized fault diagnosis model; the original vibration data of the piston pump is input into the optimized fault diagnosis model, the identification precision of fault types is obtained, and the original vibration signal of the 2D pump is used as a data source, so that the purpose of nondestructive state monitoring can be achieved.

Description

Two-dimensional piston pump fault diagnosis method based on local cut space arrangement and gating circulation network

Technical Field

The invention belongs to the technical field of intelligent fault diagnosis of hydraulic elements, and relates to a two-dimensional piston pump fault diagnosis method based on local cutting space arrangement and a gating circulation network.

Background

Microminiature hydraulic components are key foundations for aircraft, robots, high-end equipment, and the like. How to realize high power-to-weight ratio and microminiaturization of the structure at the same time, thereby obtaining the competitive advantage relative to electric transmission which is rapidly developed in recent years, and being one of the hot research problems in the field of flow control. The technology is blocked in the field abroad, and most of research works around the structure of the traditional pump valve at home and abroad are scaled down, which is not satisfactory in various aspects such as performance, reliability, efficiency, power-weight ratio and the like, but the present situation is solved by the appearance of a two-dimensional (2D) piston pump. Along with the rapid development of signal processing technology and artificial intelligence, the application of the artificial intelligence to the fault diagnosis analysis of a two-dimensional (2D) piston pump is expected to become a sharp tool for the data processing and the fault diagnosis of mechanical equipment in order to reduce the influence on the working efficiency after the pump breaks down, and a new thought is provided for the intelligent fault diagnosis and the health management of the 2D pump.

Currently, the fault diagnosis of the current hydraulic equipment mainly comprises two types of methods based on a model and data driving. Accurate modeling is very difficult in a model method, and meanwhile, the problems of weak generalization capability, low fault diagnosis precision and the like exist. The data driving method overcomes the inherent defects of the model method, and the obtained information is deeply mined by collecting various data signals and using an artificial intelligence algorithm to complete the evaluation of a nonlinear system. As the amount of data acquired by industrial sensors has exploded, the increase in data dimension also creates a vast amount of uncertainty data and fuzzy data. The deep learning has great breakthrough in many fields by virtue of the strong capability of autonomously learning nonlinear data representation and pattern recognition, is combined with manifold theory and applied to dimension reduction of nonlinear feature sets, can improve the accuracy of high-dimensional nonlinear fault feature recognition and reduce fault diagnosis sensitivity to a certain extent, but has less application in intelligent fault diagnosis of 2D pumps.

In recent years, there are a plurality of deep learning algorithms, however, their advantages and disadvantages are different. The cyclic neural network overcomes the defect that the output of the traditional neural network only depends on the current input, but has the problems of gradient explosion and disappearance. The long-term and short-term memory network solves the inherent problems of the cyclic neural network by updating the network node information. However, the long-period memory network has the problems of more model parameters, slower training speed and the like, and the gating circulation unit is an improved version of the long-period memory network, so that the prediction accuracy is ensured, and meanwhile, the model parameters are fewer, and the training speed is faster. Therefore, a two-dimensional piston pump fault diagnosis method based on local cut space arrangement and a gating circulation network is provided.

Disclosure of Invention

In order to solve the problems, the invention provides the following technical scheme: a piston pump fault diagnosis method based on local cut space arrangement and gating circulation network comprises the following steps:

and (3) data acquisition: collecting vibration signals reflecting the normal state and the fault state of the piston pump as original state data;

data preprocessing: optimizing a variation modal decomposition method by using a genetic algorithm, and carrying out noise reduction reconstruction on original vibration data of the piston pump by adopting the optimized variation modal decomposition method;

constructing a total fault sample set based on the original vibration data after noise reduction and reconstruction by using a time-frequency analysis method, and dividing the total fault sample set into a training set and a testing set according to a proportion based on a time-frequency domain characteristic sample set of a health state;

model training and optimizing: constructing a fault diagnosis model for diagnosing faults of the piston pump, taking a training sample set as input of the diagnosis model, taking a state label as output of the fault diagnosis model, training the diagnosis model,

optimizing and training the fault diagnosis model through fusion characteristics of a local cut space arrangement method to obtain an optimized fault diagnosis model;

fault diagnosis: and inputting the original vibration data of the piston pump into an optimized fault diagnosis model, and outputting a state label value by the optimized fault diagnosis model to further obtain the identification precision of fault types.

Further: the piston pump adopts a two-dimensional piston pump

Further: the fault categories include: health, cylinder wear, piston wear, and cam wear.

Further: : the fault diagnosis model includes: an input layer, a hidden layer and an output layer,

the input layer inputs a data set constructed based on the original vibration data after noise reduction and reconstruction into the hidden layer;

the hidden layer performs fault diagnosis on time-frequency characteristics of original vibration signals of the 2D pump based on the gating circulation network model, and when different modal characteristics are found, fault accuracy identification can be performed according to the corresponding modal characteristics;

the output layer outputs the fault accuracy of the identified corresponding modal feature;

the input layer: setting the length of a single sample as 3000, dividing the number of samples in each state as 400, totaling 1600 groups, defining a training model as a single hidden layer, wherein the training model comprises 100 hidden units, the characteristic dimension of an input sequence is 12, and inputting x at the moment t _t The relevant input weight is 300 multiplied by 12, the data set comprises time sequence data of 9 speakers, each sequence to be identified has 12 characteristics, and the total number of the training sets is 270 and the total number of the testing sets is 370;

the hidden layer: the hidden layer comprises a GRU unit, the state of the GRU unit is adjusted by using a refresh gate and a reset gate, and the state information h is hidden at the moment t-1 _t-1 The associated weights are 300 x 100.

Further: : the method for denoising and reconstructing the original vibration data of the piston pump by using the variation modal decomposition method comprises the following steps:

and collecting fault characteristic indexes of the reconstructed signals under four operating states of the piston pump from the angles of time domain analysis, frequency domain analysis and time frequency analysis.

Further: the optimization process of the variation modal decomposition method by using the genetic algorithm comprises the following steps:

s21: let the time sequence be x= { X (1), X (2),. The term, X (N) }, length N, then constitute an m-dimensional vector:

X(i)＝{x(i),x(i+1),...,x(i+m-1)}，(i＝1,...,N-m+1) (1)

s22: defining the distance between X (i) and X (j) as d [ X (i), X (j) ], i not equal to j, and when the difference between the two sequences is the largest, the formula (2) is given:

s23: let the threshold be r (r > 0), count the number that the sequence distance is smaller than the threshold, and make the ratio with the total vector number N-m, then have formula (3):

s24: the average of the results of step S23 is given by formula (4):

s25: adding 1 to the vector dimension formed by the time sequence, and repeating the steps S21 to S24:

s26: the sample entropy of the time series is as follows:

s27: coding the decomposition layer number k and the penalty factor alpha in the variation modal decomposition method to form a chromosome in a genetic algorithm, randomly initializing a population, setting the individual capacity of the population as n, and binary coding as a ₁ ,a ₂ ,...a _n The following formula is given:

s28: let the comfort level of the ith individual be f _i Evaluating the fitness of each genetic individual by using sample entropy, outputting an optimal solution and ending if the fitness meets an optimization criterion, otherwise, performing the following steps:

s29: the individual selection is carried out by a wheel bet method, the selected individual is carried out in the next step, the unselected individual is eliminated, and the probability of each individual being selected is as follows:

s210: and (3) taking the individual selected in the step S29 as a parent, generating offspring in a crossing way, promoting the offspring to mutate at the same time, generating a new generation population, returning to the step S28, and finding out the optimal combination parameters of VMD decomposition.

Further: the process of optimizing and training the fault diagnosis model through the fusion characteristics of the local cut space arrangement method is as follows:

let x= [ X ] _i ,...,x _N ]For the sample set of the input space, there are

Wherein τ _i ∈R ^{d ×1} Is x _i Is an eigen representation of f is a mapping function, ">

Representing noise;

s31: finding sample point x in input space _i Adopting epsilon neighbor method or k neighbor method to obtain sample point x _i Is a local spreading matrix of (a);

s32: for sample point x _i Performing eigenvalue decomposition on the local scattering matrix of (2), and obtaining a local principal component;

(1) Let X _i ＝[x _i1 ,...,x _ik ]Is x _i K-nearest neighbor under Euclidean distance metric, comprising x _i By itself, there is a single sample after processing

Wherein Q is a orthonormal matrix of d columns;

(2) Post-treatment X _i Becomes as follows

S33: based on the obtained local principal component Θ= [ Θ ] ₁ ,...,Θ _n ]Solving global coordinates T _i ＝[τ _i1 ,...,τ _ik ]，τ _ij Satisfies the following formula:

wherein,,

for k tau _ij Mean value of L _i Is an unknown mapping transformation and plays a role in arrangement, and the matrix form of the formula (1) is shown as the formula (2):

wherein,,

reconstructing residual error, wherein the process requires tau obtained after dimension reduction _i Local mapping transform L _i The reconstruction residual, i.e. global optimum (3), can be minimized:

i.e. the output y at the current moment _t Is only subject to the input x at the current moment _t Hidden layer state h at last moment _t-1 Influence.

According to the 2D pump fault diagnosis method based on the local cut space arrangement and the gate control circulation network, the gate control circulation network is combined with the local cut space arrangement method, the characteristics are processed in different modes, the limitation of the method is overcome to a certain extent, when the 2D pump is in fault, a vibration source signal generates pressure waves through shell vibration and propagates in air to form vibration and noise, high-precision intelligent diagnosis of a typical fault type of the 2D pump can be achieved, after an original vibration signal is processed, 38 characteristic indexes are extracted from a time domain, a frequency domain and a wavelet domain, fusion is carried out by the local cut space arrangement method, and finally fusion characteristics are input into a fault diagnosis model to identify the fault type of the 2D pump. Has the following advantages:

1. according to the vibration signal-based two-dimensional (2D) piston pump intelligent fault diagnosis method, the original vibration signal of the 2D pump is used as a data source, and the purpose of nondestructive state monitoring can be achieved.

2. According to the vibration signal-based intelligent fault diagnosis method for the 2D pump, a Genetic Algorithm (GA) is utilized to optimize a Variation Modal Decomposition (VMD) method, the optimized Variation Modal Decomposition (VMD) method is adopted to conduct noise reduction reconstruction on original vibration data of the 2D pump, sample entropy is used as a population fitness function, a genetic algorithm is utilized to conduct iterative optimization on VMD method parameter combinations, the influence of environmental noise on the fault diagnosis accuracy of the 2D pump can be avoided, a high-dimensional and complex overall fault sample set is constructed by utilizing a time-frequency analysis method, and time domain feature information and frequency domain feature information of sensor signals can be synchronously reserved.

3. According to the intelligent fault diagnosis method for the 2D pump based on the vibration signals, disclosed by the invention, the characteristic extraction capacity and the high-precision identification capacity of a gating circulation network model for high-dimensional information are fully utilized, a time-frequency characteristic image of an original vibration signal of the 2D pump is used as the input of a diagnosis model, and the characteristic extraction of the gating circulation network is adopted, so that when different modal characteristics are found, the accurate identification can be performed according to the corresponding modal characteristic precision.

4. According to the intelligent fault diagnosis method for the 2D pump based on the vibration signal, disclosed by the invention, the local cut space arrangement method (LSTA) is used for fusing the fault characteristic data of the 2D pump, so that the learning capacity of a GRU network model is enhanced, the convergence rate of model training is improved, and the training time of the model is shortened in the training process of a diagnosis model; on the other hand, in the case of consistent parameter settings, the improvement of the model performance leads to the improvement of the final diagnosis accuracy.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to the drawings without inventive effort to a person skilled in the art.

FIG. 1 is a flow chart of an implementation of a vibration signal-based intelligent fault diagnosis method for a 2D pump according to the present invention;

FIG. 2 (a) is a graph of a collection of time domain features 1-10 of health status samples; (b) is a collection of state of health sample temporal features 11-17;

FIG. 3 (a) is a plot of a collection of frequency domain features 1-7 of health status samples; (b) is a plot of a frequency domain signature 8-9 collection of health status samples; (a) is a plot of a collection of frequency domain features 10-12 of health status samples; (a) is a plot of a collection of health status sample frequency domain features 13;

FIG. 4 is a graph of a time-frequency domain feature set for a health state sample;

FIG. 5 is a diagram of a fault diagnosis model of the present invention;

FIG. 6 is a block diagram of the internal network of the GRU of the invention;

FIG. 7 is a time domain comparison graph of vibration signals of four states of a 2D pump according to an embodiment of the present invention, wherein (a) is a healthy state, (b) is a cylinder wear failure, (c) is a piston wear failure, and (D) is a cam wear failure;

FIG. 8 is a graph of frequency domain comparisons of vibration signals for four states of a 2D pump according to an embodiment of the present invention, wherein (a) is a healthy state, (b) is a cylinder wear failure, (c) is a piston wear failure, and (D) is a cam wear failure;

FIG. 9 (a) is a time domain plot of the modal components of the example 2D pump wear sample of the present invention after GA-VMD processing, and (b) is a frequency domain plot of the modal components of the example 2D pump wear sample of the present invention after GA-VMD processing;

FIG. 10 shows the accuracy and loss of model training of a measured 2D pump wear sample via LSTA fusion features, where (a) is model training accuracy and (b) is model training loss;

fig. 11 is a confusion matrix diagram of accuracy in identifying the wear state of a 2D pump piston according to an embodiment of the present invention.

Detailed Description

It should be noted that, without conflict, the embodiments of the present invention and features in the embodiments may be combined with each other, and the present invention will be described in detail below with reference to the drawings and the embodiments.

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. The following description of at least one exemplary embodiment is merely exemplary in nature and is in no way intended to limit the invention, its application, or uses. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.

The relative arrangement of the components and steps, numerical expressions and numerical values set forth in these embodiments do not limit the scope of the present invention unless it is specifically stated otherwise. Meanwhile, it should be clear that the dimensions of the respective parts shown in the drawings are not drawn in actual scale for convenience of description. Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail, but are intended to be part of the specification where appropriate. In all examples shown and discussed herein, any specific values should be construed as merely illustrative, and not a limitation. Thus, other examples of the exemplary embodiments may have different values. It should be noted that: like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further discussion thereof is necessary in subsequent figures.

In the description of the present invention, it should be understood that the azimuth or positional relationships indicated by the azimuth terms such as "front, rear, upper, lower, left, right", "lateral, vertical, horizontal", and "top, bottom", etc., are generally based on the azimuth or positional relationships shown in the drawings, merely to facilitate description of the present invention and simplify the description, and these azimuth terms do not indicate and imply that the apparatus or elements referred to must have a specific azimuth or be constructed and operated in a specific azimuth, and thus should not be construed as limiting the scope of protection of the present invention: the orientation word "inner and outer" refers to inner and outer relative to the contour of the respective component itself.

Spatially relative terms, such as "above … …," "above … …," "upper surface at … …," "above," and the like, may be used herein for ease of description to describe one device or feature's spatial location relative to another device or feature as illustrated in the figures. It will be understood that the spatially relative terms are intended to encompass different orientations in use or operation in addition to the orientation depicted in the figures. For example, if the device in the figures is turned over, elements described as "above" or "over" other devices or structures would then be oriented "below" or "beneath" the other devices or structures. Thus, the exemplary term "above … …" may include both orientations of "above … …" and "below … …". The device may also be positioned in other different ways (rotated 90 degrees or at other orientations) and the spatially relative descriptors used herein interpreted accordingly.

In addition, the terms "first", "second", etc. are used to define the components, and are only for convenience of distinguishing the corresponding components, and the terms have no special meaning unless otherwise stated, and therefore should not be construed as limiting the scope of the present invention.

FIG. 1 is a flow chart of an implementation of a vibration signal-based intelligent fault diagnosis method for a 2D pump according to the present invention;

as shown in fig. 1, the two-dimensional (2D) piston pump fault diagnosis method based on the local cut space arrangement and the gating circulation network of the invention comprises the following steps:

s1: and (3) data acquisition: collecting vibration signals reflecting the normal state and the fault state of the pump as original state data through a vibration sensor arranged on the 2D pump;

s2: data preprocessing: noise reduction reconstruction is carried out on original vibration data of the 2D pump by using a Variation Modal Decomposition (VMD) method optimized by a Genetic Algorithm (GA),

constructing a total fault sample set based on the original vibration data after noise reduction and reconstruction by using a time-frequency analysis method, and dividing the total fault sample set into a training set and a testing set according to a proportion based on a time-frequency domain characteristic sample set of a health state;

s3: model training and optimizing: constructing a fault diagnosis model for diagnosing faults of the piston pump, and randomly initializing all model parameters;

the training sample set is used as input of a fault diagnosis model, the state label is used as output of the fault diagnosis model, and the fault diagnosis model is trained;

evaluating the comprehensive performance of the diagnostic model by using the test sample set; including evaluating the training speed and accuracy of the fault diagnosis model,

selecting a feature fused by a local cut space arrangement method (LSTA) to perform optimization training on the model to obtain an optimized fault diagnosis model;

s4: fault diagnosis: and the vibration signal on the 2D pump is input into an optimized fault diagnosis model, and the fault diagnosis model outputs a state label value, so that a fault type identification precision result can be obtained.

S1, S2, S3 and S4 are sequentially executed;

the fault categories include: health, cylinder wear, piston wear, and cam wear.

The original vibration data of the piston pump is subjected to noise reduction reconstruction by using a variation modal decomposition method;

further, a time-frequency analysis method is utilized, a total fault sample set is constructed based on the original vibration data after noise reduction and reconstruction, and fault characteristic indexes of extracted signals are reconstructed from the time domain analysis, the frequency domain analysis and the time-frequency analysis angles of the piston pump under four operating states of health state, cylinder abrasion, piston abrasion and cam abrasion; obtaining a video feature domain feature sample set of health status

Constructing a high-dimensional and complex (1600 features with the size of 38 multiplied by 10) total fault sample set based on the original vibration data after noise reduction and reconstruction by using a time-frequency analysis method; dividing the total fault sample set into a training sample set and a test sample set according to the proportion, wherein 270 training sets and 370 test sets;

the time domain, the frequency domain and the time-frequency domain are obtained by using a time domain-frequency domain correlation formula (such as an average value, a pretilt degree and the like), and taking a health state sample as an example, the frequency domain characteristic value of a single sample is as follows. The time domain characteristic parameters have dimension characteristic parameters and non-dimension characteristic parameters, wherein the dimension has units, the non-dimension has no units, and the total number of the time domain characteristic parameters is 38.

FIG. 2 (a) is a graph of a collection of time domain features 1-10 of health status samples; (b) is a collection of state of health sample temporal features 11-17;

FIG. 3 (a) is a plot of a collection of frequency domain features 1-7 of health status samples; (b) is a plot of a frequency domain signature 8-9 collection of health status samples; (a) is a plot of a collection of frequency domain features 10-12 of health status samples; (a) is a plot of a collection of health status sample frequency domain features 13;

FIG. 4 is a graph of a time-frequency domain feature set for a health state sample;

the construction of the fault diagnosis model for diagnosing the fault of the piston pump comprises the following steps:

the training sample set is used as the input of the diagnosis model, the state label precision is used as the output of the fault diagnosis model, the diagnosis model is trained,

FIG. 5 is a diagram of a fault diagnosis model of the present invention; the fault diagnosis model constructed by the invention comprises the following components: input layer, hidden layer, output layer. The method comprises the steps of randomly initializing all fault diagnosis model parameters, taking a training sample set as input of a diagnosis model, taking identification accuracy of a state label as output of the model, and training the diagnosis model;

evaluating the comprehensive performance of the fault diagnosis model by using a test sample set;

and optimizing the fault diagnosis model by using the LSTA until the comprehensive performance of the fault model reaches the global optimum.

The fault diagnosis model includes: an input layer, a hidden layer and an output layer,

the input layer inputs a data set constructed based on the original vibration data after noise reduction and reconstruction into the hidden layer;

the hidden layer performs fault diagnosis on time-frequency characteristics of original vibration signals of the 2D pump based on the gating circulation network model, and when different modal characteristics are found, fault accuracy identification can be performed according to the corresponding modal characteristics;

the output layer outputs the fault accuracy of the identified corresponding modal feature;

the input layer: setting the length of a single sample as 3000, dividing the number of samples in each state as 400, totaling 1600 groups, defining a training model as a single hidden layer, wherein the training model comprises 100 hidden units, the characteristic dimension of an input sequence is 12, and inputting x at the moment t _t The relevant input weight is 300 multiplied by 12, the data set comprises time sequence data of 9 speakers, each sequence to be identified has 12 characteristics (different in length), and the total number of the training sets is 270 and the total number of the testing sets is 370;

the hidden layer: the hidden layer comprises a GRU unit, wherein the GRU unit internally uses an update gate (update gate) and a reset gate (reset gate) to adjust the state, and the hidden layer information h is at the time of t-1 _t-1 The associated weights are 300×100; FIG. 6 is a block diagram of the internal network of the GRU of the invention;

the method for denoising and reconstructing the original vibration data of the piston pump by using the variation modal decomposition method comprises the following steps:

and collecting fault characteristic indexes of the reconstructed signals under four operating states (health state, cylinder wear, piston wear and cam wear) of the piston pump from the angles of time domain analysis, frequency domain analysis and time frequency analysis, wherein 1600 characteristic attributes with the size of 38 multiplied by 10 are taken as characteristic parameters of a single fault sample.

And the feature index dimension reduction work is completed by combining an LTSA method.

Noise reduction reconstruction is carried out on original vibration data of the 2D pump, sample entropy is used as a population fitness function, iterative optimization is carried out on VMD method parameter combinations by using a genetic algorithm, the influence of environmental noise on the fault diagnosis accuracy of the 2D pump can be avoided, a high-dimensional and complex total fault sample set is constructed by using a time-frequency analysis method, and time domain feature information and frequency domain feature information of sensor signals can be synchronously reserved.

The optimization process of the variation modal decomposition method by using the genetic algorithm comprises the following steps:

s21: let the time sequence be x= { X (1), X (2),. The term, X (N) }, length N, then constitute an m-dimensional vector:

X(i)＝{x(i),x(i+1),...,x(i+m-1)}，(i＝1,...,N-m+1) (1)

s22: defining the distance between X (i) and X (j) as d [ X (i), X (j) ], i not equal to j, and when the difference between the two sequences is the largest, the formula (2) is given:

s23: let the threshold be r (r > 0), count the number that the sequence distance is smaller than the threshold, and make the ratio with the total vector number N-m, then have formula (3):

s24: the average of the results of step S23 is given by formula (4):

s25: adding 1 to the vector dimension formed by the time sequence, and repeating the steps S21 to S24:

s26: the sample entropy of the time series is as follows:

s27: decomposition layer number k and penalty in variational modal decomposition methodThe factor alpha is encoded to become chromosome in genetic algorithm to randomly initialize a population, the individual capacity of the population is set as n, and binary code is set as a ₁ ,a ₂ ,...a _n The following formula is given:

s28: let the comfort level of the ith individual be f _i Evaluating the fitness of each genetic individual by using sample entropy, outputting an optimal solution and ending if the fitness meets an optimization criterion, otherwise, performing the following steps:

s29: the individual selection is carried out by a wheel bet method, the selected individual is carried out in the next step, the unselected individual is eliminated, and the probability of each individual being selected is as follows:

s210: and (3) taking the individual selected in the step S29 as a parent, generating offspring in a crossing way, promoting the offspring to mutate at the same time, generating a new generation population, returning to the step S28, and finding out the optimal combination parameters of VMD decomposition.

The process of optimizing and training the fault diagnosis model through the fusion characteristics of the local cut space arrangement method (LSTA) is as follows:

let x= [ X ] _i ,...,x _N ]For the sample set of the input space, there are

Wherein τ _i ∈R ^{d ×1} Is x _i Is an eigen representation of f is a mapping function, ">

Representing noise.

S31: finding sample point x in input space _i Adopting epsilon neighbor method or k neighbor method to obtain sample point x _i Is a local spreading matrix of (a);

s32: for sample point x _i Performing eigenvalue decomposition on the local scattering matrix of (2), and obtaining a local principal component;

(1) Let X _i ＝[x _i1 ,...,x _ik ]Is x _i K-nearest neighbor under Euclidean distance metric, comprising x _i By itself, there is a single sample after processing

Wherein Q is a orthonormal matrix of d columns;

(2) Post-treatment X _i Becomes as follows

S33: based on the obtained local principal component Θ= [ Θ ] ₁ ,...,Θ _n ]Solving global coordinates T _i ＝[τ _i1 ,...,τ _ik ]，τ _ij Satisfying the formula:

wherein,,

for k tau _ij Mean value of L _i Is an unknown mapping transformation and plays a role in arrangement, and the matrix form of the formula (1) is shown as the formula (2):

wherein,,

reconstructing residual error, wherein the process requires tau obtained after dimension reduction _i Local mapping transform L _i The reconstruction residual, i.e. global optimum (3), can be minimized:

i.e. the output y at the current moment _t Is only subject to the input x at the current moment _t Hidden layer state h at last moment _t-1 Influence.

Example 1: a two-dimensional (2D) piston pump test bed is built, different fault states of the 2D pump are simulated through replacement of a cylinder body shell, a piston and a cam with abrasion faults in the test, vibration signals of the 2D pump are collected, pressure, temperature and flow are monitored, rotation speed and torque signals of a coupler are monitored, and the two-dimensional (2D) piston pump fault diagnosis method based on local cutting space arrangement and a door control circulation network is utilized for processing the signals.

The test classifies three types of parts with most obvious abrasion from the failed 2D pump, namely a cylinder shell, a piston and a cam, and as shown in fig. 7, the diagram (a) is a schematic diagram of time domain distribution of vibration signals in a health state, the diagram (b) is a schematic diagram of time domain distribution of vibration signals in a cylinder abrasion state, the diagram (c) is a schematic diagram of time domain distribution of vibration signals in a piston abrasion state and the diagram (D) is a schematic diagram of time domain distribution of vibration signals in a cam abrasion state. As can be seen from fig. 4, it is more difficult to determine different health states of the 2D pump corresponding to the pressure signal by observing the difference of the time domain waveforms.

The comparison diagrams of vibration signal frequency domains of four states of the actually measured 2D pump are shown in fig. 8, wherein (a) the diagram is a vibration signal frequency domain diagram of a healthy state, (b) the diagram is a vibration signal frequency domain diagram of a cylinder abrasion fault state, (c) the diagram is a vibration signal frequency domain diagram of a piston abrasion fault state, and (D) the diagram is a vibration signal frequency domain diagram of a cam abrasion fault state. As shown in fig. 8, the frequency component of the black circle mark is the piston rotation frequency, 103.125Hz. In order to make the samples contain as much information as possible for one revolution of the 2D pump, the length of a single sample is set to 3000, and the number of samples divided into each state is 400, which amounts to 1600 groups.

Noise reduction processing is carried out on each group of samples by using a GA-VMD algorithm, and optimal VMD decomposition parameters of each state sample are obtained as shown in table 1. Taking the 2D pump abrasion sample as an example, after being processed by GA-VMD, fig. 9 (a) is a time domain diagram of modal components of the 2D pump abrasion sample of the embodiment of the present invention after being processed by GA-VMD, and (b) is a frequency domain diagram of modal components of the 2D pump abrasion sample of the embodiment of the present invention after being processed by GA-VMD.

Table 1 optimum parameter combinations

Status information	Number of decomposition modes k	Penalty factor alpha
			Health status
	9	1960
			Wear of cylinder body	9	1488
Piston wear	10	1945
			Cam wear	8	1763

And optimizing the fault diagnosis model by using the LSTA until the comprehensive performance of the fault model reaches the global optimum.

Input layer: setting the length of a single sample as 3000, dividing the number of samples in each state as 400, totaling 1600 groups, defining a training model as a single hidden layer,the input sequence comprises 100 hidden units, the feature dimension of the input sequence is 12, and the input x is the moment t _t The relevant input weight is 300×12, the data set contains time sequence data of 9 speakers, each sequence to be identified has 12 characteristics (different lengths), and total 270 training sets and 370 test sets. Hidden layer: the GRU hidden layer state is adjusted by using an update gate (update gate) and a reset gate (reset gate) in the unit, and the hidden layer state information h is at the time of t-1 _t-1 The associated weights are 300 x 100.

In order to highlight the superiority of LSTA and GRU combined diagnosis, the fault diagnosis model is trained under the three conditions of original untreated feature, standardized treatment feature and LSTA fusion feature, the parameter indexes such as training accuracy, convergence performance, training time and the like are selected for comparison verification, the training is performed under the condition that the composition form of the input feature is consistent with the network training parameters, and the model parameter setting is shown in the table 2.

And selecting to add a dropout layer to prevent the overfitting phenomenon, wherein the solver is Adam, training is performed in a CPU environment, and the average value of multiple training is selected for avoiding the accidental training, training time and recognition accuracy. The feature of LSTA fusion processing is input to the fault diagnosis model, and the training process of the single model is shown in figure 10.

FIG. 10 shows the accuracy and loss of model training of a measured 2D pump wear sample via LSTA fusion features, where (a) is model training accuracy and (b) is model training loss;

as can be seen from fig. 10 (a), the convergence performance of the present model is optimal, and the accuracy tends to converge at about 120 iterative computations.

FIG. 11 is a confusion matrix diagram of accuracy in identifying the wear state of a 2D pump piston according to an embodiment of the present invention;

as shown in FIG. 11, the average recognition accuracy of the model reaches 99.6%, and the average training time is 39s, which is obviously superior to that of a standardized but unfused model training method.

TABLE 2 super parameter detail table of fault diagnosis model

Under the three types of conditions, the training performance and the final prediction accuracy of the fault diagnosis model are shown in table 3. The fault diagnosis method after the LSTA fusion treatment has the advantages of faster convergence speed, shorter training time and more accurate prediction precision.

TABLE 3 training performance of 2D pump diagnostic models under different conditions

Training conditions	Not standardized, not fused	Normalization process	Fusion processing
				Convergence speed	Unconverged is not converged	270 iterations convergence	120 times of iteration convergence
Training time	63s	44s	39s
				Prediction accuracy	47.1％	93.8％	99.6％

Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.

Claims (7)

1. A piston pump fault diagnosis method based on local cut space arrangement and gating circulation network is characterized by comprising the following steps:

and (3) data acquisition: collecting vibration signals reflecting the normal state and the fault state of the piston pump as original state data;

data preprocessing: optimizing a variation modal decomposition method by using a genetic algorithm, and carrying out noise reduction reconstruction on original vibration data of the piston pump by adopting the optimized variation modal decomposition method;

constructing a total fault sample set based on the original vibration data after noise reduction and reconstruction by using a time-frequency analysis method, and dividing the total fault sample set into a training set and a testing set according to a proportion based on a time-frequency domain characteristic sample set of a health state;

model training and optimizing: constructing a fault diagnosis model for diagnosing faults of the piston pump, taking a training sample set as input of the diagnosis model, taking a state label as output of the fault diagnosis model, training the diagnosis model,

optimizing and training the fault diagnosis model through fusion characteristics of a local cut space arrangement method to obtain an optimized fault diagnosis model;

fault diagnosis: and inputting the original vibration data of the piston pump into an optimized fault diagnosis model, and outputting a state label value by the optimized fault diagnosis model to further obtain the identification precision of fault types.

2. The method for diagnosing a fault of a piston pump based on a local cut space arrangement and gating cycle network according to claim 1, wherein the method comprises the following steps: the piston pump adopts a two-dimensional piston pump.

3. The method for diagnosing a fault of a piston pump based on a local cut space arrangement and gating cycle network according to claim 1, wherein the method comprises the following steps: the fault categories include: health, cylinder wear, piston wear, and cam wear.

4. The method for diagnosing a fault of a piston pump based on a local cut space arrangement and gating cycle network according to claim 1, wherein the method comprises the following steps: the fault diagnosis model includes: an input layer, a hidden layer and an output layer,

the input layer inputs a data set constructed based on the original vibration data after noise reduction and reconstruction into the hidden layer;

the hidden layer performs fault diagnosis on time-frequency characteristics of original vibration signals of the 2D pump based on the gating circulation network model, and when different modal characteristics are found, fault accuracy identification can be performed according to the corresponding modal characteristics;

the output layer outputs the fault accuracy of the identified corresponding modal feature;

the input layer: setting the length of a single sample as 3000, dividing the number of samples in each state as 400, totaling 1600 groups, defining a training model as a single hidden layer, wherein the training model comprises 100 hidden units, the characteristic dimension of an input sequence is 12, and inputting x at the moment t _t The relevant input weight is 300 multiplied by 12, the data set comprises time sequence data of 9 speakers, each sequence to be identified has 12 characteristics, and the total number of the training sets is 270 and the total number of the testing sets is 370;

the hidden layer: the hidden layer comprises a GRU unit, the state of the GRU unit is adjusted by using a refresh gate and a reset gate, and the state information h is hidden at the moment t-1 _t-1 The associated weights are 300 x 100.

5. The method for diagnosing a fault of a piston pump based on a local cut space arrangement and gating cycle network according to claim 1, wherein the method comprises the following steps: the method for denoising and reconstructing the original vibration data of the piston pump by using the variation modal decomposition method comprises the following steps:

and collecting fault characteristic indexes of the reconstructed signals under four operating states of the piston pump from the angles of time domain analysis, frequency domain analysis and time frequency analysis.

6. The method for diagnosing a fault of a piston pump based on a local cut space arrangement and gating cycle network according to claim 1, wherein the method comprises the following steps: the optimization process of the variation modal decomposition method by using the genetic algorithm comprises the following steps:

s21: let the time sequence be x= { X (1), X (2),. The term, X (N) }, length N, then constitute an m-dimensional vector:

X(i)＝{x(i),x(i+1),...,x(i+m-1)}，(i＝1,...,N-m+1) (1)

s22: defining the distance between X (i) and X (j) as d [ X (i), X (j) ], i not equal to j, and when the difference between the two sequences is the largest, the formula (2) is given:

s23: let the threshold be r (r > 0), count the number that the sequence distance is smaller than the threshold, and make the ratio with the total vector number N-m, then have formula (3):

s24: the average of the results of step S23 is given by formula (4):

s25: adding 1 to the vector dimension formed by the time sequence, and repeating the steps S21 to S24:

s26: the sample entropy of the time series is as follows:

s27: coding the decomposition layer number k and the penalty factor alpha in the variation modal decomposition method to form a chromosome in a genetic algorithm, randomly initializing a population, setting the individual capacity of the population as n, and binary coding as a ₁ ,a ₂ ,...a _n The following formula is given:

s28: let the comfort level of the ith individual be f _i Evaluating the fitness of each genetic individual by using sample entropy, outputting an optimal solution and ending if the fitness meets an optimization criterion, otherwise, performing the following steps:

s29: the individual selection is carried out by a wheel bet method, the selected individual is carried out in the next step, the unselected individual is eliminated, and the probability of each individual being selected is as follows:

s210: and (3) taking the individual selected in the step S29 as a parent, generating offspring in a crossing way, promoting the offspring to mutate at the same time, generating a new generation population, returning to the step S28, and finding out the optimal combination parameters of VMD decomposition.

7. The method for diagnosing a fault of a piston pump based on a local cut space arrangement and gating cycle network according to claim 1, wherein the method comprises the following steps: the process of optimizing and training the fault diagnosis model through the fusion characteristics of the local cut space arrangement method is as follows:

let x= [ X ] _i ,...,x _N ]For the sample set of the input space, there are

Wherein τ _i ∈R ^d×1 Is x _i Is an eigen representation of f is a mapping function, ">

Representing noise;

s31: finding sample point x in input space _i Adopting epsilon neighbor method or k neighbor method to obtain sample point x _i Is a local spreading matrix of (a);

s32: for sample point x _i Performing eigenvalue decomposition on the local scattering matrix of (2), and obtaining a local principal component;

(1) Let X _i ＝[x _i1 ,...,x _ik ]Is x _i K-nearest neighbor under Euclidean distance metric, comprising x _i By itself, there is a single sample after processing

Wherein Q is a orthonormal matrix of d columns;

(2) Post-treatment X _i Becomes as follows

S33: based on the obtained local principal component Θ= [ Θ ] ₁ ,...,Θ _n ]Solving global coordinates T _i ＝[τ _i1 ,...,τ _ik ]，τ _ij Satisfies the following formula:

wherein,,

for k tau _ij Mean value of L _i Is an unknown mapping transformation and plays a role in arrangement, and the matrix form of the formula (1) is shown as the formula (2):

wherein,,

reconstructing residual error, wherein the process requires tau obtained after dimension reduction _i Local mapping transform L _i The reconstruction residual, i.e. global optimum (3), can be minimized:

i.e. the output y at the current moment _t Is only subject to the input x at the current moment _t Hidden layer state h at last moment _t-1 Influence.

Images