CN111610026B

CN111610026B - Rotary machine fault diagnosis method based on deep clustering

Info

Publication number: CN111610026B
Application number: CN202010599985.XA
Authority: CN
Inventors: 安晶; 刘大琨; 刘聪; 徐森; 李青祝; 黄曙荣; 孙花
Original assignee: Yancheng Institute of Technology
Current assignee: Hangzhou Xingchen Zhilian Technology Co ltd
Priority date: 2020-06-28
Filing date: 2020-06-28
Publication date: 2021-06-04
Anticipated expiration: 2040-06-28
Also published as: CN111610026A

Abstract

The invention discloses a rotary machine fault diagnosis method based on deep clustering, which comprises the steps of collecting a label-free mechanical vibration signal; preprocessing the label-free mechanical vibration signal to obtain a signal data set; constructing a self-encoder structure, and pre-training the self-encoder structure; the self-encoder structure learns the initial feature representation of the data through pre-training; setting hyper-parameters of a manifold learning method and using the manifold learning method to search for more clusterible manifolds to relearn the stacked self-encoder embedding; and based on the updated embedding, finishing clustering by using a shallow clustering algorithm and evaluating results. The invention has the beneficial effects that: by adopting the shallow clustering algorithm, the calculation complexity of the model is reduced, the required time consumption is reduced, and meanwhile, the precision of the diagnosis result is ensured, so that the method can be better applied to the actual industrial scene.

Description

Rotary machine fault diagnosis method based on deep clustering

Technical Field

The invention relates to the technical field of mechanical fault diagnosis, in particular to a rotary mechanical fault diagnosis method based on deep clustering.

Background

In recent years, mechanical failure diagnosis technology has been greatly developed, and the types of data-driven failure diagnosis methods in practical application are various, including processing from the aspects of artificial intelligence, information fusion, multivariate statistical analysis, rough set, signal processing and the like. Due to the rapid development of the artificial intelligence technology in recent years, the mechanical fault diagnosis technology driven by intelligent data is also greatly developed at present, a large amount of training of models is required in the process, and the accuracy of the final diagnosis result is affected by the training effect of the models. Most of the existing diagnostic models are trained in a supervised mode, namely training samples are required to be provided with labels.

In order to improve the accuracy of the diagnosis result, a large amount of high-quality training data is needed during model training. However, in practical industrial applications, it is often difficult and expensive, and sometimes even impossible, to collect sufficient marker data, and these difficulties also greatly hinder the application of intelligent diagnostic methods.

Disclosure of Invention

This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.

The present invention has been made in view of the above-mentioned conventional problems.

Therefore, the technical problem solved by the invention is as follows: the rotary machine fault diagnosis method based on the deep clustering can obtain a good diagnosis result on label-free data, reduce the calculation complexity of a model and obtain a better effect in practical application.

In order to solve the technical problems, the invention provides the following technical scheme: a rotary machine fault diagnosis method based on deep clustering comprises the steps of collecting a label-free mechanical vibration signal; preprocessing the label-free mechanical vibration signal to obtain a signal data set; constructing a self-encoder structure, and pre-training the self-encoder structure; the self-encoder structure learns the initial feature representation of the data through pre-training; setting hyper-parameters of a manifold learning method and using the manifold learning method to search for more clusterible manifolds to relearn the stacked self-encoder embedding; and based on the updated embedding, finishing clustering by using a shallow clustering algorithm and evaluating results.

As a preferable scheme of the method for diagnosing faults of a rotary machine based on deep clustering according to the present invention, wherein: the non-tag mechanical vibration signal is acquired from a motor-driven mechanical system, the load during acquisition comprises 1, 2 or 3hp, and the acquisition position comprises a fan end, a driving end and a base.

As a preferable scheme of the method for diagnosing faults of a rotary machine based on deep clustering according to the present invention, wherein: the preprocessing includes data merging and spectral transformation, which is done by fourier transform.

As a preferable scheme of the method for diagnosing faults of a rotary machine based on deep clustering according to the present invention, wherein: the self-encoder is a stacked self-encoder, the structure of the stacked self-encoder is a five-layer structure and comprises an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer, an activation function of each layer is Relu, and the optimizer is Adam.

As a preferable scheme of the method for diagnosing faults of a rotary machine based on deep clustering according to the present invention, wherein: the dimension of an output layer of the stacked self-encoder structure is 10, the dimension of the output layer is the same as the data category of the mechanical vibration signal, and the data category comprises 10 types including 3 inner ring faults, 3 outer ring faults, 3 rolling body faults and 1 normal state.

As a preferable scheme of the method for diagnosing faults of a rotary machine based on deep clustering according to the present invention, wherein: the pre-training comprises that an encoder and a decoder of each self-encoder train each layer together, and after each layer is trained, the decoder is lost, the output of a hidden layer of the self-encoder is used as the input of the next self-encoder, and the training is continued.

As a preferable scheme of the method for diagnosing faults of a rotary machine based on deep clustering according to the present invention, wherein: the manifold learning method is UMAP, and the set UMAP hyper-parameter comprises the number of neighbors as a part, the embedded dimension of the target and the allowable minimum spacing distance between points in the embedding space.

As a preferable scheme of the method for diagnosing faults of a rotary machine based on deep clustering according to the present invention, wherein: the applied shallow clustering algorithm is GMM, and the clustering process further comprises randomly selecting a mean value of each cluster represented by k objects; solving the probability that each object belongs to each cluster according to a multivariate Gaussian density function and solving a likelihood function value of data; the mean, covariance matrix, and probability of each cluster are updated for each cluster based on the probability of each data point belonging to each cluster.

The invention has the beneficial effects that: in practical application, the method can use the original frequency domain information as model input, does not need a manual feature extraction algorithm and a large number of high-quality labeled samples, adopts a shallow clustering algorithm, reduces the calculation complexity of the model, reduces the required time consumption, and simultaneously ensures the precision of a diagnosis result, thereby being better applied to practical industrial scenes.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise. Wherein:

fig. 1 is a schematic overall flow chart of a depth clustering-based fault diagnosis method for a rotary machine according to a first embodiment of the present invention;

FIG. 2 is a diagram illustrating a stacked self-encoder structure according to a first embodiment of the present invention;

FIG. 3 is a graph showing experimental accuracy on three data sets using the method described in this example;

FIG. 4 is a graph showing experimental results on data A using different methods;

FIG. 5 is a graph showing experimental results on data B using different methods;

fig. 6 is a graph showing experimental results on data C using different methods.

Detailed Description

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, specific embodiments accompanied with figures are described in detail below, and it is apparent that the described embodiments are a part of the embodiments of the present invention, not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making creative efforts based on the embodiments of the present invention, shall fall within the protection scope of the present invention.

In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.

Furthermore, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.

The present invention will be described in detail with reference to the drawings, wherein the cross-sectional views illustrating the structure of the device are not enlarged partially in general scale for convenience of illustration, and the drawings are only exemplary and should not be construed as limiting the scope of the present invention. In addition, the three-dimensional dimensions of length, width and depth should be included in the actual fabrication.

Meanwhile, in the description of the present invention, it should be noted that the terms "upper, lower, inner and outer" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, and are only for convenience of describing the present invention and simplifying the description, but do not indicate or imply that the referred device or element must have a specific orientation, be constructed in a specific orientation and operate, and thus, cannot be construed as limiting the present invention. Furthermore, the terms first, second, or third are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.

The terms "mounted, connected and connected" in the present invention are to be understood broadly, unless otherwise explicitly specified or limited, for example: can be fixedly connected, detachably connected or integrally connected; they may be mechanically, electrically, or directly connected, or indirectly connected through intervening media, or may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.

Example 1

Referring to the schematic diagram of fig. 1, there is shown an overall flowchart of a method for diagnosing faults of a rotary machine based on deep clustering according to this embodiment, which specifically includes the following steps,

s1: collecting a label-free mechanical vibration signal;

the non-tag mechanical vibration signal can be acquired through a motor-driven mechanical system, the load during acquisition comprises 1, 2 or 3hp, and the acquisition position comprises a fan end, a driving end and a base. The sampling frequency in this embodiment is set at 48kHz, and there are four types OF mechanical bearings, normal (N) and three failure types, including outer race failure (OF), inner race failure (IF) and Roller Failure (RF), which are divided into three severity levels for each OF the three failure types, including failure diameters OF 0.007 inches, 0.014 inches and 0.021 inches, respectively, for a total OF 10 health states.

S2: preprocessing the label-free mechanical vibration signal to obtain a signal data set;

specifically, the signal data set includes a plurality of samples from the acquired signals after preprocessing, the preprocessing includes data combination and spectrum conversion, wherein the spectrum conversion can be realized by performing fast fourier transform on the signals, and the acquired vibration signals are one-dimensional long signals, for example, from top to bottom, each 2400 points serves as a sample, and the signals are combined. The samples are combined into 2400 x 2000 sample matrices and fourier transformed into 1200 x 2000 sample matrices.

S3: constructing a self-encoder structure, and pre-training the self-encoder structure;

specifically, the self-encoder in this embodiment is preferably a stacked self-encoder, the stacked self-encoder is a stacked automatic encoder structure, referring to the schematic diagram of fig. 2, the stacked self-encoder structure constructed in this embodiment is a five-layer structure, and includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer, and the activation function of each layer is Relu, and the optimizer is Adam. The number of neurons of the input layer is the number of dimensions of samples in the signal data set, the number of dimensions of the first hidden layer, the second hidden layer and the third hidden layer is 500, 500 and 2000 respectively, the number of dimensions of the output layer is determined by the number of clusters, namely the number of types of the signal data set is 10 in the embodiment, the number of dimensions of the output layer of the stacked self-encoder structure is the same as the number of types of data, therefore, the number of dimensions of the output layer is also 10, and the types of data comprise 3 inner ring faults, 3 outer ring faults, 3 rolling body faults and 1 normal state. The activation function for all layers is Relu, the optimizer is Adam, and the maximum number of training is 1000. The pre-training is to connect a plurality of automatic encoders to form each hidden layer of the stacked automatic encoder, so as to realize the layer-by-layer extraction of the fault characteristics.

In the pre-training, an encoder and a decoder of each self-encoder train each layer together, and after each layer is trained, the decoder is discarded, the output of a hidden layer of the self-encoder is used as the input of the next self-encoder and continues to train, and after the pre-training, the decoder network is discarded.

S4: the self-encoder structure learns the initial characteristic representation of the data through pre-training; in this embodiment, a signal data set is input into a network structure of a stacked self-encoder, and an initial feature representation of the data is obtained through a pre-training process.

S5: setting hyper-parameters of a manifold learning method and using the manifold learning method to search for more clusterible manifolds to relearn the stacked self-encoder embedding; in this embodiment, the manifold learning method is preferably UMAP, and the initial feature output by the stacked self-encoder structure further learns the clustering manifold representation by the UMAP local manifold learning method. Wherein the hyper-parameters of the UMAP are set to include the number of neighbors as a local, the dimension of target embedding, and the minimum separation distance allowed between points in the embedding space.

Specifically, there are three important parameters in the UMAP that affect the performance, namely, three hyper-parameters set in the present embodiment. The first is to consider the number of neighbors as local, the parameter represents the trade-off between the granularity of preserving the local structure and the granularity of capturing the global structure, since what needs to be implemented in practical application is to integrate the local structure into the embedding, a lower value is usually selected for the number of neighbors, and the parameter is set to 20 in the model of the embodiment; the second is the dimension of target embedding, which is set to the number of clusters in this embodiment, i.e., the dimension of target embedding is 10; the third is the minimum separation distance allowed between points in the embedding space, where the set value of the minimum separation distance can more accurately capture the real manifold structure, but the lower value may also result in dense clouds, making visualization difficult, in this embodiment, since the main purpose is not to require visualization, but to more accurately represent the real manifold, the minimum separation distance in UMAP is set to 0 in this example.

And (3) taking the 10-dimensional feature representation output after the network structure of the stacked self-encoder is processed as input, further searching for a more-clustering manifold by a UMAP local manifold learning method to relearn the self-encoder embedding, wherein the output result is also the 10-dimensional feature representation.

S6: and based on the updated embedding, finishing clustering by using a shallow clustering algorithm and evaluating results.

Specifically, in this embodiment, the shallow clustering algorithm is preferably a GMM, which is a probabilistic clustering method, and first, k objects are randomly selected to represent a mean value of each cluster, a covariance matrix of each cluster is guessed, and the probability of each cluster is assumed to be equal when an initial state is assumed; then, the probability that each object belongs to each cluster is solved according to a multivariate Gaussian density function, and a likelihood function value of data is solved; finally, the mean, covariance matrix, and probability of each cluster are updated based on the probability of each data point belonging to each cluster. And iterating the above two steps continuously until the algorithm converges, and assigning the objects to the cluster with the highest probability according to the probability that each object belongs to each cluster. And clustering the result obtained by the UMAP local manifold learning method through a shallow clustering algorithm GMM to obtain a fault diagnosis result, and finishing evaluation.

Scene one:

in order to verify the advantages of the method for diagnosing the fault of the rotary machine based on the deep clustering provided by the embodiment in practical application, the method provided by the embodiment and the conventional method are respectively adopted to carry out experiments under the same conditions, and the method specifically comprises the following steps:

bearing fault data for experimental verification was obtained from the CWRU (university of kasseiki) bearing data center, where vibration signals at different fault locations and under different health conditions were selected, and in particular, see table 1 below, data set a contains 2000 samples of 10 load-bearing health states, 1hp and 1772rpm, respectively, and the compositions of data sets B and C are similar to data set a. Each health state has 200 samples, each sample contains 2400 data points, 2400 fourier coefficients are obtained after FFT of each sample, and the first 1200 coefficients are used in each sample because the coefficients are symmetric.

Table 1: 48kHz bearing data set parameter table

In the experiment, because the class label of the bearing data is known, the matching degree of the clustering result and the known class label is quantified by using the Accuarcy and NMI of the clustering, and the algorithm is comprehensively evaluated. And the popular evaluation indexes Accuarcy and NMI values in the field of machine learning are selected, so that the matching degree of the clustering result and the real class label can be effectively measured, and the range is 0 to 1. The larger the value, the more the clustering result conforms to the real category label. When the clustering result matches the true category label, Accuarcy and NMI reach a maximum of 1. Accuarcy is used to compare the obtained tag with the true tag of the data, and the Accuarcy and NMI values are defined as follows,

wherein, y_iAs a genuine label, c_iDistributing the clusters obtained by the algorithm, wherein n is the data quantity, delta () is an indication function, and map () is a function representing the reapportion distribution of the optimal class mark so as to ensure the correct statistics; y is a real label, C is a cluster label, H (Y) is the entropy of Y, and I (Y; C) is the mutual information between Y and C.

The conventional methods adopted in the experiments include: (1) GMM, (2) K-Means, (3) SC (spectral clustering), (4) ISOMAP + GMM, (5) LLE + GMM, (6) tSNE + GMM, (7) UMAP + GMM, (8) AE + GMM, (9) DEC, (10) AE + ISOMAP + GMM, (11) AE + LLE + GMM, and (12) AE + tSNE + GMM, wherein (1) - (3) are common shallow clustering methods, (4) - (8) are manifold learning + shallow clustering methods, (9) are deep-embedding clustering methods, and (10) - (12) are AE + manifold learning + shallow clustering methods; the method provided by this embodiment is AE + UMAP + GMM, which may be referred to as E2LMC method for short,

in order to reduce the influence of random factors, 10 times of experiments are performed on each data set by using the E2LMC method of the present embodiment, and the results of the experiments on the three data sets A, B and C are shown in fig. 3, it can be seen that the Accuracy of each data set is above 97%, and the Accuracy (Accuracy) of the a data set is even above 99%. Experimental results show that the method has good clustering performance and stability.

Referring to table 2 below, the results of comparing the accuracy and NMI under different methods are shown in table 2,

table 2: accuracy and NMI value lookup tables under each method

It can be seen that the end result is more accurate in this embodiment method than in the conventional method, so more local methods should be better choices when applied to the bearing dataset and when applied to the self-encoder embedding. Further, UMAP that focuses on local structures but better captures global structures than tSNE can obtain the best clusterable manifold and obtain better clustering performance.

In addition, when the fault diagnosis problem is solved by actually using the deep clustering algorithm, time is another important factor to be considered, and the following experiment is carried out,

the experimental machine adopts an Intel core i7 processor, a video card NVIDIA GeForce 940MX and a memory 8 GB. The stack-type self-encoder is a stack automatic encoder structure, and trains 1000 epochs; the UMAP and tSNE methods are used for manifold learning, and the experiments do not use early stopping in stacked self-encoder training. Referring to table 3 below, table 3 shows the time taken by each part of the method of the present embodiment in seconds and the total time taken for each data set.

Table 3: the running time and the total time of each part in the method of the embodiment

It can be seen that, for the data sets A, B and C, the method provided by the embodiment can complete clustering within about 1160 seconds, and the result shows that the speed of processing high-dimensional data by the UMAP method is fast, so that the effectiveness and feasibility of the method in practical application are further verified, and clustering can be completed faster.

In order to further understand the clustering performance of the method for the bearing data sets, the dimension is set to be 2, the clustering results of the three data sets are visualized, and the results on the three data sets are respectively shown in fig. 4-6. It can be seen that, in both methods, each data is aggregated into 10 clusters, and compared with the AE + tSNE + GMM method, the method provided by this embodiment has better intra-class compactness and inter-class separability, thereby obtaining higher clustering performance.

It should be recognized that embodiments of the present invention can be realized and implemented by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable memory. The methods may be implemented in a computer program using standard programming techniques, including a non-transitory computer-readable storage medium configured with the computer program, where the storage medium so configured causes a computer to operate in a specific and predefined manner, according to the methods and figures described in the detailed description. Each program may be implemented in a high level procedural or object oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language. Furthermore, the program can be run on a programmed application specific integrated circuit for this purpose.

Further, the operations of processes described herein can be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and/or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) collectively executed on one or more processors, by hardware, or combinations thereof. The computer program includes a plurality of instructions executable by one or more processors.

Further, the method may be implemented in any type of computing platform operatively connected to a suitable interface, including but not limited to a personal computer, mini computer, mainframe, workstation, networked or distributed computing environment, separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, and the like. Aspects of the invention may be embodied in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optically read and/or write storage medium, RAM, ROM, or the like, such that it may be read by a programmable computer, which when read by the storage medium or device, is operative to configure and operate the computer to perform the procedures described herein. Further, the machine-readable code, or portions thereof, may be transmitted over a wired or wireless network. The invention described herein includes these and other different types of non-transitory computer-readable storage media when such media include instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. The invention also includes the computer itself when programmed according to the methods and techniques described herein. A computer program can be applied to input data to perform the functions described herein to transform the input data to generate output data that is stored to non-volatile memory. The output information may also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including particular visual depictions of physical and tangible objects produced on a display.

As used in this application, the terms "component," "module," "system," and the like are intended to refer to a computer-related entity, either hardware, firmware, a combination of hardware and software, or software in execution. For example, a component may be, but is not limited to being: a process running on a processor, an object, an executable, a thread of execution, a program, and/or a computer. By way of example, both an application running on a computing device and the computing device can be a component. One or more components can reside within a process and/or thread of execution and a component can be localized on one computer and/or distributed between two or more computers. In addition, these components can execute from various computer readable media having various data structures thereon. The components may communicate by way of local and/or remote processes such as in accordance with a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and/or across a network such as the internet with other systems by way of the signal).

It should be noted that the above-mentioned embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.

Claims

1. A rotary machine fault diagnosis method based on deep clustering is characterized in that: comprises the steps of (a) preparing a mixture of a plurality of raw materials,

collecting a label-free mechanical vibration signal;

preprocessing the label-free mechanical vibration signal to obtain a signal data set;

constructing a self-encoder structure, and pre-training the self-encoder structure;

the self-encoder structure learns the initial feature representation of the data through pre-training;

setting hyper-parameters of a manifold learning method and searching for more clustered manifolds by using the manifold learning method to relearn stacked self-encoder embedding;

the manifold learning method is UMAP, and the set UMAP hyper-parameter comprises the number of neighbors as parts, the embedded dimension of the target and the allowable minimum spacing distance between points in an embedding space;

based on the updated embedding, completing clustering by using a shallow clustering algorithm, and evaluating results;

the applied shallow clustering algorithm is GMM, and the clustering process further comprises randomly selecting a mean value of each cluster represented by k objects; solving the probability that each object belongs to each cluster according to a multivariate Gaussian density function and solving a likelihood function value of data; the mean, covariance matrix, and probability of each cluster are updated for each cluster based on the probability of each data point belonging to each cluster.

2. The rotary machine fault diagnosis method based on deep clustering according to claim 1, characterized in that: the non-tag mechanical vibration signal is acquired from a motor-driven mechanical system, the load during acquisition comprises 1, 2 or 3hp, and the acquisition position comprises a fan end, a driving end and a base.

3. The rotary machine fault diagnosis method based on deep clustering according to claim 1 or 2, characterized in that: the preprocessing includes data merging and spectral transformation, which is done by fourier transform.

4. The rotary machine fault diagnosis method based on deep clustering according to claim 3, characterized in that: the stacked self-encoder structure is a five-layer structure and comprises an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer, the activation function of each layer is Relu, and an optimizer is Adam.

5. The rotary machine fault diagnosis method based on deep clustering according to claim 4, characterized in that: the dimension of an output layer of the stacked self-encoder structure is 10, the dimension of the output layer is the same as the data category of the mechanical vibration signal, and the data category comprises 10 types including 3 inner ring faults, 3 outer ring faults, 3 rolling body faults and 1 normal state.

6. The method for diagnosing faults of rotating machinery based on deep clustering according to claim 4 or 5, wherein: the pre-training comprises that an encoder and a decoder of each self-encoder train each layer together, and after each layer is trained, the decoder is lost, the output of a hidden layer of the self-encoder is used as the input of the next self-encoder, and the training is continued.