CN111382792B - Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation - Google Patents
Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation Download PDFInfo
- Publication number
- CN111382792B CN111382792B CN202010155523.9A CN202010155523A CN111382792B CN 111382792 B CN111382792 B CN 111382792B CN 202010155523 A CN202010155523 A CN 202010155523A CN 111382792 B CN111382792 B CN 111382792B
- Authority
- CN
- China
- Prior art keywords
- sparse
- dictionary
- double
- rolling bearing
- fault diagnosis
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000005096 rolling process Methods 0.000 title claims abstract description 113
- 238000003745 diagnosis Methods 0.000 title claims abstract description 89
- 238000000034 method Methods 0.000 title claims abstract description 71
- 238000012549 training Methods 0.000 claims abstract description 42
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 37
- 239000013598 vector Substances 0.000 claims abstract description 31
- 230000036541 health Effects 0.000 claims description 21
- 230000008569 process Effects 0.000 claims description 10
- 238000012216 screening Methods 0.000 claims description 10
- 230000009977 dual effect Effects 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000012360 testing method Methods 0.000 abstract description 8
- 230000000052 comparative effect Effects 0.000 description 23
- 230000000694 effects Effects 0.000 description 8
- 238000002474 experimental method Methods 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 239000010410 layer Substances 0.000 description 3
- 238000000605 extraction Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 239000002356 single layer Substances 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000000306 component Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 239000008358 core component Substances 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000556 factor analysis Methods 0.000 description 1
- 230000002349 favourable effect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007786 learning performance Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2433—Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2136—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on sparsity criteria, e.g. with an overcomplete basis
Abstract
The invention belongs to the technical field of fault diagnosis of rolling bearings. The invention discloses a rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation, and the method comprises the following steps of S1, training a rolling bearing vibration signal by adopting a double-sparse dictionary learning algorithm to obtain a double-sparse dictionary; step S2, obtaining the decomposition coefficients of the vibration signals of the modeled rolling bearings of different fault types under the double sparse dictionaries and taking the decomposition coefficients as feature vectors; step S3, inputting the characteristic vector obtained in the step S2 into a deep belief network for training and learning to obtain a rolling bearing vibration signal fault diagnosis model; and step S4, inputting the vibration signal of the rolling bearing to be tested containing the fault information into a fault diagnosis model for fault identification, and completing fault diagnosis. By adopting the fault diagnosis method for the vibration signal of the rolling bearing, the higher diagnosis precision and the higher accuracy stability can be obtained, the training and testing time of a deep belief network can be greatly reduced, and the fault diagnosis efficiency is improved.
Description
Technical Field
The invention belongs to the technical field of rolling bearing fault diagnosis, and particularly relates to a rolling bearing fault diagnosis method based on combination of sparse representation of a double-sparse dictionary and a deep belief network.
Background
The rolling bearing is used as the most widely used key part in the rotary machinery, the running state of the rolling bearing has direct influence on the comprehensive performance of the whole machine, and once a fault occurs, shutdown maintenance or even major accidents cannot be avoided. The mechanical vibration signal can reflect the state information of the rolling bearing, so that the fault diagnosis of the vibration signal in the working state of the rolling bearing is an effective method for reducing the downtime and the maintenance cost and ensuring the production safety.
However, since the mechanical vibration signal inevitably contains some noise information, which has non-stationarity and non-linearity, the method of using the characteristic of the subjective selection of the researcher not only wastes time and labor, but also the extracted characteristic parameter sometimes cannot represent the characteristic of the signal itself well, and the discrimination is poor; for mechanical fault diagnosis, the proportion of fault information in one period to the whole vibration signal is relatively small, when fault diagnosis is carried out classification and identification, the parts which cannot represent the fault information in the time domain signals are useless, and if the components which do not contribute to fault classification can be reduced or removed, more concise fault information is obtained, and the fault diagnosis is more efficient.
Sparse representation is a hotspot in the field of signal processing in recent years, and the theory indicates that by constructing a reasonable dictionary model, signals can be approximated by linear combinations of a small number of dictionary atoms. Most of the research on dictionary construction is mainly divided into two types: an analysis dictionary and a learning dictionary. The analysis dictionary is a highly structured mathematical model, has the characteristic of quick solving, but simultaneously has the defect of poor signal adaptability, and comprises a wavelet dictionary, a DCT (discrete cosine transformation) dictionary, a curvelet dictionary and the like; the learning dictionary is an overcomplete dictionary obtained by training a group of sample signals by adopting a machine learning algorithm, so that the self-adaption is good, but the non-structuralization causes difficulty in solving, and the dictionary comprises PCA, MOD and K-SVD.
Therefore, the dictionary learning efficiency is low due to the problems existing in the mechanical vibration signal processing in the rolling bearing fault diagnosis process by adopting the conventional sparse representation of the analysis dictionary or the learning dictionary, and the precision and the accuracy of the fault diagnosis of the whole rolling bearing are further influenced.
Disclosure of Invention
In order to improve the diagnosis precision and accuracy of the rolling bearing fault, the invention provides a rolling bearing fault diagnosis method based on double sparse dictionary sparse representation. The fault diagnosis method of the rolling bearing specifically comprises the following steps:
step S1, training the mechanical vibration signal of the modeled rolling bearing by adopting a double-sparse dictionary learning method to obtain a double-sparse dictionary;
step S2, obtaining the decomposition coefficient of the mechanical vibration signal of the modeling rolling bearing under the double sparse dictionaries and taking the decomposition coefficient as a feature vector;
step S3, inputting the characteristic vector obtained in the step S2 into a deep belief network for training and learning to obtain a rolling bearing fault diagnosis model;
and step S4, inputting the mechanical vibration signal of the rolling bearing to be tested into the rolling bearing fault diagnosis model for fault identification, and completing fault diagnosis.
Preferably, in step S1, firstly, a double sparse dictionary learning method is adopted to train the mechanical vibration signal of the modeled rolling bearing to obtain a double sparse sub-dictionary, then, the obtained double sparse sub-dictionary is subjected to double sparse sub-dictionary atom screening optimization, and atoms with low contribution rate are removed to obtain a final double sparse dictionary.
Further preferably, in step S1, the double-sparse sub-dictionary is optimized by using a sparse decomposition algorithm, and the mechanical vibration signal of the modeled rolling bearing and the absolute value of the maximum inner product of the atoms of the corresponding double-sparse sub-dictionary are used as the atom screening standard of the double-sparse sub-dictionary to perform atom screening, and then the screened candidate atoms are sequentially arranged and combined in sequence to form the final double-sparse dictionary.
Preferably, in step S1, the specific process of obtaining the dual sparse dictionary is as follows:
step S11, training the mechanical vibration signal of the modeled rolling bearing based on a learning dictionary algorithm to obtain a base dictionary;
step S12, solving a sparse coefficient matrix of the mechanical vibration signal of the modeled rolling bearing under the base dictionary by using a sparse decomposition algorithm to serve as a sparse dictionary;
and step S13, solving and obtaining the double-sparse dictionary through a double-sparse dictionary learning algorithm according to the base dictionary and the sparse dictionary.
Further preferably, in step S11, the learning dictionary algorithm is a K-SVD algorithm.
Preferably, the OMP algorithm is selected as the sparse decomposition algorithm.
Preferably, the modeling rolling bearing mechanical vibration signal is composed of rolling bearing mechanical vibration signals of one or more different health states; wherein, when the mechanical vibration signal of the modeling rolling bearing is composed of rolling mechanical vibration signals of various health states,
in step S1, training mechanical vibration signals of rolling bearings in different health states to obtain corresponding double sparse dictionaries respectively;
in the step S2, respectively obtaining the decomposition coefficients of the mechanical vibration signals of the rolling bearings in different health states in the double sparse dictionaries, and respectively taking the decomposition coefficients as feature vectors;
in the step S3, the feature vectors obtained for the mechanical vibration signals of the rolling bearing in different health states are input to the deep belief network for training and learning, and a rolling bearing fault diagnosis model capable of identifying different health states is obtained.
Further preferably, the modeling rolling bearing mechanical vibration signal may be any one or more OF normal condition signal data (N), inner ring fault signal data (IF), outer ring fault signal data (OF) or rolling fault signal data (RF).
When the rolling bearing fault diagnosis method based on the sparse representation of the double sparse dictionaries is adopted to carry out rolling bearing fault diagnosis processing, the rolling bearing fault diagnosis method has the following beneficial technical effects:
1. in the invention, the double-sparse dictionary learning algorithm is adopted, so that the learning dictionary and the fixed dictionary are combined by utilizing the double-sparse dictionary learning algorithm, the respective advantages of the two dictionary learning algorithms are inherited, and the high-efficiency sparse representation capability is obtained. Meanwhile, sparse representation feature vectors of the original vibration signals under the double sparse dictionaries are used as deep neural network learning target data, so that input data dimensionality of a deep belief network model is reduced, neural network complexity is reduced, and time for overall network training is greatly saved.
2. Compared with the traditional intelligent fault diagnosis, the rolling bearing fault diagnosis method based on the double-sparse dictionary sparse representation and the deep belief network has the advantages that the problems of low efficiency, high time consumption and the like of feature extraction in the process of feature extraction and in the process of processing big data are solved, the double-sparse dictionary sparse representation is combined with the rolling bearing fault diagnosis method, the double-sparse dictionary atoms are further simplified, the secondary atoms with low use frequency in the redundant dictionary are removed, and accordingly sparse feature dimensions are compressed to the maximum extent; and then, the simplified sparse representation characteristic signals are combined with the deep belief network to carry out efficient identification and diagnosis of the fault, so that higher diagnosis precision and accuracy stability are obtained, the training and testing time of the deep belief network is greatly reduced, and the fault diagnosis efficiency is improved.
Drawings
FIG. 1 is a schematic flow chart of a rolling bearing fault diagnosis method based on double sparse dictionary sparse representation according to the present invention;
FIG. 2 is a schematic flow chart of obtaining a double sparse dictionary by using a double sparse dictionary learning method according to the present invention;
FIG. 3 is a graph comparing the diagnostic accuracy obtained in 15 experiments for example 1, comparative example 2, and comparative example 3, respectively;
FIG. 4 is a graph comparing separability of an original signal and a sparsely represented signal of mechanical vibration of a rolling bearing;
FIG. 5 is a graph comparing the diagnostic accuracy and training and diagnostic time obtained in 15 experiments for example 1 and comparative example 4, respectively.
Detailed Description
The technical solution of the present invention will be further described in detail with reference to the accompanying drawings and embodiments.
Referring to fig. 1, the rolling bearing fault diagnosis method based on dual sparse dictionary sparse representation provided by the invention comprises the following steps:
and step S1, training the mechanical vibration signal of the modeled rolling bearing by adopting a double-sparse dictionary learning algorithm to obtain a double-sparse dictionary. The mechanical vibration signal of the modeled rolling bearing is a vibration signal determined by fault types, and the vibration signals of the rolling bearing in different fault states can be acquired through the acceleration sensor.
With reference to fig. 2, in the present invention, a specific process for obtaining a dual-sparse dictionary by using a dual-sparse dictionary learning algorithm is as follows:
and step S11, training the mechanical vibration signal of the modeled rolling bearing based on the learning dictionary algorithm to obtain a base dictionary.
And step S12, solving a sparse coefficient matrix of the mechanical vibration signal of the modeled rolling bearing under the base dictionary by using a sparse decomposition algorithm to serve as a sparse dictionary.
And step S13, according to the base dictionary and the sparse dictionary, solving through a double sparse dictionary learning algorithm to obtain a double sparse dictionary.
In the process of obtaining the base dictionary through the learning dictionary algorithm, the K-SVD dictionary is excellent in vibration signal learning performance and widely used in the field of dictionary learning, so that the K-SVD algorithm is preferably used for training the mechanical vibration signal of the modeling rolling bearing to obtain the base dictionary.
And step S2, obtaining the decomposition coefficient of the mechanical vibration signal of the modeled rolling bearing under the double sparse dictionaries and taking the decomposition coefficient as a feature vector. In the invention, an Orthogonal Matching Pursuit (OMP) algorithm is selected as a sparse decomposition algorithm to solve the decomposition coefficient of the mechanical vibration signal of the modeled rolling bearing under the double sparse dictionaries and is used as a feature vector. Similarly, other sparse decomposition algorithms, such as the matching pursuit algorithm (MP algorithm), may be used according to different situations.
And step S3, inputting the characteristic vector obtained in the step S2 into a deep belief network for training and learning, and obtaining a rolling bearing fault diagnosis model.
At the moment, the deep learning training is carried out on the characteristic vector of the mechanical vibration signal of the rolling bearing with the specific fault type through the deep belief network, so that a fault diagnosis model is established, and the better classification and identification effect of the mechanical vibration signal of the rolling bearing corresponding to the fault type is achieved.
And step S4, inputting the mechanical vibration signal of the rolling bearing to be tested into the rolling bearing fault diagnosis model for fault identification, and completing fault diagnosis.
In the step S3, the deep belief network training learning process finds that, because the double sparse dictionaries obtained in the step S1 have redundancy of dictionary atoms, a large number of dictionary atoms are not used or the use frequency is very low during sparse decomposition, and the existence of the atoms increases the dimensionality of sparse representation features obtained during subsequent sparse decomposition of the mechanical vibration signal of the modeled rolling bearing, which is not beneficial to the training and testing of the subsequent deep belief network, so that the dictionary atoms need to be screened to remove atoms with low contribution rate.
Preferably, in the process of obtaining the double sparse dictionaries, firstly, a double sparse dictionary learning method is adopted to train mechanical vibration signals of the modeled rolling bearing to obtain the double sparse sub-dictionaries; and then, carrying out screening optimization on the atoms of the obtained double-sparse sub-dictionary, removing the atoms with low contribution rate, and obtaining the final double-sparse dictionary, so that the effect of reducing the dimension of the obtained characteristic vector when processing according to the subsequent step S2 is achieved, the deep belief network training learning time is further shortened, and the efficiency is improved.
Further, the specific operation of optimizing the double-sparse sub-dictionary by adopting the sparse decomposition algorithm is as follows: firstly, using mechanical vibration signals of a to-be-decomposed modeling rolling bearing and the maximum inner product absolute value of atoms of a corresponding double-sparse sub-dictionary as a double-sparse sub-dictionary atom screening standard to carry out atom screening, and then sequentially arranging and combining screened alternative atoms in sequence to form a final double-sparse dictionary.
At the moment, the mechanical vibration signal of the rolling bearing to be decomposed and the maximum inner product absolute value of the atoms of the corresponding double sparse sub-dictionaries are used as the atom screening standard of the double sparse sub-dictionaries, so that the first few atoms with the highest use frequency can be used as the alternative atoms of the double sparse sub-dictionaries, and the alternative atoms are sequentially arranged and combined in sequence to obtain the optimized double sparse dictionaries. Therefore, dictionary atom redundancy optimization processing of the double-sparse dictionary obtained by directly adopting the double-sparse dictionary learning algorithm can be realized, and the dimensionality of the feature vector obtained by the sparse decomposition algorithm in the follow-up process is reduced, so that the training learning speed of the deep belief network on the feature vector is improved, the training time is reduced, and the diagnosis efficiency is improved.
In addition, in the rolling bearing fault diagnosis method based on the sparse representation of the double sparse dictionaries, the mechanical vibration signal of the modeled rolling bearing can be a rolling bearing mechanical vibration signal in a healthy state, and can also be composed of rolling bearing mechanical vibration signals in various different healthy states.
When the mechanical vibration signal of the modeled rolling bearing is composed of rolling mechanical vibration signals of various different health states, in step S1, the mechanical vibration signals of the rolling bearing of different health states are trained respectively, so as to obtain corresponding double sparse dictionaries respectively; in step S2, obtaining the decomposition coefficients of the mechanical vibration signals of the rolling bearings in different health states under the double sparse dictionaries respectively and using the decomposition coefficients as feature vectors respectively; in step S3, the feature vectors obtained from the mechanical vibration signals of the rolling bearings in different health states are input to the deep belief network for training and learning, so as to obtain the rolling bearing fault diagnosis models capable of identifying different health states.
The modeling rolling bearing mechanical vibration signal can be any one or more OF signal data (N) under normal conditions, inner ring fault signal data (IF), outer ring fault signal data (OF) or rolling fault signal data (RF).
The Deep Belief Network (DBN) is a multi-hidden-layer probability generation type artificial neural network constructed by stacking a plurality of Restricted Boltzmann Machines (RBMs), and the core component RBM of the deep belief network is an algorithm of a probability distribution model based on energy. At the moment, the distributed energy of the sparse points of the sparse representation feature vectors of different types obtained by the double sparse comprehensive dictionaries is concentrated and is matched with the characteristic of limited Boltzmann machine feature learning, so that a good classification and identification effect can be obtained when fault diagnosis is carried out by combining the sparse representation feature vectors with a deep belief network, and effective identification and diagnosis of different faults are realized.
Next, the effect of performing rolling bearing fault diagnosis by using the rolling bearing fault diagnosis method based on dual sparse dictionary sparse representation proposed by the present invention is described by comparing through experiments.
The bearing fault data of the bearing test center database of the American Kaiser university of storage are selected for experiments, the experiments are carried out on the basis of a PC platform with an Intel (R) core (TM) i5-4590 CPU @3.30GHz processor, 8.00GB memory and a Windows 1064-bit operating system, and MATLAB R2017b is used as software. The data types used are 4 types, namely signal data (N), inner ring fault (IF), outer ring fault (OF) and Rolling Fault (RF) data under normal conditions, measuring points are all located at a driving end, the applied load is 3 horsepower, the reference rotating speed is 1730r/min, the sampling frequency is 12kHz, and the fault degree is 0.18 mm. To ensure the validity of the experimental data, the length of each sample signal is set to 1024, i.e. the sample signal in two complete vibration cycles is included. The used data comprises training set data and test set data, the training set is used for building a rolling bearing fault diagnosis model, the test set data is used for verifying the rolling bearing fault diagnosis model, and the specific data condition is shown in table 1.
TABLE 1
In the conventional intelligent diagnosis method for mechanical faults, fault classification based on time-frequency domain statistical features combined with SVMs and classification based on Back Propagation Neural Network (BPNN) of single layer and deep layer are most commonly used. Therefore, on the premise of the original vibration signal with the same dimension, the embodiment adopts the diagnosis method provided by the invention, the comparative example 1 is a diagnosis method which adopts 14 time domain statistical characteristic parameters of the fault signal and combines SVM, the comparative example 2 is a deep BPNM diagnosis method, and the comparative example 3 is a single-layer BPNM diagnosis method.
In embodiment 1, firstly, according to mechanical vibration signals of rolling bearings of different types, corresponding double sparse dictionaries are obtained through training, and the influence of various parameters of the double sparse dictionaries on diagnosis precision is evaluated to determine optimal parameter setting; then, further optimizing redundant atoms by using the maximum inner product of the original rolling bearing mechanical vibration signal and the dictionary atoms to form a final double sparse dictionary; then, carrying out sparse decomposition on mechanical vibration signals of all types of rolling bearings under a double-sparse dictionary by adopting an OMP algorithm to obtain sparse representation characteristic vectors; then constructing a deep belief network model, gradually training to obtain relevant parameters of the model, and obtaining a rolling bearing fault diagnosis model capable of identifying different health states; finally, the test set signals are used for classification and identification.
When the double sparse dictionaries are adopted to carry out sparse decomposition on the vibration signals, parameters influencing the classification effect are analyzed through a single factor analysis method: the method comprises the steps of reasonably selecting and determining sparsity of characteristic coefficients in sparse representation, namely, fixing other parameters, only considering the influence of single parameter change on diagnosis precision, analyzing the influence of each type of parameter on the diagnosis precision by taking a classification precision average value of 15 experiments and a standard deviation when each parameter value is changed, and determining each parameter value of the double sparse dictionaries to ensure that an over-complete dictionary with more ideal performance is obtained, and is favorable for fault classification.
The deep belief network needs to pay attention to several parameters when used for sample classification: RBM number, hidden layer neuron number, learning rate, momentum, and network training times. Generally, the smaller the learning rate value is, the better the learning effect is, but the network training time can be increased, the value range is usually 0.01-0.8, the learning rate is 0.01 in the embodiment, the rest parameter values are determined through experimental analysis, the average value of 15 experiments and the corresponding standard deviation of the average value are taken as each experimental result for analysis, and finally, the relevant parameters are determined, so that the rolling bearing fault diagnosis model is obtained.
The results of the 15 experiments were obtained for example 1, comparative example 2 and comparative example 3, respectively, and the average diagnostic accuracy results shown in fig. 3 and the standard deviations corresponding to table 2 were obtained.
TABLE 2
As shown in fig. 3, the results obtained by the method of example 1 all have high diagnosis accuracy, and the curve of the diagnosis result is similar to a straight line, which indicates that the method has good accuracy stability. The method based on comparative example 1 has relatively high diagnosis results, but has poorer diagnosis accuracy and accuracy stability than those of example 1, but has obvious advantages compared with the rest of comparative examples 2 and 3. The method of comparative example 3 has relatively low diagnosis accuracy and obvious fluctuation; the method of comparative example 2 has a somewhat higher diagnostic accuracy than the method of comparative example 3, but with a greater degree of variability.
As can be seen from the data in Table 2, the method of example 1 has the highest average diagnostic accuracy, which is 98.12%, and the method of comparative example 3 has the lowest average diagnostic accuracy, which is only 49.91%. From the aspect of diagnostic stability, the method of example 1 has the highest stability with a standard deviation of 0.33%, the method of comparative example 2 has the worst stability with a standard deviation of diagnostic accuracy as high as 6.99%.
Rolling bearing vibration signals based on different health condition categories have respective characteristics, the method of the embodiment 1 utilizes the characteristic that the rolling bearing vibration signals only have the highest matching degree with double sparse dictionary atoms obtained by self signal training, and further eliminates redundant atoms with low contribution rate when sparse decomposition is carried out in an over-complete dictionary, so that a dictionary model becomes more simplified, the dimension of sparse representation characteristic coefficient of original rolling bearing vibration signals with the length of 1024 under the dictionary is only 80, and because the signals are sparse characteristics represented by sparse characteristics, the sparse characteristic points are only 10, so a small amount of data can represent obvious characteristic information of the signals different from other signals, meanwhile, the method of the embodiment 1 combines a deep belief network with strong information mining capability, and through learning of sparse representation characteristic vectors, higher differential information identification capability is obtained, and thus, higher diagnosis accuracy can be achieved. On the contrary, the statistical characteristics based on the time-frequency domain in the comparative example 1 are difficult to accurately and comprehensively reflect the characteristics of the original signal, so that the average diagnosis accuracy rate is slightly low when the SVM method is used for diagnosis; in the method of the comparative example 2, the network is trained by adopting a back propagation algorithm, so that the whole network is poor in stability, generalization and the like, and the diagnosis accuracy and stability are not high; the method of comparative example 3 has limited ability to learn the features of high-dimensional original signals due to the limitation of the shallow structure, and thus has low diagnostic accuracy.
By integrating the experimental results and analysis, the method of embodiment 1 has greater advantages than the conventional intelligent fault diagnosis method, and can directly realize accurate diagnosis of health conditions by using a sparse representation feature obtained by sparse decomposition of mechanical vibration signals of the rolling bearing under a double sparse dictionary by using a deep belief network.
Further, comparative example 4 in which the failure diagnosis was performed using a deep belief network while directly using the original signal was introduced, and the diagnosis effect was compared with that of the method of example 1.
First, compared with the method for fault diagnosis using a deep belief network when the original signal is directly adopted in comparative example 4, the method of example 1 combines a double-sparse dictionary learning algorithm, and takes the sparse representation feature vector of the original vibration signal under the double-sparse dictionary as the target data of deep belief network learning. By respectively randomly taking a sample from the sample signals of various health states and obtaining the sparse representation feature vector of the sample in the double sparse dictionaries to perform feature classifiability comparison, a separability comparison graph of the original signal and the sparse representation signal as shown in fig. 4 is obtained.
As shown in fig. 4, the sparse representation feature vector obtained by the dual sparse dictionary sparse decomposition has many advantages compared to the original signal. Firstly, compared with an original vibration signal, the feature vector obtained after sparse representation is greatly reduced in dimensionality, the length of an original signal 1024 is changed into 80, and the signal dimensionality is compressed by dozens of times; secondly, the data volume of the signals changes, the original signals have numerical values at each sampling point, but only sparse values are arranged at ten sparse points in the sparse representation characteristic vectors, and the rest positions are all zero, so that the calculation amount during subsequent neural network training and testing is greatly reduced, and the network training and diagnosis efficiency is improved; finally, it can be visually observed from the graph that compared with the disordered form of the original signals, the sparse feature vector of each type of fault signals after sparse decomposition has more unique and obvious features, the sparse point distribution of the sparse feature vector is obviously different from that of other types of signals, and each type of signals is only decomposed by most atoms of a dictionary obtained by training of the same type of signals during sparse decomposition, for example, most of the sparse points of normal signals are distributed between 0 and 20, most of the sparse points of the fault signals in the inner circle are distributed between 20 and 40, and the rest of the signals have the same distribution characteristics by analogy. Therefore, the characteristics of sparse decomposition signals under the dual-sparse dictionary model are beneficial to the aspects of training of the deep belief network and improvement of the recognition efficiency.
Next, the method of example 1 and the method of comparative example 4 were used to perform fault diagnosis to obtain a result comparison chart shown in fig. 5, and 15 experimental average diagnosis results and corresponding standard deviations and training and recognition time data shown in table 3 were obtained.
TABLE 3
As can be seen from fig. 5 and table 3, the diagnosis accuracy of the method of example 1 is higher than that of the method of comparative example 4, and the fault diagnosis accuracy is relatively stable, so that the network training time is greatly reduced while the diagnosis accuracy is ensured; when the method of comparative example 4 is adopted, the diagnosis result takes longer and has larger fluctuation due to the high dimensionality of the original data. Therefore, in the method of the embodiment 1, by means of the sparse decomposition of the proposed double-sparse dictionary model on the original signals, the obtained sparse representation features simplify the feature complexity among different types of signals, thereby greatly improving the diagnosis accuracy and reducing the time for network training and testing.
Claims (4)
1. A rolling bearing fault diagnosis method based on double sparse dictionary sparse representation is characterized by comprising the following steps:
step S1, training the vibration signal of the modeled rolling bearing by adopting a double-sparse dictionary learning method to obtain a double-sparse dictionary;
step S2, obtaining the decomposition coefficient of the mechanical vibration signal of the modeling rolling bearing under the double sparse dictionaries and taking the decomposition coefficient as a feature vector;
step S3, inputting the characteristic vector obtained in the step S2 into a deep belief network for training and learning to obtain a rolling bearing fault diagnosis model;
step S4, inputting the mechanical vibration signal of the rolling bearing to be tested into a rolling bearing fault diagnosis model for fault identification, and completing fault diagnosis;
in the step S1, firstly, a double sparse dictionary learning method is adopted to train the mechanical vibration signal of the modeled rolling bearing to obtain a double sparse sub-dictionary, then, the obtained double sparse sub-dictionary is subjected to double sparse sub-dictionary atom screening optimization, and atoms with low contribution rate are removed to obtain a final double sparse dictionary;
in the step S1, a sparse decomposition algorithm is used to optimize the double-sparse sub-dictionary, firstly, atom screening is performed by using the maximum inner product absolute value of the modeling rolling bearing mechanical vibration signal and the corresponding double-sparse sub-dictionary atoms as a double-sparse sub-dictionary atom screening standard, and then the screened alternative atoms are sequentially arranged and combined in sequence to form a final double-sparse dictionary;
in step S1, the specific process of obtaining the dual sparse dictionary is as follows: step S11, training the mechanical vibration signal of the modeled rolling bearing based on a learning dictionary algorithm to obtain a base dictionary; step S12, solving a sparse coefficient matrix of the mechanical vibration signal of the modeled rolling bearing under the base dictionary by using a sparse decomposition algorithm to serve as a sparse dictionary; step S13, according to the base dictionary and the sparse dictionary, solving through a double-sparse-dictionary learning algorithm to obtain a double-sparse dictionary;
the modeling rolling bearing mechanical vibration signal consists of one or more rolling bearing mechanical vibration signals in different health states; when the modeling rolling bearing mechanical vibration signal is composed of rolling mechanical vibration signals of different health states, in the step S1, training is performed on the rolling bearing mechanical vibration signals of different health states, and corresponding double sparse dictionaries are respectively obtained; in the step S2, respectively obtaining the decomposition coefficients of the mechanical vibration signals of the rolling bearings in different health states in the double sparse dictionaries, and respectively taking the decomposition coefficients as feature vectors; in the step S3, the feature vectors obtained for the mechanical vibration signals of the rolling bearings in different health states are input to the deep belief network for training and learning, and the rolling bearing fault diagnosis models for identifying different health states are obtained.
2. The rolling bearing fault diagnosis method based on the double-sparse-dictionary sparse representation according to claim 1, wherein in the step S11, a K-SVD algorithm is adopted as a learning dictionary algorithm.
3. The rolling bearing fault diagnosis method based on double sparse dictionary sparse representation according to claim 1, wherein OMP algorithm is selected as sparse decomposition algorithm.
4. The rolling bearing fault diagnosis method based on the double-sparse dictionary sparse representation according to claim 1, wherein the mechanical vibration signal OF the modeled rolling bearing is any one or more OF signal data N, inner ring fault signal data IF, outer ring fault signal data OF or rolling fault signal data RF under normal conditions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010155523.9A CN111382792B (en) | 2020-03-09 | 2020-03-09 | Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010155523.9A CN111382792B (en) | 2020-03-09 | 2020-03-09 | Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111382792A CN111382792A (en) | 2020-07-07 |
| CN111382792B true CN111382792B (en) | 2022-06-14 |
Family
ID=71219956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010155523.9A Expired - Fee Related CN111382792B (en) | 2020-03-09 | 2020-03-09 | Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111382792B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115524150B (en) * | 2022-09-13 | 2024-04-05 | 西安交通大学 | Rotary machine fault detection method based on sparse time domain synchronous average |
| CN116558828B (en) * | 2023-07-10 | 2023-09-15 | 昆明理工大学 | Rolling bearing health state assessment method based on autocorrelation coefficient sparsity characteristic |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105139028A (en) * | 2015-08-13 | 2015-12-09 | 西安电子科技大学 | SAR image classification method based on hierarchical sparse filtering convolutional neural network |
| CN108388692A (en) * | 2018-01-17 | 2018-08-10 | 西安交通大学 | Rolling Bearing Fault Character extracting method based on layering sparse coding |
| CN108507789A (en) * | 2018-05-31 | 2018-09-07 | 燕山大学 | The sparse diagnostic method of rolling bearing fault based on the weak orthogonal matching pursuit of mean random |
| CN109060350A (en) * | 2018-09-04 | 2018-12-21 | 重庆大学 | A kind of Rolling Bearing Fault Character extracting method dictionary-based learning |
| CN110114834A (en) * | 2016-11-23 | 2019-08-09 | 通用电气公司 | Deep learning medical system and method for medical procedure |
| CN110348491A (en) * | 2019-06-20 | 2019-10-18 | 燕山大学 | Rolling bearing fault recognition methods based on study dictionary and singular value decomposition |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7552385B2 (en) * | 2001-05-04 | 2009-06-23 | International Business Machines Coporation | Efficient storage mechanism for representing term occurrence in unstructured text documents |
-
2020
- 2020-03-09 CN CN202010155523.9A patent/CN111382792B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105139028A (en) * | 2015-08-13 | 2015-12-09 | 西安电子科技大学 | SAR image classification method based on hierarchical sparse filtering convolutional neural network |
| CN110114834A (en) * | 2016-11-23 | 2019-08-09 | 通用电气公司 | Deep learning medical system and method for medical procedure |
| CN108388692A (en) * | 2018-01-17 | 2018-08-10 | 西安交通大学 | Rolling Bearing Fault Character extracting method based on layering sparse coding |
| CN108507789A (en) * | 2018-05-31 | 2018-09-07 | 燕山大学 | The sparse diagnostic method of rolling bearing fault based on the weak orthogonal matching pursuit of mean random |
| CN109060350A (en) * | 2018-09-04 | 2018-12-21 | 重庆大学 | A kind of Rolling Bearing Fault Character extracting method dictionary-based learning |
| CN110348491A (en) * | 2019-06-20 | 2019-10-18 | 燕山大学 | Rolling bearing fault recognition methods based on study dictionary and singular value decomposition |
Non-Patent Citations (2)
| Title |
|---|
| Sparse Deep Stacking Network for Fault Diagnosis of Motor;C. Sun;《IEEE Transactions on Industrial Informatics》;20180326;第14卷(第07期);3261-3270 * |
| 基于稀疏差异深度信念网络的绝缘子故障识别算法;高强;《电测与仪表》;20160321;第53卷(第01期);19-25 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111382792A (en) | 2020-07-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xu et al. | Imbalanced fault diagnosis of rotating machinery via multi-domain feature extraction and cost-sensitive learning | |
| Huang et al. | Rolling bearing fault diagnosis and performance degradation assessment under variable operation conditions based on nuisance attribute projection | |
| CN107368809B (en) | A kind of bearing fault sorting technique based on rarefaction representation and dictionary learning | |
| CN105275833B (en) | CEEMD (Complementary Empirical Mode Decomposition)-STFT (Short-Time Fourier Transform) time-frequency information entropy and multi-SVM (Support Vector Machine) based fault diagnosis method for centrifugal pump | |
| CN112417954B (en) | Bearing fault mode diagnosis method and system for small sample data set | |
| Wang et al. | A method for rapidly evaluating reliability and predicting remaining useful life using two-dimensional convolutional neural network with signal conversion | |
| CN109827777B (en) | Rolling bearing fault prediction method based on partial least square method extreme learning machine | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN111382792B (en) | Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation | |
| CN112633098B (en) | Rotary machine fault diagnosis method, system and storage medium | |
| CN112577743B (en) | Rolling bearing fault diagnosis method based on maximum local boundary criterion | |
| CN110647911A (en) | Bearing fault diagnosis method based on principal component analysis and deep belief network | |
| CN113052271B (en) | Biological fermentation data prediction method based on deep neural network | |
| de Paula Monteiro et al. | A hybrid prototype selection-based deep learning approach for anomaly detection in industrial machines | |
| Yin et al. | Isomap and Deep Belief Network-Based Machine Health Combined Assessment Model. | |
| Guo et al. | A method of rolling bearing fault diagnose based on double sparse dictionary and deep belief network | |
| CN115375026A (en) | Method for predicting service life of aircraft engine in multiple fault modes | |
| Sadoughi et al. | A deep learning approach for failure prognostics of rolling element bearings | |
| CN115290326A (en) | Rolling bearing fault intelligent diagnosis method | |
| Hu et al. | Fault diagnosis based on multi-scale redefined dimensionless indicators and density peak clustering with geodesic distances | |
| CN114139639A (en) | Fault classification method based on self-walking neighborhood preserving embedding | |
| Jianhui et al. | New fault diagnosis method for rolling bearing based on PCA | |
| Hao et al. | New fusion features convolutional neural network with high generalization ability on rolling bearing fault diagnosis | |
| CN114964776A (en) | Wheel set bearing fault diagnosis method based on MSE and PSO-SVM | |
| CN114186593A (en) | Bearing fault identification method, device, equipment and storage medium under noise condition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220614 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |