CN109034028B - Symbolic and TF-IDF-based mechanical equipment fault feature extraction method - Google Patents

Symbolic and TF-IDF-based mechanical equipment fault feature extraction method Download PDF

Info

Publication number
CN109034028B
CN109034028B CN201810779843.4A CN201810779843A CN109034028B CN 109034028 B CN109034028 B CN 109034028B CN 201810779843 A CN201810779843 A CN 201810779843A CN 109034028 B CN109034028 B CN 109034028B
Authority
CN
China
Prior art keywords
vibration time
idf
vibration
subsequence
sequence
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.)
Active
Application number
CN201810779843.4A
Other languages
Chinese (zh)
Other versions
CN109034028A (en
Inventor
丁承君
冯玉伯
朱雪宏
王鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Taihua Hongye Tianjin Robot Technology Research Institute Co ltd
Hebei University of Technology
Original Assignee
Taihua Hongye Tianjin Robot Technology Research Institute Co ltd
Hebei University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Taihua Hongye Tianjin Robot Technology Research Institute Co ltd, Hebei University of Technology filed Critical Taihua Hongye Tianjin Robot Technology Research Institute Co ltd
Priority to CN201810779843.4A priority Critical patent/CN109034028B/en
Publication of CN109034028A publication Critical patent/CN109034028A/en
Application granted granted Critical
Publication of CN109034028B publication Critical patent/CN109034028B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
    • 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
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering

Abstract

A mechanical equipment fault feature extraction method based on symbolization and TF-IDF comprises the following steps: converting the acquired vibration signal to be diagnosed into a time sequence, and intercepting the time sequence to obtain a subsequence of the vibration time sequence; loading the vibration time subsequence to a normal vibrator sequence set, and symbolizing the loaded vibrator sequence set; and performing differential feature extraction on the symbolized vibrator sequence set by adopting a TF-IDF method. According to the technical scheme, the symbolization and TF-IDF methods are combined to extract the fault features of the mechanical equipment, so that the accuracy and the effectiveness of vibration signal feature extraction in fault diagnosis of the mechanical equipment are improved, the classification capability is good, and the mechanical equipment features can be extracted better and faster.

Description

Symbolic and TF-IDF-based mechanical equipment fault feature extraction method
Technical Field
The invention relates to the technical field of mechanical equipment fault extraction, in particular to a method for extracting mechanical equipment fault characteristics based on symbolization and TF-IDF.
Background
The complex and large-scale of mechanical equipment causes huge economic loss and even casualties once a system fails, so that the fault diagnosis and the abnormal detection of the mechanical equipment are effective means for ensuring the safe, reliable and efficient operation of the equipment. Whether important assemblies in mechanical equipment break down or not directly influences the health condition of the whole machine, so that the damage of parts in the equipment can be found in time by monitoring and detecting the faults of the mechanical equipment, and the failure of the parts and the damage of the machine are avoided. The fault detection of the parts in the equipment can be realized by effectively extracting and analyzing the vibration signal characteristics, and the vibration signals of the parts in the equipment are often mixed with very large noise and have non-stationarity characteristics, so that accurate and effective extraction of the fault characteristic information of the parts is a key point and a difficult point in fault diagnosis of mechanical equipment.
Chinese patent publication No.: CN105738722A discloses a method for diagnosing faults of an electromechanical actuator of an aircraft,
the method comprises the steps of performing symbolization processing on an amplitude time sequence of bus current on the side of an inverter of the electromechanical actuator, taking the calculated information entropy value and the maximum amplitude of the bus current time sequence as two characteristic quantities of fault diagnosis of the electromechanical actuator, then establishing a standard fault characteristic matrix based on a training sample, and judging the fault type of the sample to be detected by calculating the distance between the sample to be detected and the fault characteristic matrix by adopting an improved Parks clustering algorithm. It can be seen that the diagnostic method has the following problems:
firstly, the diagnosis method needs to sample 6 states of the electromechanical actuator respectively, at least 50 groups of bus current time needs to be collected in each running state, fault diagnosis characteristic quantities need to be extracted from the bus current time respectively after the bus current time is measured, the sampling engineering quantity is large, and the consumed time is large.
Secondly, the diagnostic method uses the maximum amplitude of the bus current time sequence and the characteristic quantity of the sign dynamics information entropy seat fault diagnosis, the calculation amount is large, and the data extraction efficiency of the diagnostic method is reduced.
Thirdly, the diagnostic method changes the essential characteristics of the electromechanical actuator during sampling, so that various parameters of the electromechanical actuator are changed, the acquired data do not belong to the same condition, and the fault data extraction efficiency is low.
Disclosure of Invention
Therefore, the invention provides a mechanical equipment fault feature extraction method based on symbolization and TF-IDF, which is used for overcoming the problem of low part fault feature extraction efficiency in the prior art.
In order to achieve the above object, the present invention provides a method for extracting fault characteristics of mechanical equipment based on symbolization and TF-IDF, comprising:
converting the acquired vibration signal to be diagnosed into a vibration time sequence, intercepting the vibration time sequence, and obtaining a vibration time subsequence after interception;
loading the vibration time subsequence to a normal vibration time subsequence set, and symbolizing the loaded vibration time subsequence set;
and performing differential feature extraction on the symbolized vibration time subsequence set by adopting a TF-IDF method.
Further, when the vibration signal to be diagnosed is converted into a vibration time sequence, firstly, the vibration displacement, the speed and the acceleration in the vibration signal to be diagnosed pass through corresponding sensors and are collected according to the time sequence to obtain a vibration time sequence S; after the acquisition is finished, a sliding window with a fixed size is selected to intercept the vibration time sequence S so as to form a vibration time subsequence
Figure BDA0001732289940000022
Further, the vibration time series S is: when measuring the corresponding physical quantity of the vibration signal, the vibration signal is arranged in a time sequence, and S ═ S (S ═ S)1,s2,…,sn) (1), wherein n is the length of the vibration time series S, and n is a positive integer; the vibration time sequence subsequence
Figure BDA0001732289940000023
Comprises the following steps: a subsequence of length k in said vibration time series S,
Figure BDA0001732289940000021
wherein j is more than or equal to 1 and less than or equal to n-k + 1.
Further, the length of the sliding window is k, and the step length is m, where k and m are positive integers.
Further, when symbolizing the vibration time series, the method includes:
discretizing data in the vibration time sequence to form a small number of numerical values;
after discretization, a particle method is used for describing a nonlinear non-stationary system, and the essential characteristics of periodicity, correlation and the like in the system are kept stable while description is carried out;
after the description is finished, the vibration time sequence is symbolized by using an equal interval method, and each symbol is called a word.
Further, when the vibration time series is coded by the equal interval method, the vibration time series S with the length n is calculated (S) first1,s2,…,sn) Minimum and mean value SminAnd maximum value SmaxAnd calculating the value range [ S ] of the original vibration time sequencemin,Smax]；
After the calculation is finished, dividing the value range into k intervals with equal size according to a formula (3),
Figure BDA0001732289940000031
wherein C isiA "word" that is the vibration time series;
after the processing of formula (3), each value in the vibration time series corresponds to a word, so that the vibration time series is converted into a symbol time series.
Furthermore, the TF-IDF is a weighting technique commonly used in the consultative search; the method for providing the vibration time sequence TF-IDF according to the thought of the TF-IDF method comprises the following steps: when some of the signed values in the set of vibration time subsequences occur with high frequency in one vibration time subsequence (high TF value) and rarely occur in the other vibration time subsequences (low IDF value), the signed values are considered to represent the typical characteristics of the sequence, and the signed vibration time sequences are classified according to the characteristic.
Further, the word frequency TF of the vibration time series is the number of times that a given word is in the tokenized subsequence, a certain word C in the tokenized subsequenceiThe importance of can be expressed as:
Figure BDA0001732289940000032
wherein n isi,jIs the word CiIn the subsequence djAnd the denominator represents the subsequence djThe sum of the occurrence times of all the words in the list;
the inverse file frequency IDF of the vibration time sequence is a word CiThe measure of general importance, the IDF of a particular word, can be obtained by dividing the total number of signed subsequences by the number of subsequences containing the word and taking the logarithm of the result:
Figure BDA0001732289940000033
where | D | is the total number of subsequences in the time series.
Further, after the TF-IDF method is adopted for differential feature extraction, Euclidean distance is adopted for sequence feature similarity comparison of the extracted features, and if the Euclidean distance numerical value of the fault subsequence to be diagnosed is separated from the normal vibration time subsequence TF-IDF feature value, the feature extracted by TF-IDF is effective.
Further, after the TF-IDF method is adopted for differential feature extraction, Principal Component Analysis (PCA) is adopted for extracting the extracted features, and a KMeans method is used for clustering, if the intra-class sample distance and the inter-class sample distance in different fault subsequences to be diagnosed are relatively small, and normal vibration data are more compact relative to fault data, the TF-IDF extracted features are proved to be effective.
Compared with the prior art, the method has the advantages that symbolization and TF-IDF are combined in the technical scheme of the method for extracting the fault features of the mechanical equipment, so that the accuracy and effectiveness of extracting the vibration signal features in fault diagnosis of the mechanical equipment are improved, the classification capability is good, and the fault features of the mechanical equipment can be extracted better and faster; and the method uses an unsupervised learning method to solve the problem that the fault data with the label in the supervised learning method is difficult to obtain.
Furthermore, the vibration signals are collected by using the corresponding sensors, the time sequence S is obtained according to the sequence, the complex signals are converted into data, preparation can be made for subsequent symbolization, and the extraction efficiency of the extraction method on the fault features of the parts is improved.
Furthermore, the invention extracts a plurality of vibration time subsequences from the vibration time sequence S
Figure BDA0001732289940000041
By extracting a plurality of vibration time subsequences, preparation is made for subsequent symbolization and extraction of fault features, and the extraction efficiency of the extraction method for the fault features of the parts is further improved.
Furthermore, the sliding window for intercepting the vibration time subsequences is of a fixed size, so that the intercepted vibration time subsequences are all the same in length and step length, the uniformity of the symbols can be guaranteed after symbolization, the data can be displayed more visually in subsequent word extraction, and the extraction efficiency of the extraction method for the fault features of the parts is further improved.
Furthermore, when the TF-IDF method is used for extracting words from the symbolic time subsequence, the data in the time subsequence is discretized, the particle method is used for describing the system, and the intrinsic characteristics such as periodicity, relevance and the like in the system are not changed during description, so that the intrinsic characteristics in the system are not changed due to operation during description of the system, the time subsequence after description is the data under the same characteristic, the stability of the system and the data is maintained, and the extraction efficiency of the extraction method for the fault characteristics of the parts is further improved.
Furthermore, the value range of the vibration time sequence is calculated and divided into a plurality of intervals with equal size, so that the vibration time sequence is converted into the symbol time sequence, the TF-IDF method can be used for extracting the fault characteristics of the symbol time sequence, and the extraction efficiency of the extraction method on the fault characteristics of the parts is further improved.
Furthermore, the TF value and the IDF value of the symbolic vibration time sequence are respectively calculated by using corresponding formulas, the symbolic numerical value of the typical feature in the symbolic vibration time subsequence is extracted according to the two parameters, and the vibration time sequence is subjected to fault classification according to the symbolic numerical value, so that the extraction efficiency of the extraction method on the fault feature of the part is further improved.
Drawings
FIG. 1 is a flow chart of the present invention for on-line monitoring of mechanical equipment failure based on symbolization and TF-IDF;
FIG. 2 is a flow chart of the fault diagnosis of mechanical equipment based on symbolization and TF-IDF according to the present invention;
FIG. 3 is a schematic view of an experimental apparatus for a rolling bearing according to an embodiment of the present invention;
FIG. 4 is a diagram of the sequence of the original vibrator and the sequence of the symbolic vibrator in the experimental apparatus according to the embodiment of the present invention;
FIG. 5 is a TF-IDF distribution diagram of a partial subsequence of a symbolic vibration sequence in an experimental apparatus according to an embodiment of the present invention;
FIG. 6 is a Euclidean distance between a fault subsequence at different positions and a reference subsequence in the experimental apparatus according to the embodiment of the present invention;
FIG. 7 is a diagram illustrating symbolic subsequence clustering analysis in the experimental apparatus according to an embodiment of the present invention.
Detailed Description
In order that the objects and advantages of the invention will be more clearly understood, the invention is further described below with reference to examples; it will be understood by those skilled in the art that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Fig. 1 shows a process of online monitoring of a fault of a mechanical device according to the present invention, in which a sequence 1 is a vibration sequence to be diagnosed, and a sequence 2 is a normal vibration sequence set. When in diagnosis, a vibrator sequence to be diagnosed (usually a vibration signal segment monitored by a vibration sensor) is added into a normal vibrator sequence set, the vibrator sequence set is symbolized, differential feature extraction is carried out by a vibration sequence TF-IDF method, sequence feature similarity comparison is carried out on extracted features by adopting Euclidean distances, and if the Euclidean distance numerical values of the feature values of different fault sub-sequences to be diagnosed and the normal vibrator sequence TF-IDF have separability, the extracted features of the TF-IDF are effective. The method is mainly used for judging whether mechanical equipment has faults or not on line, adding fault sequences into vibrator sequence sets with different types of faults and starting a fault diagnosis process.
Fig. 2 is a flow chart of fault diagnosis of mechanical equipment according to the present invention, in which a sequence 3 is a vibration sequence to be diagnosed, and a sequence 4 is a set of different types of vibration sequences. Adding fault subsequences to be diagnosed into fault vibrator sequence sets of different types, performing feature extraction on the vibrator sequence sets through symbolization and TF-IDF methods, and finally performing category attribution by using PCA principal components and a Kmeans clustering method to verify the effectiveness of feature extraction. If the clustering method can perform good clustering through the TF-IDF characteristics, the extracted characteristics can distinguish different types of bearing faults.
Example 1
Fig. 3 is a schematic structural diagram of an experimental apparatus for a rolling bearing according to an embodiment of the present invention, and includes an induction motor 1, a motor fan end 2, a motor driving end 3, a vibration sensor 4, a driving end bearing 5, a coupling 6, a torque sensor 7, and a load cell 8. In the embodiment, a bearing vibration test open data set provided by Kaiser university of West university of America is selected as the reference vibrator sequence set; the data set is a reference data set of a new method for testing and verifying the equipment state and fault, and is acquired based on the flow shown in fig. 2.
Specifically, the experimental device comprises a three-way induction motor 1, a dynamometer 8 capable of generating rated load is arranged on the right side of the motor 1, and the three-way induction motor and the dynamometer are respectively connected with a coupling 6 and a coupling 6The torque sensor 7 is connected, the motor driving end 3 is provided with a bearing 5 to be tested, and the vibration sensor 4 is arranged above the motor driving end 3. The SKF6205-2RSJEM type bearing is adopted in the experiment, single-point defects are respectively manufactured on the inner bearing, the outer bearing and the rolling body in an electric spark machining mode, and the defect size is 1.75 multiplied by 10-4And m is selected. Load 0kw, pick-up frequency 12 kHz. The 3 types of failure data and 1 type of normal data are divided into sub-sequences with the length of 4096, 25 sub-sequences are taken from each type of data, and the obtained original vibrator sequence and the obtained symbolic vibrator sequence are shown in fig. 4.
It will be appreciated that in practical applications, the subsequence length and number need to be determined according to the computing power of the device, but the two values generally tend to be larger values to satisfy the theorem of large numbers, so that the statistical characteristics are stable.
Fig. 4 is a diagram showing the sequence of the vibrator and the sequence of the symbolic vibrator in the normal state and the occurrence of defects of the bearing in the experimental apparatus according to the embodiment of the present invention. For defect size of 1.75 × 10-4And m, selecting normal vibration data and defect vibration data of the inner bearing, the rolling body and the outer bearing. The four vibration time series were subjected to the symbolization processing, and 64 symbol types were set, and the result is shown in fig. 4. The left side of the figure is the original vibrator time sequence diagram, and the right side is the symbolic vibrator sequence diagram. Compared with the original vibrator sequence, the symbolized vibrator sequence is more granular, simultaneously filters out small changes in the original sequence, and keeps the characteristics of periodicity, correlation and the like in the original sequence.
It is understood that in practical use, the symbol type in each symbolic vibration time series needs to be adjusted according to experience or algorithm results.
FIG. 5 is a TF-IDF distribution diagram of a partial subsequence of a symbolic vibration sequence of the experimental apparatus according to an embodiment of the present invention. Processing the symbolic vibrator sequence set through a TF-IDF algorithm, and extracting one of the sub-sequences of each fault type to display the distribution condition of TF-IDF values of the sub-sequences;
specifically, the abscissa in the figure represents the symbol type, each type corresponds to a numerical value, and the ordinate represents the TF-IDF value of each symbol. According to the graph, the TF-IDF values of the four states approximately follow a normal distribution, wherein the TF-IDF distribution division of the normal state A of the rolling bearing and the TF-IDF distribution division of the rolling element fault C are not obvious, but show relatively obvious difference in a symbol interval of 0-20; and the peak values of the TF-IDF distribution and the central point positions of the inner bearing fault B and the outer bearing fault D are obviously different.
Please refer to fig. 6, which shows the euclidean distances between the features of the reference subsequence and the fault subsequence at different positions in the experimental apparatus according to the embodiment of the present invention. After the vibrator sequence when the bearing is defective is processed according to the fault monitoring processing flow method, the distance between the vibrator sequence and the reference subsequence is measured by using the Euclidean distance with any normal state vibrator sequence as the reference, and the result is shown in FIG. 6; it can be seen from the figure that the distance between the vibrator sequences of the same kind of defect and the reference subsequence feature is basically in a straight line state, and the vibrator sequences of different fault positions and the reference subsequence feature are not intersected with each other and keep a certain distance interval.
According to the method, the four types of data are subjected to statistical processing, so that the extracted TF-IDF characteristics can be used for online monitoring of bearing faults; in the actual online monitoring of the bearing fault, a corresponding threshold value can be set according to the principle, and whether the bearing has the fault or not is judged.
It is to be understood that the first point of the normal state normal in fig. 6 is the reference state itself, so its euclidean distance is zero and is not considered in the analysis.
Please refer to fig. 7, which illustrates a cluster analysis of the symbolic subsequences in the experimental apparatus according to the embodiment of the present invention. For the test data, symbolization processing, TF-IDF feature extraction and PCA principal component extraction (in the embodiment, PCA extracts only principal components of 2 dimensions) are performed on the test data, and the data are clustered by using a KMeans method. It can be seen from the figure that the four types of data can be well clustered by using the fault diagnosis process, the intra-class sample distance is relatively small, the inter-class sample distance is large, and the normal vibration data is more compact than the fault data. According to the result of the cluster analysis, the TF-IDF feature extraction method can be effectively applied to fault detection of the experimental device.
Example 2
The experimental device is the link mechanism in this embodiment, and wherein the quarter butt is connected through the connecting rod with the stock, just the connecting rod respectively with stock and quarter butt are articulated, the quarter butt with the junction of connecting piece is equipped with the belt, the belt links to each other with the motor.
Before the experiment, the link mechanism is operated, data acquisition is carried out on the link mechanism by using a corresponding sensor to obtain a vibration time sequence and a vibration time subsequence, and the subsequence is symbolized and calculated to obtain an initial symbolized vibration time subsequence.
In the experiment, single-point defects are respectively manufactured at the belt connecting part of the short rod and the junction of the short rod main body and the short rod and the connecting rod in an electric spark machining mode, and the defect size is 1.3 multiplied by 10-4And m is selected. Load 0kw, pick-up frequency 12 kHz. And (3) dividing the 3 types of fault data and the 1 type of normal data into subsequences with the length of 6144, and obtaining an original vibrator sequence and a symbolic vibrator sequence by taking 40 subsequences of each type of data.
For defect size of 1.3 × 10-4And m, selecting normal vibration data and defect vibration data of the belt end, the rod body and the hinged end. The four vibration time series are symbolized, and 80 symbol types are set. Compared with the original vibrator sequence, the symbolized vibrator sequence is more granular, simultaneously filters out small changes in the original sequence, and keeps the characteristics of periodicity, correlation and the like in the original sequence.
Processing the symbolic vibrator sequence set through a TF-IDF algorithm, and extracting one of the subsequences of each fault type to display the distribution condition of the TF-IDF value; it can be concluded that the TF-IDF values in the four states follow approximately normal distributions, and that the peak values of the TF-IDF distributions and the center point positions of the belt end and hinged end faults are significantly different.
Processing the vibrator sequence when the short rod is defective according to a fault monitoring processing flow method, and measuring the distance between the vibrator sequence and a reference subsequence by using an Euclidean distance on the basis of any normal state vibrator sequence; the distance between the vibrator sequences of the same type of defects and the reference subsequence feature is basically in a linear state, and the vibrator sequences at different fault positions and the reference subsequence feature are not intersected with each other and keep a certain distance interval.
According to the method, the four types of data are subjected to statistical processing, so that the extracted TF-IDF characteristics can be used for online monitoring of the faults of the link mechanism; in the actual online fault monitoring, a corresponding threshold value can be set according to the principle, and whether each rod of the two-rod mechanism has a fault or not is judged.
For the test data, symbolization processing, TF-IDF feature extraction and PCA principal component extraction (in the embodiment, PCA extracts only principal components of 2 dimensions) are performed on the test data, and the data are clustered by using a KMeans method. The fault diagnosis process can well cluster the four types of data, and the normal vibration data is more compact relative to the fault data. According to the result of the cluster analysis, the TF-IDF feature extraction method can be effectively applied to fault detection of the link mechanism.
So far, the technical solutions of the present invention have been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of the present invention is obviously not limited to these specific embodiments. Equivalent changes or substitutions of related technical features can be made by those skilled in the art without departing from the principle of the invention, and the technical scheme after the changes or substitutions can fall into the protection scope of the invention.

Claims (8)

1. A mechanical equipment fault feature extraction method based on symbolization and TF-IDF is characterized by comprising the following steps:
converting the acquired vibration signal to be diagnosed into a vibration time sequence, intercepting the vibration time sequence, and obtaining a vibration time subsequence after interception;
loading the vibration time subsequence to a normal vibration time subsequence set, and symbolizing the loaded vibration time subsequence set;
carrying out differential feature extraction on the signed vibration time subsequence set by adopting a TF-IDF method;
when the vibration time series is symbolized, the method comprises the following steps:
discretizing data in the vibration time sequence to form a small number of numerical values;
after discretization, describing a nonlinear non-stationary system by using a particle method, and keeping the essential characteristics of periodicity and correlation in the system stable while describing;
after the description is finished, symbolizing the vibration time sequence by using an equal interval method, and calling each symbol as a word;
when the vibration time series is symbolized by using an equal interval method, the vibration time series S with the length n is firstly calculated as (S)1，s2，…，sn) Minimum and mean value SminAnd maximum value SmaxAnd calculating the value range [ S ] of the original vibration time sequencemin，Smax]；
After the calculation is finished, dividing the value range into k intervals with equal size according to a formula (3),
Figure FDA0003070276750000011
wherein C isiA "word" that is the vibration time series;
after the processing of formula (3), each value in the vibration time series corresponds to a word, so that the vibration time series is converted into a symbol time series.
2. The mechanical device of claim 1 based on symbolization and TF-IDFThe barrier feature extraction method is characterized in that when a vibration signal to be diagnosed is converted into a vibration time sequence, firstly, vibration displacement, speed and acceleration in the vibration signal to be diagnosed pass through corresponding sensors and are collected according to the time sequence to obtain a vibration time sequence S; after the acquisition is finished, a sliding window with a fixed size is selected to intercept the vibration time sequence S so as to form a vibration time subsequence
Figure FDA0003070276750000012
3. The method of claim 2, wherein the vibration time sequence S is: when measuring the corresponding physical quantity of the vibration signal, the vibration signal is arranged in a time sequence, and S ═ S (S ═ S)1，s2，…，sn) (1), wherein n is the length of the vibration time series S, and n is a positive integer; the vibration time sequence subsequence
Figure FDA0003070276750000021
Comprises the following steps: a subsequence of length k in said vibration time series S,
Figure FDA0003070276750000022
wherein j is more than or equal to 1 and less than or equal to n-k + 1.
4. The method of claim 2, wherein the sliding window has a length of k and a step size of m, where k and m are positive integers.
5. The method for extracting fault characteristics of mechanical equipment based on symbolization and TF-IDF as claimed in claim 1, wherein said TF-IDF is a weighting technique commonly used in consultancy; the method for providing the vibration time sequence TF-IDF according to the thought of the TF-IDF method comprises the following steps: when some of the signed values in the set of vibration time subsequences occur with high frequency in one vibration time subsequence and rarely occur in other vibration time subsequences, the signed values are considered to represent the typical characteristics of the sequences, and the signed vibration time sequences are classified according to the characteristic characteristics.
6. The method of claim 5, wherein the frequency of the vibration time sequence (TF) is the number of times that a given word is in a sequence of symbols (A, C, or a combination thereofiThe importance of can be expressed as:
Figure FDA0003070276750000023
wherein n isi，jIs the word CiIn the subsequence djAnd the denominator represents the subsequence djThe sum of the occurrence times of all the words in the list;
the inverse file frequency IDF of the vibration time sequence is a word CiThe measure of general importance, the IDF of a particular word, can be obtained by dividing the total number of signed subsequences by the number of subsequences containing the word and taking the logarithm of the result:
Figure FDA0003070276750000024
where | D | is the total number of subsequences in the time series.
7. The method for extracting fault characteristics of mechanical equipment based on symbolization and TF-IDF as claimed in claim 6, wherein after the TF-IDF method is used for differential characteristic extraction, Euclidean distance is used for sequence characteristic similarity comparison of the extracted characteristics, and if the fault subsequence to be diagnosed has separation from the Euclidean distance value of the TF-IDF characteristic value of the normal vibration time subsequence, the extracted characteristics of TF-IDF are valid.
8. The method as claimed in claim 6, wherein after the TF-IDF method is used for differential feature extraction, the PCA principal component extraction is used for the extracted features and the KMeans method is used for clustering.
CN201810779843.4A 2018-07-16 2018-07-16 Symbolic and TF-IDF-based mechanical equipment fault feature extraction method Active CN109034028B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810779843.4A CN109034028B (en) 2018-07-16 2018-07-16 Symbolic and TF-IDF-based mechanical equipment fault feature extraction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810779843.4A CN109034028B (en) 2018-07-16 2018-07-16 Symbolic and TF-IDF-based mechanical equipment fault feature extraction method

Publications (2)

Publication Number Publication Date
CN109034028A CN109034028A (en) 2018-12-18
CN109034028B true CN109034028B (en) 2021-09-24

Family

ID=64643283

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810779843.4A Active CN109034028B (en) 2018-07-16 2018-07-16 Symbolic and TF-IDF-based mechanical equipment fault feature extraction method

Country Status (1)

Country Link
CN (1) CN109034028B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113959555B (en) * 2020-07-20 2024-03-26 中国石油天然气股份有限公司 Method and device for intercepting vibration signal
CN116975545A (en) * 2023-07-31 2023-10-31 长沙穗城轨道交通有限公司 Platform door abnormality detection method and device, electronic equipment and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103528836A (en) * 2013-09-22 2014-01-22 重庆科技学院 Rotary machine fault diagnosis method based on symbolic dynamics disable word mode
CN104931263A (en) * 2015-06-18 2015-09-23 东南大学 Bearing fault diagnosis method based on symbolic probabilistic finite state machine
CN105004523A (en) * 2015-08-04 2015-10-28 潍坊学院 Rolling bearing state monitoring method based on weighted similarity measure

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130297256A1 (en) * 2012-05-04 2013-11-07 Jun Yang Method and System for Predictive and Conditional Fault Detection

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103528836A (en) * 2013-09-22 2014-01-22 重庆科技学院 Rotary machine fault diagnosis method based on symbolic dynamics disable word mode
CN104931263A (en) * 2015-06-18 2015-09-23 东南大学 Bearing fault diagnosis method based on symbolic probabilistic finite state machine
CN105004523A (en) * 2015-08-04 2015-10-28 潍坊学院 Rolling bearing state monitoring method based on weighted similarity measure

Also Published As

Publication number Publication date
CN109034028A (en) 2018-12-18

Similar Documents

Publication Publication Date Title
CN111562108A (en) Rolling bearing intelligent fault diagnosis method based on CNN and FCMC
CN110823576B (en) Mechanical anomaly detection method based on generation of countermeasure network
CN108956111B (en) Abnormal state detection method and detection system for mechanical part
GB2476246A (en) Diagnosing an operation mode of a machine
CN108106846B (en) A kind of rolling bearing fault damage extent identification method
CN116756595B (en) Conductive slip ring fault data acquisition and monitoring method
CN109034028B (en) Symbolic and TF-IDF-based mechanical equipment fault feature extraction method
CN106597160B (en) Electronic equipment fault detection method and device
CN111678699B (en) Early fault monitoring and diagnosing method and system for rolling bearing
CN103389430A (en) Oil-immersed type transformer fault detection method based on Bayesian discrimination theory
CN111964909A (en) Rolling bearing operation state detection method, fault diagnosis method and system
Yuan et al. Early stage internal short circuit fault diagnosis for lithium-ion batteries based on local-outlier detection
CN110108474B (en) Online monitoring and evaluating method and system for operation stability of rotary machine
CN108982106B (en) Effective method for rapidly detecting kinetic mutation of complex system
Ginart et al. Automated feature selection for embeddable prognostic and health monitoring (PHM) architectures
CN113051092B (en) Fault diagnosis method based on optimized kernel density estimation and JS divergence
CN114739671A (en) Bearing fault diagnosis method based on improved generalized S transformation
CN113933054A (en) Rolling bearing fault monitoring and diagnosing method, equipment, medium and processor
CN116685967A (en) Method and device for detecting anomalies in a machine or machine component
Zhang et al. Applied sensor fault detection and validation using transposed input data PCA and ANNs
Lv et al. A weight recognition method for movable objects in sealed cavity based on supervised learning
CN112352165B (en) Method for monitoring circuit breaker, device using method and Internet of things
Priva et al. Bearing health condition monitoring using time-domain acoustic signal features
Lim et al. Motor fault detection method for vibration signal using FFT residuals
Irgat et al. Comparative Evaluation of Machine Learning Methods for the Detection of the Eccentricity Faults of Induction Motors by Using Vibration Signals

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
CB03 Change of inventor or designer information

Inventor after: Ding Chengjun

Inventor after: Zhu Xuehong

Inventor after: Feng Yubo

Inventor after: Wang Xin

Inventor before: Ding Chengjun

Inventor before: Feng Yubo

Inventor before: Zhu Xuehong

Inventor before: Wang Xin

CB03 Change of inventor or designer information