CN110967184B - Gearbox fault detection method and system based on vibration signal distribution characteristic recognition - Google Patents
Gearbox fault detection method and system based on vibration signal distribution characteristic recognition Download PDFInfo
- Publication number
- CN110967184B CN110967184B CN201911221599.0A CN201911221599A CN110967184B CN 110967184 B CN110967184 B CN 110967184B CN 201911221599 A CN201911221599 A CN 201911221599A CN 110967184 B CN110967184 B CN 110967184B
- Authority
- CN
- China
- Prior art keywords
- data set
- data
- distribution
- gearbox
- fault detection
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
- G01M13/02—Gearings; Transmission mechanisms
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M17/00—Testing of vehicles
- G01M17/007—Wheeled or endless-tracked vehicles
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention provides a gearbox fault detection method and system based on vibration signal distribution characteristic identification, and particularly relates to the technical field of data pattern identification. The method can comprise the following steps: collecting vibration signals generated by a gearbox in real time in the driving process of an automobile, and establishing a data set to be detected; constructing a fault detection model based on the established data set to be detected and a preset sample data set reflecting the normal working state of the gearbox; based on the constructed fault detection model, fitting and analyzing the distribution structure of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement, and carrying out fault detection on the gearbox by identifying abnormal data patterns in the data set to be detected; and outputting a diagnosis result of the fault type of the gearbox based on the identification of the abnormal data mode distribution characteristics. The scheme provided by the invention can quickly detect the real-time fault of the gearbox in the working state, and has stronger robustness in the complex working environment influenced by multiple vibration sources.
Description
Technical Field
The invention relates to the technical field of data pattern recognition, in particular to a gearbox fault detection method and system based on vibration signal distribution characteristic recognition.
Background
The safety problem of the automobile, which is one of the most popular vehicles at present, has been a hot issue concerned in various subject fields. The gearbox is used as a main kinetic energy transmission device of the automobile, the engine is ensured to be in the optimal power performance state by coordinating the rotating speed of the engine and the actual running speed of wheels, and the stability of the running condition of the gearbox directly influences the normal running of the whole automobile. Therefore, how to accurately identify the working state of the gearbox and find the abnormality in time is a key research for ensuring the safe driving of the automobile.
The equipment fault diagnosis technology is an emerging science which is used for acquiring related analysis data, determining fault reasons and improving corresponding solutions by monitoring the running state of equipment. At present, with the development of artificial intelligence technology, the research direction of fault detection has been changed to an intelligent computing technology based on machine learning, fuzzy logic, deep learning, image recognition and the like, and a data-driven intelligent diagnosis system is established. The data-driven-oriented fault diagnosis method can be divided into frequent pattern mining based on association rules and rare pattern mining based on anomaly detection. Data-driven fault diagnosis research is mainly focused on semi-supervised anomaly detection methods that detect unknown fault patterns through limited normal sample data tags.
One common limitation of existing detection methods is that they have difficulty detecting anomalies that occur in real time in the same background as normal data. Namely, how to monitor faults generated in real time sharply and accurately in the continuous operation of the gearbox. The data that make up these abnormal patterns are normal in nature, but are abnormal when they appear together as a set. Instead, it may erroneously identify some normal data that falls within the low probability density range as an anomaly. Such abnormal patterns are difficult to find if only individual data of unlabeled samples are examined one by one.
Disclosure of Invention
Technical problem to be solved
Aiming at the defects of the prior art, the invention provides a gearbox fault detection method and system based on vibration signal distribution characteristic identification, and solves the problem of monitoring gearbox fault during real-time running of a vehicle.
(II) technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
according to one aspect of the application, a gearbox fault detection method based on vibration signal distribution characteristic identification is provided, and comprises the following steps:
collecting vibration signals generated by a gearbox in real time in the driving process of an automobile, and establishing a data set to be detected; the data set to be tested comprises normal vibration data generated under the normal operation state of the gearbox and abnormal vibration data caused by faults;
constructing a fault detection model based on the data set to be detected and a preset sample data set reflecting the normal working state of the gearbox;
fitting and analyzing the distribution structure of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement for the constructed fault detection model, and carrying out fault detection on the gearbox by identifying an abnormal data mode in the data set to be detected;
and outputting the detection result of the fault type of the gearbox based on the identification of the abnormal data pattern distribution characteristics.
Optionally, before acquiring a vibration signal generated by the transmission in real time during the driving of the vehicle and establishing the data set to be measured, the method further includes:
collecting normal vibration signals generated under the normal operation state of the gearbox, and adding a label to a data set formed by the normal vibration signals;
and establishing a normal sample data set by a random sampling method based on a data set consisting of the normal vibration signals added with the labels.
Optionally, constructing a fault detection model based on the data set to be detected and a preset sample data set reflecting a normal operating state of the transmission, including:
fitting the data distribution of the sample data set by using a mixed multivariate Gaussian distribution function to obtain a first distribution function of the sample data set, and estimating a first parameter in the first distribution function based on a maximum likelihood algorithm;
fitting the data distribution of the data set to be detected by using a mixed multivariate Gaussian distribution function to obtain a second distribution function of the data set to be detected, and estimating a second parameter in the second distribution function based on a maximum likelihood algorithm;
constructing the fault detection model based on the first distribution function, the first parameter, the second distribution function and the second parameter; the fault detection model is a mixed distribution function model.
Optionally, before constructing the fault detection model based on the first distribution function, the first parameter, the second distribution function, and the second parameter, the method further includes:
and carrying out iterative processing on the maximum likelihood equation based on an immobile point iterative method and solving the first parameter and the second parameter.
Optionally, fitting a data distribution of the sample data set by a first formula;
the first formula is:
D(Ss)＝fs(x|θs)
wherein f issRepresenting a first distribution function; thetasRepresenting a first parameter; d (S)s) Representing the data distribution of the sample data set.
Optionally, fitting the data distribution of the data set to be tested by a second formula;
the second formula is:
D(St)＝F(x|θt)
wherein F represents a second distribution function; thetatRepresents a second parameter; d (S)t) Representing the data distribution of the data set under test.
Optionally, constructing the fault detection model by a third formula;
the third formula is:
wherein F represents a second distribution function; f. ofaA third distribution function representing an abnormal data pattern; f. ofsRepresenting a first distribution function;
representing the proportion of abnormal data; thetasRepresenting a first parameter; thetatRepresents a second parameter; thetaaA third parameter representing a third distribution function.
Optionally, for the constructed fault detection model, fitting and analyzing the distribution structure of the sample data set and the data set to be detected by using the data distribution characteristics as the metric, and performing fault detection on the transmission by identifying an abnormal data pattern in the data set to be detected includes:
and fitting and analyzing the distribution structure of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement for the constructed fault detection model, obtaining a third distribution function of the abnormal data mode according to a third formula, and identifying the abnormal data mode in the data set to be detected based on the third distribution function to carry out fault detection on the gearbox.
According to another aspect of the application, a gearbox fault detection system based on vibration signal distribution characteristic identification is provided, and comprises:
the data set acquisition module to be detected is configured to acquire vibration signals generated by the gearbox in real time in the driving process of the automobile and establish a data set to be detected; the data set to be tested comprises normal vibration data generated under the normal operation state of the gearbox and abnormal vibration data caused by faults;
the model building module is configured to build a fault detection model based on the data set to be tested and a preset sample data set reflecting the normal working state of the gearbox;
the abnormal vibration data identification module is configured to fit and analyze the distribution structure of the sample data set and the data set to be detected by taking data distribution characteristics as measurement for the constructed fault detection model, and carry out fault detection on the gearbox by identifying an abnormal data mode in the data set to be detected;
and the identification result output module is configured to output the detection result of the gearbox fault type based on the identification of the abnormal data pattern distribution characteristics.
(III) advantageous effects
The invention provides a gearbox fault detection method and system based on vibration signal distribution characteristic identification. Compared with the prior art, the method has the following beneficial effects:
1. when the gearbox continuously works, faults generated in real time can be quickly and effectively detected;
2. under the complex working environment influenced by multiple vibration sources, the abnormity of the gearbox can be accurately identified, and the robustness is strong.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
FIG. 1 is a flow chart of a gearbox fault detection method based on vibration signal distribution characteristic identification according to an embodiment of the application;
FIG. 2 is a schematic of a WLY CVT25 continuously variable transmission;
FIG. 3 is a schematic diagram of a transmission test component according to an embodiment of the invention;
FIG. 4 is a schematic structural diagram of a gearbox fault detection system based on vibration signal distribution characteristic identification according to an embodiment of the application;
FIG. 5 is a schematic structural diagram of a gearbox fault detection system based on vibration signal distribution characteristic identification according to another embodiment of the application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described, and it is obvious that the described embodiments are a part of the embodiments of the present invention, but not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the application provides the gearbox fault detection method and system based on vibration signal distribution characteristic identification, so that the working state of the gearbox can be sensitively and accurately monitored and abnormity can be timely found in the continuous working process of the gearbox, the problem of monitoring the gearbox fault in the real-time running of a vehicle is solved, and the safe running of the vehicle is ensured.
In order to solve the technical problems, the general idea of the embodiment of the application is as follows:
firstly, acquiring a vibration signal generated in real time by a gearbox in the driving process of an automobile, and establishing a data set to be detected; then constructing a fault detection model based on the data set to be detected and a preset sample data set reflecting the normal working state of the gearbox, fitting and analyzing the distribution structures of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement, and carrying out fault detection on the gearbox by identifying an abnormal data mode in the data set to be detected; and finally, outputting a detection result of the fault type of the gearbox based on the identification of the abnormal data pattern distribution characteristics.
In order to better understand the technical solution, the technical solution will be described in detail with reference to the drawings and the specific embodiments.
The development of equipment fault diagnosis technology from the late 20 th century is an emerging technology of interdisciplinary and fusion of multiple disciplines. Early researches are mainly based on modeling and analyzing fault mechanisms by an expert system, such as fault mechanism researches, detection and diagnosis technology researches, reliability analysis researches, durability evaluation and the like. With the development of communication and sensing technologies, fault detection research is changed into a method based on signal analysis technologies, such as wavelet analysis, Fourier transform, energy spectrum and the like, and the gearbox fault signals are subjected to noise reduction and decomposition based on high-resolution decomposition and reconstruction capabilities of the fault detection research, so that periodic impact components in the fault signals are separated. At present, with the development of artificial intelligence technology, more and more fault detection is changed to be based on data driving, and an intelligent diagnosis system is developed by introducing technologies such as knowledge reasoning, fuzzy logic, neural network, image recognition and the like. Under the complex environment, the multi-source heterogeneous data can be accurately processed, and unknown fault modes can be better identified.
Currently, data-driven fault diagnosis research is mainly focused on a semi-supervised anomaly detection method for detecting unknown fault modes through limited normal sample data labels. A common limitation of these existing detection methods is that they have difficulty detecting anomalies that occur in real time in the same background as normal data. Namely, how to monitor faults generated in real time sharply and accurately in the continuous operation of the gearbox. The data that make up these abnormal patterns are normal in nature, but are abnormal when they appear together as a set. Such abnormal patterns are difficult to find if only individual data of unlabeled samples are examined one by one. Instead, it may erroneously identify some normal data that falls within the low probability density range as an anomaly. Furthermore, the development of gearbox fault diagnostics is mostly based on the characteristic analysis of vibration signals. The data characteristics of the vibration signal can be divided into two categories: time domain and frequency domain. The characteristics of each category are complex and variable and are susceptible to other sources of vibration during vehicle travel.
In order to solve the above-mentioned problems, embodiments of the present application provide a method for monitoring a failure of a transmission, preferably a gearbox, based on vibration signal distribution characteristic identification. Generally, vibration signals of bearings and gears in an automobile gearbox basically follow a normal distribution, but when a fault occurs, parameters of a probability density function of data distribution of the vibration signals are changed remarkably. Therefore, in the embodiment of the present application, the overall distribution trend of the data is selected as a measure, and according to the distribution of the labeled normal sample data set, the unknown and unexpected data distribution is found through the collective outlier detection, even though the data in the abnormal distribution is not necessarily abnormal per se.
FIG. 1 is a flow chart of a gearbox fault detection method based on vibration signal distribution characteristic identification according to an embodiment of the application. Referring to fig. 1, a gearbox fault detection method based on vibration signal distribution characteristic identification provided by an embodiment of the present application may include:
step S101: collecting vibration signals generated by a gearbox in real time in the driving process of an automobile, and establishing a data set to be detected;
step S102: constructing a fault detection model based on a data set to be detected and a preset sample data set reflecting the normal working state of the gearbox;
step S103: fitting and analyzing the distribution structure of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement for the constructed fault detection model, and carrying out fault detection on the gearbox by identifying abnormal data patterns in the data set to be detected;
step S104: and outputting a detection result of the fault type of the gearbox based on the identification of the abnormal data pattern distribution characteristics.
Generally speaking, gearbox failure detection first collects signals of various operating states of the gearbox, and accordingly establishes a data set to be tested, as described in step S101. It should be noted that the data set to be measured established at this time includes both normal vibration data generated in a normal operating state of the transmission and abnormal vibration data caused by a fault, so that the abnormal vibration data included therein can be subsequently and effectively identified to identify the fault of the transmission.
Next, referring to step S102, a fault detection model is constructed based on the data set to be detected and a preset sample data set reflecting the normal operating state of the transmission. That is to say, in order to effectively identify abnormal vibration data in the data set to be detected, in an optional embodiment of the present invention, before the data set to be detected is established, a normal vibration signal generated in a normal operation state of the transmission may be collected, and a tag is added to the data set formed by the normal vibration signal; and establishing a normal sample data set by a random sampling method based on a data set consisting of the normal vibration signals added with the labels. This is the comparison data that is referenced for the next step in analyzing the test data set to identify abnormal vibration data.
After the fault detection model is established, the data set to be detected and the sample data set are input into the fault detection model for detection, the distribution structure of the sample data set and the distribution structure of the data set to be detected are fitted and analyzed by taking the data distribution characteristics as measurement, and the gearbox is subjected to fault detection by identifying abnormal data modes in the data set to be detected.
In an optional embodiment of the present invention, when constructing the fault detection model, a first distribution function of the sample data set may be obtained by fitting the data distribution of the sample data set using a formula (1) using a mixed multivariate gaussian distribution function, and a first parameter in the first distribution function is estimated based on a maximum likelihood algorithm, where the calculation formula is as follows:
D(Ss)＝fs(x|θs) (1)
wherein:
fsrepresenting a first distribution function;
θsrepresenting a first parameter;
D(Ss) Representing the data distribution of the sample data set.
Fitting the data distribution of the data set to be measured by using a mixed multivariate Gaussian distribution function and applying a formula (2) to obtain a second distribution function of the data set to be measured, and estimating a second parameter in the second distribution function based on a maximum likelihood algorithm, wherein the calculation formula is as follows:
D(St)＝F(x|θt) (2)
wherein:
f represents a second distribution function;
θtrepresents a second parameter;
D(St) Representing the data distribution of the data set under test.
Then, based on the similarity of data distribution, a fault detection model is constructed through a formula (3) to detect whether the data set to be detected contains a fault mode, and the calculation formula is as follows:
wherein:
f represents a second distribution function;
faa third distribution function representing an abnormal data pattern;
fsrepresenting a first distribution function;
θsrepresenting a first parameter;
θtrepresents a second parameter;
θarepresenting a third parameter of the third distribution function.
The fault detection model is constructed based on the first distribution function, the first parameter, the second distribution function, the second parameter, the third distribution function and the parameters; the fault detection model is a mixed distribution function model.
Optionally, the solving process of the first parameter, the second parameter, and the third parameter may be obtained by performing iterative processing calculation on the maximum likelihood equation based on an invariant point iteration method.
As mentioned above, the normal vibration data in the sample data set has a sample label, while the vibration data to be tested in the data set to be tested is unlabeled. Therefore, the present embodiment can find out those unknown and unexpected data distributions by collective outlier detection according to the distribution of the labeled normal sample data set. While calculating the proportion of abnormal data
In the process, normal vibration data in the data set to be detected can be identified based on the sample data set of the label, the rest is abnormal vibration data, and the proportion of the abnormal vibration data in the total data to be detected can be used for obtaining the proportion of the abnormal vibration data
According to the formula (1) and the formula (2), a mixed multivariate Gaussian distribution mixed function pair is used for the sample data set SsAnd a data set S to be testedtThe data distributions of (a) were fitted separately. Secondly, based on the maximum likelihood algorithmTo estimate parameters within the mixture distribution function. And finally, solving the maximum likelihood estimation function by adopting an immobile point iterative algorithm.
A Mixture Gaussian Distribution (Mixture Gaussian Distribution) refers to a linear combination of multiple Gaussian Distribution functions, and the linearly combined Distribution is used to describe data in the whole set. It is often used to address the situation where data under the same collection contains multiple different distributions.
In the n-dimensional Euclidean space, a multivariate Gaussian distribution formed by mixing K Gaussian distributions is described according to a formula (4) and a formula (5), and the calculation formula is as follows:
wherein:
N(x|μk；∑k) A probability density function representing a mixed gaussian distribution;
μkrepresents the mean value;
Σ represents a symmetric semi-positive definite covariance matrix;
|∑kl represents a determinant of a matrix;
f represents a distribution function;
θ represents a parameter of the distribution function;
d (S) represents the data distribution of the data set;
λkexpressing respective mixing coefficients of K Gaussian distributions satisfying lambdakNot less than 0 and
for a sample data set SsThe distribution function is measured according to the formula (6), and the calculation formula is as follows:
wherein:
D(Ss) Representing a data distribution of the sample data set;
fsrepresenting a first distribution function;
θs＝(λj；μj；∑j) Representing the parameters to be estimated by maximum likelihood;
N(x|μj；∑j) A probability density function representing a sample data set distribution;
λjrepresents the mixing coefficients of the J Gaussian distributions;
j represents the number of gaussian distributions in the mixing function.
For the data set S to be testedtThe distribution function is measured according to equation 7, which is calculated as follows:
wherein:
D(St) A data distribution representing a dataset under test;
f represents a second distribution function;
θtrepresents a second parameter;
fsrepresenting a first distribution function;
θs＝(λj；μj；∑j) Representing the parameters to be estimated by maximum likelihood;
N(x|μq；∑q) A probability density function representing a sample data set distribution;
j represents the number of Gaussian distributions in the mixing function;
q denotes the data set StThe number of gaussian distributions used to fit the anomalous distributions.
The Maximum Likelihood algorithm (MLE for short) is a statistical method based on the Maximum Likelihood principle, and is a statistical method based on the Maximum Likelihood principle that a certain random sample is known to meet a certain probability distribution, but specific parameters are unclear, and parameter estimation refers to observing results through a plurality of tests, and using the result of the Maximum probability to deduce the approximate value of the parameters.
For a continuous data sequence S, the distribution function is known as f (x | θ), but the parameter θ in the function is unknown. If sequence S is present1＝(X1,X2…Xn) Is a sample of S and the distribution function
Is known, the parameters in the function f (x | θ) can be estimated from the samples using a maximum likelihood algorithm. The method comprises the following specific steps:
if point Y ═ x1,x2,…,xn) Is a sequence S1Then the probability p that any point in the sequence S falls on the adjacent side of Y can be expressed as
Constructing a likelihood function based on the sequence S according to the formula (8):
the core idea of the maximum likelihood estimation method is as follows: according to equation (9), the search results in a probability function
Parameters for obtaining maximum values
Due to the sequence S1Is a sample of the sequence S, the distribution function being determined by the parameter when p has a maximum
The data distribution of the sequence S is best fitted.
Since the logarithmic function lnL (θ) is an increasing function of the original function L (θ), i.e., the extreme point of the function lnL (θ) is also the extreme point of the function L (θ), the problem of extremizing the likelihood function can be converted into extremizing the function lnL (θ). Not only can the multiplication operation in the function be converted into the addition operation, but also the problem of floating point number overflow is avoided. Thus, for the data set S described in equation (6) and equation (7)sAnd a test data set StThe likelihood function can be constructed according to equation (10) and equation (11), respectively, by solving the parameters
And
a best fit to the data set distribution is achieved.
The parameters for constructing the above functions are solved by an iterative method.
The fixed point iteration is a method for expressing an implicit equation by a set of explicit equations through successive approximation, namely, the equations are continuously corrected by using an approximate solution until the final convergence. The stationary point iteration method is an effective method for solving a highly nonlinear numerical problem, and is widely used for solving equations in many engineering mathematics fields due to the good mathematical properties and mature theorem.
In many problems, although the existence of stationary points can be proved, in numerical calculation, it is often difficult to really find out an accurate solution thereof, for example, x cannot be obtained2-an exact solution of 2 ═ 0. Therefore, in order to ensure the convergence rate of the algorithm, the embodiment of the present application introduces an approximate stationary point concept, and if an accurate stationary point is not found when the search reaches a preset iteration number, the approximate stationary point with the highest accuracy in the iteration is selected as a result. The main concepts and principles of the algorithm are as follows:
definition 1: let X be a subset of Rn, if for each point X in X there is a certain f (X) e X to which it corresponds, then f is a self-mapping of X, denoted as f: x → X.
Definition 2: let X be a non-empty set, f: x → X is its self-map. If X ∈ X exists and f (X ═ X) is satisfied, X is called an exact fixed point of f.
Definition 3: let (X, ρ) be a metric space and T: X → X be a mapping. If L ∈ [0, 1) exists, such that any X, y ∈ X, there is ρ (T (X), T (y) ≦ L ρ (X, y), then T is said to be a compression mapping on X.
Definition 4: approximate motionless point: let ε be any positive number, if for the compressed mapping T: X → X, | X-f (X) | represents the modulus of the medium vector X-f (X) of the n-dimensional Euclidean space Rn. If there is a point x that satisfies | x ^ f (x ^ i) | < epsilon, then x is said to be an approximate stationary point of f.
Theorem 1: the Banach stationary point theorem, also called the compressive mapping theorem, (X, ρ) is a non-empty complete metric space, and T: X → X is a compressive mapping, so that T has a unique stationary point in X. The Banach stationary point theorem states the existence and uniqueness of the solution of the stationary point equation t (x) ═ x.
Theorem 2: for any compression mapping T: X → X, let X be a bounded discrete non-empty set, i.e. any X belongs to X, and X is greater than or equal to a and less than or equal to b, if the following two conditions are satisfied: (1) for any X epsilon X, T is more than or equal to a and less than or equal to b (X); (2) there is a normal number L <1, for any X, y ∈ X, | T (X) -T (y) | < L | X-y |; then T is within the bounded discrete non-empty set and there is a unique immobile point x.
And (3) performing characteristic analysis on the vibration data to be detected by combining the sample data set through the fault detection model, and identifying abnormal vibration data in the vibration data to be detected according to an abnormal data distribution function obtained by the formula (3).
The abnormal distribution function measures the data distribution in the data set to be measured which is significantly different from the data distribution in the sample data set, i.e. the data distribution obeying unknown allocation. These data, subject to unknown distributions, are identified as collective outliers that reflect abnormal data patterns that may be generated by faults during operation of the transmission.
In data analysis, an outlier is an observation that deviates too much from other observations, causing one to suspect that it was generated by a different mechanism. According to different characteristics of the outliers, the outliers can be divided into point outlier constant, situation outliers and collective outliers.
If one data differs significantly from the other data in the measure of its target characteristic, the data is called a point outlier. Context outliers refer to data that deviates significantly from normal patterns in a particular context. A point outlier is a special type of contextual outlier if the entire data set is considered to be contextual or if the contextual attributes are empty. Detection methods based on them are mainly focused on analyzing whether individual data exhibit anomalous behavior. A collective anomaly refers to a group of related data whose overall behavioral attributes will deviate significantly from the entire data set when they occur together in a pattern, but individual data in the set may not be anomalous by itself.
Finally, as described in step S104, a transmission failure detection recognition result is output based on the abnormal vibration data.
Overall, in the gearbox fault diagnosis based on vibration data analysis, an abnormal data pattern and normal data are sequentially generated under the same background, each data forming the abnormal pattern may not have an abnormal representation, and detection cannot be performed by comparing unmarked data and sample data one by one. The vibration data collected by the gearbox under normal operating conditions are generated by the same mechanism, which can be assumed to follow the same distribution. The transmission is rated for a small proportion of abnormal data patterns caused by faults during normal operation, as compared to the normal data scale. Based on the method, a normal data sample distribution function is fitted, then a mixed function is constructed to fit a data set to be tested, and a fault mode is diagnosed according to the mixed function.
The method described in the above embodiments is described in detail below with reference to a preferred embodiment. The embodiments are merely examples for illustrating the embodiments of the present disclosure, and the scope of the claims of the embodiments of the present disclosure should not be limited thereto.
The WLY CVT25 continuously variable transmission newly mounted by a certain automobile manufacturing company is selected as a test object, as shown in FIG. 2, the transmission can match an engine with the displacement of 1.5L-2.0L, the torque coverage range is 250 N.m, the dry weight is 84.5Kg, the axial length is 365mm, the bearing center distance is 197mm, and the speed ratio is 7.07. The experimental data is based on vibration signals of the gearbox acquired by the sensor under different working states, and the acquisition frequency is once every 15 seconds. The entire experimental process was completed in the enterprise test booth and all procedures were run on the MTLAB version 7.0.
1. Test data set
The data required by the experiment of the scheme consists of 3 parts, and a sample data set S in a normal working statesAbnormal working state data set SaData set S to be detected of unknown operating statet。
Sample data set Ss: respectively continuously working three qualified gearboxes with the same model for 24 hours under the same load condition, merging the generated three data sets to form a data set S under the normal working statenormal. From the data set SnormalRandomly selecting 10 independent hours of data, and merging the data into a sample data set S in a normal working states。
Abnormal operating state data set Sa: as shown in fig. 3, the main power shaft, and the most important components of 4 gearboxes including the power shaft, the driving gear and the driven gear are selected as targets for fault detection in the experiment. Replacing conforming parts in gearboxes with cracked parts one by oneAnd tested under the same load. For each crack component, the vibration data of the gearbox can be collected for 6 hours and used as an abnormal working state data set Sa。
In addition, because the difference of vibration signals generated by worn old parts is not obvious compared with cracks, in order to test the sensitivity of the detection method provided by the embodiment of the invention, the worn old parts can be adopted to replace qualified parts in the gearbox one by one, the test is carried out under the same load, and vibration data of 6 hours are collected. For clarity of presentation of these data sets, reference is made to Table 1 so as not to cause unnecessary confusion. Table 1 is a data set of abnormal operating states of the respective components, as follows:
TABLE 1
| Power shaft | Driven shaft | Driving gear | Driven gear | |
| Crack(s) | Sa1(C) | Sa2(C) | Sa3(C) | Sa4(C) |
| Wear and tear | Sa1(A) | Sa2(A) | Sa3(A) | Sa4(A) |
To-be-detected data set S with unknown working statet: data set S under collected normal operating conditionsnormalIn which different classes of abnormal working state data sets S are added in sequenceaForming a data set S to be detectedt. Based on the assumption that the fault mode only occupies a small part of the whole data set, the proportion of the abnormal data set is controlled to be less than 5% of the normal working data set, namely, the data size of not more than 3 hours is randomly selected from the collected various abnormal data sets, and the data set SnormalAnd (6) merging. Finally 8 test data sets are formed, constituting a proportion of each of their data sets.
Table 2 shows the specific information of 8 datasets to be measured, as follows:
TABLE 2
2. Test results
For a sample data set SsAnd an abnormal data set SaAnd fitting data distribution based on the detection algorithm provided by the embodiment of the invention. According to the detection framework, the distribution parameters of the sample data set continue to participate in subsequent analysis, and the distribution parameters and proportion of the abnormal data set serve as data labels of the abnormality, are used for comparing with the detection result of the embodiment of the invention, and do not directly participate in the detection of the data set S to be detected in the unknown working statetAnd (4) during detection.
Fitting the sample data set SsProbability of (2)The parameter of the density function is thetasμ ═ 0.0126; σ ═ 36.432. Data set St1～St8The detection results are shown in table 3, the distribution function and the proportion of the fault data detected based on the method are compared with the actual label, and all the results are accurate to three decimal places. Table 3 compares the failure detection results with the actual labels, as follows:
TABLE 3
As can be seen from the experimental results in table 3, when the method provided by the embodiment of the present invention detects the failure mode of each test data set, the conformity degree of each parameter of the data distribution function is more than 90%, and particularly, when worn old parts are identified, the method still exhibits high sensitivity, so that the validity of the detection method provided by the present invention can be proved. Furthermore, based on the detection results, the following conclusions can also be drawn:
1) bearing failure is more pronounced than with gears;
2) the failure of the driving part is more obvious than that of the driven part;
3) cracked parts fail to a greater extent than worn old parts.
Based on the same inventive concept, as shown in fig. 4, an embodiment of the present application further provides a transmission fault detection system 400 based on vibration signal distribution characteristic identification, including:
the data set acquisition module 410 to be detected is configured to acquire vibration signals generated by the gearbox in real time in the driving process of the automobile and establish a data set to be detected; the data set to be tested comprises normal vibration data generated under the normal operation state of the gearbox and abnormal vibration data caused by faults;
the model construction module 420 is configured to construct a fault detection model based on a data set to be detected and a preset sample data set reflecting the normal working state of the gearbox;
the abnormal vibration data identification module 430 is configured to fit and analyze the distribution structure of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement for the constructed fault detection model, and perform fault detection on the gearbox by identifying the abnormal data mode in the data set to be detected;
and the identification result output module 440 is configured to output a detection result of the gearbox fault type based on the identification of the abnormal data pattern distribution characteristics.
In an alternative embodiment of the present invention, as shown in fig. 5, the system may further include a sample data set collecting module 450, which may be configured to:
collecting normal vibration signals generated under the normal operation state of the gearbox, and adding labels to a data set consisting of the normal vibration signals;
and establishing a normal sample data set by a random sampling method based on a data set consisting of the normal vibration signals added with the labels.
In an optional embodiment of the invention, the model building module 440 may be further configured to:
fitting the data distribution of the sample data set by using a mixed multivariate Gaussian distribution function to obtain a first distribution function of the sample data set, and estimating a first parameter in the first distribution function based on a maximum likelihood algorithm;
fitting the data distribution of the data set to be measured by using a mixed multivariate Gaussian distribution function to obtain a second distribution function of the data set to be measured, and estimating a second parameter in the second distribution function based on a maximum likelihood algorithm;
constructing a fault detection model based on the first distribution function, the first parameter, the second distribution function and the second parameter; the fault detection model is a mixed distribution function model.
In an optional embodiment of the invention, the model building module 440 may be further configured to:
and iterating and solving the first parameter and the second parameter based on the fixed point iteration method.
In an optional embodiment of the present invention, the data distribution of the sample data set is fitted by a first formula;
the first formula is:
D(Ss)＝fs(x|θs)
wherein f issRepresenting a first distribution function; thetasRepresenting a first parameter; d (S)s) Representing the data distribution of the sample data set.
In an optional embodiment of the present invention, the data distribution of the data set to be measured is fitted by a second formula;
the second formula is:
D(St)＝F(x|θt)
wherein F represents a second distribution function; thetatRepresents a second parameter; d (S)t) Representing the data distribution of the data set under test.
In an optional embodiment of the present invention, the fault detection model is constructed by a third formula;
the third formula is:
wherein F represents a second distribution function; f. ofaA third distribution function representing an abnormal data pattern; f. ofsRepresenting a first distribution function;
representing the proportion of abnormal data; thetasRepresenting a first parameter; thetatRepresents a second parameter; thetaaA third parameter representing a third distribution function.
In an optional embodiment of the present invention, the abnormal vibration data identification module 430 may be further configured to:
and fitting and analyzing the distribution structure of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement for the constructed fault detection model, obtaining a third distribution function of the abnormal data mode according to the third formula, and identifying the abnormal data mode in the data set to be detected based on the third distribution function to carry out fault detection on the gearbox.
The embodiment of the invention provides a high-efficiency gearbox fault detection method and system based on vibration signal distribution characteristic identification, wherein the gearbox fault detection method based on vibration signal distribution characteristic identification is used as a semi-supervision method based on collective outlier detection, and a vibration signal data stream generated during gearbox working is detected based on a labeled normal data sample and by taking data distribution as measurement. Those data patterns that follow an unknown distribution will be identified as collective outliers that reflect abnormal data patterns in the transmission that may be caused by a fault. In addition, in the method provided by the embodiment of the invention, a mixed function is constructed to fit the data set to be tested based on fitting of the normal data sample distribution function, the data distribution in the data set to be tested, which is obviously different from the data sample distribution, is identified as a collective outlier, and the fault mode is diagnosed according to the collective outlier.
In summary, compared with the prior art, the method has the following beneficial effects:
1. when the gearbox continuously works, faults generated in real time can be quickly and effectively detected;
2. under the complex working environment influenced by multiple vibration sources, the abnormity of the gearbox can be accurately identified, and the robustness is strong.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (6)
1. A gearbox fault detection method based on vibration signal distribution characteristic identification comprises the following steps:
collecting vibration signals generated by a gearbox in real time in the driving process of an automobile, and establishing a data set to be detected; the data set to be tested comprises normal vibration data generated under the normal operation state of the gearbox and abnormal vibration data caused by faults;
constructing a fault detection model based on the data set to be detected and a preset sample data set reflecting the normal working state of the gearbox;
fitting and analyzing the distribution structure of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement for the constructed fault detection model, and carrying out fault detection on the gearbox by identifying an abnormal data mode in the data set to be detected;
outputting a detection result of the gearbox fault type based on the identification of the abnormal data pattern distribution characteristics;
gather the vibration signal that the gearbox produced in real time at the automobile driving in-process, before establishing the data set that awaits measuring, still include:
collecting normal vibration signals generated under the normal operation state of the gearbox, and adding a label to a data set formed by the normal vibration signals;
establishing a normal sample data set by a random sampling method based on a data set consisting of the normal vibration signals added with labels;
constructing a fault detection model based on the data set to be detected and a preset sample data set reflecting the normal working state of the gearbox, wherein the fault detection model comprises the following steps:
fitting the data distribution of the sample data set by using a mixed multivariate Gaussian distribution function to obtain a first distribution function of the sample data set, and estimating a first parameter in the first distribution function based on a maximum likelihood algorithm;
fitting the data distribution of the data set to be detected by using a mixed multivariate Gaussian distribution function to obtain a second distribution function of the data set to be detected, and estimating a second parameter in the second distribution function based on a maximum likelihood algorithm;
constructing the fault detection model based on the first distribution function, the first parameter, the second distribution function and the second parameter; wherein the fault detection model is a mixed distribution function model;
constructing the fault detection model through a third formula;
the third formula is:
wherein F represents a second distribution function; f. ofaA third distribution function representing an abnormal data pattern; f. ofsRepresenting a first distribution function;
representing the proportion of abnormal data; thetasRepresenting a first parameter; thetatRepresents a second parameter; thetaaA third parameter representing a third distribution function.
2. The method of claim 1, wherein prior to constructing the fault detection model based on the first distribution function, the first parameter, the second distribution function, and the second parameter, further comprising:
and carrying out iterative processing on the maximum likelihood equation based on an immobile point iterative method and solving the first parameter and the second parameter.
3. The method of claim 1, wherein the data distribution of the sample data set is fitted by a first formula;
the first formula is:
D(Ss)＝fs(x|θs)
wherein f issRepresenting a first distribution function; thetasRepresenting a first parameter; d (S)s) Representing the data distribution of the sample data set.
4. The method of claim 3, wherein the data distribution of the dataset under test is fitted by a second formula;
the second formula is:
D(St)＝F(x|θt)
wherein F represents a second distribution function; thetatRepresents a second parameter; d (S)t) Representing the data distribution of the data set under test.
5. The method of claim 1, wherein the step of fitting and analyzing the distribution structure of the sample data set and the data set to be tested by using the data distribution characteristics as metrics to the constructed fault detection model, and performing fault detection on the gearbox by identifying abnormal data patterns in the data set to be tested comprises the following steps:
and fitting and analyzing the distribution structure of the sample data set and the data set to be detected by taking the data distribution characteristics as measurement for the constructed fault detection model, obtaining a third distribution function of the abnormal data mode according to a third formula, and identifying the abnormal data mode in the data set to be detected based on the third distribution function to carry out fault detection on the gearbox.
6. A gearbox fault detection system based on vibration signal distribution characteristic identification comprises:
the data set acquisition module to be detected is configured to acquire vibration signals generated by the gearbox in real time in the driving process of the automobile and establish a data set to be detected; the data set to be tested comprises normal vibration data generated under the normal operation state of the gearbox and abnormal vibration data caused by faults;
the model building module is configured to build a fault detection model based on the data set to be tested and a preset sample data set reflecting the normal working state of the gearbox;
the abnormal vibration data identification module is configured to fit and analyze the distribution structure of the sample data set and the data set to be detected by taking data distribution characteristics as measurement for the constructed fault detection model, and carry out fault detection on the gearbox by identifying an abnormal data mode in the data set to be detected;
an identification result output module configured to output a detection result of the gearbox fault type based on identification of the abnormal data pattern distribution characteristics;
gather the vibration signal that the gearbox produced in real time at the automobile driving in-process, before establishing the data set that awaits measuring, still include:
collecting normal vibration signals generated under the normal operation state of the gearbox, and adding a label to a data set formed by the normal vibration signals;
establishing a normal sample data set by a random sampling method based on a data set consisting of the normal vibration signals added with labels;
constructing a fault detection model based on the data set to be detected and a preset sample data set reflecting the normal working state of the gearbox, wherein the fault detection model comprises the following steps:
fitting the data distribution of the sample data set by using a mixed multivariate Gaussian distribution function to obtain a first distribution function of the sample data set, and estimating a first parameter in the first distribution function based on a maximum likelihood algorithm;
fitting the data distribution of the data set to be detected by using a mixed multivariate Gaussian distribution function to obtain a second distribution function of the data set to be detected, and estimating a second parameter in the second distribution function based on a maximum likelihood algorithm;
constructing the fault detection model based on the first distribution function, the first parameter, the second distribution function and the second parameter; wherein the fault detection model is a mixed distribution function model;
constructing the fault detection model through a third formula;
the third formula is:
wherein F represents a second distribution function; f. ofaA third distribution function representing an abnormal data pattern; f. ofsRepresenting a first distribution function;
representing the proportion of abnormal data; thetasRepresenting a first parameter; thetatRepresents a second parameter; thetaaA third parameter representing a third distribution function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911221599.0A CN110967184B (en) | 2019-12-03 | 2019-12-03 | Gearbox fault detection method and system based on vibration signal distribution characteristic recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911221599.0A CN110967184B (en) | 2019-12-03 | 2019-12-03 | Gearbox fault detection method and system based on vibration signal distribution characteristic recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110967184A CN110967184A (en) | 2020-04-07 |
| CN110967184B true CN110967184B (en) | 2021-06-11 |
Family
ID=70032713
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911221599.0A Active CN110967184B (en) | 2019-12-03 | 2019-12-03 | Gearbox fault detection method and system based on vibration signal distribution characteristic recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110967184B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117033950B (en) * | 2023-10-08 | 2023-12-26 | 华中科技大学 | GIS isolating switch mechanical fault on-line diagnosis method and device |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103049653A (en) * | 2012-12-17 | 2013-04-17 | 南京航空航天大学 | G0 distribution parameter maximum likelihood estimation method based on EM (Expectation-Maximization) algorithm |
| KR20140071016A (en) * | 2012-12-03 | 2014-06-11 | 엘지디스플레이 주식회사 | Panel Inspection Method |
| KR20140120716A (en) * | 2013-04-04 | 2014-10-14 | 한국기술교육대학교 산학협력단 | Defect detection method in heterogeneously textured surface |
| CN104376211A (en) * | 2014-11-14 | 2015-02-25 | 合肥工业大学 | Method for assessing measurement uncertainty of coordinate measuring machine |
| CN104392040A (en) * | 2014-11-20 | 2015-03-04 | 合肥工业大学 | Product inspection misjudgment rate computing method based on measurement uncertainty |
| CN104636496A (en) * | 2015-03-04 | 2015-05-20 | 重庆理工大学 | Hybrid clustering recommendation method based on Gaussian distribution and distance similarity |
| CN107942308A (en) * | 2017-10-19 | 2018-04-20 | 西安电子科技大学 | Adaptive Rao detection methods based on gamma texture under complex Gaussian environment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2568178C (en) * | 2004-07-02 | 2009-11-24 | Vibration Research Corporation | System and method for simultaneously controlling spectrum and kurtosis of a random vibration |
| CN100460845C (en) * | 2007-02-07 | 2009-02-11 | 北京航空航天大学 | Method for generating non Gaussian random vibration pumping signal and device thereof |
-
2019
- 2019-12-03 CN CN201911221599.0A patent/CN110967184B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20140071016A (en) * | 2012-12-03 | 2014-06-11 | 엘지디스플레이 주식회사 | Panel Inspection Method |
| CN103049653A (en) * | 2012-12-17 | 2013-04-17 | 南京航空航天大学 | G0 distribution parameter maximum likelihood estimation method based on EM (Expectation-Maximization) algorithm |
| KR20140120716A (en) * | 2013-04-04 | 2014-10-14 | 한국기술교육대학교 산학협력단 | Defect detection method in heterogeneously textured surface |
| CN104376211A (en) * | 2014-11-14 | 2015-02-25 | 合肥工业大学 | Method for assessing measurement uncertainty of coordinate measuring machine |
| CN104392040A (en) * | 2014-11-20 | 2015-03-04 | 合肥工业大学 | Product inspection misjudgment rate computing method based on measurement uncertainty |
| CN104636496A (en) * | 2015-03-04 | 2015-05-20 | 重庆理工大学 | Hybrid clustering recommendation method based on Gaussian distribution and distance similarity |
| CN107942308A (en) * | 2017-10-19 | 2018-04-20 | 西安电子科技大学 | Adaptive Rao detection methods based on gamma texture under complex Gaussian environment |
Non-Patent Citations (2)
| Title |
|---|
| Pseudo maximum likelihood estimation of structural models involving fixed-point problems;Victor Aguirregabiria;《science direct》;20040930;全文 * |
| 机械故障信号统计建模及其故障诊断方法的研究;李长宁;《中国博士学位论文全文数据库 工程科技Ⅱ辑》;20110815;第17-49页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110967184A (en) | 2020-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109460618B (en) | Rolling bearing residual life online prediction method and system | |
| Jiang et al. | Fault diagnosis for rolling bearing using a hybrid hierarchical method based on scale-variable dispersion entropy and parametric t-SNE algorithm | |
| US8725456B1 (en) | Decomposition technique for remaining useful life prediction | |
| Yan et al. | Fisher’s discriminant ratio based health indicator for locating informative frequency bands for machine performance degradation assessment | |
| CN113469060A (en) | Multi-sensor fusion convolution neural network aeroengine bearing fault diagnosis method | |
| Chen et al. | Generalized open-set domain adaptation in mechanical fault diagnosis using multiple metric weighting learning network | |
| CN111562108A (en) | Rolling bearing intelligent fault diagnosis method based on CNN and FCMC | |
| CN206504869U (en) | A kind of rolling bearing fault diagnosis device | |
| CN112785091B (en) | Method for carrying out fault prediction and health management on oil field electric submersible pump | |
| Zhao et al. | A novel deep fuzzy clustering neural network model and its application in rolling bearing fault recognition | |
| CN114705432B (en) | Method and system for evaluating health state of explosion-proof motor bearing | |
| CN114091504A (en) | Rotary machine small sample fault diagnosis method based on generation countermeasure network | |
| CN114861349B (en) | Rolling bearing RUL prediction method based on model migration and wiener process | |
| CN115221973A (en) | Aviation bearing fault diagnosis method based on enhanced weighted heterogeneous ensemble learning | |
| CN116625686A (en) | On-line diagnosis method for bearing faults of aero-engine | |
| Zheng et al. | An unsupervised transfer learning method based on SOCNN and FBNN and its application on bearing fault diagnosis | |
| Chou et al. | SHM data anomaly classification using machine learning strategies: A comparative study | |
| CN115375026A (en) | Method for predicting service life of aircraft engine in multiple fault modes | |
| CN110967184B (en) | Gearbox fault detection method and system based on vibration signal distribution characteristic recognition | |
| CN112069621B (en) | Method for predicting residual service life of rolling bearing based on linear reliability index | |
| Nemani et al. | Health index construction with feature fusion optimization for predictive maintenance of physical systems | |
| Gao et al. | Long-term Temporal Attention Neural Network with Adaptive Stage Division for Remaining Useful Life Prediction of Rolling Bearings | |
| Aremu et al. | Kullback-leibler divergence constructed health indicator for data-driven predictive maintenance of multi-sensor systems | |
| CN114371677B (en) | Industrial process state monitoring method based on spectrum radius-interval principal component analysis | |
| Du et al. | Graph neural network-based early bearing fault detection |
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 |