CN114046993A - A state evaluation method of slewing bearing based on multi-feature parameter fusion and FCM-HMM - Google Patents

A state evaluation method of slewing bearing based on multi-feature parameter fusion and FCM-HMM Download PDF

Info

Publication number
CN114046993A
CN114046993A CN202111217266.8A CN202111217266A CN114046993A CN 114046993 A CN114046993 A CN 114046993A CN 202111217266 A CN202111217266 A CN 202111217266A CN 114046993 A CN114046993 A CN 114046993A
Authority
CN
China
Prior art keywords
slewing bearing
state
fcm
hmm
evaluation
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.)
Pending
Application number
CN202111217266.8A
Other languages
Chinese (zh)
Inventor
王�华
姜烨飞
乾钦荣
傅航
张磊
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.)
Nanjing Tech University
Suote Transmission Equipment Co Ltd
Original Assignee
Nanjing Tech University
Suote Transmission Equipment Co Ltd
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 Nanjing Tech University, Suote Transmission Equipment Co Ltd filed Critical Nanjing Tech University
Priority to CN202111217266.8A priority Critical patent/CN114046993A/en
Publication of CN114046993A publication Critical patent/CN114046993A/en
Pending legal-status Critical Current

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
    • G01M13/04—Bearings
    • 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
    • G01M13/04—Bearings
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2133—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on naturality criteria, e.g. with non-negative factorisation or negative correlation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Acoustics & Sound (AREA)
  • Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)

Abstract

本发明提供了一种基于多特征参数融合与FCM‑HMM的回转支承状态评估方法，包括如下步骤：对多物理信号进行预处理，采用强化局部均值分解对原始信号降噪；提取信号时域、频域、时频域多领域特征；为避免特征冗余或重叠对后续评估过程的干扰，提出综合评价指标筛选优胜特征；基于等距离映射算法构建多物理信号健康指标，有效表征回转支承的性能退化趋势；结合模糊C均值与隐马尔科夫模型，识别回转支承退化状态转移过程，确定回转支承早期故障点及失效预警点。本方法采用回转支承全寿命加速试验数据进行模型验证，有效划分回转支承退化状态。

Figure 202111217266

The invention provides a slewing bearing state evaluation method based on multi-feature parameter fusion and FCM-HMM, comprising the following steps: preprocessing multi-physical signals, using enhanced local mean decomposition to denoise original signals; extracting signal time domain, Multi-domain features in frequency domain and time-frequency domain; in order to avoid the interference of feature redundancy or overlapping on the subsequent evaluation process, a comprehensive evaluation index is proposed to screen out the winning features; based on the equidistant mapping algorithm, a multi-physical signal health index is constructed to effectively characterize the performance of the slewing bearing Degradation trend; combined with fuzzy C-mean value and hidden Markov model, identify the transition process of the slewing bearing degradation state, and determine the early failure points and failure warning points of the slewing bearing. This method uses the full-life accelerated test data of the slewing bearing for model verification, and effectively divides the degradation state of the slewing bearing.

Figure 202111217266

Description

Slewing bearing state evaluation method based on multi-feature parameter fusion and FCM-HMM
Technical Field
The invention relates to the field of parameter evaluation, in particular to a slewing bearing state evaluation method based on multi-feature parameter fusion and FCM-HMM.
Background
The slewing bearing is used as a core component of a rotary mechanical complete machine, is widely applied to various large engineering practice fields, and the operational reliability of the slewing bearing directly influences the health state of mechanical equipment. The method for accurately evaluating the service life state of the slewing bearing is researched, guidance is provided for active operation and maintenance, economic and safety problems caused by slewing bearing faults or failures are effectively reduced, and the method has important engineering significance. By means of the rapid development of an artificial intelligence technology, a data driving method does not need to depend on a damage mechanism of a slewing bearing, and reliable data driving modeling becomes a mainstream method for study of scholars. The evaluation method based on data driving has strong advantages when aiming at mechanical systems such as slewing bearings with unclear or complex damage mechanisms.
Due to the fact that the actual working environment is severe, effective information in monitored sensing signals is often covered by a large amount of noise, the original characteristic extraction of the signals is affected, particularly, the slewing bearing runs at a low speed, vibration signals are weak, and therefore effective noise reduction reconstruction processing needs to be carried out on the signals. The feature extraction of the signal can reduce the calculation amount and highlight the change trend of the signal, and is one of the indispensable steps in the life state evaluation. However, the phenomenon of redundancy or overlapping of various signal characteristics often exists among the various signal characteristics, and subsequent information fusion is easily interfered, so that the influence of the characteristic redundancy on an evaluation result can be reduced to a certain extent by effectively screening the characteristics.
The state evaluation of the slewing bearing still has several problems at present: firstly, the feature extraction of the signal not only needs to meet the comprehensive requirement, but also needs to avoid the interference of feature redundancy or overlapping on the subsequent evaluation process; secondly, the health state of the slewing bearing is monitored in real time, and higher efficiency is needed for determining state transfer; and thirdly, the damage information of the slewing bearing is difficult to be comprehensively acquired by a single vibration signal.
Disclosure of Invention
The invention provides an FCM-HMM modeling strategy aiming at the characteristics of complex actual working conditions and weak acquired signals of the slewing bearing and combining the existing problems in the estimation of the service life state of the slewing bearing so as to efficiently identify the progressive degradation process in the operation process of the slewing bearing.
The invention is directed to a high-reliability slewing bearing state evaluation method under complex working conditions. Preprocessing multiple physical signals, and denoising original signals by adopting enhanced local mean decomposition; extracting multi-domain characteristics of a signal time domain, a signal frequency domain and a signal time-frequency domain; in order to avoid the interference of characteristic redundancy or overlapping on a subsequent evaluation process, a comprehensive evaluation index is provided to screen out the winning characteristics; multiple physical signal health indexes are constructed based on an equidistant mapping algorithm, and the performance degradation trend of the slewing bearing is effectively represented; and identifying the degradation state transition process of the slewing bearing by combining the fuzzy C mean value and the hidden Markov model, and determining an early fault point and a failure early warning point of the slewing bearing.
In order to achieve the purpose, the invention adopts the following technical scheme:
a slewing bearing state evaluation method based on multi-feature parameter fusion and FCM-HMM is characterized by comprising the following steps:
step 1, denoising an original signal: acquiring original signals of the slewing bearing of the service part through acceleration, temperature and torque sensors, and reducing noise by using a Robust Local Mean Decomposition (RLMD);
step 2, feature extraction and screening: extracting multi-domain characteristics of a signal time domain, a signal frequency domain and a signal time-frequency domain, and providing comprehensive evaluation indexes to screen out superior characteristics in order to avoid the interference of characteristic redundancy or overlapping on a subsequent evaluation process;
step 3, constructing multiple physical signal health indexes: the global characteristics are reserved by a nonlinear dimension reduction method, and an Isometric Mapping (ISOMAP) algorithm is selected to establish the health indexes of the vibration, temperature and torque signals of the slewing bearing;
step 4, establishing a state evaluation model: based on multiple physical signal health indexes and in combination with a Fuzzy C-means (FCM) and Hidden Markov Model (HMM), namely an FCM-HMM joint modeling strategy, a state evaluation Model is established, and recognition of a slewing bearing degradation state transition process is achieved.
In the step 1, the enhanced Local Mean Decomposition technology is an improved enhanced algorithm based on Local Mean Decomposition (LMD) proposed in recent years, and by optimizing three key parts of boundary conditions, envelope evaluation and screening stop criteria in the LMD algorithm, the optimal step size and the screening iteration number are determined in a self-adaptive manner, and the algorithm performance is improved.
In the step 2, the characteristic fields are as follows:
Figure BDA0003311169860000021
Figure BDA0003311169860000031
wherein, the values from (r) to (r) are time domain characteristics,
Figure BDA0003311169860000032
to
Figure BDA0003311169860000033
For the purpose of the frequency domain characterization,
Figure BDA0003311169860000034
is a time-frequency domain feature; wherein, X is a noise-reduced signal, xn is the nth point of X, f i represents the power spectrum frequency at the moment i, p i represents the power spectrum amplitude, and Ci (t) is the inherent mode function of signal decomposition; selecting monotonicity indexes and track difference indexes to quantitatively and effectively screen the characteristics of all physical signals of the slewing bearing; because the two quantitative evaluation indexes respectively represent the characteristic attributes from different angles, in order to comprehensively and visually screen the winning characteristics under the same physical signal, a mixed evaluation mode of linearly superposing the two quantitative evaluation indexes, namely a comprehensive evaluation index Score, is provided, and specifically:
Score＝map min max(Mon)+map min max(Div)；
in order to avoid negative influence on the screening result caused by the value domain difference of the two quantitative indexes, normalization processing is uniformly carried out before linear superposition, namely mapminmax operation in a formula.
In the step 3, an equidistant mapping algorithm is selected to establish the health indexes of the vibration, temperature and torque signals of the slewing bearing, and the problem to be solved in the dimensionality reduction algorithm is the estimation of intrinsic dimensionality, namely, the minimum embedding dimensionality of the low-dimensional space is determined under the condition of fully retaining the original information of high-dimensional data; reasonable eigen-dimension estimation is crucial. Using the classical dimensionality estimation method-Maximum Likelihood Estimation (MLE); and (3) realizing maximum likelihood estimation of the intrinsic dimension by establishing a likelihood function of the neighbor distance.
In step 4, the FCM-HMM model is established in two steps: dividing a state matrix and establishing a degradation state identification library; the specific steps of degradation state evaluation based on the FCM-HMM are as follows:
1) dividing a state matrix; carrying out unsupervised clustering on the observation sequence by adopting an FCM method, and dividing a state matrix corresponding to the state number K; initial parameters in FCM need to be preset, the clustering number is determined by a full-life acceleration test result, certain subjectivity is achieved, and meanwhile, the initial membership degree is randomly selected, so that the accurate judgment of the category of the critical point is difficult to realize; therefore, the state matrix is selected in proportion to construct an observation vector On, so that the influence of a critical point On the model is avoided; in addition, the observation sequence is a full life cycle sequence, the input type of the observation sequence is researched, and the degradation state evaluation result based on different observation sequences is discussed in a contrast manner;
2) establishing a degradation state identification library; initializing model parameters based On the constructed observation vector On, continuously optimizing the model parameters by adopting a Baum-welch algorithm, and obtaining an optimal model HMMn (lambda n) of each state; the established FCM-HMM degradation state recognition library is used for dividing the degradation stage of the current observation sequence and determining the state transition process;
3) evaluating the current degradation state of the slewing bearing; inputting the current observation sequence into an FCM-HMM degradation state recognition library, calculating the log-likelihood probability p (O | λ n) output by each model HMMn (λ n) by adopting a forward-backward algorithm, comparing n ═ arg max [ p (O | λ n) ], wherein n is the current degradation state of the slewing bearing; if n is equal to K, the slewing bearing is shown to enter a rapid decline stage of the whole life cycle, and a failure and blocking phenomenon may occur at a certain moment later, so that failure early warning should be performed at the stage, and economic loss caused by sudden failure of equipment is avoided.
Compared with the prior art, the invention has the beneficial effects that:
1. the RLMD method adopted by the invention is used for carrying out noise reduction processing on the signals, thereby effectively reducing the interference of noise components. And a multi-field feature extraction mode is adopted to construct a high-dimensional feature vector, so that the integrity of the signal is ensured. The provided comprehensive evaluation index screens multiple characteristics, and avoids the interference of characteristic redundancy or overlapping on a subsequent evaluation process.
2. The invention adopts the multi-physical signal health index constructed based on the ISOMAP algorithm, solves the problem that the damage information of the slewing bearing is difficult to be comprehensively acquired by a single vibration signal, effectively represents the performance degradation trend of the slewing bearing and lays a foundation for the subsequent state evaluation of the slewing bearing.
The FCM-HMM modeling strategy provided by the invention can more effectively divide the degradation state of the slewing bearing, greatly shorten the evaluation time and provide a good basis for judging the state transition of the slewing bearing in real time.
The FCM method provided by the invention solves the problem that the HMM needs to preset the state number of the model and the limiting conditions of the corresponding observation sequence, and determines the early fault point and the failure early warning point of the slewing bearing by utilizing the dynamic modeling capability of the HMM. The FCM-HMM method can effectively divide the degradation state of the slewing bearing, greatly shortens the evaluation time, and provides a good basis for judging the state of the slewing bearing in real time.
Drawings
FIG. 1 is a multi-physical signal pre-processing framework of the method of the present invention;
FIG. 2 is a diagram of multiple physical raw vibration signals in the present invention;
FIG. 3 is a graph of multiple physical raw temperature signals in accordance with the present invention;
FIG. 4 is a graph of multiple physical raw torque signals in accordance with the present invention;
FIG. 5 is a vibration comprehensive evaluation index diagram in the present invention;
FIG. 6 is a comprehensive evaluation index chart of temperature in the present invention;
FIG. 7 is a torque integrated evaluation index chart according to the present invention;
FIG. 8 is a schematic diagram of the vibration signal health indicator of the present invention;
FIG. 9 is a schematic diagram of a temperature signal health indicator in accordance with the present invention;
FIG. 10 is a schematic illustration of a torque signal health indicator in accordance with the present invention;
FIG. 11 is a diagram of FCM clustering results in the present invention;
FIG. 12 is a diagram showing the result of FCM-HMM evaluation in the present invention;
FIG. 13 is a schematic diagram of the state transition process of the original vibration signal in the present invention;
FIG. 14 is a schematic diagram illustrating a state transition process of an original temperature signal according to the present invention;
FIG. 15 is a schematic diagram of the state transition process of the raw torque signal in the present invention;
FIG. 16 is a diagram illustrating a state transition process of the multiple physical quantity health indicator according to the present invention.
Detailed Description
The technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings, and it is to be understood that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments, and all other embodiments obtained by those skilled in the art without any inventive work based on the embodiments of the present invention belong to the scope of the present invention.
As shown in fig. 1 to 16, the present embodiment describes a slewing bearing state evaluation method based on multi-feature parameter fusion and FCM-HMM, which includes the following steps:
step (1), denoising an original signal: acquiring original signals of the slewing bearing of the service part through acceleration, temperature and torque sensors, and reducing noise by using a Robust Local Mean Decomposition (RLMD);
step (2), feature extraction and screening: extracting multi-domain characteristics of a signal time domain, a signal frequency domain and a signal time-frequency domain, and providing comprehensive evaluation indexes to screen out superior characteristics in order to avoid the interference of characteristic redundancy or overlapping on a subsequent evaluation process;
step (3), constructing multiple physical signal health indexes: the global characteristics are reserved by a nonlinear dimension reduction method, and an Isometric Mapping (ISOMAP) algorithm is selected to establish the health indexes of the vibration, temperature and torque signals of the slewing bearing;
step (4), establishing a state evaluation model: based on multiple physical signal health indexes and in combination with a Fuzzy C-means (FCM) and Hidden Markov Model (HMM), namely an FCM-HMM joint modeling strategy, a state evaluation Model is established, and recognition of a slewing bearing degradation state transition process is achieved.
The enhanced Local Mean Decomposition in the step (1) is an improved enhancement algorithm based on Local Mean Decomposition (LMD) proposed in recent years, and by optimizing three key parts of boundary conditions, envelope evaluation and screening stopping standards in the LMD algorithm, the self-adaptive determination of the optimal step size and the screening iteration times is realized, and the algorithm performance is improved. The original signal is as in fig. 2. The test is ended by the dead locking of the rotation support, and the disassembly and inspection result shows that all parts are damaged in different degrees, so that the collected test data are effective. Because there are more interference in the test process, the background noise is serious, the fault signal is weak and is easy to be covered by the external interference component, so the whole change is not obvious, but the amplitude presents a slow increasing trend, which accords with the decline condition. The temperature signal shows a fluctuation rising trend due to the external environment temperature change and day-night temperature difference, and maintenance processing such as bolt adjustment, lubrication and the like is carried out near the 8 th day, so that the temperature and torque signals are reduced to some extent near the temperature and torque signals, and the temperature and torque are gradually increased due to the fading condition after restarting and finally enter a failure stage. Compared with temperature and torque signals, the vibration signal has relatively better anti-interference performance, and is commonly used for fault diagnosis and state evaluation of mechanical equipment.
The characteristic field in the step (2) is as followsTable 1 shows: phi to r are time domain features,
Figure BDA0003311169860000061
to
Figure BDA0003311169860000062
For the purpose of the frequency domain characterization,
Figure BDA0003311169860000063
is a time-frequency domain feature. Wherein X is a noise-reduced signal, XnN-th point of X, fiRepresenting power spectrum frequency, p, at time iiRepresenting the amplitude of the power spectrum, Ci(t) is the natural mode function of the signal decomposition. And (3) quantitatively and effectively screening the characteristics of each physical signal of the slewing bearing by using a monotonicity index and a track difference index. Because the two quantitative evaluation indexes respectively represent the characteristic attributes from different angles, in order to comprehensively and visually screen the winning characteristics under the same physical signal, a mixed evaluation mode of linearly superposing the two quantitative evaluation indexes, namely a comprehensive evaluation index Score, is provided. In order to avoid negative influence on the screening result caused by the value domain difference of the two quantitative indexes, normalization processing is uniformly carried out before linear superposition, namely mapminmax operation in a formula. The comprehensive evaluation indexes of the vibration, temperature and torque signals are shown in figure 3.
Score＝map min max(Mon)+map min max(Div)
And (3) selecting an equidistant mapping algorithm to establish the health indexes of the vibration, temperature and torque signals of the slewing bearing, wherein the problem to be solved in the dimensionality reduction algorithm is the estimation of intrinsic dimensionality, namely, the minimum embedding dimensionality of the low-dimensional space is determined under the condition of fully retaining the original information of high-dimensional data. Reasonable eigen-dimension estimation is crucial. The classical dimensionality estimation method, Maximum Likelihood Estimation (MLE), is used. And (3) realizing maximum likelihood estimation of the intrinsic dimension by establishing a likelihood function of the neighbor distance. The vibration, temperature and torque signal health indexes are shown in figure 4.
The FCM-HMM model establishment in the step (4) is divided into two steps: dividing the state matrix and establishing a degradation state identification library. The specific steps of degradation state evaluation based on the FCM-HMM are as follows:
1) the state matrix is partitioned. And carrying out unsupervised clustering on the observation sequence by adopting an FCM method, and dividing a state matrix corresponding to the state number K. Initial parameters in FCM need to be preset, the clustering number is determined by a full-life acceleration test result, certain subjectivity is achieved, and meanwhile, accurate judgment of the class of the critical point is difficult to achieve due to random selection of the initial membership degree. Therefore, the state matrix is selected in proportion to construct an observation vector On, and the influence of the critical point On the model is avoided. In addition, the observation sequence is a full life cycle sequence, the input type of the observation sequence is researched, and the degradation state evaluation result based on different observation sequences is discussed in a contrast mode.
2) And establishing a degradation state identification library. Model parameters are initialized based On the constructed observation vector On, and the model parameters are continuously optimized by adopting a Baum-welch algorithm to obtain an optimal model HMMn (lambda n) of each state. And the established FCM-HMM degradation state recognition library is used for dividing the degradation stage of the current observation sequence and determining the state transition process.
3) And evaluating the current degradation state of the slewing bearing. Inputting the current observation sequence into an FCM-HMM degradation state recognition library, calculating the log-likelihood probability p (O | λ n) output by each model HMMn (λ n) by adopting a forward-backward algorithm, comparing n ═ arg max [ p (O | λ n) ], and obtaining n as the current degradation state of the slewing bearing. If n is equal to K, the slewing bearing is indicated to enter a rapid decline stage of the whole life cycle, and a failure and locking phenomenon may occur at a certain moment later, so that failure early warning should be performed at the stage.
Dividing a state matrix corresponding to the four states by adopting an FCM (fuzzy C-means) method, wherein the number of categories is 4, m is 2, and the maximum iteration number is 150; the stop threshold is ∈ 10-5, and the clustering result is shown in fig. 5. As can be seen from the figure, the life cycle of the slewing bearing can be effectively divided into four stages based on the multiple physical signal health indexes, and the failure stage is obviously distinguished from the normal stage. However, the FCM clustering result tends to be stable after multiple operations, and meanwhile, the category of the critical point is difficult to accurately distinguish through random selection of the initial clustering center, so that the critical point is avoided, 50% of each state matrix is selected to construct an observation vector On, and a degraded state identification library HMMn (lambda n) is established. And (4) inputting the multiple physical signal health indexes established in the step (3) into a degradation state identification library HMMn (lambda n), wherein the state evaluation result of the full life cycle sequence is shown in FIG. 6. As can be seen from the figure, the difference degree of the output probabilities p (O | λ n) of different models HMMn (λ n) calculated by adopting a forward-backward algorithm is obvious, the HMM model with the maximum output probability can be clearly judged, and the corresponding state of the model is the current degradation stage of the slewing bearing. The slewing bearing degradation state was effectively partitioned based on the FCM-HMM model, and the results are shown in table 1.
TABLE 1 degradation status partitioning results
Figure BDA0003311169860000081
In conclusion, the slewing bearing state evaluation method based on the FCM-HMM can realize the state recognition of the current observation sequence, the state transfer process has good interpretability, and reference is provided for realizing the online state evaluation of equipment.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the preferred embodiments of the invention and described in the specification are only preferred embodiments of the invention and are not intended to limit the invention, and that various changes and modifications may be made without departing from the novel spirit and scope of the invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (5)

1.一种基于多特征参数融合与FCM-HMM的回转支承状态评估方法，其特征在于，包括以下步骤：1. a slewing bearing state assessment method based on multi-feature parameter fusion and FCM-HMM, is characterized in that, comprises the following steps: 步骤1、原始信号降噪：通过加速度、温度和扭矩传感器对服役件回转支承采集原始信号，并利用强化局部均值分解技术(Robust Local Mean Decomposition,RLMD)进行降噪；Step 1. Noise reduction of the original signal: collect the original signal of the slewing bearing of the service piece through acceleration, temperature and torque sensors, and use the enhanced local mean decomposition technology (Robust Local Mean Decomposition, RLMD) to reduce noise; 步骤2、特征提取与筛选：提取信号时域、频域、时频域多领域特征，为避免特征冗余或重叠对后续评估过程的干扰，提出综合评价指标筛选优胜特征；Step 2. Feature extraction and screening: Extract the multi-domain features of the signal in time domain, frequency domain, and time-frequency domain. In order to avoid the interference of redundant or overlapping features on the subsequent evaluation process, a comprehensive evaluation index is proposed to screen out the winning features; 步骤3、构建多物理信号健康指标：以非线性降维手段保留全局特征，选用等距离映射算法(Isometric Mapping,ISOMAP)建立回转支承振动、温度、扭矩信号健康指标；Step 3. Construct multi-physical signal health indicators: retain global features by means of nonlinear dimensionality reduction, and select Isometric Mapping (ISOMAP) to establish slewing bearing vibration, temperature, and torque signal health indicators; 步骤4、建立状态评估模型：基于多物理信号健康指标，并结合模糊C均值(Fuzzy C-means,FCM)与隐马尔科夫模型(Hidden Markov Model,HMM)即FCM-HMM的联合建模策略，建立状态评估模型，实现对回转支承退化状态转移过程的识别。Step 4. Establish a state evaluation model: based on multi-physical signal health indicators, combined with Fuzzy C-means (FCM) and Hidden Markov Model (HMM), that is, a joint modeling strategy of FCM-HMM , establish a state evaluation model, and realize the identification of the transition process of the degraded state of the slewing bearing. 2.根据权利要求1所述的基于多特征参数融合与FCM-HMM的回转支承状态评估方法，其特征在于：所述步骤1中，所述强化局部均值分解技术是近几年提出的基于局部均值分解(Local Mean Decomposition,LMD)的改进增强算法，通过优化LMD算法中的边界条件、包络评估和筛选停止标准三大关键部分，实现自适应确定最佳步长大小及筛选迭代次数，提高算法性能。2. the slewing bearing state evaluation method based on multi-feature parameter fusion and FCM-HMM according to claim 1, is characterized in that: in described step 1, described strengthening local mean value decomposition technology is based on local mean value proposed in recent years. The improved and enhanced algorithm of Local Mean Decomposition (LMD) realizes the adaptive determination of the optimal step size and the number of screening iterations by optimizing the three key parts of the boundary conditions, envelope evaluation and screening stop criteria in the LMD algorithm. Algorithm performance. 3.根据权利要求1所述的基于多特征参数融合与FCM-HMM的回转支承状态评估方法，其特征在于：所述步骤2中，所述特征领域具体如下：3. the slewing bearing state evaluation method based on multi-feature parameter fusion and FCM-HMM according to claim 1, is characterized in that: in described step 2, described characteristic field is specifically as follows:
Figure FDA0003311169850000011
Figure FDA0003311169850000011
Figure FDA0003311169850000021
Figure FDA0003311169850000021
 其中，①至⑩为时域特征，
Figure FDA0003311169850000022
至
Figure FDA0003311169850000023
为频域特征，
Figure FDA0003311169850000024
为时频域特征；其中，X为降噪后信号，xn为X的第n个点，fi代表i时刻功率谱频率，pi代表功率谱振幅，Ci(t)为信号分解的固有模态函数；选用单调性指标和轨迹差异性指标定量地对回转支承各物理信号特征进行有效筛选；由于这两种量化评价指标分别从不同角度表征特征属性，为了全面且直观筛选相同物理信号下的优胜特征，提出将两者线性叠加的混合评价方式，即综合评价指标Score，具体为：Among them, ① to ⑩ are time domain features,
Figure FDA0003311169850000022
to
Figure FDA0003311169850000023
is the frequency domain feature,
Figure FDA0003311169850000024
is the time-frequency domain feature; where X is the signal after noise reduction, x n is the nth point of X, f i represents the frequency of the power spectrum at time i , pi represents the amplitude of the power spectrum, and C i (t) is the signal decomposition Intrinsic modal function; the monotonicity index and the trajectory difference index are used to quantitatively and effectively screen the characteristics of each physical signal of the slewing bearing; since these two quantitative evaluation indicators characterize the characteristic attributes from different angles, in order to comprehensively and intuitively screen the same physical signal According to the winning features below, a hybrid evaluation method that linearly superimposes the two is proposed, that is, the comprehensive evaluation index Score, specifically: Score＝map min max(Mon)+map min max(Div)；Score=map min max(Mon)+map min max(Div); 为了避免两种量化指标的值域差异对筛选结果造成的负面影响，在线性叠加前统一进行归一化处理，即式中mapminmax操作。In order to avoid the negative impact of the difference in the range of the two quantitative indicators on the screening results, normalization is performed uniformly before the linear superposition, that is, the mapminmax operation in the formula. 4.根据权利要求1所述的基于多特征参数融合与FCM-HMM的回转支承状态评估方法，其特征在于：所述步骤3中，选用等距离映射算法建立回转支承振动、温度、扭矩信号健康指标，而降维算法中需首要解决的问题在于本征维数的估计，即在充分保留高维数据原始信息的条件下，确定低维空间的最小嵌入维度；合理的本征维数估计至关重要；使用经典维数估计方法——极大似然估计法(MLE)；通过建立近邻距离的似然函数，实现本征维度的极大似然估计。4. the slewing bearing state assessment method based on multi-feature parameter fusion and FCM-HMM according to claim 1, is characterized in that: in described step 3, selects equidistance mapping algorithm to establish slewing bearing vibration, temperature, torque signal health The first problem to be solved in the dimensionality reduction algorithm is the estimation of the intrinsic dimension, that is, to determine the minimum embedding dimension of the low-dimensional space under the condition of fully retaining the original information of the high-dimensional data; a reasonable intrinsic dimension is estimated to It is very important to use the classical dimension estimation method - Maximum Likelihood Estimation (MLE); by establishing the likelihood function of the nearest neighbor distance, the maximum likelihood estimation of the intrinsic dimension is realized. 5.根据权利要求1所述的基于多特征参数融合与FCM-HMM的回转支承状态评估方法，其特征在于：所述步骤4中，FCM-HMM模型的建立分为两个步骤：划分状态矩阵和建立退化状态识别库；基于FCM-HMM的退化状态评估具体步骤如下：5. the slewing bearing state evaluation method based on multi-feature parameter fusion and FCM-HMM according to claim 1, is characterized in that: in described step 4, the establishment of FCM-HMM model is divided into two steps: divide state matrix and establish a degradation state identification library; the specific steps of degradation state evaluation based on FCM-HMM are as follows: 1)划分状态矩阵；对观测序列采用FCM方法进行无监督聚类，划分状态数K对应的状态矩阵；FCM中初始参数需预设，聚类数由全寿命加速试验结果确定，具有一定的主观性，同时初始隶属度随机选取，很难实现临界点类别的准确判定；因此，按比例对状态矩阵进行选取，构造观测向量On，避免临界点对模型的影响；另外，观测序列为全寿命周期序列，对于观测序列的输入类别进行研究，对比讨论基于不同观测序列的退化状态评估结果；1) Divide the state matrix; use the FCM method to perform unsupervised clustering on the observation sequence, and divide the state matrix corresponding to the number of states K; the initial parameters in the FCM need to be preset, and the number of clusters is determined by the results of the full-life accelerated test, which has a certain subjective At the same time, the initial membership degree is randomly selected, so it is difficult to accurately determine the critical point category; therefore, the state matrix is selected proportionally, and the observation vector On is constructed to avoid the impact of the critical point on the model; in addition, the observation sequence is a full life cycle. Sequence, study the input category of the observation sequence, and compare and discuss the evaluation results of the degradation state based on different observation sequences; 2)建立退化状态识别库；基于构造的观测向量On，初始化模型参数，采用Baum-welch算法不断优化模型参数，获取每个状态的最优模型HMMn(λn)；建立的FCM-HMM退化状态识别库用于划分当前观测序列的退化阶段，确定状态转移过程；2) Establish a degradation state identification library; initialize the model parameters based on the constructed observation vector On, and use the Baum-welch algorithm to continuously optimize the model parameters to obtain the optimal model HMMn(λn) for each state; the established FCM-HMM degradation state identification The library is used to divide the degradation stage of the current observation sequence and determine the state transition process; 3)评估回转支承当前退化状态；将当前观测序列输入至FCM-HMM退化状态识别库中，采用前-后向算法计算每个模型HMMn(λn)输出的对数似然概率p(O|λn)，比较n＝arg max[p(O|λn)]，n即为回转支承当前退化状态；若n＝K，表明回转支承已进入全寿命周期的快速衰退阶段，在此后的某一时刻可能发生失效卡死现象，故此阶段应进行失效预警，避免因设备突然失效造成的经济损失。3) Evaluate the current degradation state of the slewing bearing; input the current observation sequence into the FCM-HMM degradation state recognition library, and use the forward-backward algorithm to calculate the log-likelihood probability p(O|λn) of the output of each model HMMn(λn) ), compare n=arg max[p(O|λn)], n is the current degradation state of the slewing bearing; if n=K, it means that the slewing bearing has entered the rapid decay stage of the whole life cycle, and it may be possible at a certain time after that The phenomenon of failure and stuck occurs, so the failure warning should be carried out at this stage to avoid economic losses caused by the sudden failure of the equipment.
CN202111217266.8A 2021-10-19 2021-10-19 A state evaluation method of slewing bearing based on multi-feature parameter fusion and FCM-HMM Pending CN114046993A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111217266.8A CN114046993A (en) 2021-10-19 2021-10-19 A state evaluation method of slewing bearing based on multi-feature parameter fusion and FCM-HMM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111217266.8A CN114046993A (en) 2021-10-19 2021-10-19 A state evaluation method of slewing bearing based on multi-feature parameter fusion and FCM-HMM

Publications (1)

Publication Number Publication Date
CN114046993A true CN114046993A (en) 2022-02-15

Family

ID=80205512

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111217266.8A Pending CN114046993A (en) 2021-10-19 2021-10-19 A state evaluation method of slewing bearing based on multi-feature parameter fusion and FCM-HMM

Country Status (1)

Country Link
CN (1) CN114046993A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114813119A (en) * 2022-03-18 2022-07-29 大连理工大学 A performance degradation evaluation method of micro-turbine bearing based on multi-source information fusion
CN115758277A (en) * 2022-11-30 2023-03-07 重庆忽米网络科技有限公司 Online health state evaluation method for rotary equipment
CN118364247A (en) * 2024-06-20 2024-07-19 青岛哈尔滨工程大学创新发展中心 Method for predicting service life of guide tube propeller of vector propeller

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106885697A (en) * 2017-03-17 2017-06-23 华东交通大学 The performance degradation assessment method of the rolling bearing based on FCM HMM
CN109506938A (en) * 2018-11-21 2019-03-22 昆明理工大学 Based on the failure extracting method for improving wavelet packet threshold denoising and local mean value decomposition
CN111751117A (en) * 2020-04-10 2020-10-09 中国人民解放军92228 部队 Special vehicle health state assessment method and device
CN112001632A (en) * 2020-08-25 2020-11-27 中国船舶重工集团海装风电股份有限公司 Wind turbine generator gearbox bearing performance degradation state evaluation method
CN113469107A (en) * 2021-07-16 2021-10-01 辽宁大学 Bearing fault diagnosis method fusing spatial density distribution

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106885697A (en) * 2017-03-17 2017-06-23 华东交通大学 The performance degradation assessment method of the rolling bearing based on FCM HMM
CN109506938A (en) * 2018-11-21 2019-03-22 昆明理工大学 Based on the failure extracting method for improving wavelet packet threshold denoising and local mean value decomposition
CN111751117A (en) * 2020-04-10 2020-10-09 中国人民解放军92228 部队 Special vehicle health state assessment method and device
CN112001632A (en) * 2020-08-25 2020-11-27 中国船舶重工集团海装风电股份有限公司 Wind turbine generator gearbox bearing performance degradation state evaluation method
CN113469107A (en) * 2021-07-16 2021-10-01 辽宁大学 Bearing fault diagnosis method fusing spatial density distribution

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
WANG，SAISAI等: "Degradation evaluation of slewing bearing using HMM and improved GRU", 《MEASUREMENT》 *
乔宁国: "基于多传感器数据融合的高速列车传动系统故障诊断与健康状态预测", 《中国博士学位论文全文数据库 工程科技Ⅱ辑》 *
周建民等: "基于非线性降维和模糊均值聚类的滚动轴承的性能退化在线评估方法", 《机械设计与研究》 *
田淑华等: "基于多特征集融合与多变量支持向量回归的回转支承剩余寿命评估", 《南京工业大学学报(自然科学版)》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114813119A (en) * 2022-03-18 2022-07-29 大连理工大学 A performance degradation evaluation method of micro-turbine bearing based on multi-source information fusion
CN115758277A (en) * 2022-11-30 2023-03-07 重庆忽米网络科技有限公司 Online health state evaluation method for rotary equipment
CN118364247A (en) * 2024-06-20 2024-07-19 青岛哈尔滨工程大学创新发展中心 Method for predicting service life of guide tube propeller of vector propeller

Similar Documents

Publication Publication Date Title
Feng et al. Digital twin enabled domain adversarial graph networks for bearing fault diagnosis
CN114046993A (en) A state evaluation method of slewing bearing based on multi-feature parameter fusion and FCM-HMM
Wei et al. A novel intelligent method for bearing fault diagnosis based on affinity propagation clustering and adaptive feature selection
Li et al. Bearing fault diagnosis using fully-connected winner-take-all autoencoder
Yang et al. Bearing fault automatic classification based on deep learning
Li et al. Bearing fault feature selection method based on weighted multidimensional feature fusion
CN112257530B (en) Rolling bearing fault diagnosis method based on blind signal separation and support vector machine
Ma et al. Scattering transform and LSPTSVM based fault diagnosis of rotating machinery
CN110823576B (en) Mechanical anomaly detection method based on generation of countermeasure network
CN111506862B (en) Rolling bearing fault diagnosis method based on multisource weighting integrated transfer learning
Zhao et al. A novel unsupervised directed hierarchical graph network with clustering representation for intelligent fault diagnosis of machines
CN104713728B (en) Large slewing bearing residual life online prediction method based on multidimensional data drive
Zhang et al. A novel data-driven method based on sample reliability assessment and improved CNN for machinery fault diagnosis with non-ideal data
CN114742115B (en) Method for constructing fault diagnosis model of rolling bearing and diagnosis method
CN109932174A (en) A fault diagnosis method for gearboxes based on multi-task deep learning
Chen et al. A visualized classification method via t-distributed stochastic neighbor embedding and various diagnostic parameters for planetary gearbox fault identification from raw mechanical data
Yan et al. Fault diagnosis of rolling‐element bearing using multiscale pattern gradient spectrum entropy coupled with Laplacian score
CN113112000B (en) Construction method and application of multilayer convolutional neural network model
Attoui Novel fast and automatic condition monitoring strategy based on small amount of labeled data
Xu et al. Constructing a health indicator for bearing degradation assessment via an unsupervised and enhanced stacked autoencoder
Decker et al. Does your model think like an engineer? explainable ai for bearing fault detection with deep learning
Tan et al. Fault diagnosis method for rolling bearing based on VMD and improved SVM optimized by METLBO
CN114235408B (en) Bearing Fault Diagnosis Method and System Based on Improved Cascaded Variational Mode Decomposition
Fan et al. Performance degradation assessment of rolling bearing cage failure based on enhanced CycleGAN
CN118518359B (en) A bearing fault diagnosis method based on dual domain adaptive neural network

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20220215

RJ01 Rejection of invention patent application after publication