CN109141881B - Rotary machine health assessment method of deep self-coding network - Google Patents

Rotary machine health assessment method of deep self-coding network Download PDF

Info

Publication number
CN109141881B
CN109141881B CN201810736521.1A CN201810736521A CN109141881B CN 109141881 B CN109141881 B CN 109141881B CN 201810736521 A CN201810736521 A CN 201810736521A CN 109141881 B CN109141881 B CN 109141881B
Authority
CN
China
Prior art keywords
health
value
features
self
deep self
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810736521.1A
Other languages
Chinese (zh)
Other versions
CN109141881A (en
Inventor
贾民平
佘道明
许飞云
胡建中
黄鹏
鄢小安
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.)
Southeast University
Original Assignee
Southeast University
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 Southeast University filed Critical Southeast University
Priority to CN201810736521.1A priority Critical patent/CN109141881B/en
Publication of CN109141881A publication Critical patent/CN109141881A/en
Application granted granted Critical
Publication of CN109141881B publication Critical patent/CN109141881B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G01M13/045—Acoustic or vibration analysis
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01K—MEASURING TEMPERATURE; MEASURING QUANTITY OF HEAT; THERMALLY-SENSITIVE ELEMENTS NOT OTHERWISE PROVIDED FOR
    • G01K7/00—Measuring temperature based on the use of electric or magnetic elements directly sensitive to heat ; Power supply therefor, e.g. using thermoelectric elements
    • G01K7/02—Measuring temperature based on the use of electric or magnetic elements directly sensitive to heat ; Power supply therefor, e.g. using thermoelectric elements using thermoelectric elements, e.g. thermocouples
    • 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
    • G01M13/00—Testing of machine parts
    • G01M13/02—Gearings; Transmission mechanisms
    • G01M13/028—Acoustic or vibration analysis
    • 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
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks

Abstract

The invention discloses a rotating machinery health assessment method of a depth self-coding network, which comprises the following steps of 1, collecting vibration signals; step 2, extracting original features; step 3, performing feature dimension reduction by adopting a deep self-encoding network DAE; step 4, feature selection; step 5, adopting an unsupervised SOM algorithm to construct health indexes; and 6, evaluating the health indexes by adopting a fusion evaluation criterion based on a genetic algorithm. The invention combines the advantages of strong feature extraction capability of deep learning and combines the depth self-coding with the minimum quantization error method. In addition, a fusion evaluation criterion based on genetic algorithm is provided for the problem that an evaluation criterion based on one metric often has a deviation. The method can accurately evaluate the health state of the rotary machine, can be widely applied to the health evaluation of the rotary machine in the fields of chemical industry, metallurgy, electric power, aviation and the like, can accurately describe the dynamic process of performance degradation of the parts, and can predict the residual life.

Description

Rotary machine health assessment method of deep self-coding network
Technical Field
The invention relates to a rotating machine health assessment technology, in particular to a rotating machine health assessment method of a deep self-coding network.
Background
Due to the development of advanced sensor and computer technologies, which accumulate a large amount of condition monitoring data in industrial production, data-driven methods are widely used in bearing prediction because they can quantify the degradation process using the condition monitoring data, rather than building an accurate system model that is not readily available.
In general, a data-driven prediction method generally consists of three steps: data acquisition, health index construction and residual life prediction. Health indicators attempt to identify and quantify history and ongoing degradation processes by extracting feature information from the acquired data. Therefore, the quality of the constructed health indicators directly affects the effectiveness of the data-driven prediction method to a large extent. From this perspective, it is important to construct a health index that effectively reflects the degradation of mechanical equipment. Rotating parts, such as bearings, gears, rotors and the like, which are common in industrial fields, are important components in rotating machinery, and the health condition of the rotating parts directly influences whether the rotating machinery can normally operate. The critical parts are damaged seriously, which can cause production shutdown and bring huge economic loss, therefore, the accurate evaluation of the health condition of the critical parts is of great significance for the safe and reliable operation of the equipment.
According to the construction strategy of the health indexes, the existing rotating machinery health indexes can be divided into two types: physical health indicators and virtual health indicators. The time-frequency domain feature extraction method is usually a method for calculating dimensional and dimensionless parameter indexes of time-frequency domain waveforms, is the most convenient feature extraction method for mechanical vibration signals, and the parameter indexes are closely related to the operating speed, load and other working conditions of mechanical equipment. However, it is often not convincing to evaluate the degradation process of the rotating machine by means of only one characteristic, and there is a certain deviation. Principal Component Analysis (PCA), one of the most popular dimension reduction techniques at present, is often applied in the process of constructing virtual health indicators. However, PCA is a linear dimensionality reduction method, and the bearing degradation process is a nonlinear degradation process, and therefore cannot be used for accurately evaluating the rotating machine health index. In addition, health indexes constructed by combining the PCA and the time-frequency domain feature extraction method are to be improved in monotonicity, trend and robustness.
Disclosure of Invention
The invention provides a rotating machine health assessment method of a deep self-coding network, aiming at the defects of the prior art, and the rotating machine health assessment method of the deep self-coding network can improve the quality of the constructed health index, effectively assess the health condition of the rotating machine and further improve the accuracy of the residual life prediction of the rotating machine.
In order to solve the technical problems, the invention adopts the technical scheme that:
a rotating machine health assessment method of a deep self-coding network comprises the following steps.
Step 1, vibration signal acquisition: vibration signals of key parts of a rotating machine are acquired.
Step 2, extracting original features: and (3) performing primary extraction on the life-cycle original characteristics of the key components on the vibration signals acquired in the step (1).
Step 3, feature dimension reduction: and (3) taking the original features extracted in the step (2) as the input of a deep self-coding network (DAE), and carrying out nonlinear dimensionality reduction on the original features by the deep self-coding network (DAE) to obtain a compressed vector Z.
And 4, selecting characteristics: and (4) sorting the features in the compressed vector Z obtained in the step (3) according to the trend values, and selecting the features with the trend value larger than 0.8 to form a feature subset.
Step 5, health indexes are constructed: and 4, fusing the feature subsets selected in the step 4 into a one-dimensional health value by adopting an unsupervised SOM algorithm to obtain a health index of the whole life of the key parts of the rotary machine.
Step 6, health index evaluation: and (5) evaluating the health indexes constructed in the step (5) by adopting a fusion evaluation criterion based on a genetic algorithm.
In step 6, the specific method for evaluating the health index by adopting the fusion evaluation criterion comprises the following steps: and searching the maximum value of the fitness function by using a genetic algorithm as a standard for evaluating the health index, wherein the larger the fitness function value is, the better the constructed health index is.
Maximum value of fitness function
Figure BDA0001722167110000021
The calculation formula is expressed as follows:
Figure BDA0001722167110000022
Figure BDA0001722167110000023
wherein the content of the first and second substances,
Figure BDA0001722167110000024
Figure BDA0001722167110000025
Figure BDA0001722167110000026
Y(tk)＝YT(tk)+XR(tk) (6)
wherein, fitness is moderate function value, corr (Y (t)k),T(tk) Mon (Y (t)) as a trend valuek) Has a monotonicity value of rob (Y (t)k) ) is a robustness value; y (t)k) Indicates a health index, YT(tk) Trend part, X, representing health indexR(tk) Is a random part of the health indicator, K represents the length of the time series,
Figure BDA0001722167110000027
and
Figure BDA0001722167110000028
respectively, the health index Y (t) from 1 to Kk) And a time vector T (T)k) The mean value of (a); dY (t)k) Representing the derivative of the health indicator at time K, YstartAnd YendRespectively represent health indexes Y (t)k) Start and end values of, omegaiAnd a weight representing each evaluation index value.
In step 5, the one-dimensional health value formed by fusion is the minimum quantization error MQE value, and the smaller the value MQE is, the closer the current state is to the reference health state is; MQE, the larger the value, the farther the current state deviates from the baseline health state.
The minimum quantization error MQE is calculated as:
MQE＝||h-wBMU|| (7)
in the formula, wBMURepresenting the optimal matching unit, wherein h is a real-time state feature vector; the BMU is the best matching neuron, and the Euclidean distance between the weight vector of the best matching neuron and the input vector in the SOM algorithm is the minimum.
Key components of rotary machines include bearings, gears or rotors.
In step 2, the life-cycle original features of the key components comprise 16 time domain features, 13 frequency domain features, 17 time domain features and 2 features based on a trigonometric function; the 2 characteristics based on the trigonometric function are an inverse-triangular hyperbolic cosine standard and an inverse-triangular hyperbolic sine standard deviation respectively.
The invention has the following beneficial effects:
1. the trend values corr (Y, T), the monotonicity value mon (Y) and the robustness value rob (Y) of the health index constructed by the method are all larger than a single-layer self-coding model and a traditional PCA dimension reduction method. The reason is that deep self-encoding has the ability to greedy learn the non-linear characteristics of the bearing's full life, layer by layer, strongly. The redundant characteristics are effectively removed, and the defect of the traditional linear dimension reduction is overcome.
2. The health index with a single characteristic cannot completely reflect the actual degradation state of the rotary machine, and the fusion of multiple health indexes based on the MQE method provided by the invention can better represent the actual degradation state of the rotary machine.
3. The health index evaluation criterion based on a measurement is often deviated, the multi-evaluation criterion problem is designed into a combined constraint optimization problem, and the proposed fusion evaluation criterion based on the genetic algorithm is more persuasive than a single evaluation criterion.
4. The method can accurately evaluate the health state of the rotary machine, is simple and feasible, and can be widely applied to the health evaluation and the residual life prediction of the rotary machine in the fields of chemical industry, metallurgy, electric power, aviation and the like.
Drawings
FIG. 1 is a flow chart of a method for evaluating health of a rotating machine in a deep self-coding network according to the present invention.
FIG. 2 shows a time domain waveform of a bearing vibration signal.
Fig. 3 shows a relationship diagram of a trend value and a feature number when the feature of the bearing is selected.
FIG. 4 shows a bearing life-cycle degradation health indicator curve.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and specific preferred embodiments.
As shown in FIG. 1, a method for evaluating health of a rotating machine of a deep self-coding network comprises the following steps.
Step 1, vibration signal acquisition: vibration signals of key parts of a rotating machine are acquired.
Key components of rotary machines include bearings, gears or rotors, etc.
The method for acquiring the vibration signals of the key components is the prior art, and in the application, the bearings are taken as an example, and the following preferred method is adopted for acquisition.
Step 11, bearing installation: the four bearings are simultaneously installed on a bearing life strengthening test bed, the type of the bearing life strengthening test bed is preferably ABLT-1A, and the bearing life strengthening test bed comprises a test head, a test head seat, a transmission system, a loading system, a lubricating system, an electrical control system and a test and data acquisition system. The bearing life strengthening test bed can be used for simultaneously mounting four bearings to carry out accelerated fatigue life tests.
Step 12, distribution of an acquisition system: the data acquisition system of the bearing life strengthening test bed comprises four thermocouples and three acceleration sensors, wherein the four thermocouples are used for picking up temperature signals of four bearing outer rings, the three acceleration sensors respectively pick up vibration signals on three rigid body shells, the first acceleration sensor is used for acquiring bearing data of 1 station, the second acceleration sensor is used for acquiring bearing data of 2 stations and 3 stations, and the 3 rd acceleration sensor is used for acquiring bearing data of 4 stations.
Step 13, bearing loading: the loading conditions for the test are shown in table 1, where the dynamic load rating of the 6308 bearing is 42.3kN, and the actual weight is 30kg, that is, the dynamic load rating of each bearing is 15 kN.
TABLE 1 test conditions for full Life test
Figure BDA0001722167110000041
Step 14, vibration signal acquisition: in the loading process, the acquisition system automatically acquires vibration signals of the four bearings. At full load, as shown in table 2, the bearing was run for 9 hours, and finally the tester was shut down because the root mean square of vibration reached the shut down threshold. (the rms vibration value at full load was set to 5.0 and the shutdown threshold was set to 20). Then, the rolling bodies are obviously peeled off after the bearing is cut by linear cutting, and the time domain waveform of the vibration signal of the tested 6308 bearing is shown in fig. 2.
TABLE 2 radial load conditions
Figure BDA0001722167110000042
Step 2, extracting original features: and (3) performing primary extraction on the life-cycle original characteristics of the key components on the vibration signals acquired in the step (1). The preliminary extraction method is the prior art, and the preliminary extracted life-cycle original features of the key components mainly comprise time domain, frequency domain, time-frequency domain and other features.
The full-life original features of the key components comprise 16 time domain features, 13 frequency domain features, 17 time-frequency domain features and 2 trigonometric function-based features. The 2 characteristics based on the trigonometric function are an inverse-triangular hyperbolic cosine standard and an inverse-triangular hyperbolic sine standard deviation respectively.
And 3, reducing the dimension of the features.
The deep self-coding network DAE is an artificial neural network for learning high-efficiency coding proposed by Hinton, and the purpose of data dimension reduction can be achieved by obtaining compressed coding of a data set through learning. The deep self-coding network can gradually convert concrete feature vectors into abstract feature vectors.
In the application, the original features extracted in the step 2 are used as the input of the deep self-coding network DAE, and the deep self-coding network DAE performs nonlinear dimensionality reduction on the original features to obtain the compressed vector Z.
The method specifically comprises the following steps:
and 31, inputting the preliminarily extracted life-cycle original features into the deep self-coding network DAE.
And step 32, the deep self-coding network DAE learns the nonlinear characteristics of the whole life of the bearing layer by layer, and then the whole deep neural network is pre-trained.
The deep self-coding network is a neural network composed of a plurality of layers of self-encoders, and the output of the self-encoder of the previous layer is used as the input of the self-encoder of the next layer. For an n-layer depth self-coding neural network, assuming that W (k,1), W (k,2), b (k,1), b (k,2) are used to represent the W (1), W (2), b (1), b (2) parameters corresponding to the kth self-coder, the coding process of the self-coding neural network is to perform the coding steps of each layer of self-coders in the order from front to back:
a(k)＝f(z(k))
wherein z is(k+1)＝W(k,1)a(k)+b(k,1)
Wherein f (-) is an activation function, a(k)The output of the k layer of the DAE coding layer.
Similarly, the decoding process of the deep neural network is to execute the decoding steps of the self-encoder of each layer in the order from back to front:
a(n+k)＝f(z(n+k))
wherein z is(n+k+1)＝W(n+k,2)a(n+k)+b(n+k,2)
Wherein f (-) is an activation function, W(n+k,2)，b(n+k,2)Weights and offsets for the n + k th layer of the DAE network decoding layer, a(n+k)Is the output of the n + k th layer of the decoding layer. a is(n)Is the activation value of the deepest hidden unit and is a higher order representation of the input value.
The DAE trains each layer of the network in turn by a layer-by-layer greedy training method, and thus pre-trains the entire deep neural network.
And step 33, finally performing global fine adjustment. The intermediate compressed vector Z is the feature obtained by nonlinear dimensionality reduction of the DAE.
The fine tuning is a strategy of deep learning, and can improve the performance of the deep self-coding neural network. In the fine tuning process, all layers of the whole self-coding neural network are regarded as a model, and the parameters in the model are uniformly corrected. A common way to do global trimming is to back-propagate the error.
And 4, selecting characteristics: and (4) sorting the features in the compressed vector Z obtained in the step (3) according to the trend values, and selecting the features with the trend value larger than 0.8 to form a feature subset. In the present invention, the compressed vector feature selection of 6308 bearing is shown in FIG. 3.
The specific process of feature selection comprises the following steps.
And step 41, calculating a trend value of each feature according to the features in the compressed vector Z obtained in the step 3, wherein the trend value calculation method preferably adopts the formula (3) in the step 6 for calculation.
And step 42, sorting the calculated trend values of each feature according to the trend values.
Step 43, selecting the features with the trend value greater than 0.8 to form a feature subset, and selecting the 1 st, 2 nd, and 5 th features corresponding to the present example, as shown in fig. 3.
Step 5, health indexes are constructed: and 4, fusing the feature subsets selected in the step 4 into a one-dimensional health value by adopting an unsupervised SOM algorithm to obtain a health index of the whole life of the key parts of the rotary machine.
The health index construction method specifically comprises the following steps.
Step 51, training the SOM structure by using the characteristic value of the rotating mechanical part in normal as input data.
Step 52, training the SOM neural network to reach a steady state.
Step 53, calculating the real-time status and the best matching unit wBMUI.e. the health of the rotating machine part at that time.
In the step 5, the one-dimensional health value formed by fusion is the minimum quantization error MQE value, and the smaller the value MQE is, the closer the current state is to the reference health state is; MQE, the larger the value, the farther the current state deviates from the baseline health state.
The minimum quantization error MQE is calculated as:
MQE＝||h-wBMU|| (7)
in the formula, wBMUExpressing an optimal matching unit, and obtaining by training the SOM neural network by adopting a formula (8); h is a real-time state feature vector, namely the feature with the selected trend value larger than 0.8; the BMU is the best matching neuron, and the Euclidean distance between the weight vector of the best matching neuron and the input vector in the SOM algorithm is the minimum.
The SOM neural network can transform an input pattern of arbitrary dimensions to a discrete space of one or two dimensions in a topologically ordered manner. When the Euclidean distance between the weight vector of a certain neuron and the input vector is minimum, the neuron is called as a best matching neuron BMU.
In the field of winning neurons, the weights of all neurons are adjusted appropriately to increase the discrimination function value for the input mode, i.e. the weight vector of the output layer neurons is changed with the input vector. The formula for updating the neuron weight in the field is shown as formula (8).
wi(n+1)＝wi(n)+η(n)hj,i(x)(n)h-wi(n)) (8)
In the formula, wi(n) is the weight vector of neuron i at time n; w is ai(n +1) weight vector of neuron i at time n +1, η (n) is learning rate and gradually decreases with time n, hj,i(x)And (n) is a predefined domain function, and h is a real-time state feature vector.
Step 6, health index evaluation: and (5) evaluating the health indexes constructed in the step (5) by adopting a fusion evaluation criterion based on a genetic algorithm.
In actual operation and use, the bearing is subjected to continuous accumulation of various stresses in a use environment, so that physical or chemical performance degradation such as fatigue damage, aging, looseness, corrosion, stress deformation and the like occurs. This performance degradation is considered from a quantitative point of view and manifests itself as a gradual deviation of the performance parameters of the bearing from the normal range. Usually, a single evaluation index cannot completely reflect the actual degradation state of the bearing and is often deviated, and the fusion of multiple health indexes (the invention selects the characteristic with the trend value larger than 0.8 for fusion) can better represent the actual degradation state of the bearing.
In step 6, the specific method for evaluating the health index by adopting the fusion evaluation criterion is as follows: and searching the maximum value of the fitness function by using a genetic algorithm as a standard for evaluating the health index, wherein the larger the fitness function value is, the better the constructed health index is.
Maximum value of fitness function
Figure BDA0001722167110000071
The calculation formula is expressed as follows:
Figure BDA0001722167110000072
Figure BDA0001722167110000073
wherein the content of the first and second substances,
Figure BDA0001722167110000074
Figure BDA0001722167110000075
Figure BDA0001722167110000076
Y(tk)＝YT(tk)+XR(tk) (6)
wherein, fitness is moderate function value, corr (Y (t)k),T(tk) Mon (Y (t)) as a trend valuek) Has a monotonicity value of rob (Y (t)k) ) is a robustness value; y (t)k) Expressing the health index for evaluating the bearing degradation process, and subtracting the best fit line from the health index by least square fitting to obtain X in formula (5)R(tk)；YT(tk) Trend part, X, representing health indexR(tk) Is a random part of the health indicator, K represents the length of the time series,
Figure BDA0001722167110000077
and
Figure BDA0001722167110000078
respectively, the health index Y (t) from 1 to Kk) And a time vector T (T)k) The mean value of (a); dY (t)k) Representing the derivative of the health indicator at time K, YstartAnd YendRespectively represent health indexes Y (t)k) Start and end values of, omegaiAnd a weight representing each evaluation index value.
In addition, No. of in the formula (4) represents the number of points in the health index at which the derivative is greater than or less than 0
The health index curve in the present invention is shown in fig. 4.
The monotonicity value, the trend value, the robustness value and the fusion evaluation criterion value of the health index curve calculated by the method are shown in table 3, and AE represents the health index constructed by a single-layer self-encoding method.
TABLE 3 evaluation index values of the three methods
Figure BDA0001722167110000081
The trend value, monotonicity value and robustness value of the health index constructed by the method are all larger than those of a single-layer self-coding model and a traditional PCA dimension reduction method. The actual degradation process of the bearing can be reflected.
In a word, the method combines the advantages of strong feature extraction capability of Deep learning, combines Deep Auto-Encoder (DAE) and Minimum Quantization Error (MQE) methods, uses a Deep self-encoding model to compress and extract original features, sorts the compressed features according to trends, selects features with large trends, and uses the Minimum Quantization Error method to construct health indexes. In addition, the invention provides a fusion evaluation criterion based on genetic algorithm, aiming at the problem that the evaluation criterion based on one metric often has deviation. The method can accurately evaluate the health state of the rotary machine, can be widely applied to the health evaluation of the rotary machine in the fields of chemical industry, metallurgy, electric power, aviation and the like, can accurately describe the dynamic process of performance degradation of the parts, and can predict the residual life.
Although the preferred embodiments of the present invention have been described in detail, the present invention is not limited to the details of the embodiments, and various equivalent modifications can be made within the technical spirit of the present invention, and the scope of the present invention is also within the scope of the present invention.

Claims (5)

1. A rotating machine health assessment method of a deep self-coding network is characterized by comprising the following steps: the method comprises the following steps:
step 1, vibration signal acquisition: collecting vibration signals of key parts of a rotating machine;
step 2, extracting original features: performing primary extraction on the life-cycle original characteristics of the key components on the vibration signals acquired in the step 1;
step 3, feature dimension reduction: taking the original features extracted in the step (2) as the input of a deep self-encoding network DAE, and carrying out nonlinear dimensionality reduction on the original features by the deep self-encoding network DAE to obtain a compressed vector Z;
and 4, selecting characteristics: sorting the features in the compressed vector Z obtained in the step (3) according to the trend values, and selecting the features with the trend values larger than 0.8 to form a feature subset;
step 5, health indexes are constructed: fusing the feature subsets selected in the step 4 into a one-dimensional health value by adopting an unsupervised SOM algorithm to obtain a health index of the whole life of the key part of the rotary machine;
step 6, health index evaluation: evaluating the health index constructed in the step 5 by adopting a fusion evaluation criterion based on a genetic algorithm; the specific method for evaluating the health indexes by adopting the fusion evaluation criterion comprises the following steps: searching the maximum value of the fitness function by using a genetic algorithm as a standard for evaluating the health index, wherein the larger the fitness function value is, the better the constructed health index is; maximum value max of fitness functionY∈ΩThe fitness calculation formula is expressed as follows:
maxY∈Ωfitness＝ω1corr(Y(tk)，T(tk))+ω2mon(Y(tk))+ω3rob(Y(tk)) (1)
Figure FDA0002295973100000011
wherein the content of the first and second substances,
Figure FDA0002295973100000012
Figure FDA0002295973100000013
rob(Y(tk))＝exp(-std(XR(tk))/mean(|Ystart-Yend|)) (5)
Y(tk)＝YT(tk)+XR(tk) (6)
wherein, fitness is moderate function value, corr (Y (t)k)，T(tk) Mon (Y (t)) as a trend valuek) Has a monotonicity value of rob (Y (t)k) ) is a robustness value; y (t)k) Indicates a health index, YT(tk) Indicates health indexTrend part of (1), XR(tk) Is a random part of the health indicator, K represents the length of the time series,
Figure FDA0002295973100000014
and
Figure FDA0002295973100000015
respectively, the health index Y (t) from 1 to Kk) And a time vector T (T)k) The mean value of (a); dY (t)k) Representing the derivative of the health indicator at time K, YstartAnd YendRespectively represent health indexes Y (t)k) Start and end values of, omegaiAnd a weight representing each evaluation index value.
2. The method for health assessment of rotating machinery of a deep self-coding network according to claim 1, wherein: in step 5, the one-dimensional health value formed by fusion is the minimum quantization error MQE value, and the smaller the value MQE is, the closer the current state is to the reference health state is; MQE, the larger the value, the farther the current state deviates from the baseline health state.
3. The method for health assessment of rotating machinery of a deep self-coding network according to claim 2, wherein: the minimum quantization error MQE is calculated as:
MQE＝||h-wBMU|| (7)
in the formula, wBMhRepresenting the optimal matching unit, wherein h is a real-time state feature vector; the BMU is the best matching neuron, and the Euclidean distance between the weight vector of the best matching neuron and the input vector in the SOM algorithm is the minimum.
4. The method for health assessment of rotating machinery of a deep self-coding network according to claim 1, wherein: key components of rotary machines include bearings, gears or rotors.
5. The method for health assessment of rotating machinery of a deep self-coding network according to claim 1, wherein: in step 2, the life-cycle original features of the key components comprise 16 time domain features, 13 frequency domain features, 17 time domain features and 2 features based on a trigonometric function; the 2 trigonometric function-based features are an inverse trigonometric hyperbolic cosine standard deviation and an inverse trigonometric hyperbolic sine standard deviation, respectively.
CN201810736521.1A 2018-07-06 2018-07-06 Rotary machine health assessment method of deep self-coding network Active CN109141881B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810736521.1A CN109141881B (en) 2018-07-06 2018-07-06 Rotary machine health assessment method of deep self-coding network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810736521.1A CN109141881B (en) 2018-07-06 2018-07-06 Rotary machine health assessment method of deep self-coding network

Publications (2)

Publication Number Publication Date
CN109141881A CN109141881A (en) 2019-01-04
CN109141881B true CN109141881B (en) 2020-03-31

Family

ID=64799777

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810736521.1A Active CN109141881B (en) 2018-07-06 2018-07-06 Rotary machine health assessment method of deep self-coding network

Country Status (1)

Country Link
CN (1) CN109141881B (en)

Families Citing this family (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110108456A (en) * 2019-04-16 2019-08-09 东南大学 A kind of rotating machinery health evaluating method of depth convolutional neural networks
CN110132554B (en) * 2019-04-17 2020-10-09 东南大学 Rotary machine fault diagnosis method based on deep Laplace self-coding
CN110175170B (en) * 2019-04-19 2023-11-21 平安科技（深圳）有限公司 Data optimization processing method, device, computer equipment and storage medium
CN110787024B (en) * 2019-06-26 2021-07-20 东南大学 Shoulder joint rehabilitation exoskeleton mechanism adopting unpowered compensation joints
CN111639842B (en) * 2020-05-20 2023-06-20 湖北博华自动化系统工程有限公司 Equipment health assessment method, assessment system and equipment health prediction method
CN111795819B (en) * 2020-06-12 2021-06-22 燕山大学 Gear box fault diagnosis method integrating vibration and current signal collaborative learning
CN112152633B (en) * 2020-09-30 2023-05-16 石家庄铁道大学 Improved convolution self-encoder method and device for vibration signal compression
CN112446415A (en) * 2020-10-09 2021-03-05 山东中医药大学 Fusion subtraction automatic encoder algorithm for image feature extraction
CN112347898B (en) * 2020-11-03 2024-04-09 重庆大学 Rolling bearing health index construction method based on DCAE neural network
CN112633331A (en) * 2020-12-07 2021-04-09 东南大学 Self-encoder-based thermal power plant auxiliary machine health state assessment method
CN112816191B (en) * 2020-12-28 2022-07-29 哈尔滨工业大学 Multi-feature health factor fusion method based on SDRSN
CN112766219A (en) * 2021-01-31 2021-05-07 石家庄铁道大学 Bearing health assessment method and device based on depth separable convolution self-encoder
CN113051809A (en) * 2021-03-12 2021-06-29 中国人民解放军海军工程大学 Virtual health factor construction method based on improved restricted Boltzmann machine
CN112926505B (en) * 2021-03-24 2022-11-11 重庆大学 Rotating machine health index construction method based on DTC-VAE neural network
CN113033102B (en) * 2021-03-30 2023-04-07 西安电子科技大学 Unsupervised learning-based method for evaluating health of cutter head of shield tunneling machine under complex stratum
CN112966400B (en) * 2021-04-23 2023-04-18 重庆大学 Centrifugal fan fault trend prediction method based on multi-source information fusion
CN113434970B (en) * 2021-06-01 2023-01-17 北京交通大学 Health index curve extraction and service life prediction method for mechanical equipment
CN113569903B (en) * 2021-06-09 2024-04-09 西安电子科技大学 Method, system, equipment, medium and terminal for predicting cutter abrasion of numerical control machine tool
CN113418700B (en) * 2021-06-23 2023-04-07 太原理工大学 Intelligent sensor and health state monitoring method for mining belt conveyor transmission system
CN113627088B (en) * 2021-08-23 2024-04-09 上海交通大学 Machine performance degradation evaluation method and system based on gene programming and data fusion
CN114167838B (en) * 2021-12-03 2023-07-07 西安电子科技大学 Multi-scale health assessment and fault prediction method for servo system
CN116449135B (en) * 2023-04-19 2024-01-30 北京航空航天大学 Method and system for determining health state of electromechanical system component and electronic equipment
CN117633517A (en) * 2024-01-25 2024-03-01 南京工业大学 Bearing health index evaluation method based on depth self-encoder and local linear embedding

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005252167A (en) * 2004-03-08 2005-09-15 Tottori Univ Method for diagnosing abnormality of oil transformer
WO2008036921A3 (en) * 2006-09-21 2008-08-21 Impact Technologies Llc Systems and methods for predicting failure of electronic systems and assessing level of degradation and remaining useful life
CN101937207A (en) * 2010-08-27 2011-01-05 上海交通大学 Intelligent visual monitoring and diagnosing method of mechanical equipment state
CN102768115A (en) * 2012-06-27 2012-11-07 华北电力大学 Method for dynamically monitoring health status of wind turbine gearbox in real time
CN105129109A (en) * 2015-09-30 2015-12-09 北京航空航天大学 Method for evaluating health of aircraft aileron actuator system based on multi-fractal theory and self-organizing map (SOM) network
CN105975749A (en) * 2016-04-28 2016-09-28 上海交通大学 Bearing health assessment and prediction method and system
CN107941537A (en) * 2017-10-25 2018-04-20 南京航空航天大学 A kind of mechanical equipment health state evaluation method

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2005252167A (en) * 2004-03-08 2005-09-15 Tottori Univ Method for diagnosing abnormality of oil transformer
WO2008036921A3 (en) * 2006-09-21 2008-08-21 Impact Technologies Llc Systems and methods for predicting failure of electronic systems and assessing level of degradation and remaining useful life
CN101937207A (en) * 2010-08-27 2011-01-05 上海交通大学 Intelligent visual monitoring and diagnosing method of mechanical equipment state
CN102768115A (en) * 2012-06-27 2012-11-07 华北电力大学 Method for dynamically monitoring health status of wind turbine gearbox in real time
CN105129109A (en) * 2015-09-30 2015-12-09 北京航空航天大学 Method for evaluating health of aircraft aileron actuator system based on multi-fractal theory and self-organizing map (SOM) network
CN105975749A (en) * 2016-04-28 2016-09-28 上海交通大学 Bearing health assessment and prediction method and system
CN107941537A (en) * 2017-10-25 2018-04-20 南京航空航天大学 A kind of mechanical equipment health state evaluation method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于深度学习的轴承健康因子无监督构建方法;赵光权等;《仪器仪表学报》;20180630;第39卷(第6期);第82-87页 *

Also Published As

Publication number Publication date
CN109141881A (en) 2019-01-04

Similar Documents

Publication Publication Date Title
CN109141881B (en) Rotary machine health assessment method of deep self-coding network
CN110866314B (en) Method for predicting residual life of rotating machinery of multilayer bidirectional gate control circulation unit network
Ma et al. Discriminative deep belief networks with ant colony optimization for health status assessment of machine
CN110378052B (en) Equipment residual life prediction method considering future working conditions based on cyclic neural network
CN107368809B (en) A kind of bearing fault sorting technique based on rarefaction representation and dictionary learning
Zhang et al. Feature selection for high-dimensional machinery fault diagnosis data using multiple models and radial basis function networks
CN110132554B (en) Rotary machine fault diagnosis method based on deep Laplace self-coding
CN103115789B (en) Second generation small-wave support vector machine assessment method for damage and remaining life of metal structure
Ma et al. Bearing degradation assessment based on weibull distribution and deep belief network
CN108304348B (en) Bearing residual life prediction method based on binary wiener process
CN112633098B (en) Rotary machine fault diagnosis method, system and storage medium
CN114429153A (en) Lifetime learning-based gearbox increment fault diagnosis method and system
CN112834211A (en) Fault early warning method for transmission system of wind turbine generator
CN113570138A (en) Method and device for predicting residual service life of equipment of time convolution network
CN109710955A (en) Based on LCD-recurrence quantification analysis rolling bearing fault diagnosis and health evaluating method
CN111337244A (en) Method and device for monitoring and diagnosing faults of input shaft of fan gearbox
CN114462508A (en) Power transformer health state assessment method based on multi-mode neural network
CN112949402A (en) Fault diagnosis method for planetary gear box under minimum fault sample size
CN114091525A (en) Rolling bearing degradation trend prediction method
Sadoughi et al. A deep learning approach for failure prognostics of rolling element bearings
CN111382792B (en) Rolling bearing fault diagnosis method based on double-sparse dictionary sparse representation
Melendez et al. Self-supervised Multi-stage Estimation of Remaining Useful Life for Electric Drive Units
Tang et al. A fault diagnosis method using Interval coded deep belief network
CN116611184A (en) Fault detection method, device and medium for gear box
CN116227586A (en) Meta learning fault diagnosis method and device based on depth residual error shrinkage prototype 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
GR01 Patent grant
GR01 Patent grant