CN113505639B - Rotary machine multi-parameter health state assessment method based on TPE-XGBoost - Google Patents
Rotary machine multi-parameter health state assessment method based on TPE-XGBoost Download PDFInfo
- Publication number
- CN113505639B CN113505639B CN202110590308.6A CN202110590308A CN113505639B CN 113505639 B CN113505639 B CN 113505639B CN 202110590308 A CN202110590308 A CN 202110590308A CN 113505639 B CN113505639 B CN 113505639B
- Authority
- CN
- China
- Prior art keywords
- tpe
- parameter
- xgboost
- model
- sample
- 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
- 238000000034 method Methods 0.000 title claims abstract description 78
- 230000036541 health Effects 0.000 title claims abstract description 30
- 238000012549 training Methods 0.000 claims abstract description 34
- 238000005457 optimization Methods 0.000 claims abstract description 19
- 238000001914 filtration Methods 0.000 claims abstract description 6
- 238000009826 distribution Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 9
- 239000000203 mixture Substances 0.000 claims description 7
- 238000009827 uniform distribution Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000003745 diagnosis Methods 0.000 abstract description 4
- 238000013210 evaluation model Methods 0.000 abstract 2
- 238000010801 machine learning Methods 0.000 description 11
- 238000012544 monitoring process Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 9
- 230000015556 catabolic process Effects 0.000 description 7
- 238000006731 degradation reaction Methods 0.000 description 7
- 230000003862 health status Effects 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000001960 triggered effect Effects 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000001050 lubricating effect Effects 0.000 description 1
- 239000010687 lubricating oil Substances 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000003921 oil Substances 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000009751 slip forming Methods 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- 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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Probability & Statistics with Applications (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
A rotary machine multi-parameter health state assessment method based on TPE-XGBoost includes the steps that firstly, a Butterworth filter is used for filtering vibration signals collected on mechanical equipment to obtain two parts higher than 1KHz and lower than 1KHz, and then five vibration sensitive parameters including an effective value, kurtosis factors, peak factors, distortion factors and margin factors are extracted; then performing training sample equalization by using an SMOTE-ENN method; and finally, performing iterative optimization on the XGBoost model appointed hyper-parameter space by using a TPE optimization algorithm, and training under the optimal parameters to obtain an evaluation model. Two actual industrial cases prove that the evaluation model obtained by the invention can output state probability, can fully reflect the health state information of mechanical equipment, and improves the fault diagnosis accuracy.
Description
Technical Field
The invention relates to a health assessment method of rotating machinery, in particular to a TPE-XGBoost-based multi-parameter fusion rotating machinery health assessment method.
Background
Automatic identification of equipment faults is an important research field in the field of industrial Internet. In general, large-sized equipment comprises a plurality of middle-sized and small-sized rotating machine parts, and the health status of the middle-sized and small-sized rotating machines is accurately predicted to be indispensable to the healthy and stable operation of the large-sized equipment. The sensor is installed on the rotating machinery equipment, a plurality of parameters can be monitored, and whether the equipment fails or not can be identified by monitoring certain specific parameters. Fault diagnosis engineers observe a plurality of parameters, analyze and integrate information conveyed by the parameters when diagnosing the equipment, so that the health state of the equipment is accurately estimated. In general, the more parameters that are observed, the more accurate the state of health of the device is reflected by these parameters. But this increases the difficulty of the work of engineers and reduces the monitoring efficiency to some extent. In addition, in order to find out the tiny faults of the machine early, the health state monitoring method can also predict the operation data in a short time in the future according to the previous operation data so as to realize early warning in time.
In order to solve the problems of increased working difficulty and reduced monitoring efficiency caused by simultaneous observation of a large number of parameters, some methods construct a specific health index to represent the health state of equipment according to the observed multiple parameters, and the methods often have high requirements on expert knowledge of engineers. Generally, the more the observed parameters are, the more accurate the reflected health status of the equipment is, but this further increases the difficulty of work and reduces the monitoring efficiency and effectiveness.
In order to improve the problems of increased difficulty in operation and reduced monitoring efficiency due to the simultaneous observation of a large number of parameters, some methods have attempted to construct a certain specified index to reflect the operating state of the device. This method includes both simple and complex exponential methods. The simple exponential method mainly extracts specific characteristic parameters from an original vibration signal to construct a health exponential model, and the complex exponential method is mostly developed based on experimental data, for example, a test method is adopted to calculate the cross-correlation coefficient of a normal bearing vibration signal and a bearing signal with known performance degradation, and the cross-correlation coefficient is used as a health index to represent the health state of a bearing and used for predicting the residual life of the bearing. However, the data sets used for modeling and comparison are experimental data, and thus model generalization may be low.
In recent years, machine learning methods have proven to be effective in solving the above-described problems. For example, the detection and diagnosis of faults are realized by utilizing an ANN technology aiming at the degradation of components and the mechanical abrasion in a production system; and identifying the fault mode by using a support vector machine and a clustering method. The method is well verified in various industrial cases. However, the traditional statistics based on the progressive theory at the basis of the neural network algorithm can only theoretically guarantee the identification performance when the number of learning samples tends to infinity. The machine learning methods such as the support vector machine have higher requirements on the model hyper-parameters, and meanwhile, the problem of unbalanced training samples cannot be effectively solved.
Disclosure of Invention
The invention aims to solve the problems of poor performance, low super-parameter optimization efficiency and low prediction precision of a model when training samples are unbalanced in the prior art. The method provided by the invention is an automatic machine learning method integrating an SMOTE-ENN method, a TPE (Tree-structured Parzen Estimator, tree Parzen estimation) optimization algorithm and XGBoost (eXtreme Gradient Boosting, extremum gradient lifting Tree). The method can effectively solve the problems of unbalanced training samples and few fault samples in the research in the field, constructs a high-performance training set, optimizes the model super-parameters by using a TPE optimization algorithm, and avoids tedious and low-efficiency super-parameter debugging. According to the method, the health state of the in-service equipment can be estimated on line through offline training of the existing full life cycle data samples.
The method mainly comprises the following specific processes in actual application:
(1) Offline training:
a) And acquiring the full life cycle operation data of the equipment, and carrying out a small number of labels on the state of the obvious equipment by combining an expert experience method.
b) Using a high-low pass filtering method to separate two parts of waveform data below 1KHz and above 1KHz for each time period waveform, and extracting five vibration sensitive time domain parameters for the two parts of waveform data: an effective value, a kurtosis factor, a peak factor, a skew factor, and a margin factor.
c) And expanding the labeled data set by using an SMOTE-ENN method to obtain a training set and a testing set of the model.
d) Specifying a target hyper-parameter space for XGBoost optimization, and optimizing an XGBoost model on a training set and a testing set by using a TPE method.
e) And selecting an optimal super-parameter combination to construct an XGBoost model, and training on a training set and a testing set.
f) And storing the obtained optimal model structure.
(2) Online assessment
a) And extracting the same characteristic parameters from the actually monitored operation data.
b) And inputting the characteristic parameters into a model trained in an offline stage, and obtaining probability values x of the equipment in different states.
d) If the probability of x belonging to the early warning state is rapidly increased and breaks through 80%, the device is gradually separated from the normal state, and early failure occurs, and an early warning and alarming mechanism is triggered; if the probability of x belonging to the fault state is rapidly increased and breaks through 80%, the device is indicated to enter a late fault state, an emergency alarm mechanism is triggered, and workers are further guided to operate the device.
2. The invention adopts the following technical scheme:
a rotary machine multi-parameter health state assessment method based on TPE optimization and XGBoost model combines a SMOTE-ENN sample equalization method, a TPE optimization algorithm and an XGBoost model, and utilizes a known full life cycle sample of equipment to assess the health state of the equipment on line.
(1) The SMOTE-ENN method is very suitable for the conditions of unbalanced samples, few fault samples and large normal samples in the fault diagnosis field. The SMOTE-ENN method combines an SMOTE oversampling method and a nearest neighbor (KNN) algorithm, and is mainly used for solving the problem of poor model training effect caused by unbalanced training samples in the machine learning field.
In practical engineering applications, full life cycle samples of known devices are easy to obtain, however, in this case, the vast majority of samples are sample data of devices in a healthy state. For normal samples, the data of equipment degradation and serious fault states are small in proportion, and training of the model directly on the training set can lead to model inertization, namely, higher precision can be obtained by judging that the equipment is in a normal state, so that the machine learning model construction is not facilitated.
Therefore, the training sample construction method can effectively solve the model inerting phenomenon caused by unbalance of the training sample by utilizing the good sample equalization method of SMOTE-ENN, can better construct a machine learning model and better evaluate the health state of equipment.
(2) The basic principle of the TPE optimization algorithm is as follows: the model hyper-parameter configuration space is converted into non-parameter density distribution, and the configuration space can be represented by uniform distribution, logarithmic uniform distribution, quantitative logarithmic uniform distribution and classification variables. The TPE algorithm converts it: uniformly distributing, truncating the Gaussian mixture model, logarithmically uniformly distributing, exponentially truncating the Gaussian mixture model, classifying, and classifying by weight distribution, wherein priori knowledge is continuously formed in the super-parameter optimization process so as to obtain the optimal super-parameter configuration of the model more efficiently.
In the training stage, a main hyper-parameter search range of a model and an objective function optimized by a TPE algorithm are specified first. And continuously iterating the machine learning model in a training sample constructed by the SMOTE-ENN method, and finishing optimization after the iteration times are finished or after the appointed condition is reached. And after the optimization is finished, obtaining the optimal super-parameters of the corresponding machine learning model.
(2) The XGBoost model belongs to an integrated model in a machine learning model, is an improved boosting algorithm of GBDT, and has the advantages of high training speed, high prediction accuracy and the like.
Inputting the model hyper-parameters obtained through the TPE optimization algorithm into the XGBoost, and training and learning in a training sample. At the same time, to prevent model over-training, resulting in over-fitting, an early stop mechanism should also be introduced. After the model is learned, the XGBoost model with the optimal structure can be obtained.
And inputting the monitoring data of the in-service equipment into the model to obtain probability values of the equipment in different health states. The smaller the normal state probability, the more the device operation deviates from the normal state.
3. The invention has the following advantages and outstanding effects:
the method provided by the invention can be effectively applied to equipment health monitoring under the conditions of small samples, few samples and unbalanced samples; the method can predict different states of the equipment and give out probability values of the equipment in different states; the method provided by the invention can avoid the complex super-parameter optimization process of the machine learning model and the problem of low prediction performance of the machine learning model caused by poor super-parameter selection; the method provided by the invention has novel thought, combines the advantages of SMOTE-ENN, TPE and XGBoost models, and enables the equipment state evaluation to be more accurate; the high-frequency data and the low-frequency data are separated by using a filtering method, and corresponding vibration sensitive characteristic parameters are extracted, so that the model has strong robustness and generalization. In practical engineering application, the method successfully predicts that a certain practical rotating machine enters a degradation period from a normal state to a failure period under the condition that an expert cannot predict the actual rotating machine efficiently and accurately by a traditional method.
In conclusion, the invention is simple and feasible, has high flexibility and wide application range, and is convenient to apply to engineering practice.
Drawings
FIG. 1 is a training flow chart of a rotary machine multi-parameter health state assessment method based on TPE-XGBoost according to the present invention.
FIG. 2 is a timing chart of extracted feature parameters of a rotary machine multi-parameter health assessment method embodiment-A4005 pump based on TPE-XGBoost according to the present invention.
FIG. 3 is a schematic diagram of sample equalization using the SMOTE-ENN method of the rotary machine multi-parameter health status assessment method based on TPE-XGBoost of the present invention.
FIG. 4 is a graph showing the results of an application of the pump of example-A4005 of a method for evaluating the health status of a rotating machine based on TPE-XGBoost.
Fig. 5 is a timing chart of extracted characteristic parameters from three groups of three bearings in an IMS center double row ball bearing life test, which is an embodiment of a rotary machine multi-parameter health state evaluation method based on TPE-XGBoost.
Fig. 6 is an application result of three groups of third bearings in an IMS center double row ball bearing life test, which is an embodiment of a rotary machine multi-parameter health state evaluation method based on TPE-XGBoost.
Detailed Description
The invention is further described below in connection with examples. The scope of the invention is not limited by these examples. The details of the specific working principle of the invention will be further described with reference to the accompanying drawings.
The TPE-XGBoost-based health prediction method designed by the invention is applied to an actual A4005 pump fault case, the whole offline training process is shown in figure 1, the data acquisition interval is 8 seconds, the sampling frequency is 25600Hz, and the length of each file data point is 16384.
Vibration sensor data for a complete life cycle is first collected from old pumps of the same model as the pump. The data set is acquired by a sensor mounted on the old pump, which contains all the data information of the old pump from just being put into operation to the time of failure. Filtering and feature extraction work is carried out on the full life cycle data, and five vibration sensitive time domain parameters of two parts of data (high and low) higher than 1KHz and lower than 1KHz are extracted: the effective value (RMS), kurtosis factor (KUR), peak factor (CF), SKEW factor (SKEW), margin factor (CL), constitute a ten-dimensional feature vector, as shown in fig. 2. And (3) marking a small amount of samples in the full life cycle data by using expert experience to obtain three state labels of normal, degradation and failure. There are 4100 samples in the normal state, 300 samples in the degraded state, and 100 samples in the fault state, which are serious unbalanced samples.
The training samples are equalized using the SMOTE-ENN method. The SMOTE method (Synthetic Minority OversamplingTechnique) uses a self-service method and a K-nearest neighbor method to generate new data similar to a few classes of observations based on feature space to reduce the error of the classifier. The method comprises the following specific steps:
let A denote a minority class, take X i E, A, calculating the distance from E to all samples in the minority class sample set A by taking Euclidean distance as a standard to obtain X i Is selected randomly from the nearest neighbor samples, namely X ij (j=1, 2, …, n); at X i And X is ij (j=1, 2, …, n) to construct a new minority class sample Y j This procedure is equivalent to randomly selecting one on the line of two samples.
Y j ＝X i +rand(0,1)×(X ij -X i )
Where rand (0, 1) represents a random number within the interval (0, 1). A schematic of the synthetic sample is shown in fig. 3. The SMOTE-ENN method is a combination of SMOTE and ENN methods. Firstly, generating a new minority class sample by using an SMOTE method, obtaining a new data set ND, predicting each sample in the new data set ND by using a K nearest neighbor (K=5) method, and eliminating the sample if the predicted result is different from the actual class label, so as to obtain a training set. The training set obtained has 3350 normal samples, 4058 degraded samples and 4077 fault samples.
Designating a hyper-parameter space of the XGBoost model, and performing iterative optimization on the XGBoost model by using a TPE method. TPE converts the hyper-parametric space into a non-parametric density distribution, modeling the p (x|y) process. The conversion mode is that the uniform distribution is converted into truncated Gaussian mixture distribution, the logarithmic uniform distribution is converted into exponential Gaussian mixture distribution and the discrete distribution is converted into heavy weighted discrete distribution of 3 kinds. By using different observations (x 1 ,x 2 ,…,x k ) Alternatively, the super-parameter set of the TPE may use learning algorithms of different densities. The density is defined as:
where l (x) is defined by the observation { x } i An objective function F (x) of less than y * G (x) is composed of the density of the observed values { x } i An objective function F (x) of } is greater than or equal to y * Is a density composition of (a). TPE algorithm uses y * The quantile γ as the observation y. By maintaining an ordered list of observations in observation domain H, the runtime of each iteration of the TPE algorithm can scale linearly in the |h| and the optimized feature dimension, where the desired boost (EI) is:
finally, γ=p (y<y * ) And the above two equations are brought into the following equation, so that each iteration returns an x that yields the maximum EI point * 。
p(x)＝∫p(x|y)p(y)dy＝γl(x)+(1-γ)g(x)
The objective function for each iteration of the XGBoost model is specified as:
wherein:
where T is the tree of leaf child nodes, j is the index of each leaf node, i represents the ith sample in the dataset, y i To be a true value of the value,for the predicted value in the t-1 iteration, the set of samples contained on the leaf with index j is defined as I j Gamma is a self-determined parameter controlling the number of leaves. The formula for performing super parameter optimization by using the TPE method is as follows:
wherein F (x) represents an objective function of XGBoost; x is the parameter at which F (x) achieves the best result, x represents the hyper-parameter space.
After obtaining the optimal model structure and parameters in the offline training state, extracting the same characteristics from the service equipment operation data, and inputting the same characteristics into the model, so as to obtain probability values of the model in three health states. The experimental results obtained are shown in FIG. 4. In this original pump monitoring system, the report is displayed about 1 hour (sampling point 8700) before the failure occurs. In the results obtained by this method, the degradation signal is shown at the sampling point 5434, which is advanced by about 8 hours.
Meanwhile, the method is applied to the full-life experiment of the IMS center double-row ball bearing. The set of bearing life-time experimental data is provided by the IMS center of the university of cincinnati. The test bearing model is RexnordZA-2115 double row ball bearing. The bearing life test is divided into three groups of four bearings, and the test working conditions are as follows: shaft speed 2000rpm, radial load 6000lbs. Lubricating by adopting lubricating oil, and arranging a magnetic plug on an oil return pipeline. When the deposit adhered to the magnetic plug exceeds a certain level, the test is stopped. The data acquisition interval was 10 minutes, with 20480 data points per file length.
And analyzing the total life data of the third group of No. three (# 3-3) bearings, wherein the total life data is 6321 data files, and all the collected characteristic parameters are shown in fig. 5, so that the outer ring faults finally occur. And (3) using an expert experience method to label the model, and using the method to train the TPE-XGBoost model to obtain an optimal model.
The full life data of the second group of bearing #2-1 was analyzed by applying the model trained on bearing #3-3 full life cycle sample, and the analysis results obtained are shown in fig. 6 (black line indicates low-report line). The system sends out degradation alarm signals in advance around the sample point 400, sends out strong fault signals when the sample point 560, and then alarms all the time, so that the system is consistent with the actual situation.
The practical case research shows that the method has good generalization performance, can early warn failure trend in advance, and has practical application value.
Claims (2)
1. The rotary machine multi-parameter health state assessment method based on TPE-XGBoost comprises multi-parameter extraction, sample equalization and equipment health state assessment of a TPE-XGBoost model, and is characterized in that: the multi-parameter extraction method comprises five vibration sensitive parameters obtained after high-low pass filtering, and specifically comprises the following steps:
firstly, collecting vibration sensor data of a complete life cycle from an old pump of the same model as the pump; the data set is acquired by a sensor arranged on the old pump, and comprises all data information from the time of operation of the old pump to the time of fault replacement; filtering and feature extraction work is carried out on the full life cycle data, and five vibration sensitive time domain parameters of two parts of data higher than 1KHz and lower than 1KHz are extracted: the effective value, the kurtosis factor, the peak value factor, the skewness factor and the margin factor form a ten-dimensional characteristic vector;
firstly, generating a new minority class sample by using an SMOTE method, obtaining a new data set ND, predicting each sample in the new data set ND by using a K nearest neighbor method, and removing the sample if the predicted result is different from an actual class label to obtain a training set;
the sample equalization is to use the SMOTE-ENN method to balance training samples, and specifically comprises the following steps:
firstly, generating a new minority class sample by using an SMOTE method, obtaining a new data set ND, predicting each sample in the new data set ND by using a K nearest neighbor method, and removing the sample if the predicted result is different from an actual class label to obtain a training set;
the TPE-XGBoost model is characterized in that a TPE algorithm is firstly used for carrying out super-parameter optimization on a specified parameter space, and then the XGBoost model with an optimal super-parameter structure is trained to complete equipment health state assessment, and the method specifically comprises the following steps:
designating a hyper-parameter space of the XGBoost model, and performing iterative optimization on the XGBoost model by using a TPE method; TPE converts the super-parameter space into non-parameter density distribution, and models the p (x|y) process; the conversion mode is that the uniform distribution is converted into truncated Gaussian mixture distribution, the logarithmic uniform distribution is converted into exponential Gaussian mixture distribution and the discrete distribution is converted into heavy weighted discrete distribution of 3 kinds; by using different observations (x 1 ,x 2 ,…,x k ) Performing replacement processing, wherein the super-parameter group of the TPE uses learning algorithms with different densities; the density is defined as:
where l (x) is defined by the observation { x } i An objective function F (x) of less than y * G (x) is composed of the density of the observed values { x } i Target of }The function F (x) is greater than or equal to y * Is composed of the density of (3); TPE algorithm uses y * A quantile γ as an observation value y; by maintaining an ordered list of observations in observation domain H, the runtime of each iteration of the TPE algorithm can scale linearly in the |h| and the optimized feature dimension, where the desired boost (EI) is:
finally, γ=p (y<y * ) And the above two equations are brought into the following equation, so that each iteration returns an x that yields the maximum EI point * ；
p(x)＝∫p(x|y)p(y)dy＝γl(x)+(1-γ)g(x)
The objective function for each iteration of the XGBoost model is specified as:
wherein:
where T is the tree of leaf child nodes, j is the index of each leaf node, i represents the ith sample in the dataset, y i To be a true value of the value,for the predicted value in the t-1 iteration, the set of samples contained on the leaf with index j is defined as I j Gamma is a self-determined parameter controlling the number of leaves; the formula for performing super parameter optimization by using the TPE method is as follows:
x * ＝arg min x∈χ F(x)
wherein F (x) represents an objective function of XGBoost; x is x * Is the parameter when F (x) obtains the best result, x represents the super parameter space;
after obtaining the optimal model structure and parameters in the offline training state, extracting the same characteristics from the service equipment operation data, and inputting the same characteristics into the model, so as to obtain probability values of the model in three health states.
2. The method of claim 1, wherein: the waveform is filtered by using a Butterworth filter to obtain two parts of higher than 1KHz and lower than 1KHz, and five vibration sensitive parameters including an effective value, a kurtosis factor, a peak factor, a skewness factor and a margin factor are extracted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110590308.6A CN113505639B (en) | 2021-05-28 | 2021-05-28 | Rotary machine multi-parameter health state assessment method based on TPE-XGBoost |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110590308.6A CN113505639B (en) | 2021-05-28 | 2021-05-28 | Rotary machine multi-parameter health state assessment method based on TPE-XGBoost |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113505639A CN113505639A (en) | 2021-10-15 |
| CN113505639B true CN113505639B (en) | 2024-03-22 |
Family
ID=78008596
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110590308.6A Active CN113505639B (en) | 2021-05-28 | 2021-05-28 | Rotary machine multi-parameter health state assessment method based on TPE-XGBoost |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113505639B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114707706A (en) * | 2022-03-17 | 2022-07-05 | 广西科学院 | Diesel engine fault diagnosis method based on unbalanced data |
| CN114969990B (en) * | 2022-08-02 | 2022-10-14 | 中国电子科技集团公司第十研究所 | Multi-model fused avionic product health assessment method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN109063308A (en) * | 2018-07-26 | 2018-12-21 | 北京航空航天大学 | A kind of health evaluating method based on depth quantum learning |
| CN110685868A (en) * | 2019-10-25 | 2020-01-14 | 长沙理工大学 | Wind turbine generator fault detection method and device based on improved gradient elevator |
| CN111695209A (en) * | 2020-05-13 | 2020-09-22 | 东南大学 | Rotary machine small sample health assessment method driven by meta-depth learning |
| CN112507479A (en) * | 2020-11-19 | 2021-03-16 | 电子科技大学 | Oil drilling machine health state assessment method based on manifold learning and softmax |
-
2021
- 2021-05-28 CN CN202110590308.6A patent/CN113505639B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN109063308A (en) * | 2018-07-26 | 2018-12-21 | 北京航空航天大学 | A kind of health evaluating method based on depth quantum learning |
| CN110685868A (en) * | 2019-10-25 | 2020-01-14 | 长沙理工大学 | Wind turbine generator fault detection method and device based on improved gradient elevator |
| CN111695209A (en) * | 2020-05-13 | 2020-09-22 | 东南大学 | Rotary machine small sample health assessment method driven by meta-depth learning |
| CN112507479A (en) * | 2020-11-19 | 2021-03-16 | 电子科技大学 | Oil drilling machine health state assessment method based on manifold learning and softmax |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113505639A (en) | 2021-10-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ren et al. | Multi-bearing remaining useful life collaborative prediction: A deep learning approach | |
| CN102112933B (en) | Error detection method and system | |
| CN106769052B (en) | A kind of mechanical system rolling bearing intelligent failure diagnosis method based on clustering | |
| Sugumaran et al. | Fault diagnostics of roller bearing using kernel based neighborhood score multi-class support vector machine | |
| Sugumaran et al. | Automatic rule learning using decision tree for fuzzy classifier in fault diagnosis of roller bearing | |
| CN109781411B (en) | Bearing fault diagnosis method combining improved sparse filter and KELM | |
| Deutsch et al. | Using deep learning based approaches for bearing remaining useful life prediction | |
| CN110263846A (en) | The method for diagnosing faults for being excavated and being learnt based on fault data depth | |
| Guo et al. | Deep convolution feature learning for health indicator construction of bearings | |
| CN113255848B (en) | Water turbine cavitation sound signal identification method based on big data learning | |
| CN113505639B (en) | Rotary machine multi-parameter health state assessment method based on TPE-XGBoost | |
| CN114295377B (en) | CNN-LSTM bearing fault diagnosis method based on genetic algorithm | |
| CN112765890B (en) | Dynamic domain adaptive network-based multi-working-condition rotating machine residual life prediction method | |
| CN110110768A (en) | Fault Diagnosis of Roller Bearings based on Concurrent Feature study and multi-categorizer | |
| CN114358124B (en) | New fault diagnosis method for rotary machinery based on deep countermeasure convolutional neural network | |
| CN111538311B (en) | Flexible multi-state self-adaptive early warning method and device for mechanical equipment based on data mining | |
| CN110738255A (en) | device state monitoring method based on clustering algorithm | |
| Lee et al. | Unsupervised anomaly detection of the gas turbine operation via convolutional auto-encoder | |
| Hou et al. | A novel rolling bearing fault diagnosis method based on adaptive feature selection and clustering | |
| Son et al. | Deep learning-based anomaly detection to classify inaccurate data and damaged condition of a cable-stayed bridge | |
| CN110728377B (en) | Intelligent fault diagnosis method and system for electromechanical equipment | |
| CN114118219A (en) | Data-driven real-time abnormal detection method for health state of long-term power-on equipment | |
| Xu et al. | A two-stage estimation method based on Conceptors-aided unsupervised clustering and convolutional neural network classification for the estimation of the degradation level of industrial equipment | |
| Techane et al. | Rotating machinery prognostics and application of machine learning algorithms: Use of deep learning with similarity index measure for health status prediction | |
| Kizito et al. | The application of random forest to predictive maintenance |
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 |