CN111852837A - Clustering-based reciprocating compressor self-adaptive fault monitoring method - Google Patents
Clustering-based reciprocating compressor self-adaptive fault monitoring method Download PDFInfo
- Publication number
- CN111852837A CN111852837A CN202010672515.1A CN202010672515A CN111852837A CN 111852837 A CN111852837 A CN 111852837A CN 202010672515 A CN202010672515 A CN 202010672515A CN 111852837 A CN111852837 A CN 111852837A
- Authority
- CN
- China
- Prior art keywords
- data
- point
- alarm
- distance
- points
- 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.)
- Granted
Links
Images
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F04—POSITIVE - DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS FOR LIQUIDS OR ELASTIC FLUIDS
- F04B—POSITIVE-DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS
- F04B51/00—Testing machines, pumps, or pumping installations
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F04—POSITIVE - DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS FOR LIQUIDS OR ELASTIC FLUIDS
- F04B—POSITIVE-DISPLACEMENT MACHINES FOR LIQUIDS; PUMPS
- F04B49/00—Control, e.g. of pump delivery, or pump pressure of, or safety measures for, machines, pumps, or pumping installations, not otherwise provided for, or of interest apart from, groups F04B1/00 - F04B47/00
- F04B49/10—Other safety measures
Abstract
The invention discloses a clustering-based reciprocating compressor self-adaptive fault monitoring method, which comprises the following steps: an off-line modeling stage and an on-line monitoring stage; the hardware components involved are: multi-source sensors (displacement, acceleration, velocity), field servers. According to the clustering-based reciprocating compressor self-adaptive fault monitoring method provided by the invention, reciprocating working condition processes are classified, intelligent early warning limits are set for all working conditions, operators are timely reminded of carrying out subsequent treatment when suspected faults occur, and economic loss is avoided or reduced. Meanwhile, the requirements of real-time performance and accuracy can be met while multi-type fault detection can be carried out. The key point is that the data is divided into clusters under different working conditions through self-adaptive clustering, and the clusters are obtained through the relation among cluster data points.
Description
Technical Field
The invention relates to the field of online mechanical fault monitoring, in particular to a clustering-based reciprocating compressor self-adaptive fault monitoring method.
Background
The reciprocating compressor is an indispensable key unit in the field of petrochemical industry, and is mainly applied to oil refining, oil extraction, gas transportation and other purposes. Due to the complex structure of the reciprocating machine, signals of all parts in the operation state are mutually excited and interfered, and fault signals of a specific part are often submerged in noise. In addition, the reciprocating machine has various failure reasons during operation, mainly including air valve type failure, filler type failure, piston failure and transmission system failure, which are reflected differently in different physical quantities, and it is difficult to evaluate the failure degree of all types by using a single physical quantity. Because the fault characteristics of the reciprocating engine are relatively complex, the current domestic detection standard of the reciprocating engine only has the intensity evaluation index [ GB/T12779-91 ] aiming at the vibration speed signal of the whole engine, the comprehensive consideration of a heterogeneous signal of the whole engine is not carried out, and a specific unified standard is not provided aiming at a certain fault characteristic. In addition, the single hard index is adopted to ignore the difference between reciprocating mechanical units. If the alarm threshold is too high, the system reaction is slow, the hysteresis is large, and small faults are ignored; otherwise, the threshold limit is too low, the system is too sensitive, false alarms occur frequently, and waste of resources and time is caused. Therefore, the step-by-step intelligent threshold value alarm limit is necessary to be divided according to the fault degree of the working condition when the reciprocating engine operates, and the accuracy rate of online monitoring alarm can be obviously improved.
Disclosure of Invention
In order to solve the problems in the prior art, the embodiment of the invention provides a clustering-based reciprocating compressor self-adaptive fault monitoring method. The technical scheme is as follows:
in one aspect, a clustering-based adaptive fault monitoring method for a reciprocating compressor is provided, which includes: the method comprises the steps of off-line modeling and on-line monitoring;
the offline modeling comprises the following steps:
step 1: data acquisition is carried out through multi-source sensors at different positions on the reciprocating machine, and the data are led into a field server and transmitted to a middleware;
step 2: taking data from the middleware through a TCP/IP protocol, wherein the data comprises a cylinder body vibration signal, a piston rod displacement signal and a process quantity signal, and a main analysis object is a vibration signal;
and step 3: initializing data and carrying out preliminary wavelet denoising;
and 4, step 4: obtaining the characteristics of the vibration signal data, and obtaining 6 time domain characteristics of standard deviation, root mean square, kurtosis, skewness, peak value and peak value, 5 time domain factors, 5 frequency domain characteristics of center of gravity frequency, variance frequency, mean square frequency, power spectrum energy and power spectrum peak value, and 6 chaotic characteristics of approximate entropy, sample entropy, fuzzy entropy, Teager operator energy, Teager operator peak value and LZ complexity;
And 5: respectively normalizing a time domain, a frequency domain and the chaotic characteristic set, then carrying out PCA to obtain a characteristic matrix of the vibration signal after dimension reduction, taking the characteristic matrix as a training sample, and recording corresponding parameters;
step 6: calculating the optimal cluster number of the clusters by adopting DaviesBouldin and Calinski Harabasz indexes; clustering the training samples into a plurality of classes by adopting a K-means clustering mode, carrying out second normalization in the classes, and recording normalization parameters;
and 7: extracting the edge of each cluster, dividing the data in each cluster into safety data, dangerous data and edge data by adopting LOF (low level of detail);
and 8: independently taking an edge sample set to perform modeling of OCSVM, and storing the parameters and SVM models obtained in the steps 5 and 6 into an on-site server;
the online monitoring comprises the following steps:
step a: data acquisition is carried out through a multi-source sensor, and the data are led into a field server;
step b: initializing data and carrying out preliminary wavelet denoising;
step c: the same as the step 4 of the off-line modeling stage, three feature groups are obtained for the real-time data;
step d: normalizing and reducing dimensions of the feature group by using the parameters obtained in the step 5 of the off-line modeling stage;
Step e: normalizing by using the intra-class normalization parameters in the step 6 of the off-line modeling stage, solving the near neighbor of the real-time data point by adopting a KNN method, and outputting the cluster to which the point belongs if the point is a safe point; if the point is a dangerous point or an edge point, calculating an LOF value of a point adjacent to the point; additionally, if the point is an edge point, calling an SVM model of the edge of the cluster to which the point belongs to perform single classification on the point to obtain a clear outlier judgment result;
step f: classifying the outlier degree of the real-time detection points according to the LOF value of the outlier and the LOF threshold division limit of the current cluster, and recording the corresponding alarm result;
step g: and the alarm processing program feeds back the final result to a field personnel end after the alarm mechanism is triggered by the data alarm result.
Further, the normalization mode is Zscore,
wherein
Mean and standard deviation of x, respectively. The recording parameters were specifically the normalized Mean1, standard deviation Std1, and Pca scoring matrix Coeff.
Further, the normalization is Zscore or Zscore by Mad (Median absolute difference), and the formula of Zscore-Mad normalization is x' ═ x-media)/Mad, where media is the x Median and Mad is the x Median absolute difference. When the data in the clusters are very different, a Zscore-Mad normalization mode is adopted.
Further, the specific process of the LOF algorithm is as follows:
1. defining a K distance
The k distance of the data object q is defined as the distance from the nearest k point to the data object q in the data set, and is denoted as k-distance (q), and the k distance is referred to as Euclidean distance;
k distance neighborhood
The set of data points in the data set having a distance to the data object q of no more than k, i.e. Nk-distance (q) ({ p ∈ D { q } | D (p, q) ≦ k-distance (q) }. d (p, q) refers to the Euclidean distance between p and q;
3. reachable distance
p and q are any two points in the data set, and the reachable distance from p to q is defined as: reach-distk (p, q) ═ max { d (p, q), k-distance (q) };
4. local achievable density
The local reachable density of q refers to the inverse of the average reachable distance of q to all points in the neighborhood, and is calculated as follows:
wherein, | Nk(q) | is the number of points in the k neighborhood of q. If lrdk(q) the larger the density of q, the more normal the point q is;
5. local outlier factor
If LOF > 1, the difference between the q point density and the whole data density is large, namely, the outlier; if LOF is close to 1, the point q is normal;
6. edge extraction
Two outlier ratio control limits of p0 and p1 are set, wherein 0< p0< p1< 1.
And sequencing the size of the dimensionless LOF value to obtain a LOF sequence of the current data point from small to large.
Taking th0 ═ n × p0 and th1 ═ n × p1, wherein n is the length of the sequence, i.e., the number of data points;
data points in the sequence below th0 belong to the safety set, and this part should be included in the LOF value floating above or below 1; data points in the sequence with higher ranking than th0 belong to an unsafe set, where the data below th1 part has approached the boundary of the data, called a hazard set, representing the transition of data points from safe to outlier, and the data above th1 part has been at the boundary of the full data set, called an edge set.
Additionally, if the components in the data are complex and the outliers of individual points in the edge set are high, the noise points and the outliers should be separated by dividing the edge set.
Furthermore, the dimension reduction method is to perform zero-mean value on the feature set and then multiply the feature set by a PCA (principal component analysis) score matrix Coeff.
Further, a hardware alarm strategy is to accumulate alarm data points in a time window, wherein each alarm point is 1 point, if the alarm points continuously appear, the next alarm point is 2 times of the previous alarm point, and if the alarm accumulation score in the time window exceeds the alarm limit, the alarm is triggered and recorded; no alarm is given for a period of time after the alarm, but the current alarm score is recorded.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
according to the clustering-based reciprocating compressor self-adaptive fault monitoring method provided by the invention, reciprocating working condition processes are classified, intelligent early warning limits are set for all working conditions, operators are timely reminded of carrying out subsequent treatment when suspected faults occur, and economic loss is avoided or reduced. Meanwhile, the requirements of real-time performance and accuracy can be met while multi-type fault detection can be carried out. The key point is that the data is divided into clusters under different working conditions through self-adaptive clustering, and the clusters are obtained through the relation among cluster data points.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
FIG. 1 is an off-line modeling flow chart of a cluster-based reciprocating compressor adaptive fault monitoring method of an embodiment of the present invention;
FIG. 2 is a flow chart of an on-line monitoring method of a cluster-based reciprocating compressor adaptive fault monitoring according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
The invention provides a clustering-based reciprocating compressor self-adaptive fault monitoring method, which comprises the following steps of, referring to fig. 1-2: an off-line modeling stage and an on-line monitoring stage;
in the present embodiment, the hardware components include: multi-source sensors (displacement, acceleration, velocity), field servers.
The off-line modeling phase, see fig. 1:
step 1: data acquisition is carried out through multi-source sensors at different positions on the reciprocating machine, and the data are led into a field server and transmitted to the middleware.
Step 2: and taking data from the middleware through a TCP/IP protocol, wherein the data comprises a cylinder vibration signal, a piston rod displacement signal and a process quantity signal, and a main analysis object is a vibration signal.
And step 3: initializing data and carrying out preliminary wavelet denoising.
And 4, step 4: the method comprises the steps of obtaining characteristics of vibration signal data, and obtaining 6 time domain characteristics of standard deviation, root mean square, kurtosis, skewness, peak value and peak-peak value, 5 time domain characteristics of 5 time domain factors, 5 frequency domain characteristics of center-of-gravity frequency, variance frequency, mean square frequency and the like, and 6 chaotic characteristics of approximate entropy, sample entropy, fuzzy entropy, LZ complexity and the like.
And 5: respectively normalizing the time domain, the frequency domain and the chaotic characteristic set, then carrying out PCA to obtain a characteristic matrix of the vibration signal after dimension reduction, taking the characteristic matrix as a training sample, and recording corresponding parameters.
Further, the normalization mode is Zscore,
wherein
Mean and standard deviation of x, respectively. The recording parameters were specifically the normalized Mean1, standard deviation Std1, and Pca scoring matrix Coeff.
Step 6: the optimal cluster number for clustering was calculated using DaviesBouldin and Calinski Harabasz indices. And clustering the training samples into a plurality of classes by adopting a K-means clustering mode, carrying out second normalization in the classes, and recording normalization parameters.
Further, the normalization is Zscore or Zscore by Mad (Median absolute difference), and the formula of Zscore-Mad normalization is x' ═ x-media)/Mad, where media is the x Median and Mad is the x Median absolute difference. When the data in the clusters are very different, a Zscore-Mad normalization mode is adopted.
And 7: and extracting the edge of each cluster, and dividing the data in the class of each cluster into safety data, dangerous data and edge data by adopting an LOF (local outlier factor) method.
Further, the specific process of the LOF algorithm is as follows:
1. Defining a K distance
The k-distance of the data object q is defined as the distance from the nearest k-th point to the data object q in the data set, and is denoted as k-distance (q), and is referred to as Euclidean distance.
K distance neighborhood
The set of data points in the data set having a distance to the data object q of no more than k, i.e. Nk-distance (q) ({ p ∈ D { q } | D (p, q) ≦ k-distance (q) }. d (p, q) refers to the Euclidean distance between p and q.
3. Reachable distance
p and q are any two points in the data set, and the reachable distance from p to q is defined as: reach-distk (p, q) ═ max { d (p, q), k-distance (q) }.
4. Local achievable density
The local reachable density of q refers to the inverse of the average reachable distance of q to all points in the neighborhood, and is calculated as follows:
wherein, | Nk(q) | is the number of points in the k neighborhood of q. If lrdkThe larger (q) indicates the higher density of q and the more normal the q point.
5. Local outlier factor
If LOF > 1, it indicates that the q-point density is far from the overall data density, i.e., the outliers. LOF is close to 1, then point q is more normal.
6. Edge extraction
Two outlier ratio control limits of p0 and p1 are set, wherein 0 < p0 < p1 < 1.
And sequencing the size of the dimensionless LOF value to obtain a LOF sequence of the current data point from small to large.
Let th 0-n p0 and th 1-n p1, where n is the length of the sequence, i.e., the number of data points.
Data points in the sequence below th0 belong to the safety set, and this part should be included in the LOF value floating above or below 1; data points in the sequence with higher ranking than th0 belong to an unsafe set, where the data below th1 part has approached the boundary of the data, called a hazard set, representing the transition of data points from safe to outlier, and the data above th1 part has been at the boundary of the full data set, called an edge set.
Additionally, if the components in the data are complex and the outliers of individual points in the edge set are high, the noise points and the outliers should be separated by dividing the edge set.
And 8: and (4) independently taking the edge sample set to perform modeling of the OCSVM, and storing the parameters and the SVM model obtained in the steps (5) and (6) into the field server.
On-line detection phase, see fig. 2:
step 1: and data acquisition is carried out through a multi-source sensor, and the data is imported into a field server.
Step 2: initializing data and carrying out preliminary wavelet denoising.
And step 3: as in step 4 of the off-line phase, three feature sets are obtained for the real-time data.
And 4, step 4: and (5) normalizing and reducing the dimension of the feature group by using the parameters obtained in the off-line step 5.
Furthermore, the dimension reduction method is to perform zero-mean value on the feature set and then multiply the feature set by a PCA (principal component analysis) score matrix Coeff.
And 5: normalizing by using the in-class normalization parameters in the off-line step 6 again, then solving the near neighbor of the real-time data point by adopting a KNN method, and if the point is a safe point, outputting the cluster to which the point belongs; if the point is a dangerous point or an edge point, the LOF value of the point adjacent to the point is obtained. Additionally, if the point is an edge point, calling an SVM model of the edge of the cluster to which the point belongs to perform single classification on the point to obtain a clear outlier judgment result.
Step 6: and classifying the outlier degree of the real-time detection points according to the LOF value of the outlier and the LOF threshold division limit of the current cluster, and recording the corresponding alarm result.
And 7: and the alarm processing program feeds back the final result to a field personnel end after the alarm mechanism is triggered by the data alarm result.
Further, the hardware alarm strategy is to accumulate alarm data points in a time window, wherein each alarm point is 1 point, if the alarm points continuously appear, the later alarm point is 2 times of the former alarm point, and if the alarm accumulation score in the time window exceeds the alarm limit, the alarm is triggered and recorded. No alarm is given for a period of time after the alarm, but the current alarm score is recorded.
The technical scheme provided by the embodiment of the invention has the following beneficial effects:
according to the clustering-based reciprocating compressor self-adaptive fault monitoring method provided by the invention, reciprocating working condition processes are classified, intelligent early warning limits are set for all working conditions, operators are timely reminded of carrying out subsequent treatment when suspected faults occur, and economic loss is avoided or reduced. Meanwhile, the requirements of real-time performance and accuracy can be met while multi-type fault detection can be carried out. The key point is that the data is divided into clusters under different working conditions through self-adaptive clustering, and the clusters are obtained through the relation among cluster data points.
The present invention is not limited to the above preferred embodiments, and any modifications, equivalent replacements, improvements, etc. within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. A clustering-based reciprocating compressor adaptive fault monitoring method is characterized by comprising the following steps: performing off-line modeling and on-line monitoring;
the offline modeling comprises the following steps:
step 1: data acquisition is carried out through multi-source sensors at different positions on the reciprocating machine, and the data are led into a field server and transmitted to a middleware;
Step 2: taking data from the middleware through a TCP/IP protocol, wherein the data comprises a cylinder body vibration signal, a piston rod displacement signal and a process quantity signal, and a main analysis object is a vibration signal;
and step 3: initializing data and carrying out preliminary wavelet denoising;
and 4, step 4: obtaining the characteristics of the vibration signal data, and obtaining 6 time domain characteristics of standard deviation, root mean square, kurtosis, skewness, peak value and peak value, 5 time domain factors, 5 frequency domain characteristics of center of gravity frequency, variance frequency, mean square frequency, power spectrum energy and power spectrum peak value, and 6 chaotic characteristics of approximate entropy, sample entropy, fuzzy entropy, Teager operator energy, Teager operator peak value and LZ complexity;
and 5: respectively normalizing a time domain, a frequency domain and the chaotic characteristic set, then carrying out PCA to obtain a characteristic matrix of the vibration signal after dimension reduction, taking the characteristic matrix as a training sample, and recording corresponding parameters;
step 6: calculating the optimal cluster number of the clusters by adopting DaviesBouldin and Calinski Harabasz indexes; clustering the training samples into a plurality of classes by adopting a K-means clustering mode, carrying out second normalization in the classes, and recording normalization parameters;
and 7: extracting the edge of each cluster, dividing the data in each cluster into safety data, dangerous data and edge data by adopting LOF (low level of detail);
And 8: independently taking an edge sample set to perform modeling of OCSVM, and storing the parameters and SVM models obtained in the steps 5 and 6 into an on-site server;
the online monitoring comprises the following steps:
step 1: data acquisition is carried out through a multi-source sensor, and the data are led into a field server;
step 2: initializing data and carrying out preliminary wavelet denoising;
and step 3: the same as the step 4 of the off-line modeling stage, three feature groups are obtained for the real-time data;
and 4, step 4: normalizing and reducing dimensions of the feature group by using the parameters obtained in the step 5 of the off-line modeling stage;
and 5: normalizing by using the intra-class normalization parameters in the step 6 of the off-line modeling stage, solving the near neighbor of the real-time data point by adopting a KNN method, and outputting the cluster to which the point belongs if the point is a safe point; if the point is a dangerous point or an edge point, calculating an LOF value of a point adjacent to the point; additionally, if the point is an edge point, calling an SVM model of the edge of the cluster to which the point belongs to perform single classification on the point to obtain a clear outlier judgment result;
step 6: classifying the outlier degree of the real-time detection points according to the LOF value of the outlier and the LOF threshold division limit of the current cluster, and recording the corresponding alarm result;
And 7: and the alarm processing program feeds back the final result to a field personnel end after the alarm mechanism is triggered by the data alarm result.
2. The method of claim 1, wherein the normalization is by Zscore,
wherein
σxThe mean and standard deviation of x, respectively; the recording parameters were specifically the normalized Mean1, standard deviation Std1, and Pca scoring matrix Coeff.
3. The method of claim 2, wherein the normalization is Zscore or Zscore by Mad, the formula for Zscore-Mad normalization being x' ═ x-media)/Mad, where media is the x Median and Mad is the absolute Median of x; when the data in the clusters are very different, a Zscore-Mad normalization mode is adopted.
4. The method of claim 3, wherein the LOF algorithm is specifically characterized by:
1. defining a K distance
The k distance of the data object q is defined as the distance from the nearest k point to the data object q in the data set, and is denoted as k-distance (q), and the k distance is referred to as Euclidean distance;
k distance neighborhood
The set of data points in the data set having a distance to the data object q of no more than k, i.e. Nk-distance (q) ({ p ∈ D { q } | D (p, q) ≦ k-distance (q)) }; d (p, q) refers to the Euclidean distance between p and q;
3. Reachable distance
p and q are any two points in the data set, and the reachable distance from p to q is defined as: reach-distk (p, q) ═ max { d (p, q), k-distance (q) };
4. local achievable density
The local reachable density of q refers to the inverse of the average reachable distance of q to all points in the neighborhood, and is calculated as follows:
wherein, | Nk(q) | is the number of points in the k neighborhood of q; if lrdk(q) the larger the density of q, the more normal the point q is;
5. local outlier factor
If LOF > 1, the difference between the q point density and the whole data density is large, namely, the outlier; if LOF is close to 1, the point q is normal;
6. edge extraction
Setting two outlier ratio control limits of p0 and p1, wherein 0< p0< p1< 1;
obtaining a LOF sequence of the current data point from small to large by sequencing the size of the LOF dimensionless value;
taking th0 ═ n × p0 and th1 ═ n × p1, wherein n is the length of the sequence, i.e., the number of data points;
data points in the sequence below th0 belong to the safety set, and this part should be included in the LOF value floating above or below 1; data points in the sequence with the ranking order higher than th0 belong to an unsafe set, wherein the data below th1 part already approaches the boundary of the data, called a danger set, representing the transition of data points from safe to outlier, and the data above th1 part already lies at the boundary of the full data set, called an edge set;
Additionally, if the components in the data are complex and the outliers of individual points in the edge set are high, the noise points and the outliers should be separated by dividing the edge set.
5. The method of claim 4, wherein the dimension reduction is performed by zero-averaging the feature set and then multiplying by a PCA score matrix Coeff.
6. The method of claim 5, wherein the hardware alarm strategy is to accumulate alarm data points within a time window, each alarm point is 1 point, if the alarm points continuously appear, the latter alarm point is 2 times of the former alarm point, if the alarm accumulation score within the time window exceeds the alarm limit, the alarm is triggered and recorded; no alarm is given for a period of time after the alarm, but the current alarm score is recorded.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010672515.1A CN111852837B (en) | 2020-07-14 | 2020-07-14 | Clustering-based reciprocating compressor self-adaptive fault monitoring method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010672515.1A CN111852837B (en) | 2020-07-14 | 2020-07-14 | Clustering-based reciprocating compressor self-adaptive fault monitoring method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111852837A true CN111852837A (en) | 2020-10-30 |
| CN111852837B CN111852837B (en) | 2022-02-08 |
Family
ID=72984378
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010672515.1A Active CN111852837B (en) | 2020-07-14 | 2020-07-14 | Clustering-based reciprocating compressor self-adaptive fault monitoring method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111852837B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113740066A (en) * | 2021-11-08 | 2021-12-03 | 中国空气动力研究与发展中心设备设计与测试技术研究所 | Early fault detection method for compressor bearing |
| CN114046873A (en) * | 2021-11-17 | 2022-02-15 | 国家电网有限公司 | Reactor vibration monitoring system based on LOF-FCM fuzzy clustering algorithm |
| CN116304835A (en) * | 2023-03-31 | 2023-06-23 | 北京博华信智科技股份有限公司 | AI-based dynamic equipment working condition monitoring management method, system, equipment and medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102797671A (en) * | 2011-05-25 | 2012-11-28 | 中国石油大学(北京) | Fault detection method and device of reciprocating compressor |
| CN104121178A (en) * | 2013-04-27 | 2014-10-29 | 青岛科技大学 | Fault diagnosis system and method for fuel delivery pump |
| CN107632592A (en) * | 2017-09-01 | 2018-01-26 | 南通大学 | Nonlinear time-varying procedure fault monitoring method based on efficient recursion core pivot element analysis |
| CN107861492A (en) * | 2017-09-25 | 2018-03-30 | 湖州师范学院 | A kind of broad sense Non-negative Matrix Factorization fault monitoring method based on nargin statistic |
-
2020
- 2020-07-14 CN CN202010672515.1A patent/CN111852837B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102797671A (en) * | 2011-05-25 | 2012-11-28 | 中国石油大学(北京) | Fault detection method and device of reciprocating compressor |
| CN104121178A (en) * | 2013-04-27 | 2014-10-29 | 青岛科技大学 | Fault diagnosis system and method for fuel delivery pump |
| CN107632592A (en) * | 2017-09-01 | 2018-01-26 | 南通大学 | Nonlinear time-varying procedure fault monitoring method based on efficient recursion core pivot element analysis |
| CN107861492A (en) * | 2017-09-25 | 2018-03-30 | 湖州师范学院 | A kind of broad sense Non-negative Matrix Factorization fault monitoring method based on nargin statistic |
Non-Patent Citations (2)
| Title |
|---|
| 苗刚: "往复活塞式压缩机关键部件的故障诊断方法研究及应用", 《中国优秀博硕士学位论文全文数据库（博士） 工程科技II辑》 * |
| 陈永国: "《公共管理定量分析方法》", 30 September 2006, 上海交通大学出版社 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113740066A (en) * | 2021-11-08 | 2021-12-03 | 中国空气动力研究与发展中心设备设计与测试技术研究所 | Early fault detection method for compressor bearing |
| CN113740066B (en) * | 2021-11-08 | 2022-02-08 | 中国空气动力研究与发展中心设备设计与测试技术研究所 | Early fault detection method for compressor bearing |
| CN114046873A (en) * | 2021-11-17 | 2022-02-15 | 国家电网有限公司 | Reactor vibration monitoring system based on LOF-FCM fuzzy clustering algorithm |
| CN116304835A (en) * | 2023-03-31 | 2023-06-23 | 北京博华信智科技股份有限公司 | AI-based dynamic equipment working condition monitoring management method, system, equipment and medium |
| CN116304835B (en) * | 2023-03-31 | 2023-08-29 | 北京博华信智科技股份有限公司 | AI-based dynamic equipment working condition monitoring management method, system, equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111852837B (en) | 2022-02-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111852837B (en) | Clustering-based reciprocating compressor self-adaptive fault monitoring method | |
| CN111353482B (en) | LSTM-based fatigue factor recessive anomaly detection and fault diagnosis method | |
| CN110895526A (en) | Method for correcting data abnormity in atmosphere monitoring system | |
| CN109729090B (en) | Slow denial of service attack detection method based on WEDMS clustering | |
| CN111652461A (en) | Aero-engine continuous health state evaluation method based on SAE-HMM | |
| WO2023197461A1 (en) | Gearbox fault early warning method and system based on working condition similarity evaluation | |
| CN113505655B (en) | Intelligent bearing fault diagnosis method for digital twin system | |
| CN110057406B (en) | Multi-scale self-adaptive mechanical equipment trend early warning method | |
| CN112860183B (en) | Multisource distillation-migration mechanical fault intelligent diagnosis method based on high-order moment matching | |
| CN111717753A (en) | Self-adaptive elevator fault early warning system and method based on multi-dimensional fault characteristics | |
| CN114894468A (en) | Chaos detection-based early weak fault diagnosis method for planetary gear box | |
| CN116538092B (en) | Compressor on-line monitoring and diagnosing method, device, equipment and storage medium | |
| CN111934903A (en) | Docker container fault intelligent prediction method based on time sequence evolution genes | |
| CN115524002B (en) | Operation state early warning method, system and storage medium of power plant rotating equipment | |
| CN110837953A (en) | Automatic abnormal entity positioning analysis method | |
| CN112783138B (en) | Intelligent monitoring and abnormity diagnosis method and device for processing stability of production line equipment | |
| CN115788771A (en) | Intelligent operation and maintenance management system of wind power generation system based on Internet of things technology | |
| CN112836570B (en) | Equipment abnormality detection method utilizing Gaussian noise | |
| CN115460061A (en) | Health degree evaluation method and device based on intelligent operation and maintenance scene | |
| CN114004337A (en) | Abnormal working condition early warning model based on sharing loss and attention network | |
| CN114861749A (en) | Low-sample bearing fault diagnosis method based on depth prototype network | |
| CN110543908B (en) | Control chart pattern recognition method based on dynamic observation window | |
| CN113822565A (en) | Method for graded and refined analysis of time-frequency characteristics of fan monitoring data | |
| CN112466322A (en) | Electromechanical device noise signal feature extraction method | |
| Handayani et al. | Anomaly detection in vessel sensors data with unsupervised learning technique |
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 |