[go: up one dir, main page]

CN103246277A - Relative-transform-based industrial process monitoring method of information increment matrix - Google Patents

Relative-transform-based industrial process monitoring method of information increment matrix Download PDF

Info

Publication number
CN103246277A
CN103246277A CN2013101065672A CN201310106567A CN103246277A CN 103246277 A CN103246277 A CN 103246277A CN 2013101065672 A CN2013101065672 A CN 2013101065672A CN 201310106567 A CN201310106567 A CN 201310106567A CN 103246277 A CN103246277 A CN 103246277A
Authority
CN
China
Prior art keywords
matrix
fault
local
moment
variables
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN2013101065672A
Other languages
Chinese (zh)
Inventor
文成林
苑天琪
胡玉成
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.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi 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 Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN2013101065672A priority Critical patent/CN103246277A/en
Publication of CN103246277A publication Critical patent/CN103246277A/en
Pending legal-status Critical Current

Links

Images

Landscapes

  • Testing Or Calibration Of Command Recording Devices (AREA)

Abstract

本发明属于过程监控领域，主要涉及一种基于相对化变换的信息增量矩阵的工业过程监控方法。已有的信息增量矩阵的过程监控方法，虽然能有效地克服传统主元分析方法因存在严重的模式复合效应而不能辨识故障的不足，但是其自身忽略了量纲的影响，常造成在一些系统中起重要作用的变量因其自身绝对值较小而不能检测出绝对值更小的故障，而这些重要变量的很小故障通常又对系统的安全和稳定起着非常关键的作用。本发明依据各变量在实际生产过程中的重要程度，在对系统进行相对化变换的基础上，提出了一种改进的信息增量矩阵的过程监控方法，该方法在保持能有效检测出原有故障的同时，又能检测出原系统一些虽然较小但却起重要作用的变量所发生的故障。

Figure 201310106567

The invention belongs to the field of process monitoring, and mainly relates to an industrial process monitoring method based on a relativized transformed information increment matrix. Although the existing process monitoring method of information increment matrix can effectively overcome the shortcomings of the traditional principal component analysis method that cannot identify faults due to the serious mode compound effect, it ignores the influence of dimensions, which often results in some The variables that play an important role in the system cannot detect faults with smaller absolute values because of their small absolute values, and the small faults of these important variables usually play a very critical role in the safety and stability of the system. According to the importance of each variable in the actual production process, and on the basis of relativizing the system, the present invention proposes an improved process monitoring method of the information increment matrix, which can effectively detect the original At the same time, it can detect the faults of some variables that are small but play an important role in the original system.

Figure 201310106567

Description

基于相对化变换的信息增量矩阵的工业过程监控方法Industrial Process Monitoring Method Based on Information Incremental Matrix of Relativization Transformation

技术领域 technical field

本发明属于过程监控领域，主要涉及一种基于相对化变换的信息增量矩阵的工业过程监控方法。 The invention belongs to the field of process monitoring, and mainly relates to an industrial process monitoring method based on a relativized transformed information increment matrix.

背景技术 Background technique

现代工业过程中通常有大量的被测变量，如流速、浓度、温度、压力等。这些变量处于一定的波动范围，对于生产过程的正常运行、保证产品质量的一致性、可靠性来说是至关重要的。而对于操作人员来说，同时对大量的过程变量进行监控则是比较困难的。多变量统计过程监控对于检测和诊断各种工业过程中出现的异常状况非常有效，包括化工过程、高分子过程、微电子制造过程等。在多变量统计过程监控中常用的基于主元分析(PCA)以及相对主元分析(RPCA)的监控方法因存在严重的模式复合效应，难以解释故障由哪个变量引起, 且实时性不好。基于全局协方差矩阵的信息增量矩阵的监控方法以及在此基础上的改进—基于局部数据的移动窗口协方差矩阵的信息增量矩阵的监控方法，对于一些系统中起重要作用的变量因本身绝对值较小而不能检测出绝对值更小的故障，通常这些重要变量的很小故障又对系统的安全和稳定起着关键作用，若不能及时将其诊断并排除就会对系统的正常运行产生巨大影响，甚至会造成灾难性的事故。 There are usually a large number of measured variables in modern industrial processes, such as flow rate, concentration, temperature, pressure, etc. These variables are in a certain fluctuation range, which is very important for the normal operation of the production process, ensuring the consistency and reliability of product quality. For operators, it is more difficult to monitor a large number of process variables at the same time. Multivariate statistical process monitoring is very effective for detecting and diagnosing abnormal conditions in various industrial processes, including chemical processes, polymer processes, microelectronics manufacturing processes, etc. The monitoring methods based on principal component analysis (PCA) and relative principal component analysis (RPCA), which are commonly used in multivariate statistical process monitoring, have serious mode compound effects, making it difficult to explain which variable causes the fault, and the real-time performance is not good. The monitoring method of the information increment matrix based on the global covariance matrix and the improvement on this basis - the monitoring method of the information increment matrix based on the moving window covariance matrix of the local data, for some variables that play an important role in the system due to itself The absolute value is small and the faults with smaller absolute values cannot be detected. Usually, the small faults of these important variables play a key role in the safety and stability of the system. If they cannot be diagnosed and eliminated in time, the normal operation of the system will be affected. have a huge impact, and even cause catastrophic accidents.

发明内容 Contents of the invention

为解决上述问题，本发明提出了基于相对化变换的信息增量矩阵的工业过程监控方法。该方法首先根据系统的先验信息分析和确定各变量的重要程度，在能量守恒的准则下，赋予系统各变量相应的权值；其次将相邻两个实时更新的协方差矩阵做差运算，得到随时间变化的信息增量矩阵序列；最后计算出基于局部采样数据信息的信息增量均值与动态阈值，进而确定各变量对故障的影响程度，以实现故障诊断和识别，从而实现对整个过程的监控。其具体内容如下： In order to solve the above problems, the present invention proposes an industrial process monitoring method based on the information increment matrix of relativization transformation. This method first analyzes and determines the importance of each variable according to the prior information of the system, and assigns the corresponding weights to each variable of the system under the principle of energy conservation; secondly, the difference operation is performed on two adjacent real-time updated covariance matrices, Obtain the information increment matrix sequence that changes with time; finally calculate the average value and dynamic threshold of information increment based on local sampling data information, and then determine the degree of influence of each variable on the fault, so as to realize fault diagnosis and identification, so as to realize the whole process monitoring. Its specific content is as follows:

设一个多变量系统的数据矩阵                                               

Figure 2013101065672100002DEST_PATH_IMAGE002
为 Assuming a data matrix for a multivariate system
Figure 2013101065672100002DEST_PATH_IMAGE002
for

 

Figure 2013101065672100002DEST_PATH_IMAGE004
                                (1)
Figure 2013101065672100002DEST_PATH_IMAGE004
(1)

其中每一行对应一个变量，每一列对应于一个采样样本，

Figure 2013101065672100002DEST_PATH_IMAGE006
代表系统状态变量的个数，
Figure 2013101065672100002DEST_PATH_IMAGE008
为每个变量的采样数目。 where each row corresponds to a variable, and each column corresponds to a sampling sample,
Figure 2013101065672100002DEST_PATH_IMAGE006
represents the number of system state variables,
Figure 2013101065672100002DEST_PATH_IMAGE008
The number of samples for each variable.

首先对原始矩阵进行相对化变换： First relativize the original matrix:

Figure 2013101065672100002DEST_PATH_IMAGE010
              (2)
Figure 2013101065672100002DEST_PATH_IMAGE010
(2)

                           (3) (3)

则称式(2)是对原始数据矩阵

Figure 2013101065672100002DEST_PATH_IMAGE014
所做的相对化处理，
Figure 2013101065672100002DEST_PATH_IMAGE016
是相应的相对化算子，
Figure 2013101065672100002DEST_PATH_IMAGE018
是相对化变换后的数据矩阵。式(3)中的
Figure 2013101065672100002DEST_PATH_IMAGE020
称为比重因子，它是根据实际系统而定的先验信息，
Figure 884793DEST_PATH_IMAGE020
作用在
Figure 2013101065672100002DEST_PATH_IMAGE022
上，体现了相应变量
Figure 739616DEST_PATH_IMAGE022
在系统中的相对重要程度。 Then it is said that formula (2) is the original data matrix
Figure 2013101065672100002DEST_PATH_IMAGE014
The relativization done,
Figure 2013101065672100002DEST_PATH_IMAGE016
is the corresponding relativization operator,
Figure 2013101065672100002DEST_PATH_IMAGE018
is the data matrix after relativization transformation. In formula (3)
Figure 2013101065672100002DEST_PATH_IMAGE020
is called the specific gravity factor, which is a priori information based on the actual system,
Figure 884793DEST_PATH_IMAGE020
work on
Figure 2013101065672100002DEST_PATH_IMAGE022
above, reflecting the corresponding variable
Figure 739616DEST_PATH_IMAGE022
relative importance in the system.

 然后求出

Figure 207769DEST_PATH_IMAGE018
的均值向量
Figure 2013101065672100002DEST_PATH_IMAGE024
and then find
Figure 207769DEST_PATH_IMAGE018
The mean vector of
Figure 2013101065672100002DEST_PATH_IMAGE024

                                    (4) (4)

和的协方差矩阵

Figure 2013101065672100002DEST_PATH_IMAGE028
and The covariance matrix of
Figure 2013101065672100002DEST_PATH_IMAGE028

Figure 2013101065672100002DEST_PATH_IMAGE030
                     (5)
Figure 2013101065672100002DEST_PATH_IMAGE030
(5)

假设当前时刻为

Figure 165153DEST_PATH_IMAGE008
，记包含有故障的采样数据时刻为
Figure 2013101065672100002DEST_PATH_IMAGE032
，记正常采样数据(不包含故障)的时刻为
Figure 2013101065672100002DEST_PATH_IMAGE034
，且满足 Suppose the current moment is
Figure 165153DEST_PATH_IMAGE008
, record the time of sampled data containing faults as
Figure 2013101065672100002DEST_PATH_IMAGE032
, record the time of normal sampling data (excluding faults) as
Figure 2013101065672100002DEST_PATH_IMAGE034
, and satisfy

Figure 2013101065672100002DEST_PATH_IMAGE036
                               (6)
Figure 2013101065672100002DEST_PATH_IMAGE036
(6)

在时刻

Figure 49932DEST_PATH_IMAGE008
,从正常采样数据集
Figure 2013101065672100002DEST_PATH_IMAGE038
中连续选取
Figure 2013101065672100002DEST_PATH_IMAGE040
个采样数据，并形成相应的局部数据矩阵
Figure 2013101065672100002DEST_PATH_IMAGE042
at the moment
Figure 49932DEST_PATH_IMAGE008
, from the normally sampled dataset
Figure 2013101065672100002DEST_PATH_IMAGE038
continuous selection
Figure 2013101065672100002DEST_PATH_IMAGE040
sampling data, and form the corresponding local data matrix
Figure 2013101065672100002DEST_PATH_IMAGE042

Figure 2013101065672100002DEST_PATH_IMAGE044
                          (7)
Figure 2013101065672100002DEST_PATH_IMAGE044
(7)

当下一个时刻 即

Figure 2013101065672100002DEST_PATH_IMAGE046
时刻到来后，就形成的相应局部数据矩阵
Figure 2013101065672100002DEST_PATH_IMAGE048
when the next moment is
Figure 2013101065672100002DEST_PATH_IMAGE046
After the time arrives, the corresponding local data matrix is formed
Figure 2013101065672100002DEST_PATH_IMAGE048

Figure 2013101065672100002DEST_PATH_IMAGE050
                          (8)
Figure 2013101065672100002DEST_PATH_IMAGE050
(8)

这里 here

Figure 2013101065672100002DEST_PATH_IMAGE052
                   (9)
Figure 2013101065672100002DEST_PATH_IMAGE052
(9)

根据公式(5)-(6)可求出

Figure 2013101065672100002DEST_PATH_IMAGE054
和
Figure 2013101065672100002DEST_PATH_IMAGE056
。 According to the formula (5)-(6), it can be obtained
Figure 2013101065672100002DEST_PATH_IMAGE054
and
Figure 2013101065672100002DEST_PATH_IMAGE056
.

定义基于局部采样数据信息的局部信息增量矩阵为，计算得到 Define the local information increment matrix based on local sampling data information as , calculated to get

Figure 2013101065672100002DEST_PATH_IMAGE060
             (10)
Figure 2013101065672100002DEST_PATH_IMAGE060
(10)

则基于局部采样数据信息的局部信息增量均值

Figure 2013101065672100002DEST_PATH_IMAGE062
为  Then the local information increment mean value based on the local sampling data information
Figure 2013101065672100002DEST_PATH_IMAGE062
for

Figure 2013101065672100002DEST_PATH_IMAGE064
                          (11)
Figure 2013101065672100002DEST_PATH_IMAGE064
(11)

记固定窗口长度为

Figure 509776DEST_PATH_IMAGE040
的局部正常采样数据动态阈值为
Figure 2013101065672100002DEST_PATH_IMAGE066
Note that the fixed window length is
Figure 509776DEST_PATH_IMAGE040
The local normal sampling data dynamic threshold of
Figure 2013101065672100002DEST_PATH_IMAGE066

Figure 2013101065672100002DEST_PATH_IMAGE068
                           (12)
Figure 2013101065672100002DEST_PATH_IMAGE068
(12)

通过比较某一时刻的局部采样数据信息的局部信息增量均值与系统动态阈值的特定关系大小即可检测系统在该时刻是否发生异常: By comparing the specific relationship between the local information incremental mean value of the local sampling data information at a certain moment and the system dynamic threshold, it is possible to detect whether the system is abnormal at this moment:

1)若

Figure 2013101065672100002DEST_PATH_IMAGE070
，则表示系统在该时刻有故障发生。此时，为保证故障检测的准确性，在检测下一时刻系统是否发生故障时，要去掉当前发生故障时刻的数据，即滑动窗口位置不发生变化。                               1) if
Figure 2013101065672100002DEST_PATH_IMAGE070
, it means that there is a fault in the system at this moment. At this time, in order to ensure the accuracy of fault detection, when detecting whether the system is faulty at the next moment, the data at the current time when the fault occurs should be deleted, that is, the position of the sliding window does not change.

2)若

Figure 2013101065672100002DEST_PATH_IMAGE072
，则表示系统该时刻没有故障发生。此时，滑动窗口继续采样，来进行下一个时刻的检测。 2) if
Figure 2013101065672100002DEST_PATH_IMAGE072
, it means that there is no fault in the system at this moment. At this time, the sliding window continues to sample to detect the next moment.

当检测出系统已经发生故障后，需要进一步计算各变量对总体的贡献率

Figure 2013101065672100002DEST_PATH_IMAGE074
，以判断故障来源于哪个或者哪几个变量： When it is detected that the system has failed, it is necessary to further calculate the contribution rate of each variable to the overall
Figure 2013101065672100002DEST_PATH_IMAGE074
, to determine which variable or variables the fault comes from:

Figure 2013101065672100002DEST_PATH_IMAGE076
                         (13)
Figure 2013101065672100002DEST_PATH_IMAGE076
(13)

但在某些情况下，由于变量之间的相关性，会出现几个同时较大的情况，需要对其进行进一步的检测，即  But in some cases, due to the correlation between variables, several At the same time, in larger cases, further testing is required, that is,

Figure 2013101065672100002DEST_PATH_IMAGE080
                          (14)
Figure 2013101065672100002DEST_PATH_IMAGE080
(14)

通过式(14)，可以判断出故障由哪几个变量共同作用，能够更好的识别故障，从而实现有效的过程监控。 Through formula (14), it can be judged which variables are responsible for the fault, which can better identify the fault, so as to realize effective process monitoring.

本发明的有益效果： Beneficial effects of the present invention:

本发明能有效减少误报率，同时对于一些虽然较小但却起重要作用的变量所发生的故障也能够准确地诊断出来。 The invention can effectively reduce the false alarm rate, and at the same time, it can accurately diagnose the faults of some variables that are small but play an important role.

附图说明 Description of drawings

图1为本发明过程监控方法的流程图； Fig. 1 is the flowchart of process monitoring method of the present invention;

图2A为物料C供料温度发生故障后本发明方法的故障检测效果图。 Fig. 2A is a fault detection effect diagram of the method of the present invention after a fault occurs in the feeding temperature of material C.

图2B为物料C供料温度发生故障后本发明方法的故障诊断效果图。 Fig. 2B is a fault diagnosis effect diagram of the method of the present invention after a fault occurs in the feeding temperature of material C.

图3A为物料B的流量和物料C供料温度同时发生故障后本发明方法的故障检测效果图。 Fig. 3A is a fault detection effect diagram of the method of the present invention after the flow rate of material B and the feeding temperature of material C fail at the same time.

图3B为物料B的流量和物料C供料温度同时发生故障后本发明方法的故障诊断效果图。 Fig. 3B is a fault diagnosis effect diagram of the method of the present invention after the flow rate of material B and the feeding temperature of material C fail at the same time.

具体实施方式 Detailed ways

本发明的实施流程图如图1所示，具体实施方式如下： Implementation flow chart of the present invention is as shown in Figure 1, and specific implementation is as follows:

设一个多变量系统的数据矩阵

Figure 581112DEST_PATH_IMAGE002
为 Assuming a data matrix for a multivariate system
Figure 581112DEST_PATH_IMAGE002
for

 

Figure 187674DEST_PATH_IMAGE004
                                (1)
Figure 187674DEST_PATH_IMAGE004
(1)

其中每一行对应一个变量，每一列对应于一个采样样本，

Figure 869454DEST_PATH_IMAGE006
代表系统状态变量的个数，
Figure 623783DEST_PATH_IMAGE008
为每个变量的采样数目。 where each row corresponds to a variable, and each column corresponds to a sampling sample,
Figure 869454DEST_PATH_IMAGE006
represents the number of system state variables,
Figure 623783DEST_PATH_IMAGE008
The number of samples for each variable.

首先对原始矩阵进行相对化变换： First relativize the original matrix:

              (2) (2)

Figure 8814DEST_PATH_IMAGE012
                           (3)
Figure 8814DEST_PATH_IMAGE012
(3)

则称式(2)是对原始数据矩阵

Figure 438658DEST_PATH_IMAGE014
所做的相对化处理，
Figure 617967DEST_PATH_IMAGE016
是相应的相对化算子，
Figure 572717DEST_PATH_IMAGE018
是相对化变换后的数据矩阵。式(3)中的称为比重因子，它是根据实际系统而定的先验信息，
Figure 489037DEST_PATH_IMAGE020
作用在
Figure 404909DEST_PATH_IMAGE022
上，体现了相应变量
Figure 483724DEST_PATH_IMAGE022
在系统中的相对重要程度。 Then it is said that formula (2) is the original data matrix
Figure 438658DEST_PATH_IMAGE014
The relativization done,
Figure 617967DEST_PATH_IMAGE016
is the corresponding relativization operator,
Figure 572717DEST_PATH_IMAGE018
is the data matrix after relativization transformation. In formula (3) is called the specific gravity factor, which is a priori information based on the actual system,
Figure 489037DEST_PATH_IMAGE020
work on
Figure 404909DEST_PATH_IMAGE022
above, reflecting the corresponding variable
Figure 483724DEST_PATH_IMAGE022
relative importance in the system.

 然后求出

Figure 919384DEST_PATH_IMAGE018
的均值向量
Figure 379447DEST_PATH_IMAGE024
and then find
Figure 919384DEST_PATH_IMAGE018
The mean vector of
Figure 379447DEST_PATH_IMAGE024

                

Figure 330085DEST_PATH_IMAGE026
                    (4)
Figure 330085DEST_PATH_IMAGE026
(4)

和

Figure 150274DEST_PATH_IMAGE018
的协方差矩阵 and
Figure 150274DEST_PATH_IMAGE018
The covariance matrix of

Figure 632257DEST_PATH_IMAGE030
                     (5)
Figure 632257DEST_PATH_IMAGE030
(5)

假设当前时刻为

Figure 7874DEST_PATH_IMAGE008
，记包含有故障的采样数据时刻为
Figure 113977DEST_PATH_IMAGE032
，记正常采样数据(不包含故障)的时刻为
Figure 586547DEST_PATH_IMAGE034
，且满足 Suppose the current moment is
Figure 7874DEST_PATH_IMAGE008
, record the time of sampled data containing faults as
Figure 113977DEST_PATH_IMAGE032
, record the time of normal sampling data (excluding faults) as
Figure 586547DEST_PATH_IMAGE034
, and satisfy

Figure 637679DEST_PATH_IMAGE036
                               (6)
Figure 637679DEST_PATH_IMAGE036
(6)

在时刻

Figure 749861DEST_PATH_IMAGE008
,从正常采样数据集
Figure 974169DEST_PATH_IMAGE038
中连续选取
Figure 973349DEST_PATH_IMAGE040
个采样数据，并形成相应的局部数据矩阵
Figure 946115DEST_PATH_IMAGE042
at the moment
Figure 749861DEST_PATH_IMAGE008
, from the normally sampled dataset
Figure 974169DEST_PATH_IMAGE038
continuous selection
Figure 973349DEST_PATH_IMAGE040
sampling data, and form the corresponding local data matrix
Figure 946115DEST_PATH_IMAGE042

Figure 358642DEST_PATH_IMAGE044
                          (7)
Figure 358642DEST_PATH_IMAGE044
(7)

当下一个时刻 即

Figure 324324DEST_PATH_IMAGE046
时刻到来后，就形成的相应局部数据矩阵
Figure 505907DEST_PATH_IMAGE048
when the next moment is
Figure 324324DEST_PATH_IMAGE046
After the time arrives, the corresponding local data matrix is formed
Figure 505907DEST_PATH_IMAGE048

Figure 882530DEST_PATH_IMAGE050
                          (8)
Figure 882530DEST_PATH_IMAGE050
(8)

这里 here

                   (9) (9)

根据公式(5)-(6)可求出

Figure 286147DEST_PATH_IMAGE054
和
Figure 216843DEST_PATH_IMAGE056
。 According to the formula (5)-(6), it can be obtained
Figure 286147DEST_PATH_IMAGE054
and
Figure 216843DEST_PATH_IMAGE056
.

定义基于局部采样数据信息的局部信息增量矩阵为

Figure 577417DEST_PATH_IMAGE058
，计算得到 Define the local information increment matrix based on local sampling data information as
Figure 577417DEST_PATH_IMAGE058
, calculated to get

Figure 151487DEST_PATH_IMAGE060
             (10)
Figure 151487DEST_PATH_IMAGE060
(10)

则基于局部采样数据信息的局部信息增量均值

Figure 521289DEST_PATH_IMAGE062
为  Then the local information increment mean value based on the local sampling data information
Figure 521289DEST_PATH_IMAGE062
for

Figure 349567DEST_PATH_IMAGE064
                          (11)
Figure 349567DEST_PATH_IMAGE064
(11)

记固定窗口长度为

Figure 835038DEST_PATH_IMAGE040
的局部正常采样数据动态阈值为 Note that the fixed window length is
Figure 835038DEST_PATH_IMAGE040
The local normal sampling data dynamic threshold of

Figure 555049DEST_PATH_IMAGE068
                           (12)
Figure 555049DEST_PATH_IMAGE068
(12)

通过比较某一时刻的局部采样数据信息的局部信息增量均值与系统动态阈值的特定关系大小即可检测系统在该时刻是否发生异常: By comparing the specific relationship between the local information incremental mean value of the local sampling data information at a certain moment and the system dynamic threshold, it is possible to detect whether the system is abnormal at this moment:

1)若

Figure 565730DEST_PATH_IMAGE070
，则表示系统在该时刻有故障发生。此时，为保证故障检测的准确性，在检测下一时刻系统是否发生故障时，要去掉当前发生故障时刻的数据，即滑动窗口位置不发生变化。                               1) if
Figure 565730DEST_PATH_IMAGE070
, it means that there is a fault in the system at this moment. At this time, in order to ensure the accuracy of fault detection, when detecting whether the system is faulty at the next moment, the data at the current time when the fault occurs should be deleted, that is, the position of the sliding window does not change.

2)若

Figure 720637DEST_PATH_IMAGE072
，则表示系统该时刻没有故障发生。此时，滑动窗口继续采样，来进行下一个时刻的检测。 2) if
Figure 720637DEST_PATH_IMAGE072
, it means that there is no fault in the system at this moment. At this time, the sliding window continues to sample to detect the next moment.

当检测出系统已经发生故障后，需要进一步计算各变量对总体的贡献率

Figure 754452DEST_PATH_IMAGE074
，以判断故障来源于哪个或者哪几个变量： When it is detected that the system has failed, it is necessary to further calculate the contribution rate of each variable to the overall
Figure 754452DEST_PATH_IMAGE074
, to determine which variable or variables the fault comes from:

Figure 466056DEST_PATH_IMAGE076
                         (13)
Figure 466056DEST_PATH_IMAGE076
(13)

但在某些情况下，由于变量之间的相关性，会出现几个

Figure 16730DEST_PATH_IMAGE078
同时较大的情况，需要对其进行进一步的检测，即  But in some cases, due to the correlation between variables, several
Figure 16730DEST_PATH_IMAGE078
At the same time, in larger cases, further testing is required, that is,

Figure 155587DEST_PATH_IMAGE080
                          (14)
Figure 155587DEST_PATH_IMAGE080
(14)

通过式(14)，可以判断出故障由哪几个变量共同作用，能够更好的识别故障，从而实现有效的过程监控。 Through formula (14), it can be judged which variables are responsible for the fault, which can better identify the fault, so as to realize effective process monitoring.

方法试验 method test

为了验证本发明方法的有效性，考虑如下随机变量及其线性组合构建6个系统变量，对由这6个变量组成的系统的生产过程进行监控，假设此过程有3种反应物(A,B,C)和2种产物(D,E)，变量

Figure DEST_PATH_IMAGE082
代表物料A的流量，
Figure DEST_PATH_IMAGE084
代表物料B的流量，代表物料C的流量，
Figure DEST_PATH_IMAGE088
代表物料B的浓度，
Figure DEST_PATH_IMAGE090
代表物料A供料温度，
Figure DEST_PATH_IMAGE092
代表物料C供料温度。 In order to verify the effectiveness of the method of the present invention, consider the following random variables and their linear combinations to construct 6 system variables, and monitor the production process of the system made up of these 6 variables, assuming that this process has 3 kinds of reactants (A, B ,C) and 2 products (D,E), variable
Figure DEST_PATH_IMAGE082
represents the flow rate of material A,
Figure DEST_PATH_IMAGE084
represents the flow rate of material B, represents the flow rate of material C,
Figure DEST_PATH_IMAGE088
represents the concentration of material B,
Figure DEST_PATH_IMAGE090
Represents the feed temperature of material A,
Figure DEST_PATH_IMAGE092
Represents the material C feed temperature.

Figure DEST_PATH_IMAGE094
                       (16)
Figure DEST_PATH_IMAGE094
(16)

其中

Figure DEST_PATH_IMAGE096
，比重因子为2,10,1,1,1,2，固定窗口长度为
Figure DEST_PATH_IMAGE098
。下述2个实验均从故障的检测性能以及诊断性能两方面来考查本发明方法的优越性，检测时，1至800时刻超过控制限时为误报，801至1000时刻低于控制限时则为漏报。 in
Figure DEST_PATH_IMAGE096
, the specific gravity factors are 2,10,1,1,1,2, and the fixed window length is
Figure DEST_PATH_IMAGE098
. The following two experiments all examine the superiority of the method of the present invention from two aspects of fault detection performance and diagnostic performance. During detection, it is a false alarm when the time from 1 to 800 exceeds the control limit, and it is a leak when the time from 801 to 1000 is lower than the control limit. report.

(1)对测试数据中的变量x6后200个(即801-1000)样本引入幅值为3.2的恒偏差故障。仿真结果如图2A和图2B所示。 (1) For the last 200 (ie 801-1000) samples of the variable x 6 in the test data, a constant deviation fault with an amplitude of 3.2 is introduced. The simulation results are shown in Figure 2A and Figure 2B.

从图中可以看出，本发明方法能有效检测出系统发生了异常，并能诊断出是物料C供料温度发生故障。 It can be seen from the figure that the method of the present invention can effectively detect the abnormality of the system, and can diagnose that the failure of the feeding temperature of the material C occurs.

(2)对测试数据中的变量

Figure DEST_PATH_IMAGE100
和
Figure DEST_PATH_IMAGE102
后200个样本点分别引入幅值为0.1和10的恒偏差故障。仿真结果如图3A和图3B所示。 (2) For the variables in the test data
Figure DEST_PATH_IMAGE100
and
Figure DEST_PATH_IMAGE102
The last 200 sample points introduce constant deviation faults with amplitudes of 0.1 and 10, respectively. The simulation results are shown in Figure 3A and Figure 3B.

从图中可以看出，当

Figure 255128DEST_PATH_IMAGE100
和
Figure 442527DEST_PATH_IMAGE102
两个变量均发生故障时，本发明方法能有效检测出系统发生了异常，并能诊断出物料B的流量和物料C供料温度同时发生故障，说明本发明方法有很好的故障检测与诊断特性，能够很好的进行过程监控。 It can be seen from the figure that when
Figure 255128DEST_PATH_IMAGE100
and
Figure 442527DEST_PATH_IMAGE102
When both variables fail, the method of the present invention can effectively detect that an abnormality has occurred in the system, and can diagnose that the flow rate of material B and the feed temperature of material C are simultaneously faulty, indicating that the method of the present invention has good fault detection and diagnosis Features, can be very good for process monitoring.

Claims (1)

1. 基于相对化变换的信息增量矩阵的工业过程监控方法，其特征在于： 1. The industrial process monitoring method based on the information increment matrix of relativization transformation, it is characterized in that: 设一个多变量系统的数据矩阵                                               
Figure 2013101065672100001DEST_PATH_IMAGE002
为 Assuming a data matrix for a multivariate system
Figure 2013101065672100001DEST_PATH_IMAGE002
for  
Figure 2013101065672100001DEST_PATH_IMAGE004
                            (1)
Figure 2013101065672100001DEST_PATH_IMAGE004
(1) 其中每一行对应一个变量，每一列对应于一个采样样本，代表系统状态变量的个数，
Figure DEST_PATH_IMAGE008
为每个变量的采样数目； where each row corresponds to a variable, and each column corresponds to a sampling sample, represents the number of system state variables,
Figure DEST_PATH_IMAGE008
The number of samples for each variable; 首先对原始矩阵进行相对化变换： First relativize the original matrix:
Figure DEST_PATH_IMAGE010
              (2)
Figure DEST_PATH_IMAGE010
(2)                            (3) (3) 则称式(2)是对原始数据矩阵
Figure DEST_PATH_IMAGE014
所做的相对化处理，
Figure DEST_PATH_IMAGE016
是相应的相对化算子，是相对化变换后的数据矩阵；式(3)中的
Figure DEST_PATH_IMAGE020
称为比重因子，它是根据实际系统而定的先验信息，
Figure 70028DEST_PATH_IMAGE020
作用在上，体现了相应变量
Figure 442104DEST_PATH_IMAGE022
在系统中的相对重要程度； Then it is said that formula (2) is the original data matrix
Figure DEST_PATH_IMAGE014
The relativization done,
Figure DEST_PATH_IMAGE016
is the corresponding relativization operator, is the data matrix after relativization transformation; in formula (3)
Figure DEST_PATH_IMAGE020
is called the specific gravity factor, which is a priori information based on the actual system,
Figure 70028DEST_PATH_IMAGE020
work on above, reflecting the corresponding variable
Figure 442104DEST_PATH_IMAGE022
relative importance in the system;  然后求出的均值向量
Figure DEST_PATH_IMAGE024
and then find The mean vector of
Figure DEST_PATH_IMAGE024
                                     (4) (4) 和
Figure 254650DEST_PATH_IMAGE018
的协方差矩阵
Figure DEST_PATH_IMAGE028
and
Figure 254650DEST_PATH_IMAGE018
The covariance matrix of
Figure DEST_PATH_IMAGE028
Figure DEST_PATH_IMAGE030
                     (5)
Figure DEST_PATH_IMAGE030
(5) 设当前时刻为
Figure 961007DEST_PATH_IMAGE008
，记包含有故障的采样数据时刻为
Figure DEST_PATH_IMAGE032
，记正常采样数据(不包含故障)的时刻为
Figure DEST_PATH_IMAGE034
，且满足 Let the current moment be
Figure 961007DEST_PATH_IMAGE008
, record the time of sampled data containing faults as
Figure DEST_PATH_IMAGE032
, record the time of normal sampling data (excluding faults) as
Figure DEST_PATH_IMAGE034
, and satisfy
Figure DEST_PATH_IMAGE036
                               (6)
Figure DEST_PATH_IMAGE036
(6) 在时刻
Figure 317033DEST_PATH_IMAGE008
,从正常采样数据集
Figure DEST_PATH_IMAGE038
中连续选取
Figure DEST_PATH_IMAGE040
个采样数据，并形成相应的局部数据矩阵
Figure DEST_PATH_IMAGE042
at the moment
Figure 317033DEST_PATH_IMAGE008
, from the normally sampled dataset
Figure DEST_PATH_IMAGE038
continuous selection
Figure DEST_PATH_IMAGE040
sampling data, and form the corresponding local data matrix
Figure DEST_PATH_IMAGE042
Figure DEST_PATH_IMAGE044
                          (7)
Figure DEST_PATH_IMAGE044
(7) 当下一个时刻 即
Figure DEST_PATH_IMAGE046
时刻到来后，就形成的相应局部数据矩阵
Figure DEST_PATH_IMAGE048
when the next moment is
Figure DEST_PATH_IMAGE046
After the time arrives, the corresponding local data matrix is formed
Figure DEST_PATH_IMAGE048
                           (8) (8) 这里 here
Figure DEST_PATH_IMAGE052
                   (9)
Figure DEST_PATH_IMAGE052
(9) 根据公式(5)-(6)可求出
Figure DEST_PATH_IMAGE054
和
Figure DEST_PATH_IMAGE056
； According to the formula (5)-(6), it can be obtained
Figure DEST_PATH_IMAGE054
and
Figure DEST_PATH_IMAGE056
; 定义基于局部采样数据信息的局部信息增量矩阵为
Figure DEST_PATH_IMAGE058
，计算得到 Define the local information increment matrix based on local sampling data information as
Figure DEST_PATH_IMAGE058
, calculated to get
Figure DEST_PATH_IMAGE060
             (10)
Figure DEST_PATH_IMAGE060
(10) 则基于局部采样数据信息的局部信息增量均值
Figure DEST_PATH_IMAGE062
为  Then the local information increment mean value based on the local sampling data information
Figure DEST_PATH_IMAGE062
for                           (11) (11) 记固定窗口长度为
Figure 45693DEST_PATH_IMAGE040
的局部正常采样数据动态阈值为
Figure DEST_PATH_IMAGE066
Note that the fixed window length is
Figure 45693DEST_PATH_IMAGE040
The local normal sampling data dynamic threshold of
Figure DEST_PATH_IMAGE066
Figure DEST_PATH_IMAGE068
                           (12)
Figure DEST_PATH_IMAGE068
(12) 通过比较某一时刻的局部采样数据信息的局部信息增量均值与系统动态阈值的特定关系大小即可检测系统在该时刻是否发生异常： By comparing the specific relationship between the local information incremental mean value of the local sampling data information at a certain moment and the system dynamic threshold, it is possible to detect whether the system is abnormal at this moment: 1)若
Figure DEST_PATH_IMAGE070
，则表示系统在该时刻有故障发生；此时，为保证故障检测的准确性，在检测下一时刻系统是否发生故障时，要去掉当前发生故障时刻的数据，即滑动窗口位置不发生变化；                               1) if
Figure DEST_PATH_IMAGE070
, it means that the system has a fault at this moment; at this time, in order to ensure the accuracy of fault detection, when detecting whether the system is faulty at the next moment, the data at the current fault time should be deleted, that is, the position of the sliding window does not change; 2)若
Figure DEST_PATH_IMAGE072
，则表示系统该时刻没有故障发生；此时，滑动窗口继续采样，来进行下一个时刻的检测； 2) if
Figure DEST_PATH_IMAGE072
, it means that there is no fault in the system at this moment; at this time, the sliding window continues to sample for detection at the next moment; 当检测出系统已经发生故障后，需要进一步计算各变量对总体的贡献率，以判断故障来源于哪个或者哪几个变量： When it is detected that the system has failed, it is necessary to further calculate the contribution rate of each variable to the overall , to determine which variable or variables the fault comes from:
Figure DEST_PATH_IMAGE076
                         (13)
Figure DEST_PATH_IMAGE076
(13) 但在某些情况下，由于变量之间的相关性，会出现几个
Figure DEST_PATH_IMAGE078
同时较大的情况，需要对其进行进一步的检测，即  But in some cases, due to the correlation between variables, several
Figure DEST_PATH_IMAGE078
At the same time, in larger cases, further testing is required, that is,
Figure DEST_PATH_IMAGE080
                          (14)
Figure DEST_PATH_IMAGE080
(14) 通过式(14)，可以判断出故障由哪几个变量共同作用，能够更好的识别故障，从而实现有效的过程监控。 Through formula (14), it can be judged which variables are responsible for the fault, which can better identify the fault, so as to realize effective process monitoring.
CN2013101065672A 2013-03-28 2013-03-28 Relative-transform-based industrial process monitoring method of information increment matrix Pending CN103246277A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2013101065672A CN103246277A (en) 2013-03-28 2013-03-28 Relative-transform-based industrial process monitoring method of information increment matrix

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2013101065672A CN103246277A (en) 2013-03-28 2013-03-28 Relative-transform-based industrial process monitoring method of information increment matrix

Publications (1)

Publication Number Publication Date
CN103246277A true CN103246277A (en) 2013-08-14

Family

ID=48925855

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2013101065672A Pending CN103246277A (en) 2013-03-28 2013-03-28 Relative-transform-based industrial process monitoring method of information increment matrix

Country Status (1)

Country Link
CN (1) CN103246277A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103853152A (en) * 2014-03-21 2014-06-11 北京工业大学 Batch process failure monitoring method based on AR-PCA (Autoregressive Principal Component Analysis)
CN111046331A (en) * 2019-11-21 2020-04-21 中国航空无线电电子研究所 Health monitoring method based on signal abnormity
CN117571347A (en) * 2023-10-20 2024-02-20 河北白沙烟草有限责任公司 Equipment health state monitoring and evaluating method and device and electronic equipment

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101308385A (en) * 2008-07-11 2008-11-19 东北大学 Nonlinear Process Fault Detection Method Based on 2D Dynamic Kernel Principal Component Analysis
WO2012040916A1 (en) * 2010-09-29 2012-04-05 东北大学 Fault monitoring method of continuous annealing process based on recursive kernel principal component analysis

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101308385A (en) * 2008-07-11 2008-11-19 东北大学 Nonlinear Process Fault Detection Method Based on 2D Dynamic Kernel Principal Component Analysis
WO2012040916A1 (en) * 2010-09-29 2012-04-05 东北大学 Fault monitoring method of continuous annealing process based on recursive kernel principal component analysis

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
文成林: "基于相对化变换的信息增量矩阵的故障诊断方法", 《万方学术会议数据库 第23届过程控制会议》 *
文成林等: "基于信息增量矩阵的故障诊断方法", 《自动化学报》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103853152A (en) * 2014-03-21 2014-06-11 北京工业大学 Batch process failure monitoring method based on AR-PCA (Autoregressive Principal Component Analysis)
CN103853152B (en) * 2014-03-21 2016-08-17 北京工业大学 A kind of batch process fault monitoring method based on AR-PCA
CN111046331A (en) * 2019-11-21 2020-04-21 中国航空无线电电子研究所 Health monitoring method based on signal abnormity
CN117571347A (en) * 2023-10-20 2024-02-20 河北白沙烟草有限责任公司 Equipment health state monitoring and evaluating method and device and electronic equipment
CN117571347B (en) * 2023-10-20 2024-09-20 河北白沙烟草有限责任公司 Equipment health state monitoring and evaluating method and device and electronic equipment

Similar Documents

Publication Publication Date Title
CN104915568B (en) Satellite telemetering data method for detecting abnormality based on DTW
US7319939B2 (en) Predictive maintenance and management of aging of installed cables
CN111324105B (en) A Fault Detection and Separation Method for Small Sensors Based on Statistical Mahalanobis Distance
CN103776480B (en) Small fault detection method based on repeatedly rolling average and device
EP3580623B1 (en) Method for detection and isolation of faulty sensors
CN105116872A (en) Fault diagnosis method based on metric learning and time sequence during industrial process
CN106886213A (en) A kind of batch process fault detection method based on core similarity Support Vector data description
Ji et al. Diagnosis of sensor precision degradation using Kullback‐Leibler divergence
CN104142680B (en) Multiple-sensor fault diagnosis system and method based on robust input training neural network
CN103246277A (en) Relative-transform-based industrial process monitoring method of information increment matrix
CN103853144B (en) Spot sensor fault detection method based on oil recovery creation data
CN117761423A (en) Power distribution network fault prediction method
CN105389475B (en) A kind of electric power factory equipment fault detection method based on WAVELET PACKET DECOMPOSITION
CN117575027B (en) A sensor fault diagnosis method using knowledge to enhance spatiotemporal causality
CN110009033A (en) An early warning model of drilling process abnormality based on dynamic principal component analysis
CN114200914B (en) A quality-related early fault detection method based on MW-OCCA
Rahim et al. Quality monitoring of a closed-loop system with parametric uncertainties and external disturbances: a fault detection and isolation approach
CN104848883A (en) Sensor noise and fault judging method based on sparse representation
CN111473022B (en) Method for analyzing and diagnosing fault vibration of servo hydraulic system
WO2018228755A1 (en) Method of fault detection in a machine
CN104078381A (en) Method for setting specification limit of measurement machine monitoring chart
CN114707413A (en) Wafer test detection method based on long-short term memory network and sliding window
CN112114578A (en) Steady method for multi-process multivariable process online monitoring and abnormal source diagnosis
CN119641981B (en) A Fault Identification Method for Electric Actuator
KR20160039078A (en) Method for analyzing signal of power plant pipe

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C02 Deemed withdrawal of patent application after publication (patent law 2001)
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20130814