CN101526433B - Method for automatically diagnosing faults of monitoring system - Google Patents

Abstract

The invention relates to a method for automatically diagnosing faults of a monitoring system. The method comprises the following steps: analyzing the association degree of sequences of different measuring values; building a fuzzy relation matrix by utilizing the association degrees of the measuring value sequences; and carrying out fuzzy clustering analysis on the measuring value sequences to obtain the reason of measuring value abnormity, and judging the operation state of the monitoring system. The method uses a computer program to carry out quantitative analysis on relations between each measuring point and each measuring value so as to ensure that the monitoring system can diagnose whether the abnormal measuring values are initiated by self faults or not so as to avoid unnecessary economic and social cost investment brought by incorrect alarm caused by incorrect judge to be convenient to adopt measures timely to maintain the monitoring system, so the intelligent level and robustness of the structure monitoring system are improved, and the practicability is greater.

Description

Method for automatically diagnosing faults of monitoring system

Technical field

The invention belongs to the structure monitoring field, be specifically related to a kind of analysis and judgement method for the monitoring system own health status.

Background technology

The long term monitoring of large scale structure obtains paying attention to day by day at present, and system can break down sometimes in observation process, and particularly the isostructural monitoring periods of bridge generally all is longer than the term of validity of monitoring equipment, so the damage problem of monitoring equipment is difficult to avoid.The damage of monitoring equipment often is accompanied by the generation of unusual measured value, and the generation of abnormal information mainly contains 2 kinds of reasons: 1. really caused by the deterioration of structure itself; 2. monitoring system faults itself.Just the false alarm phenomenon may occur if mistake is used as the textural anomaly processing to the monitoring system faults itself, therefore unusual genetic analysis is follow-up analysis and base of prediction to the sensor measured value, and whether its accuracy is related to whole monitoring effectively reliable.To the unusual judgement more complicated of measured value, have only in actual applications at present some qualitatively principle can help to analyze these exceptional values and whether cause by the monitoring system self problem, as: malformation destroys a progressivity usually; Vector in the structural response is general to keep certain directivity etc., though can judge some tangible faults of monitoring system by these principles, then powerless for the system failure more hidden performance on measured value.In fact, be arranged between each measuring point measured value that the position is leaned on nearerly in the same works and exist certain contact, the distortion of structure generally can not only limit to influence certain isolated measuring point with destruction, but certain scope is arranged, be arranged in same engineering position, exist certain contact between each measuring point measured value that the position is leaned on nearerly.So, when exceptional value is analyzed to the consecutive point observed reading relatively be necessary, but the quantification of analytical approach at present still do not have to(for) the relevance between measured value diagnoses the measured value abnormal occurrence, therefore being necessary to develop a kind of new method remedies this deficiency.

Summary of the invention

Be difficult to avoid failure problems though the present invention is directed to present structure monitoring system, but owing to the not enough present situation that has influence on the reliability of monitoring system of intelligent level, principle based on " when the monitoring system operate as normal; each measuring point measured value that structural change reply position is leaned on nearerly is all influential " is quantized the contact between measured value with grey relational grade, and has proposed the structure monitoring system fault self-diagnosis method based on clustering analysis of grey relational grade between measured value.

The technical scheme that realizes the object of the invention is: method for automatically diagnosing faults of monitoring system comprises the following steps:

(1) sequence of different measured values is carried out correlation analysis

The computing method of grey relational grade are many between sequence, adopt the absolute degree of association to analyze as example here.If in the original data sequence, x _iAnd x _jBe respectively reference sequences and comparative sequences (also being the measured value sequence of monitoring system measuring point I and measuring point J a period of time), that is:

X _i＝{x _i(k)|k＝1，2，...，n} (1)

X _j＝{x _j(k)|k＝1，2，...，n} (2)

To the comparison between the different sequences of length, can will lack the sequence polishing with methods such as interpolation, like this degree of association value there is certain influence certainly.Can ask its initial point pulverised to resemble X to two identical sequences of length _l ⁰:

$X_{\mathrm{<figure-callout\; id="0"\; label="l"\; filenames="H2009100309902E00012.png"\; state="\{\{state\}\}">l</figure-callout>}}^0=(x_l^0\left(1\right),x_l^0\left(2\right),...,x_l^0\left(n\right)),(l=i,j)---\left(3\right)$

Wherein $x_l^0\left(k\right)=x_l\left(k\right)-x_l\left(1\right),$ Can try to achieve the absolute degree of association of two sequences by formula (4):

$r_{\mathrm{ij}}=\frac{1+\left|S_i\right|+\left|S_j\right|}{1+\left|S_i\right|+\left|S_j\right|+|S_j-S_i|}---\left(4\right)$

In the formula

$S_l={\&Integral;}_1^n(X_l-x_l\left(1\right))\mathrm{dt}={\&Integral;}_1^{\mathrm{<figure-callout\; id="0"\; label="n\; X\; l"\; filenames="H2009100309902E00012.png"\; state="\{\{state\}\}">n</figure-callout>}}{X}_l^{}{}_0^{}\mathrm{dt}\&ap;\underset{k=2}{\overset{n-1}{\mathrm{\&Sigma;}}}{x}_l^0\left(k\right)+\frac{1}{2}{x}_l^0\left(n\right),(l=i,j)$

(2) utilize degree of association member fuzzy relation matrix between the measured value sequence

Suppose that monitoring system has m measuring point, each measuring point respectively records n measured value, then this m section measured value sequence Y={Y ₁, Y ₂..., Y _mConstitute the sample vector collection of n-dimensional space, arbitrary sample vector Y _iFor:

Y _i＝{y _i1，y _i2，...，y _in} ^T

Y wherein _Ij(i=1,2 ..., m; J=1,2 ..., n) be j measured value of i measuring point.For carrying out fuzzy cluster analysis, the degree of association r that (1) step is tried to achieve _Ij(0≤r _Ij≤ 1) quantizes among the Y degree closer to each other between element in twos, with r _IjThe matrix R that forms is called fuzzy relation matrix.

(3) the measured value sequence is carried out fuzzy cluster analysis and analyze the measured value abnormal cause

According to the time dependent situation of measured value sequence cluster result, can judge the monitoring system running status, when certain measured value sequence declines to a great extent with similar other measured value serial correlation degree, and cause that cluster result is corresponding to change, even this measured value still will be classified as a class with original other similar all measured values, then some measured value in other class also be included into such originally.Variation has taken place in the relevance between the variation explanation measured value of this cluster result, can the preliminary judgement monitoring system self break down.

Beneficial effect of the present invention is:

(1) compares with the method for manually utilizing some qualitative principles that unusual measured value is judged, structure monitoring system fault self-diagnosis method based on clustering analysis of grey relational grade between measured value can carry out quantitative analysis to the contact that exists between each measuring point measured value with computer program, whether monitoring system can be caused by faults itself unusual measured value diagnoses, avoid owing to erroneous judgement causes the unnecessary economy and society cost input that false alarm brings, also be convenient in time take measures monitoring system is keeped in repair, improved the intelligent level and the robustness of structure monitoring system, so practicality is bigger.With the bridge structure monitoring system is example, if the unusual erroneous judgement of measured value that faults of monitoring system is caused is structure deterioration, then certainly will will carry out manual detection or maintenance to bridge, if this suspended traffic about 8 hours, the direct economic loss that is produced can be up to 3,000,000 yuan about.Adopt the present invention can make monitoring system can utilize software to analyze a part automatically because monitoring system owing to the unusual measured value phenomenon that faults itself causes, has reduced because the loss that erroneous judgement causes.

(2) adopt analytical approach of the present invention to be applied in the others of structure monitoring.Along with the development of monitoring technology, measuring point quantity significantly increases, and monitoring periods is shorter and shorter, causes Monitoring Data huge unusually.Carry out cluster analysis and can understand the running status that the inner link of measured value between monitoring instrument helps grasping structure, make that the formulation of materials such as form is more succinct.In addition, some instrument takes place can play one's part to the full when unusual in structure, and at its measured value at ordinary times with same detection project instrument measured value difference is very little on every side, representational instrument emphasis monitoring can be on purpose selected by cluster analysis, the storage and the intractability of monitoring materials can be reduced.

Description of drawings

Fig. 1 is the embodiment of the invention 1 substrate pore water pressure measuring point arrangenent diagram.

Fig. 2 is the measured value sequence of the embodiment of the invention 1 pore water pressure measuring point

Embodiment

Be described further below in conjunction with embodiment.

Embodiment 1

Bury pore pressure gauge 9 points under certain bridge basis underground, observation basis pore water pressure down distributes, and Figure 1 shows that substrate pore water pressure measuring point arrangenent diagram.The measured value sequence that Fig. 2 is these pore water pressure measuring points after in the April, 2002, can see among the figure that the measured value variation mainly can be divided into two stages, first stage is the stage that about 150 days of April to September, pore water pressure increased along with the foundation construction additional load.Subordinate phase dribbles and is tending towards consolidation stage of definite value for the construction pore water pressure that finishes.33131, the measured value sequence of 33231 measuring points remains near 0 value always.

(1) sequence of different measured values is carried out correlation analysis:

10 measured values in year September in April, 2002 to 2004 monitor value are divided into two sections sequences analyze, every section is continuous 5 monitoring gained measured value sequences.According to formula

$r_{\mathrm{ij}}=\frac{1+\left|S_i\right|+\left|S_j\right|}{1+\left|S_i\right|+\left|S_j\right|+|S_j-S_i|}---\left(4\right)$

Can calculate absolute degree of association matrix between two stage measured values separately respectively, this is two 9 * 9 a matrix, and ranks are counted i, j and represented 33111～33331 measuring points, matrix element r in order respectively in the matrix _IjRepresent the degree of association between the measured value sequence.

(2) utilize degree of association member fuzzy relation matrix between the measured value sequence

2002.4～2002.7 each measuring point measured value degree of association matrix [r wherein _Ij] ₁(as shown in table 1), and 2002.7～2002.9 each measuring point measured value degree of association matrix are [r _Ij] ₂(as shown in table 2).Observe two matrixes and can find that certain fluctuation can take place the degree of association between the measured value of different measuring points as time passes, but amplitude is all little, but the degree of association between measuring point that destroys and normal measuring point sharply descends.

Each measuring point measured value degree of association matrix of table 12002.4～2002.7

$\left[\begin{array}{ccccccccc}1.000& 0.973& 0.977& 0.933& 0.985& 0.990& 0.900& 0.975& 0.990\\ 0.973& 1.000& 0.953& 0.957& 0.988& 0.964& 0.922& 0.951& 0.964\\ 0.977& 0.953& 1.000& 0.915& 0.963& 0.986& 0.884& 0.997& 0.986\\ 0.933& 0.957& 0.915& 1.000& 0.946& 0.925& 0.961& 0.913& 0.925\\ 0.985& 0.988& 0.963& 0.946& 1.000& 0.975& 0.912& 0.961& 0.975\\ 0.990& 0.964& 0.986& 0.925& 0.975& 1.000& 0.892& 0.984& 0.999\\ 0.900& 0.922& 0.884& 0.961& 0.912& 0.892& 1.000& 0.882& 0.892\\ 0.975& 0.951& 0.997& 0.913& 0.961& 0.984& 0.882& 1.000& 0.984\\ 0.990& 0.964& 0.986& 0.925& 0.975& 0.999& 0.892& 0.984& 1.000\end{array}\right]$

Each measuring point measured value degree of association matrix of table 22002.7～2002.9

$\left[\begin{array}{ccccccccc}1.000& 0.976& 0.758& 0.946& 0.987& 0.758& 0.929& 0.990& 0.940\\ 0.976& 1.000& 0.746& 0.968& 0.965& 0.746& 0.951& 0.967& 0.919\\ 0.758& 0.746& 1.000& 0.731& 0.765& 1.000& 0.722& 0.763& 0.793\\ 0.946& 0.968& 0.731& 1.000& 0.935& 0.731& 0.981& 0.937& 0.893\\ 0.987& 0.965& 0.765& 0.935& 1.000& 0.765& 0.919& 0.997& 0.951\\ 0.758& 0.746& 1.000& 0.731& 0.765& 1.000& 0.722& 0.763& 0.793\\ 0.929& 0.951& 0.722& 0.981& 0.919& 0.722& 1.000& 0.921& 0.878\\ 0.990& 0.967& 0.763& 0.937& 0.997& 0.763& 0.921& 1.000& 0.948\\ 0.940& 0.919& 0.793& 0.893& 0.951& 0.793& 0.878& 0.948& 1.000\end{array}\right]$

With [r _Ij] ₂As fuzzy relation matrix, adopt the closure method to try to achieve fuzzy equivalent matrix and carry out cluster, get the horizontal λ of cut set=0.975 and measuring point can be divided into four classes, be respectively: { 33111,33121,33211,33221}, 33131, and 33231}, 33321,33331}, { 33311}.Cluster result has reflected the situation of measuring point association on space distribution substantially as can be seen, and 33131,33231 liang of destruction measuring points are also separated from other measuring point.

Claims (3)

1. method for automatically diagnosing faults of monitoring system is characterized in that, this method comprises the following steps:

(1) sequence of different measured values is carried out correlation analysis;

(2) utilize degree of association member fuzzy relation matrix between the measured value sequence;

(3) the measured value sequence is carried out fuzzy cluster analysis, analyze the measured value abnormal cause, the monitoring system running status is judged;

Adopt the absolute degree of association to analyze in the described step (1), in the described measured value sequence, x _iAnd x _jBe respectively reference sequences and comparative sequences, that is:

X _i＝{x _i(k)|k＝1，2，...，n} (1)

X _j＝{x _j(k)|k＝1，2，...，n} (2)

Ask its initial point pulverised to resemble to two identical sequences of length

$X_l^0=(x_l^0\left(1\right),x_l^0\left(2\right),...,x_l^0\left(n\right)),(l=i,j)---\left(3\right)$

Wherein

Can try to achieve the absolute degree of association of two sequences by formula (4):

$r_{\mathrm{ij}}=\frac{1+\left|S_i\right|+\left|S_j\right|}{1+\left|S_i\right|+\left|S_j\right|+|S_j-S_i|}---\left(4\right)$

In the formula

$S_l={\&Integral;}_1^n(X_l-x_l\left(1\right))\mathrm{dt}={\&Integral;}_1^n{X}_l^0\mathrm{dt}\&ap;\underset{k=2}{\overset{n-1}{\mathrm{\&Sigma;}}}{x}_l^0\left(k\right)+\frac{1}{2}{x}_l^0\left(n\right),(l=i,j);$

In the described step (2), suppose that monitoring system has m measuring point, each measuring point respectively records n measured value, then this m section measured value sequence Y={Y ₁, Y ₂..., Y _mConstitute the sample vector collection of n-dimensional space, arbitrary sample vector Y _iFor:

${\stackrel{\&OverBar;}{Y}}_i={\{y_{i1},y_{i2},...,y_{\mathrm{in}}\}}^T$

Y wherein _Ij(i=1,2 ..., m; J=1,2 ..., n) be j measured value of i measuring point;

With the degree of association r that tries to achieve in the step (1) _Ij(0≤r _Ij≤ 1), quantizes among the Y degree closer to each other between element in twos, with r _IjThe matrix R that forms is as fuzzy relation matrix.

2. fault self-diagnosis method according to claim 1, it is characterized in that, in the described step (3), if certain measured value sequence declines to a great extent with similar other measured value serial correlation degree, and cause that cluster result is corresponding to change, but then preliminary judgement monitoring system self breaks down.

3. fault self-diagnosis method according to claim 1 is characterized in that, in the described step (1), to the comparison between the different sequences of length, will lack the sequence polishing with methods such as interpolation.

Images