JP2010218301A - Failure diagnosis apparatus, failure diagnosis method, and failure diagnosis program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a failure diagnosis apparatus that estimates the state of a chemical plant through regression analysis, and automatically identifies the cause of failure when determining an abnormal state. <P>SOLUTION: The failure diagnosis apparatus performs regression analysis on the basis of a measurement value obtained by measuring a target object in a plant in order to diagnose an abnormal state in the plant, determines an estimated value of the object, and predicts the occurrence of failure by comparing the estimated value with a predetermined threshold. The failure diagnosis apparatus includes a measurement value specifying means that specifies a measurement value greatly contributing to the estimated value of the object. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an abnormality diagnosis device, an abnormality diagnosis method, and an abnormality diagnosis program that diagnose whether or not an abnormality has occurred in a plant and identify the cause of the abnormality that has occurred.

  In recent years, while chemical plants that employ process control have become more complex and specialized, operating sites have been spurred by fewer people. It is very difficult to monitor a complex and non-generalized plant with a small number of people, and there are many cases in which a chemical plant anomaly appears only when a plant stop alarm is issued. Also, abnormal operation of chemical plants can be accompanied by significant economic losses. Therefore, a system that detects an abnormality early and identifies the cause is very useful.

  In the prior art, for a chemical plant, an abnormality prediction method using principal component analysis (PCA; principal component analysis) (see, for example, Patent Document 1), or regression analysis (PLS; Partial Least Squares) There is an abnormality prediction method used (for example, see Patent Document 2). PCA and PLS are both types of multivariate analysis, and the relationship between input variables and output variables can be modeled, and output variables can be estimated using new input variables. As for PLS, it is used as a state estimation (soft sensor) of a chemical plant by utilizing a characteristic that is not affected by multicollinearity (see, for example, Patent Document 2).

  The methods described in Patent Documents 1 and 2 are methods for detecting an abnormality by estimating some measured value in a chemical plant and comparing it with an actual measured value. The method described in Patent Document 3 is a method for estimating a physical quantity that cannot be measured in real time in a chemical plant and using it for plant operation. These are equivalent techniques in that some state is estimated using multivariate analysis, and in recent years, they are used for abnormality detection and state estimation in chemical plants.

Japanese translation of PCT publication No. 2008-512792 JP 2006-350698 A JP 2007-004769 A

  By the way, in the chemical plant, in addition to detecting the abnormality early, it is important to estimate the cause of the abnormality detected early. However, the conventional technique only detects an abnormality, and there is a problem that a cause causing an abnormal state cannot be specified. Even if an abnormality can be detected in advance, it is impossible to avoid a situation such as the operation stop of the chemical plant or damage to the apparatus unless the cause is known. Many chemical plants have complicated processes overlapping, and it is difficult to identify the cause of the abnormality. In recent years, there are chemical plants that lack experienced operators, and there is a demand for a method for automatically detecting the cause of an abnormality.

  The present invention has been made in view of such circumstances, and estimates the state of a chemical plant by regression analysis such as PLS (Partial Least Squares) and automatically determines that an abnormal state has occurred. It is an object of the present invention to provide an abnormality diagnosis device, an abnormality diagnosis method, and an abnormality diagnosis program that can identify the cause of abnormality.

  According to the present invention, in order to diagnose a plant abnormality, a regression analysis is performed based on a measurement value obtained by measuring a measurement object in the plant, thereby obtaining an estimated value of the abnormality detection target, and the estimated value can be determined in advance. An abnormality diagnosis apparatus that predicts the occurrence of an abnormality by comparing the measured value with a measured threshold value, comprising: a measurement value specifying unit that specifies the measurement value that greatly contributes to the estimated value of the abnormality detection target. And

  According to the present invention, in order to diagnose a plant abnormality, a regression analysis is performed based on a measurement value obtained by measuring a measurement object in the plant, thereby obtaining an estimated value of the abnormality detection target, and the estimated value can be determined in advance. An abnormality diagnosis method in an abnormality diagnosis apparatus that predicts occurrence of an abnormality by comparing with a threshold value, and includes a measurement value identification step that identifies the measurement value that greatly contributes to the estimated value of the abnormality detection target It is characterized by that.

  According to the present invention, in order to diagnose a plant abnormality, a regression analysis is performed based on a measurement value obtained by measuring a measurement object in the plant, thereby obtaining an estimated value of the abnormality detection target, and the estimated value can be determined in advance. An abnormality diagnosis program that operates on an abnormality diagnosis device that predicts the occurrence of an abnormality by comparing with a measured threshold value, and specifies the measurement value that greatly contributes to the estimated value of the abnormality detection target The step is performed by a computer.

  According to the present invention, the regression analysis is performed based on the measurement value obtained by measuring the measurement target in the plant, thereby obtaining the estimated value of the abnormality detection target and comparing the estimated value with a predetermined threshold value. In the abnormality diagnosis device that predicts the occurrence of abnormalities, the measurement value that greatly contributes to the estimated value of the abnormality detection target is specified and output, so the cause of abnormality when an abnormality occurs in the plant must be specified Can do. Thereby, when an abnormality occurs in the plant, it is possible to obtain an effect that an abnormality avoidance measure can be taken quickly.

It is a block diagram which shows the structure of one Embodiment of this invention. It is a flowchart which shows the processing operation of the abnormality diagnosis apparatus 2 shown in FIG. It is explanatory drawing which shows an example of the plant data memorize | stored in the database 22 shown in FIG. It is a systematic diagram which shows an example of the plant 1 shown in FIG. It is explanatory drawing which contrasted and showed the predicted value of the purity of product gas, and the measured value. It is explanatory drawing which shows the intensity | strength of each measured value which contributed to the estimation result.

  Hereinafter, an abnormality diagnosis apparatus according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the embodiment. In this figure, reference numeral 1 denotes a chemical plant (hereinafter referred to as a plant) that operates by process control, and is a plant that is a target of abnormality diagnosis performed by the abnormality diagnosis apparatus according to the present invention. Reference numeral 2 denotes an abnormality diagnosis device that performs abnormality diagnosis of the plant 1 and is configured by a computer device. Reference numeral 21 denotes an actual plant data input unit for inputting measurement data (referred to as actual plant data) indicating the state of the plant output from a sensor (not shown) provided in the plant 1 during plant operation. It is. Reference numeral 22 denotes a database that stores actual plant data input by the actual plant data input unit 21 and is configured by a storage device. Reference numeral 23 denotes a PLS model determination unit that determines a PLS model from data stored in the database 22. Reference numeral 24 denotes a PLS model storage unit that stores the PLS model data determined by the PLS model determination unit 23. Reference numeral 25 denotes a PLS estimation processing unit that uses the PLS model data stored in the PLS model storage unit 24 to obtain and output an estimated value of the explained variable.

  Reference numeral 26 denotes a explained variable estimated value storage unit that stores the explained variable estimated value output from the PLS estimation processing unit 25. Reference numeral 27 denotes an abnormality determination processing unit that performs an abnormality determination by comparing actual plant data (explained variable) held in the actual plant data input unit 21 with an explanatory variable estimated value. Reference numeral 28 denotes a display unit that displays abnormality occurrence alarm information that is output when the abnormality determination processing unit 27 determines that an abnormality has occurred, and is configured by a display device. The display unit 28 may be notified of the occurrence of an alarm by voice. Reference numeral 29 denotes an abnormality cause analysis unit that analyzes and identifies the cause of an abnormality that has occurred when the abnormality determination processing unit 27 determines that an abnormality has occurred. Reference numeral 30 denotes a display unit that displays the explanatory variables specified by the abnormality cause analysis unit 29. The display unit 30 may also be used as the display unit 28.

  Next, the processing operation of the abnormality diagnosis device 2 shown in FIG. 1 will be described with reference to FIG. First, the actual plant data input unit 21 inputs actual plant data of the plant 1 (step S1) and stores it in the database 22. The actual plant data here is measurement data output from a sensor provided in the plant 1, and is, for example, a measurement value of a pressure sensor that measures a pressure in a pipe. Here, the table structure of the database 22 shown in FIG. 1 will be described with reference to FIG. As shown in FIG. 3, the database 22 is configured and stored from N pieces of time series data at times a1, a2,..., AN for m measurement points (sensor mounting positions). In FIG. 3, x is a measured value of each sensor.

  Next, the PLS model determination unit 23 determines the PLS model by dividing the plurality of actual plant data stored in the database 22 into explanatory variables used for estimation and target explained variables to be estimated, The determined PLS model data is stored in the PLS model storage unit 24 (step S2). As the explained variable, it is desirable to select a variable related to the quality and quantity of the product of the plant 1. Here, the PLS model determination process of the PLS model determination unit 23 will be described.

なお、被説明変数Ｙと説明変数Ｘは標準化処理をされているものとする。 Now, N is the number of samples (the number of time-series data), m and n are the number of explanatory variables and the number of explained variables (= variable to be estimated), respectively, and X is a data matrix of explanatory variables (N × m; (1 ) Equation), Y is the data matrix of the explained variable (N × n; Equation (2)).

It is assumed that the explained variable Y and the explanatory variable X have been standardized.

（４）式のｔはＸＸ^Ｔの特異点に対応した固有ベクトル（Ｎ次）、ｐはＸ^ＴＸの特異点に対応した固有ベクトル（ｍ次）である。 Here, it is assumed that the explained variable Y is represented by the equation (3) using the explanatory variable X and the multiple regression parameter α (m × n matrix).
Y = X · α (3)
Here, it is assumed that there is collinearity among certain variables among the m explanatory variables in X. At this time, α cannot be estimated by the method of least squares. Therefore, suppose that the hierarchy of X is r (rankX ≦ m), and the singular value decomposition can be performed as in equation (4).

(4) eigenvector t is corresponding to singularity XX ^T of formula (N order), p is the eigenvector corresponding to the singular points of ^{X T} X (m order).

ここまでの展開では、ｔはＹから独立して決められるように見えるが、ＹとＸの関係を考慮して決められる。 Substituting equation (4) into equation (3) yields equation (5).

In the development so far, t seems to be determined independently of Y, but is determined in consideration of the relationship between Y and X.

Each matrix is summarized as follows.
T = (t ₁ t ₂ ... _Tr ); N × r degree score matrix P = (p ₁ p ₂ ... P _r ); m × r order load matrix W = (W ₁ W ₂ ... W _r ); m × r degree coefficient matrix Q = (q ₁ q ₂ ... Q _r ); n × r order load matrix

The singular value decomposition of X by T and the score matrix T can be expressed by the following equations (6) and (7).
X = TP ^T (6)
T = XW ^T (7)

Therefore, the equations (6) and (7) to (8) are obtained.
^{T (P T W) = XW} ··· (8)

The matrix P ^T W is an r × r-order triangular matrix whose diagonal elements are all 1 and is regular, and therefore has an inverse matrix. Therefore, the score matrix T can be obtained as shown in Equation (9).
T = XW (P ^T WT) ⁻¹ (9)

On the other hand, since Y = TQ ^T + E, substituting equation (9) yields equation (10).
^{Y = XW (P T W)} -1 Q T + E ··· (10)

Therefore, the multiple regression parameter α can be expressed as in equation (11).
α = W (P ^T W) ⁻¹ Q ^T

  Determining the load matrixes P and Q and the coefficient matrix W satisfying the relationship of the expression (11) is a PLS model determination process.

  In the case of a process in which a future explained variable can be estimated from the past explanatory variable from the past (for example, a future explained variable can be estimated from data two hours before the present), dynamic PLS is employed. Since it is possible to estimate future explained variables, it is more preferable in terms of detecting an abnormality.

  Next, the PLS estimation processing unit 25 uses the explanatory variables of the PLS model data stored in the PLS model storage unit 24 and the actual plant data of the plant 1 held in the actual plant data input unit 21, An explanatory variable estimation process is performed (step S3). This processing can be performed by using equations (3) and (11). The PLS estimation processing unit 25 stores the explained variable estimated value in the explained variable estimated value storage unit 26.

  Next, the abnormality determination processing unit 27 includes the estimated explained variable stored in the explained variable estimated value storage unit 26 and the measured explained variable (actual plant data) held in the actual plant data input unit 21. A process of detecting an abnormality by comparison is performed (step S4). In this case, when the absolute value of the difference between the estimated explained variable and the measured explained variable exceeds a certain value, it is regarded as abnormal. The threshold at this time is preferably about several times the standard deviation of the explained variable. Further, when dynamic PLS is employed, it is possible to estimate future explained variables, so a method of comparing the estimated future explained variables with a plant alarm threshold value may be used. The abnormality determination processing unit 27 determines whether or not an abnormality is detected (step S5). If no abnormality is detected, the process returns to step S1 and repeats the process.

  On the other hand, when an abnormality is detected, the abnormality determination processing unit 27 displays alarm information indicating that an abnormality has occurred on the display unit 28 (step S6), thereby notifying the operator of the occurrence of the abnormality and analyzing the cause of the abnormality. The unit 29 is notified that an abnormality has been detected.

Upon receiving the notification of abnormality detection, the abnormality cause analysis unit 29 analyzes the cause of the abnormality (step S7). Since the prediction of the explained variable is calculated as Y = TQ ^T , if the explanatory variable that greatly contributes to the prediction result is obtained by analyzing the product of the score matrix T and the load matrix Q ^T , the cause of the abnormality Certain explanatory variables can be identified.

  Here, the processing operation in which the abnormality cause analysis unit 29 finds the explanatory variable that greatly contributes to the prediction result by performing the decomposition processing of the product of the score matrix T and the load amount matrix QT will be described.

When TQ ^T is expressed in detail, where N is the number of samples, r is the hierarchy of X, and n is the number of explained variables, equation (12) is obtained.

Therefore, the ij component of the N × n matrix is as shown in equation (13).
(T · Q ^T ) _ij = t _i1 q _j1 + t _i2 q _j2 ... + T _ir q _jr (13)

On the other hand, when the m × r matrix of W · (P ^T T) ⁻¹ is collectively set to Z and equation (9) is replaced, it can be expanded as shown in equation (14).

Now, when analyzing the cause of the abnormality of the explained variable in the k-th column in the estimated value at time h in equation (13), among the elements in equation (15), t _hr q _kr having the largest absolute value is a latent variable. Let component r be f.
Y _hk = (TQ ^T ) _hk = t _h1 q _k1 + t _h2 q _k2 ... + T _hr q _kr (15)

From equation (14) and equation (15), _htf can be expressed as equation (16) with explanatory variable X and Z component.
t _hf = x _h1 Z _1f + x _h2 Z _2f ... + x _hm Z _mf (16)

(16) In the formula, _{x hm} in a large _x hm · _{z mf} component becomes larger contribution to that described variable to estimate. When the explanatory variable that greatly contributes to the estimated value is specified, the abnormality cause analysis unit 29 displays the specified explanatory variable on the display unit 30, thereby notifying the operator of the abnormality cause candidate (step S8).

  Next, an example will be described in which the abnormality diagnosis apparatus shown in FIG. 1 performs an operation for predicting the occurrence of an abnormality and identifying the cause of the abnormality. FIG. 4 is a system diagram of a target plant that performs abnormality diagnosis using the abnormality diagnosis apparatus according to the present invention. In FIG. 4, the explained variable is a value of a purity sensor that measures the purity of the product gas. The explanatory variables are measured values of sensors provided in the system. Temperatures 1 to 3 are output from the temperature sensor shown in FIG. 4, flow rates 1 to 4 are output from the flow sensor, and pressure is output from the pressure sensor. 1 to 5 and the liquid levels 1 and 2 which are the outputs of the liquid level sensor correspond. FIG. 5 shows the result of determining the PLS model and predicting the explained variable (product gas purity) by the abnormality diagnosis apparatus shown in FIG. As shown in FIG. 5, the predicted value and the actually measured value are in good agreement until immediately before 12:00. However, at 12 o'clock, the predicted value shows an abnormal value and an abnormality is detected. Since the predicted value predicts a value after 2 hours, the predicted value at which an abnormality is detected at 12:00 is calculated at the time of 10:00. Since the measured value shows an abnormal value from 15:00, it can be seen that the abnormality can be detected five hours ago.

  FIG. 6 is a diagram illustrating a result obtained by the abnormality cause analysis unit 29 by analyzing where the predicted value abnormality at 12:00 is caused by the explanatory variable. As shown in FIG. 6, it can be seen that the strengths of pressure 4, pressure 5, liquid level 2 and pressure 3 are high. Referring to FIG. 4, it can be seen that all the measured values contributing to this result have high intensities in the vicinity of the exhaust gas cylinder. Values of pressure 4, pressure 5, liquid level 2 and pressure 3, which are explanatory variables, and sensor mounting position information are displayed on the display unit 30. The operator can know that an abnormality has occurred around the exhaust gas cylinder by viewing this display.

  As described above, the regression analysis is performed based on the measurement value obtained by measuring the measurement target in the plant, thereby obtaining the estimated value of the abnormality detection target and comparing the estimated value with a predetermined threshold value. In the abnormality diagnosis device that predicts the occurrence of an abnormality with the above, the measurement value that greatly contributes to the estimated value of the abnormality detection target is specified, and the measurement target of the specified measurement value is specified and output. It is possible to identify the cause of the abnormality when the error occurs. As a result, when an abnormality occurs in the plant, it is possible to quickly take an abnormality avoidance measure.

  A computer-readable recording medium for realizing the functions of the actual plant data input unit 21, the PLS model determination unit 23, the PLS estimation processing unit 25, the abnormality determination processing unit 27, and the abnormality cause analysis unit 29 in FIG. The abnormality diagnosis process and the abnormality cause analysis process may be performed by causing the computer system to read the program recorded in the recording medium and executing the program. The “computer system” here includes an OS and hardware such as peripheral devices. The “computer system” includes a WWW system having a homepage providing environment (or display environment). The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Further, the “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system that becomes a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, those holding programs for a certain period of time are also included.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, what is called a difference file (difference program) may be sufficient.

  DESCRIPTION OF SYMBOLS 1 ... Chemical plant, 2 ... Abnormality diagnosis apparatus, 21 ... Actual plant data input part, 22 ... Database, 23 ... PLS model determination part, 24 ... PLS model memory | storage part, 25 ... PLS estimation processing unit, 26 ... explained variable estimated value storage unit, 27 ... abnormality determination unit, 28 ... display unit, 29 ... abnormality cause analysis unit, 30 ... display unit

Claims (3)

In order to diagnose a plant abnormality, a regression analysis is performed based on a measurement value obtained by measuring the measurement target in the plant, thereby obtaining an estimated value of the abnormality detection target, and the estimated value is determined by a predetermined threshold value. An abnormality diagnosis device that predicts the occurrence of an abnormality by comparing with
An abnormality diagnosing device comprising a measured value specifying means for specifying the measured value that greatly contributes to the estimated value of the abnormality detection target. In order to diagnose a plant abnormality, a regression analysis is performed based on a measurement value obtained by measuring the measurement target in the plant, thereby obtaining an estimated value of the abnormality detection target, and the estimated value is determined by a predetermined threshold value. An abnormality diagnosis method in an abnormality diagnosis apparatus that predicts the occurrence of an abnormality by comparing with,
An abnormality diagnosis method comprising a measurement value specifying step of specifying the measurement value that greatly contributes to the estimated value of the abnormality detection target. In order to diagnose a plant abnormality, a regression analysis is performed based on a measurement value obtained by measuring the measurement target in the plant, thereby obtaining an estimated value of the abnormality detection target, and the estimated value is determined by a predetermined threshold value. An abnormality diagnosis program that operates on an abnormality diagnosis device that predicts the occurrence of an abnormality by comparing with
An abnormality diagnosis program for causing a computer to perform a measurement value specifying step for specifying the measurement value that greatly contributes to the estimated value of the abnormality detection target.

Images