JPH09282022A - Plant operation abnormality prevention system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent abnormality generation by predicting abnormality by finding a total evaluated value based on fuzzy integration while using a fuzzy measure identified by fuzzy proposition suitability and sequential identification method, and regulating the state quantity of a plant according to the predicted result. SOLUTION: A fuzzy proposition storage part 3 stores plural fuzzy propositions for which the conditions made into a language at the time of abnormality generation are expressed by membership functions defining detected values at the plural sections of the plant as evaluation items. A fuzzy proposition suitability calculation part 2 calculates the suitability of respective fuzzy propositions by applying the detected values to these fuzzy propositions. A fuzzy measure identification part 6 identifies the fuzzy measure by using the sequential identification method. An abnormality predictive part 4 performs the abnormality prediction of plant operations corresponding to this suitability and the fuzzy integration using the fuzzy measure. Then, a state quantity regulation part 7 regulates the quantity of states more than one of the plant according to the provided predicted result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system for preventing abnormal operation of a plant, and more specifically, applying a fuzzy measure / integral to detected values at a plurality of parts in a plant to predict the abnormal condition and automatically cause the abnormal condition. The present invention relates to a plant operation abnormality prevention system that performs an avoidance operation.

[0002]

2. Description of the Related Art A conventional method for coping with an abnormality occurring in plant operation has been that an operator who hears an alarm notifying the occurrence of an abnormality confirms the actual state of the abnormality and takes countermeasures against it.

[0003]

However, in the conventional methods as described above, there is a difference in the recognition of the abnormal situation due to the difference in the operator's skill level, which causes variations in the time and amount for the countermeasure operation. There is a risk that the loss may increase due to a mistaken operation. Further, as described above, the countermeasures are mainly ex-post measures, and no measures have been taken to prevent the occurrence of abnormalities.
Therefore, the object of the present invention is to predict the occurrence of an abnormality and prevent the abnormality by automatically performing an abnormality avoidance operation, thereby stabilizing the process / quality, improving productivity, and reducing the number of people. The present invention is to provide a plant operation abnormality prevention system.

[0004]

In order to achieve the above object, the first invention is verbalized in a plant operation abnormality prevention system for predicting an abnormality in plant operation from detected values in a plurality of parts constituting a plant. By applying the detected value to the fuzzy proposition, a fuzzy proposition storage means for storing a plurality of fuzzy propositions in which a situation at the time of occurrence of an abnormality is expressed by a membership function having the detected value as an evaluation item is stored. A fuzzy proposition goodness of fit calculating means for calculating the goodness of fit of each fuzzy proposition, and a fuzzy measure using a sequential identification method that minimizes the deviation between the target value of the fuzzy integration result and the total judgment value by the actual fuzzy integration model. Identifying means, fuzzy measure storing means for storing the identified fuzzy measure, the fuzzy proposition conformity, and the fuzzy measure. The fuzzy integral with Jie measure, a plant operation abnormality prevention system characterized by comprising comprises a prediction means for performing an abnormality prediction of plant operation. Further, the fuzzy integral is a plant operation abnormality prevention system including a procedure of rearranging the goodness of fit of each of the fuzzy propositions in ascending order, calculating a difference between them, and obtaining a sum of products obtained by multiplying them by a fuzzy measure corresponding to them. .

Further, in the plant operation abnormality prevention system, the detected value used as the evaluation item of the membership function includes the current state quantity of the detected value which changes with time. Further, in the plant operation abnormality prevention system, the detected value used as the evaluation item of the membership function includes the amount of change in the detected value that changes over time within the past predetermined time. Further, the plant operation abnormality prevention system according to claim 4, wherein the variation Dev (t) is calculated by Dev (t) = dat (t) -av (t-1), where t is the current time. , Dat (t) is the state quantity at the current time, and av (t-1) is the average value of the state quantity in the past predetermined time at (t-1) time. Furthermore, the above av
The state quantity mdat for obtaining (t-1) is The plant operation abnormality prevention system according to claim 5, wherein sdev (t-2) is a standard deviation of the state quantity within the past predetermined time in (t-2) time, and coef is a correction coefficient. is there.

Further, in identifying the fuzzy measure,
This is a plant operation abnormality prevention system that uses data that indicates the possibility of an abnormality occurring before the time when the abnormality actually occurred for prediction. Furthermore, the above-mentioned abnormality prediction state,
Alternatively, the plant operation abnormality prevention system adjusts the state quantity by determining that the abnormality prediction or the abnormality prediction has ended when the termination state of the abnormality prediction continues for a certain time or longer.
Further, the plant operation is a drying and classifying process of the powder and granules, and the abnormal plant operation is caused by electrostatic adhesion due to overdrying of the powder and granules or agglomeration due to agglomeration due to non-drying. This is a classification abnormality due to screen mesh clogging in the equipment, and the state quantity adjusting means is a plant operation abnormality prevention system which is a humidity control device immediately before the vibration sieving device.

[0007]

In the present invention, the detected values at a plurality of parts constituting the plant are stored in the fuzzy proposition storage unit,
By applying the situation at the time of anomaly to a plurality of fuzzy propositions expressed by membership functions, the goodness of fit of each fuzzy proposition is calculated (fuzzy proposition goodness of fit calculation means), and the goodness of fit and the sequential identification method Is detected (fuzzy measure identifying means), and the fuzzy measure stored in the fuzzy measure storage unit is used to predict an abnormality by obtaining a comprehensive evaluation value by fuzzy integration (prediction means). Further, in identifying the fuzzy measure, data indicating the possibility of occurrence of abnormality before the time when the abnormality actually occurs is used. This makes it possible to appropriately express an ambiguous evaluation structure, and to predict anomaly occurrence with a more general evaluation value that takes into consideration the mutual effects between items. Further, by adjusting one or more state quantities of the plant according to the above-mentioned prediction result, an abnormality avoiding operation can be performed (state quantity adjusting means), thereby preventing the occurrence of abnormality.

[0008]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments embodying the present invention will be described below with reference to the accompanying drawings to provide an understanding of the present invention. Note that this embodiment and the examples that follow are specific examples of the embodiment of the present invention, and do not limit the technical scope of the present invention.

Here, FIG. 1 is a diagram showing a schematic configuration of a plant operation abnormality prevention system according to the present invention, FIG. 2 is a diagram showing a schematic processing flow of the plant operation abnormality prevention system according to the present invention, and FIG. 3 is a membership. Explanatory diagram of functions, Fig. 4 is a schematic diagram showing the relationship between a fuzzy measure and a normal measure, Fig. 5 is a schematic diagram showing a fuzzy integral model, Fig. 6 is a diagram showing a flow of identification processing of a fuzzy measure, and Fig. 7 is The figure which shows the schematic structure of the drying / classifying process of the granular material which concerns on one Example of this invention,
FIG. 8 is a diagram showing items of detected values according to an embodiment of the present invention,
FIG. 9 is a diagram showing a fuzzy proposition representing a case of “overdrying side screen mesh in a dryer continuous operation mode” according to one embodiment of the present invention, and FIG. 10 is a diagram showing a “dryer circulation according to one embodiment of the present invention. The figure which shows the fuzzy proposition which represents the case of "overdrying side screen mesh clogging in operation mode", Fig. 1
1 is a diagram showing a fuzzy proposition representing a case of "dry side screen mesh clogging in a dryer continuous operation mode" according to one embodiment of the present invention, and FIG. 12 is a "dryer circulation operation according to one embodiment of the present invention. 13 is a diagram showing a fuzzy proposition representing the case of "dry side screen mesh clogging in mode".
FIG. 14 shows a membership function parameter determination method according to an embodiment of the present invention, FIG. 14 shows a fuzzy measure identification result according to an embodiment of the present invention, and FIG. 15 shows an embodiment of the present invention. FIG. 16 is a diagram showing a specific calculation example of a state total judgment value by the fuzzy integral model, FIG. 16 is a diagram showing an example of a classification abnormality prediction result according to an embodiment of the present invention, and FIG. 17 is a diagram showing an embodiment of the present invention. FIG. 18 is a diagram showing a classification abnormality prediction situation, FIG. 18 is a diagram showing a variation amount to be obtained, FIG. 19 is a diagram showing a variation amount obtained by a normal simple average, and FIG. 20 is a variation amount in the past predetermined time according to the present invention. FIG. 21 is a diagram showing a processing flow of the obtaining method, FIG. 21 is a diagram showing a difference in the calculation result when coef is changed in the obtaining method of the change amount in the past predetermined time according to the present invention, and FIG. Overview of classification abnormality prevention unit related to example FIG. 23 is a diagram showing a configuration, FIG. 23 is a diagram showing a processing flow of a humidity control vapor pressure set value calculation according to an embodiment of the present invention, and FIG. 24 is a classification abnormality during operation of a classification abnormality prevention unit according to one embodiment of the present invention. It is a figure which shows the fluctuation | variation state of a predicted value and humidity control vapor pressure.

The plant operation abnormality prevention system according to the present invention is constructed as shown in FIG. First, the online processing unit A1 includes a signal processing unit 1, a fuzzy proposition fitness calculation unit 2, a fuzzy proposition storage unit 3 connected to it, an anomaly prediction unit 4 and a fuzzy measure storage unit 5 connected to it.
It is configured by the state quantity adjusting unit 7. The off-line processing unit A2 includes a fuzzy measure identifying unit 6 and the fuzzy measure storing unit 5 connected thereto. The offline processing unit A2 operates only when the system is started up, and only the online processing unit A1 operates after the application to actual plant operation. The fuzzy proposition storage unit 3 serves as the fuzzy proposition storage unit of the present invention, the fuzzy proposition fitness calculation unit 2 serves as the fuzzy proposition fitness calculation unit, the fuzzy measure storage unit 5 serves as the fuzzy measure storage unit, and the fuzzy measure identification unit 6 serves as the fuzzy measure identification unit 6. The state quantity adjusting unit 4 corresponds to the fuzzy measure identifying means, and corresponds to the state quantity adjusting means.

Here, some ideas used for realizing this system will be described. First, in order to predict an abnormality in plant operation, a value that serves as an index for determining the abnormality is necessary. However, various factors are usually involved in abnormal plant operation in a complicated manner.
In most cases, the state quantity expressing the abnormality itself is unknown. Therefore, it is impossible to predict anomalies based on the behavior of a single state quantity. Therefore, it is necessary to make a comprehensive judgment by using the behavior of multiple state quantities involved in the occurrence of anomaly as an evaluation structure. Also, in this evaluation structure, it is necessary to quantify the state quantity that is each evaluation item, but the way of capturing each numerical value includes ambiguity in the judgment criteria for judging an abnormality, and Since the comprehensive evaluation also contains ambiguity, it is not possible to make accurate predictions with an evaluation method that simply quantifies the state quantity. On the other hand, in such a ambiguity structure, it is necessary to establish a method for evaluation based on human subjectivity.
Fuzzy measure / integral has been proposed as a method for modeling such a subjective evaluation structure of a person with high expressiveness. There is an evaluation device disclosed in Japanese Patent Application Laid-Open No. 7-244525 as an apparatus for evaluating the state of a plant using such a subjective evaluation structure of a person.

However, in the evaluation device disclosed in Japanese Patent Laid-Open No. 7-244525, the present operating condition of the plant is evaluated using the fuzzy measure / integral. The situation of occurrence is expressed by multiple fuzzy propositions (state interpretation by fuzzy propositions), and the subsequent occurrence of abnormalities is predicted by the method based on the above fuzzy measure / integral from the goodness of fit of the detected values that change over time to the fuzzy proposition. This method is different from the above-described evaluation apparatus in that an abnormality in plant operation is predicted in advance by obtaining one of the comprehensive evaluation values shown, and the abnormality avoidance operation is automatically performed according to the prediction result.

First, the state interpretation by the fuzzy proposition will be described. As mentioned above, it is difficult to directly judge the situation of abnormality occurrence directly from numerical information such as plant state quantity. Therefore, the abnormality occurrence situation is expressed linguistically from the knowledge obtained from past data and experience. Then, the ambiguity of the language is digitized with multiple fuzzy propositions to obtain the individual fitness (partial evaluation), and the state interpretation is performed by reducing redundant information that does not match the above findings. This is the state interpretation based on the above fuzzy proposition. Specifically, first, one situation in which an abnormality occurs is such that "the state amount a increases due to, the change amount b decreases accordingly, and ... Then, an abnormality occurs." Then, this is expressed by using the membership function of the fuzzy set (S1 in FIG. 2). Here, the labels LA, SA, MM represent the membership functions shown in FIG.
Numerical values in [] are parameter values ([P1, P2],
[P1, P2, P3]) is shown. In addition, this parameter value is determined using past data. Each of the fuzzy propositions defined in this way is stored in advance in the fuzzy proposition storage unit 3, and the corresponding state amount or change amount (how to obtain it will be described later) is input to this fuzzy proposition. Then, the goodness of fit (represented by a numerical value between 0 and 1) for each fuzzy proposition is obtained by the fuzzy proposition goodness-of-fit calculation unit 2 (partial evaluation value).

Next, the abnormality predicting section 4 obtains a comprehensive evaluation value by using this suitability (partial evaluation value). The fuzzy measure / integral for obtaining the comprehensive evaluation value will be described below. First of all, the fuzzy measure has been developed in the fuzzy theory as a method of handling subjective evaluation of human subjects in judgment of things, and as shown in Fig. 4, it is a measure (probability measure) that is established by ordinary additivity. Unlike the measure that does not necessarily have additivity (non-additivity measure).
That is, when A∩B = φ, μ (A∪B) ≧ μ (A) + μ (B)
Alternatively, it is a scale (function) μ that has a property such as μ (A∪B) ≦ μ (A) + μ (B) and that indicates that a certain evaluation item is considered important or that it may occur. . Next, regarding the above fuzzy measure identification method, in the present invention, a sequential identification method is used that minimizes the deviation between the target value of the fuzzy integration result and the total judgment value by the actual fuzzy integration model. This identification method is described in the above-mentioned Japanese Patent Laid-Open No.
It has already been presented in Japanese Patent No. 244525. The processing flow of this identification method is shown in FIG. In the present invention, since the abnormality of the plant operation is predicted, the past empirical data used for identifying the fuzzy measure is the data that is a predetermined time before the time when the abnormality actually occurs. The target value of the fuzzy integration result at that time was set to 0.5. That is, when the comprehensive judgment value exceeds 0.5, it is considered that an abnormality has occurred.

Assuming that the number of fuzzy propositions is four, the number of fuzzy measures used for fuzzy integration is four. However, the four fuzzy measures described above are given by the following 15 depending on the degree of goodness of fit of each proposition. Selected from the street. In other words, it is necessary to identify the following 15 fuzzy measures. μ ({1}) μ ({2}) μ ({12}) μ ({3}) μ ({13}) μ ({2 3}) μ ({1 2 3}) μ ({ 4}) μ ({1 4}) μ ({2 4}) μ ({1 2 4}) μ ({3 4}) μ ({1 3 4}) μ ({2 3 4}) μ ( {1 2 3 4})

First, a plurality of data showing the suitability of each proposition based on the past experience data as described above are prepared. FIG.
First, the fitness data of 0.27, 0.48, 0.84, and 0.36 are used as an example. First, S14-S
In step 18, the above-mentioned fuzzy integration is calculated using the fitness data. In S16, the corresponding four fuzzy measures are selected and initialized, but a value obtained by dividing 1.0 by the number of propositions is used as the initial value. Then, in S19, the target value Wj of the fuzzy integration result is obtained.
A deviation ej between (= 0.5) and the total judgment value Zj obtained in S18 is obtained. If ej is less than the convergence judgment value or the number of repetitions of this data is equal to or more than the upper limit of the number of repetitions of each data, a new calculation using another fitness data is performed until all the data are finished (S20, S23,
S24, S13 ...), and in other cases, the fuzzy measure is corrected and the fuzzy integral is recalculated (S20, S2).
1, S22, S18 ...). When all the data are repeated by repeating these steps, the total deviation te is calculated, and if it is below the convergence judgment value or if the number of repetitions of all the data is equal to or more than the upper limit value, each fuzzy measure is fixed there (S23, S25). , S26, S28), otherwise, the same calculation is repeated for all data (S23, S28).
25, S26, S27, S13, ...). As described above, the fuzzy measure used for the fuzzy integration is identified.

As described above, the fuzzy measure is identified by the fuzzy measure identifying section by the off-line processing only when the system is started up, and is stored in the fuzzy measure storing section 5. Using the fuzzy measures stored in advance as described above, the abnormality prediction unit 4 predicts the occurrence of abnormality by performing a sensitive comprehensive evaluation using all fuzzy propositions.
Here, fuzzy integration, which will be described later, is used as the method. The fuzzy integral is used for the comprehensive evaluation of the object, and unlike the conventional linear evaluation model that assumes linearity in weights, the fuzzy measure is used to express the mutual effect between items and obtain the nonlinear comprehensive evaluation. This is a sensitive calculation method. Therefore, it is possible to model a subjective evaluation structure close to that of a person. FIG. 5 shows a conceptual diagram of how to obtain an evaluation value using fuzzy integration. First, the goodness of fit of the fuzzy proposition, that is, the partial evaluation values h (xi) are rearranged in ascending order.
Next, the difference (h (xi) -h (x
i-1): However, h (x0) = 0 is calculated, and fuzzy measures (μ ({x1, x2, ... x
i}), ..., μ ({xi})) are multiplied to obtain the total sum (integration). This is the comprehensive evaluation value. At this time,
As shown in FIG. 7, the one with the larger partial evaluation value is added to the total evaluation value so as to include the possibility of the one with the smaller partial evaluation value. In this way, the mutual effects among the above items are expressed.

Next, the amount of state and the amount of change input to the fuzzy proposition will be described. When a plurality of fuzzy propositions are created as described above, as the values used to express the fuzzy propositions, the state quantity at the present time of the detected values that change with time, that is, the absolute quantity is appropriate, and It may be considered that the amount of change in the value within the past predetermined time is more appropriate. If the state quantity is appropriate, there is no problem, but if the change quantity is more appropriate, the prediction accuracy may be affected by the method of obtaining the change quantity.
Therefore, in the present invention, the amount of change is determined by the following method. The case where the amount of change is appropriate is a case where the state of change (absolute amount) at that time is the same, but the change state on the upstream side becomes a problem. That is, as shown in FIG. 18, the amount of change from the past stable state becomes a problem. In this case, a desired change amount cannot be expected as shown in FIG. 19 by a method of simply obtaining the difference from the average value for a predetermined time in the past. This is because the above average value does not show the value in the past stable state. Therefore, in the present invention, processing is performed so that the average value is as close to the past stable state as possible. That is, a data array for a predetermined time in the past for obtaining an average value was created, and the following operation was performed on the values to be put in the data array.

FIG. 20 shows an outline of the change amount calculation method. First, an array avdat [T] for the past time T samples is prepared. S41 to S43 are calculation steps of the input data mdat to avdat [T]. In S42, when the actual data (state amount) dat (t-1) is a value far from the data behavior of the past T samples, the value input to the mdat is av ± sdev / coef where coef is a correction coefficient. are doing. Array avda created in this way
The average value av is obtained from t [T] (S44), and the variation amount Dev (t) is obtained by obtaining the difference between the current state amount dat (t) and the above av. By repeating the above-described processing, the average value av always shows a value close to the past stable state, and an appropriate amount of change as shown in FIG. 18 can be obtained. Here, with respect to the correction coefficient coef and the time width T, appropriate values are used depending on the characteristics of the data to be handled. FIG. 21 shows an example showing the difference in the change amount calculation result due to the difference in the coef value. In this example, compared with A and B, when coef is small, the followability to the source waveform is improved, and the peak of the changed portion is small. Also, comparing C and D, if coef is smaller,
The amount of change at the time of recovery after the change shows a change in polarity opposite to that of the previous change. In both cases, the value of mdat input to avdat [T] is close to the peak value due to the small coef. Therefore, FIG.
For source waveforms such as, it is better to have a large coef. Further, regarding the time width T, the behavior of the state quantity is analyzed, and in the case of a behavioral change that is drastic, the time T is set to be large in order to increase the influence of the past stable period.

The comprehensive judgment value for abnormality prediction can be obtained by the method described above. Next, an automatic abnormality avoidance operation using the above-mentioned comprehensive judgment value as a judgment criterion will be described. The above-mentioned abnormality avoiding operation is performed by adjusting one or more state quantities of the plant in the state quantity adjusting section 7. When the total judgment value exceeds the threshold value, the state quantity adjusting unit 7 adjusts the set value of the state quantity from the reference set value in a direction to avoid an abnormality. At that time, if the abnormality avoidance direction cannot be judged only by the comprehensive judgment value, the partial evaluation value of each fuzzy proposition can be used. When the comprehensive judgment value becomes lower than the threshold value by performing the avoidance operation, the setting value of the state quantity is adjusted in the direction opposite to the abnormality avoiding direction, that is, in the direction of returning to the reference setting value. . Further, when the above-mentioned comprehensive judgment value exceeds or falls below the threshold value for a temporary period, it is not necessary to adjust the above-mentioned state quantity unnecessarily or to make a large adjustment at the same time as the operating condition of the plant. May destabilize. Therefore, when the total judgment value exceeds or falls below the threshold value for a certain period of time, it is desirable to adjust the state quantity in stages. It is possible to prevent abnormalities by repeating the above-described processing centering around the reference set value of the state quantity.

Next, as an example of an embodiment of the present invention, an example applied to the prevention of classification abnormality due to screen mesh clogging in the vibrating screen during classification of powder and granules in classification and classification processes will be described. The process of drying and classifying the powder and granules, which is the background of this embodiment, is configured as shown in FIG.
The slurry whose flow rate is controlled by the flow control valve FIC12 from the mixing tank (MT) 11 is dehydrated to about 20% by the dehydrator 14 and sent to the dryer 15. In the dryer 15, the powdery particles are dried by the hot air from the lower part while flowing from the inlet side (1 chamber) toward the outlet side (3 chambers). Further, not only the hot air but also the radiant heat of the internal heater 16 due to hot water is used as the dry heat source. At this time, the drying temperature is controlled by the hot air flow rate controller TIC24, 2, 3
A constant value control of the temperature of the three chambers is performed by turning the indoor heaters on and off. As it is sent in the direction of the three chambers, the gradually dried powder and granules pass through the discharge overflow weir 17 and are discharged to the vibrating screen device 20 side. The flow state of the powdery particles at this time is performed by a hot air flow operation, and the control is performed by the constant value control of the column top pressure by the flow rate controller PIC25. The powder and granules discharged from the dryer 15 are protected from static electricity by low-pressure steam blown by the operator operating the humidity controller 19 when passing through the vibration sieving device inlet duct 18. The powder and granules that have passed through the duct are classified by the vibrating and sieving device 20, and products with a predetermined particle size or smaller are discharged to the product hopper 23, and foreign substances and extraordinary products with a predetermined particle size or more are discharged from the extraordinary product discharge port 21. .

The above is the processing sequence in the normal continuous dryer operation. When the liquid level in the mixing tank (MT) 11 reaches the lower limit set value, the dryer operation mode switching three-way valve 13 is automatically operated. , The slurry is fed back to the mixing tank (MT) 11, and the dryer circulates. The abnormal state to be predicted in the present embodiment is an abnormal classification process in which an abnormally large amount of powder is discharged from the extraordinary product discharge port 21 of the vibration sieving device 20. The causes of this classification abnormality are
Adhesion to the screen mesh in the vibrating sieving device 20 due to electrostatic force due to overdrying of powder and granules and aggregation due to liquid crosslinking force due to undried powder and granules (not to mention the malfunction of the dryer 15 It is conceivable that the screen mesh is caused by the humidity controller 19 (including overhumidity etc.). If a classification abnormality occurs due to screen mesh clogging, the operator recognizes the abnormality by the abnormal alarm of the weight of powder particles on the discharge side of the vibrating sieving device and confirms the above-mentioned adjustment after confirming the state of the powder particles on the discharge site. This is dealt with by adjusting the opening of the humidity control vapor amount control valve of the humidifier 19. However, there is a difference in recognition of the discharged product state due to the difference in operator's skill level, and the discharge amount on the extraordinary product side and the humidity control work time vary. Further, there are cases in which the loss is increased by performing an erroneous operation such as a reverse operation in the humidity control operation. Therefore, the invention has been completed in order to predict the occurrence of classification abnormality due to the screen mesh clogging and perform an appropriate automatic avoidance operation in advance to prevent the classification abnormality due to the screen mesh clogging.

First, in this embodiment, the detection values (state quantity and change quantity) used for predicting the occurrence of classification abnormality due to screen mesh clogging are shown in FIG. As for these items, those that are considered to have contributed to the occurrence of classification anomalies are extracted by analyzing past data. Among them, the slurry supply three-way valve ON / OFF signal of the state quantity 58 is used for judging the above-mentioned two operating states, that is, the continuous dryer operation or the circulation operation. Either one of the state quantity and the change quantity is used as the detected value to be used, and the state quantity or the change quantity in each detected value is used according to the fuzzy proposition described later. When the detection value itself (state quantity) directly affects the agglomeration of powder particles, such as the output value of a moisture meter, the state quantity is the degree of change (change amount) in the detection value, such as dehydrator torque. For those that have a large influence, the change amount is adopted. These detected values are measured at the sites (shown by the state quantities 51 to 58) as shown in FIG. 7, respectively.

Next, it is necessary to create a fuzzy proposition for expressing the situation when a classification abnormality occurs, using the above detected values. There are four possible situations in which classification abnormalities occur, depending on the combination of the above-mentioned two operating states and the above-mentioned two causes of occurrence. Therefore, it is necessary to make independent predictions for each of these four occurrences. 9 to 12 are diagrams in which fuzzy propositions are created for each of the above four occurrence situations. These are stored in advance in the fuzzy proposition storage unit 3. A specific processing method will be described below by taking up the case of “overdrying side screen mesh clogging in the dryer continuous operation mode” shown in FIG. 9 among the above four occurrence situations. As shown in FIG. 9, first, the situation at the time of occurrence was verbalized, and this was divided into five fuzzy propositions and expressed using membership functions. The first half of each proposition is a part for judging the operating state by the three-way valve ON / OFF signal, and the latter half is used for calculating the fitness to the actual proposition. As the detection values used in each proposition, in order to express the expression contents more accurately, in Propositions 1 to 3, the change amount, Propositions 4, 5 are used.
Then, the state quantity is used.

The values of the respective parameters of the membership function in the latter half of the proposition are determined based on the values obtained by collecting the data at the time when the classification abnormality occurred in the past. The determination method is basically a method in which, as shown in FIG. 13, the maximum value of the values generated during the survey period is 0.25, the average value is 0.75, and the values corresponding to 0 and 1 at that time are adopted. Is taking. However, this is not absolute because there are cases where corrections are made while looking at the prediction results, and various other methods are possible. 5 as above
Two fuzzy propositions were created. For Proposition 5, the output values of the moisture meter used in Proposition 4 are used to correct the temperature. Therefore, the actual evaluation values used in Proposition 1 It is 4. FIG. 15 shows a fuzzy integration procedure for obtaining a comprehensive judgment value of classification anomalies based on detection values at a certain point in time in order to predict classification anomalies using the above four fuzzy propositions. Prior to that, the fuzzy measure values obtained by the online processing by the identification method described above are shown in FIG. First, the detection value at a certain time is processed by the processing method described above by the signal processing unit, and the fuzzy proposition storage unit 2 obtains the goodness of fit with respect to the fuzzy proposition from the obtained state amount and change amount. The step is S31.
Next, the goodness of fit is rearranged in ascending order (S32), and the respective differences are calculated (S33). These are multiplied by the corresponding fuzzy measures from FIG. 14 to obtain the sum, which becomes the comprehensive judgment value of the classification abnormality (S35). The above S
The processes of 31 to S35 are performed by the abnormality prediction unit.

As described above, the threshold value of the total judgment value is 0.
Since the value is set to 5, the resulting value of 0.491 is slightly below this threshold. When this comprehensive judgment value exceeds 0.5, it is predicted that a classification abnormality will occur a few minutes after that, and the classification abnormality can be prevented by an automatic avoidance operation described later. FIG. 16 shows changes in the respective propositional suitability (partial evaluation value) and prediction output (total evaluation value) when the above-described prediction processing is performed using past data. The black circle indicates the time when the classification abnormality actually occurred. In one place, there was a part where classification abnormality did not occur even though the predicted output exceeded the threshold, but in the other two places, classification abnormality actually occurred several minutes after the threshold was exceeded. Next, 4 of this embodiment
FIG. 17 shows the result of actual prediction for all one classification abnormality occurrence situation. The prediction success probability is the probability that the classification anomaly that actually occurred can be predicted, and the prediction probability is the probability that the classification anomaly has been predicted successfully. Although it is not 100%, it can be seen that there is a considerable probability of prediction. In addition, the prediction time, which is the time after the prediction until the actual classification abnormality occurs, is sufficient for performing the subsequent automatic avoidance operation, although there are variations.

Next, an automatic classification abnormality avoiding operation using the above prediction result will be described. In the present embodiment, the humidity control vapor pressure immediately before the vibrating and sieving device is used as the state quantity for the classification abnormality avoiding operation. Other state quantities such as an input current to a blower for blowing hot air may be adopted, or two or more state quantities may be combined and applied. This humidity-controlled steam pressure is adjusted by an operator during the conventional manual countermeasure operation, and by setting the humidity-controlled steam pressure as the operation state quantity, the classification abnormality prevention system can be used as an extension of the prediction system. This is because the configuration can be made in order to reduce the cost and simplify the system. FIG. 22 shows FIG.
15 is a partial view between the dryer 15 and the vibrating sieving apparatus 20 in FIG. B1 is a classification abnormality prediction unit described above, and B2 is a state quantity adjustment unit that performs an automatic classification abnormality avoiding operation. The state quantity adjusting unit B2 is composed of the humidity controller 19, a control unit 31 thereof, and a humidity steam pressure set value calculation unit 32. When the predicted value exceeds the threshold value or the predicted value that has exceeded the threshold value falls below the threshold value, the humidity control vapor pressure set value calculation section 32 receives the successive predicted values from the classification abnormality prediction section B1. The control unit 31 is instructed to change the set value of the humidity control steam pressure. At that time, although the threshold value was set to 0.5 as the predicted value of classification abnormality, the steam pressure change threshold value was set to 0.3 in this classification abnormality avoidance operation for more stable plant operation. ing. FIG. 23 shows a processing algorithm of the humidity control steam pressure set value calculation unit 32.

When the predicted value of the classification abnormality exceeds the threshold value of 0.3 for 2 minutes, the steam pressure is set to 0.0 from the reference pressure.
Adjust 1 (kg / cm ² ) and repeat. By the above operation, the predicted value fell below the threshold of 0.3, and when that state continued for 5 minutes, the steam pressure was changed to 0.01 (kg
/ Cm ² ) Return to the original pressure and repeat until the reference pressure is reached. The adjustment direction of the steam pressure depends on whether the predicted value exceeding the threshold value is on the overdry side or the non-dry side. Therefore, in the case of the present embodiment, the direction of state quantity adjustment can be determined by using only the comprehensive evaluation value as the predicted value. FIG. 24 shows the predicted value and the fluctuation state of the humidity control steam pressure when the classification abnormality prevention device by the prediction and avoidance operation of the classification abnormality described above is actually operated. It can be seen that the predicted value of classification anomaly is successfully suppressed by properly adjusting the humidity control steam pressure.

[0029]

As described above, the plant operation abnormality prevention system according to the present invention, which is configured as described above, provides a predicted value for the occurrence of an abnormality, and further automatically performs the abnormality avoidance operation according to the predicted value. By doing so, it is possible to reduce the physical and mental load on the operator, stabilize the process and quality, improve productivity, and reduce the number of employees.

[Brief description of drawings]

FIG. 1 is a diagram showing a schematic configuration of a plant operation abnormality prevention system according to the present invention.

FIG. 2 is a diagram showing a schematic processing flow of a plant operation abnormality prevention system according to the present invention.

FIG. 3 is an explanatory diagram of a membership function.

FIG. 4 is a schematic diagram showing a relationship between a fuzzy measure and a normal measure.

FIG. 5 is a schematic diagram showing a fuzzy integral model.

FIG. 6 is a diagram showing a flow of fuzzy measure identification processing.

FIG. 7 is a diagram showing a schematic configuration of a drying / classifying process of powder particles according to an embodiment of the present invention.

FIG. 8 is a diagram showing items of detected values according to an embodiment of the present invention.

FIG. 9 is a diagram showing a fuzzy proposition representing a case of “overdrying side screen mesh clogging in a dryer continuous operation mode” according to an embodiment of the present invention.

FIG. 10 is a diagram showing a fuzzy proposition representing a case of “overdrying side screen mesh clogging in a dryer circulation operation mode” according to an embodiment of the present invention.

FIG. 11 is a diagram showing a fuzzy proposition representing a case of “non-dried side screen mesh in a dryer continuous operation mode” according to an embodiment of the present invention.

FIG. 12 is a diagram showing a fuzzy proposition representing a case of “undry side screen mesh clogging in a dryer circulation operation mode” according to an embodiment of the present invention.

FIG. 13 is a diagram showing a membership function parameter determination method according to an embodiment of the present invention.

FIG. 14 is a diagram showing an identification result of a fuzzy measure according to an embodiment of the present invention.

FIG. 15 is a diagram showing a specific calculation example of a state total judgment value by a fuzzy integration model according to an embodiment of the present invention.

FIG. 16 is a diagram showing an example of a classification abnormality prediction result according to an embodiment of the present invention.

FIG. 17 is a diagram showing a classification abnormality prediction situation according to an embodiment of the present invention.

FIG. 18 is a diagram showing the amount of change to be obtained.

FIG. 19 is a diagram showing the amount of change obtained by a normal simple average.

FIG. 20 is a diagram showing a processing flow of how to obtain a change amount within a predetermined time in the past according to the present invention.

FIG. 21 is a diagram showing a difference in calculation result when coef is changed in the method of obtaining the change amount in the past predetermined time according to the present invention.

FIG. 22 is a diagram showing a schematic configuration of a classification abnormality prevention unit according to an embodiment of the present invention.

FIG. 23 is a diagram showing a processing flow of humidity control vapor pressure set value calculation according to an embodiment of the present invention.

FIG. 24 is a diagram showing a variation state of the classification abnormality predicted value and the humidity control steam pressure when the classification abnormality prevention unit according to the embodiment of the present invention is in operation.

[Explanation of symbols]

 1 ... Signal processing unit 2 ... Fuzzy proposition fitness calculation unit 3 ... Fuzzy proposition storage unit 4 ... Abnormality prediction unit 5 ... Fuzzy measure storage unit 6 ... Fuzzy measure identification unit 7 ... State quantity adjustment unit A1 ... Online processing unit A2 ... Offline Processing unit

Claims (9)

[Claims] 1. A plant operation abnormality prevention system for predicting an abnormality in plant operation from detection values in a plurality of parts constituting a plant and automatically performing an abnormality avoidance operation,
A fuzzy proposition storage means for storing a plurality of fuzzy propositions in which a verbalized situation at the time of occurrence of an abnormality is expressed by a membership function with the detected value as an evaluation item, and the detected value is applied to the fuzzy proposition. By doing so, a fuzzy proposition goodness-of-fit calculation means for calculating the goodness of fit of each fuzzy proposition, and a sequential identification method that minimizes the deviation between the target value of the fuzzy integration result and the total judgment value by the actual fuzzy integration model are obtained. Fuzzy measure identifying means used, fuzzy measure storing means for storing the identified fuzzy measure, predicting means for predicting anomaly of plant operation by fuzzy proposition conformity and fuzzy integration using the fuzzy measure And a state quantity adjusting means for adjusting one or more state quantities of the plant according to the prediction result obtained by the predicting means. Plant operation abnormality prevention system characterized by comprising in. 2. The fuzzy integration comprises the steps of rearranging the goodness of fit of each of the fuzzy propositions in ascending order, calculating the difference between them, and obtaining the sum of the products multiplied by the fuzzy measures corresponding to them. Plant operation abnormality prevention system. 3. The plant operation abnormality prevention system according to claim 1, wherein the detected value used as the evaluation item of the membership function includes the current state quantity of the detected value which changes with time. 4. The plant operation abnormality prevention according to claim 1, wherein the detected value used as the evaluation item of the membership function includes a change amount of the detected value that changes over time within a past predetermined time. system. 5. The plant operation abnormality prevention system according to claim 4, wherein the change amount Dev (t) is calculated by Dev (t) = dat (t) -av (t-1). Here, t is the current time, dat (t) is the state quantity at the current time, and av (t-1) is the average value of the state quantities within the above predetermined time in the (t-1) time. 6. A state quantity mdat for obtaining the av (t-1).
To The plant operation abnormality prevention system according to claim 5, which is calculated by Here, sdev (t-2) is the standard deviation of the state quantity in the past predetermined time in (t-2) time, and coef is a correction coefficient. 7. The plant operation according to claim 1, wherein in the identification of the fuzzy measure, data indicating a possibility of occurrence of an abnormality before the time when the abnormality actually occurs is used for prediction. Abnormality prevention system. 8. The state quantity is adjusted by judging that the abnormality prediction or the abnormality prediction is completed when the abnormality prediction state or the abnormality prediction end state continues for a certain period of time or more.
The plant operation abnormality prevention system according to any one of to 7. 9. The plant operation is a drying and classifying process of powder and granules, and the abnormal operation of the plant is classification due to electrostatic adhesion due to overdrying of the powder or granules or aggregation due to undried powder. 8. The plant operation abnormality prevention system according to any one of claims 1 to 7, wherein a classification abnormality is caused by screen mesh clogging in the vibration sieving apparatus, and the state quantity adjusting means is a humidity control apparatus immediately before the vibration sieving apparatus.