JPH0643919B2 - Phase estimation method and failure diagnosis device - Google Patents

Description

TECHNICAL FIELD The present invention relates to a data processing method and a data processing apparatus suitable for application to failure diagnosis, noise analysis and the like.

[Prior Art] Conventionally, as a data processing method used for failure diagnosis,
The following methods are known.

A method in which the upper and lower limits of the detection amount are set in advance, and if the detection amount deviates from this range, a failure is diagnosed.

A method in which a limit is set based on the deviation of the normal detection amount from the average value or the deviation from a certain standard function, and if it exceeds this limit, a failure is diagnosed.

A method in which a normal space and an abnormal space are set in a vector space having each detection item as an element, and a failure diagnosis is performed based on a position occupied by a vector corresponding to an actual detection amount in the vector space.

A knowledge engineering method in which a truth table is created in advance for each detection item in correspondence with the detection amount, and the actual detection amount is applied to this truth table to perform fault diagnosis.

[Problems to be Solved by the Invention] By the way, each of the conventional methods described above has the following drawbacks.

(1) A failure is diagnosed according to the state of one point on the time axis. That is, it does not have a processing method for temporal changes.

(2) Usually we have to believe in the value from only one detector. For example, when starting or stopping an engine, the judgment is made based only on the engine speed. Therefore, if this detector fails, the correct judgment cannot be made. In other words, the system is vulnerable to detector failure.

(3) When inputting the detection data to the knowledge engineering system, the interface was not successful. This is because the knowledge base of the knowledge engineering system is based on qualitative linguistic expressions of humans, and there is a large difference between the physical quantity indicated by the detected data and the expression of the knowledge base.

The above (1) and (2) will be further described. For example, taking an engine (operating system) as an example, even during normal operation, long-term stop, short-term stop, no load, 25% load, 50% load, 75% load, 100% load, 110% load, dangerous state, etc. There are various phase states, and the engine continues to operate while transitioning between these phase states (see FIG. 9). However, in the conventional fault diagnosis method, these phase states are not considered at all, or even when they are considered, they depend on only a detector such as an electric power meter or a horsepower meter, or a normal space or an abnormality at one point on the time axis. One of the methods was used to determine the degree of conformity to the space.

However, the engine has a dynamic characteristic, and there is an transient state in which a physical quantity having poor response such as temperature changes slowly even if the value of the power meter changes rapidly.
In such a situation, if the conventional method is applied as it is, an inconvenience such as being determined to be abnormal occurs even though the operating state is normal. Therefore, the grace period is set to be longer than the estimated duration of the transient state, and the device is not devised to be abnormal within the grace period. However, in this case, the detection will be delayed by the grace time for the actual abnormality.

In addition, some types of abnormalities are related to the transition between phase states and the history of the rate of phase change, which occurs only when the load changes abruptly. Was not considered at all.

The above (3) will be further described. A knowledge engineering system using a program such as Lisp, Prolog, or an expert system construction tool is composed of three blocks: a) knowledge acquisition unit, b) knowledge expression unit, and c) knowledge utilization unit.

The knowledge acquisition section refers to a section in which a knowledge engineer inputs knowledge acquired from experts as a knowledge base. Examples of knowledge entered here, if Shimese the operation of the engine for example, coming down a particular cylinder outlet exhaust temperature, other cylinder outlet exhaust temperature rises slightly, and if the load condition is stable , The exhaust temperature has dropped, and there is a possibility that the cylinder's fuel valve is clogged.

In this case, the underlined part means a qualitative expression against the dynamic characteristics of the engine, and conventional systems did not have such an expression. Therefore, a format was used in which human beings were asked to understand and judge this kind of expression, and to ask humans. For example, is the exhaust temperature at a specific cylinder outlet decreasing?

1. Yes 2. No 3. I don't know.

This method has the following drawbacks.

First of all, because humans must intervene, ・ Labor saving and automation cannot be achieved ・ High risk of erroneous input ・ Instantaneous processing cannot be performed because human reaction is slow ・ Variation in human sense Because of this, it is difficult to maintain consistency of input data due to individual differences and the state at that time.

Furthermore, since many expressions cannot be replaced with simple expressions, the number of expressions becomes enormous. ・ Even expressions corresponding to the same expression (for example, "come down") often differ depending on the situation. So the number of formulas becomes even more enormous.

-If you are not a knowledge base creator, you cannot fully understand the meaning of expressions.

・ It takes time and effort to make the formula.

・ Changing expressions and rules once entered is a huge task.

The present invention has been made in view of such a background, and in a device having a plurality of detectors, a highly reliable data processing which does not depend on one detection amount by comprehensively judging each detection amount. It is an object of the present invention to provide a data processing method and a data processing device capable of detecting the presence or absence of a failure in the detector itself.

It is another object of the present invention to provide a data processing method and a data processing device that perform data processing based on the transition between phase states and the rate of change thereof and facilitate an interface with a knowledge engineering system.

[Means for Solving the Problems] In order to solve the above problems, the present invention relates to (1) a plurality of types of series of detection data obtained from an object, the time when each detection data is obtained, and each detection data. A phase estimation method for estimating a phase that the object faces by processing based on an independent variable corresponding to each detection data, such as frequency, regarding the degree of change of the detection data for each type of detection data. A plurality of fuzzy sets, a plurality of fuzzy sets for the change of the phase, and rules for each type of the detection data, each fuzzy set for the degree of change of the detection data for the change of the phase And a rule associated with one of a plurality of fuzzy sets of A significant change in the detection data is extracted by comparing the amount of change in the value of the detection data with a preset value, and for each type of detection data, fuzzy corresponding to the significant change in the detection data. A process of obtaining a set and its certainty factor, and each belief of each fuzzy set of a significant change in the detected data of each type associated with the fuzzy set by the rule, for each fuzzy set with respect to the change of the aspect A fuzzy reasoning consisting of a process of obtaining a certainty factor for a fuzzy set of changes in the relevant phase by calculating an arithmetic mean of the degrees, and estimating the changes in the certain conditions and the certainty factors thereof. To do.

(2) The aspect estimation method according to claim 1, further comprising a step of classifying the degrees of change corresponding to similar detection items into groups and performing the addition-type fuzzy inference for each group. .

(3) In the situation estimation method according to any one of claims 1 and 2, fuzzy inference is applied to the detected data,
It is characterized by inferring the current position of the stage.

(4) The aspect estimation method according to claim 3 is characterized in that a transition state of the aspect position with the transition of the independent variables such as the time and frequency is obtained.

(5) The aspect estimating method according to claim 4 is characterized in that each detected data is investigated with reference to the current position of the aspect and abnormal detected data is extracted.

(6) The aspect estimating method according to claim 4 is characterized in that the fuzzy sets used for the fuzzy inference and the addition-type fuzzy inference are changed according to the current position of the aspect.

(7) The aspect estimating method according to claim 4, wherein the aspect change, the aspect position, and the transition state are output to a knowledge engineering system.

(8) Having a plurality of detectors for detecting the operating state of the device,
In a failure diagnosis device that detects a plurality of situation states that change corresponding to the detection amount of each detector, and performs failure diagnosis of the device and the detector based on this situation condition, the degree of change of the detection amount is Change detection means for extracting a significant change by comparing with a preset value, and (a) a plurality of fuzzy sets for the degree of change in the detection amount of each detector, (b) a change in the aspect And (c) rules corresponding to each type of the detection amount, and each fuzzy set regarding the degree of change of the detection amount is set to one of the plurality of fuzzy sets regarding the change of the situation. Based on a rule to be associated, for each detection amount, a fuzzy set corresponding to a significant change in the detection amount and its certainty factor are obtained, and for each fuzzy set regarding the change in the situation, the rule is set. Therefore, the addition type fuzzy which obtains the certainty factor for the fuzzy set of the change of the phase by calculating the arithmetic mean of the certainty factors of each fuzzy set of the significant changes of the detection amounts associated with the fuzzy set And a means for performing fuzzy inference on a result of the additive fuzzy inference and a portion of the degree of change that has not been subjected to the additive fuzzy inference. It is characterized in that the failure of the device and the detector is estimated based on the output of the applying means.

[Operation] According to the above means (1) and (2), a similar detection amount,
For example, the detection amount obtained from each cylinder of the engine is
Since the averaging is performed by the addition type fuzzy reasoning, even if one of the detectors is abnormal, the whole phase judgment is error-free.

Further, according to the means of (3) to (4) above, since the phase state can be accurately determined, self-excited oscillation (hunting) of the control system can be detected, and the cause thereof should be investigated to prevent hunting. Will also be possible.

According to the means of the above (5), since a detector having an abnormality can be identified, correct judgment can be made by the data excluding the data from this detector even if the certainty factor is slightly reduced.

Further, even if the amount of change is the same, if the current position of the position changes, the influence of the amount of change also changes. However, according to the means of (6) above, the fuzzy number used for fuzzy inference can be changed according to the position of the position. It is possible to more appropriately understand the situation of the situation by changing the above. Also, corrections to make the understanding of the state of the situation closer to the interpretation of experienced human beings,
Editing makes it very easy for humans to understand.

Further, the state (7) means can easily interface with the knowledge engineering system.

Further, according to the device (8), the failure diagnosis can be executed at the same level as that performed by an experienced person.

EXAMPLES Examples of the present invention will be described below with reference to the drawings.
Prior to that, fuzzy reasoning will be explained.

FIG. 3 is a diagram for explaining a specific method of fuzzy inference, FIG. 3 (a) shows a condition part, and FIG. 3 (b) shows a conclusion part.

In FIG. 3A, the horizontal axis is a variable, and N fuzzy numbers Fz1 to FzN are set according to the variable.
These fuzzy numbers Fz1 to FzN are a trapezoidal set having intersections with each other, and their height is 1, that is, 1
It is set at 00%. If, for example, the variable x takes the value shown in the figure, the variable x is the fuzzy number Fz.
3 and Fz4, and the fuzzy number Fz3 is 30.
%, And the fuzzy number Fz4 is 70%.

Next, in FIG. 3 (b), the horizontal axis is a variable representing the conclusion, and for the fuzzy numbers Fz1 to FzN, triangular fuzzy sets Gz1 to Gz1 also having N mutually intersecting portions.
GzN is preset. Here again, the height of the fuzzy sets Gz1 to GzN is 1, that is, 100%.
A region formed by intersecting the fuzzy set Gz3 with 30% level and the fuzzy set Gz with 70% level described above.
If the center of gravity of the area (shaded area in the figure) formed by intersecting 4 is taken and its horizontal axis coordinate is g, this is the conclusion value to be obtained. In this way, defining one definite value for a certain area is referred to as defuzzificatio (defuzzificatio).
n: defuzzification).

Note that the fuzzy numbers Fz1 to FzN and the fuzzy set Gz1
˜GzN is not limited to a trapezoid or a triangle, and may be any generally convex fuzzy set. Further, specific examples of the variable x include an average degree of change μ _n / n, an instantaneous degree of change ν, a change level g _ν , and g _{μν / n} , which will be described later.

Next, FIG. 4 is a diagram for explaining a method of addition-type fuzzy inference, and FIG. 4 (a) is a fuzzy number F of the antecedent part.
z1 to FzN, FIG. 6B shows regions Dz1 to DzN corresponding to the consequent part fuzzy number, and FIG. 6C shows fuzzy sets Gz1 to GzN of the conclusion part.

First, N fuzzy numbers Fz1 to FzN for each detection item
Is determined, the values of the corresponding areas Dz1 to DzN are obtained according to the variable value of that item, and finally each area Dz1
~ The arithmetic mean is calculated for each DzN, and the distribution is defined as the fuzzy sets Gz1 to GzN of the conclusion part.

FIG. 5 and FIG. 6 are diagrams for explaining the addition-type fuzzy inference of FIG. 4 in more detail and concretely by taking the operating state of the engine as an example. Engine speed as a detection item,
There are turbine speed, supply pressure, exhaust temperature, etc., and the value of these variables is g.

FIG. 5 shows only one area (for example, the area Dz1) for each item. In the figure,
The value of the area Dz1 of each item of the consequent part is determined according to the value occupied by the variable value g of each item of the antecedent part on the fuzzy number Fz1. In this figure, the area Dz1 for item 1 (for example, engine speed) is 100%, the area Dz1 for item 2 (for example, turbine speed) is 60%, the area Dz1 for item 3 (for example, supply pressure) is 0%, ... The level of the area Dz1 of the item m (for example, exhaust temperature) is set to 40%. If these are arithmetically averaged, the fuzzy set G of the conclusion part
z1 is obtained.

When this operation is executed for each area, each level is obtained for each area Dz1 to DzN as shown by the shaded area in FIG. 6, and the fuzzy set Gz of the conclusion part is obtained. Then, one specific value is determined by calculation such as taking the center of gravity. As mentioned above, this operation is called defuzzification. In this way, the level of each region Dzi is determined by the value of the variable value of each item on the fuzzy number Fzi (i = 1 to N), and the fuzzy set Gzi of the conclusion part is obtained by arithmetically averaging the levels. This is the outline of additive-type fuzzy inference.

It is needless to say that the fuzzy set includes the crisp set and thus the method shown in FIG. 7 may be used.
According to this method, the value g at the conclusion is immediately obtained. Further, instead of the above-mentioned arithmetic mean, a weighted average in which items and / or areas are weighted may be taken.

Embodiment 1 FIG. 1 is a block diagram showing the configuration of Embodiment 1 of the present invention,
2 (a) and 2 (b) are functional block diagrams thereof. In this first embodiment, the present invention is applied to failure diagnosis of an engine.

In these figures, 1-1, 1-2, ... 1-5, ...
1-m (collectively referred to as 1) is various detectors, for example, 1-1 is an engine speed detector, 1-2 is a first cylinder exhaust temperature detector, and 1-5 is a fourth cylinder. Exhaust temperature detector, 1-m is a turbine speed detector. These detectors 1-1, 1-2, ...
The outputs of 1-5 and 1-m are data collection units 2-
1, 2-2, 2-5, ... 2-m (referred to as 2 when collectively referred to, hereinafter the same).

The data collecting units 2-1, 2-2, 2-5, ... 2-m are
Each of the detectors 1-1, 1-2, ... 1-is composed of an amplifier, a sample hold circuit, an A / D converter, and the like.
5, ... 1-m analog output is converted to digital signal and output. In addition, normally, the data collection units 2-1 and 2-2
-2, 2-5, ... 2-m is also used as a single circuit, and an analog multiplexer is connected to the preceding stage of the circuit to form and output a digital signal in a time division manner. . It should be noted that these detector 1 and data collection unit 2
Does not have to be connected online to the latter part of the discussion below. For example, the data may be stored in a storage medium such as a floppy disk and read out.

Specifically, the digital output obtained as described above is supplied to the phase change understanding unit 30 and the position understanding unit 40 among the four blocks including the CPU.

The phase change understanding unit 30 comprehensively determines each detection item such as the engine speed, the exhaust temperature, and the turbine speed, and determines the following one regarding the overall phase, for example, the increase or decrease of the engine load.
This is a part for understanding the phase change R in item 3.

R1. Where it dropped sharply R2. Rapidly falling R3. Undershoot R4. Transient state after decrease R5. R6. Slightly falling R7. Stable R8. Slightly rising R9. R10. Transient state after rising R11. Overshoot R12. R13. Where it has soared In order to understand these 13 aspects R1 to R13,
The phase change understanding unit 30 is configured to execute the following three stages of calculation.

(1) First stage In this stage, considering the accuracy of the detector and the accuracy of measurement,
At the stage of investigating the sampling value of each detected amount along the time axis for each detected item and extracting a significant change, the change extraction units 31-1, 31-2, ... 31-5, ... 31- m is the part that executes this operation. The change extraction unit 31 performs the calculation described below and outputs the following three types of variables.

· N: significant change survey number until discovery · [nu: Instantaneous change degree · mu _n: degree of change should be noted that these values, as described below, is the most recent N samplings stored for each detection item.

(2) Second stage This stage is a stage in which a significant change in the motion system is detected by fuzzy inference. Each value n, obtained in the first stage,
Based on ν and μ _n , fuzzy inference is performed for each detection item as to which of the nine change levels D1 to D9 ν and μ _n / n apply, and the change levels g _ν and g _{μn / n} are obtained. Ask.

D1. Super-descent D2. Dive D3. Down D4. A little descending D5. Stable D6. On the rise D7. Rise D8. Soaring D9. Super steep rise Fuzzy inference section 32-1, 32-2, ... 32- in FIG.
5, ... 32-m is a part that executes the calculation of the second stage, and the degree of change of each detected amount is the above-mentioned change level values D1 to D.
Which of 9 is applicable, considering the characteristics of the operating system,
Fuzzy reasoning. Here, the characteristic of the operation system is the magnitude of the influence of the detected change in the physical quantity on the operation. For example, considering the exhaust gas temperature of the engine, the accuracy of the measuring system is about 3 ° C., so a change of 5 ° C. can be sufficiently captured as a change in physical quantity. However, since the exhaust temperature of the engine changes by about 10 ° C. even during the steady operation, it can be said that it is meaningless to capture 5 ° C. as a change in the state of the engine, that is, a change in the operating system. In consideration of such characteristics of the operation system, nine fuzzy numbers (corresponding to Fz1 to FzN in FIG. 3, where N = 9) are preliminarily associated with the change levels D1 to D9 for each detection item. Is set, fuzzy inference is performed to calculate a significant change level, and a change level g _ν corresponding to the instantaneous change degree _ν and an average change degree μ _n / n are calculated for each detection item.
The change level g _{μν / n} corresponding to is output (corresponding to g in the conclusion part of FIG. 3). Further, regarding changes that are not possible in the operation system, if the change levels D1 and D9 are assigned to them, an abnormality of the detector can be detected at this stage.

(3) Third stage In this stage, the change levels g _ν and g _{μn / n} obtained for each item in the second stage are combined, and the change in the current load state is the above-mentioned overall phase changes R1 to R13. It is to calculate which of the above applies.

Fuzzy inference units 33-1, 33-2, ... 33- in FIG.
5, ... 33-m, in-group addition type inference unit 34,
The addition-type fuzzy inference 35 and the defuzzification unit 36 are parts that execute this third stage.

First, each inference unit 33-i of the fuzzy inference unit 33 is
For the change levels g _ν and g _{μn / n} obtained in the step,
The fuzzy inference shown in FIG. 4 is performed. In each inference unit 33-i, two items are set in correspondence with the two change levels g _ν and g _{nν / n} in the form shown in the antecedent part of FIG. 4 (a), and further, 13 overall phase changes R1 for each item
To 13 fuzzy numbers Fz1 to Fz1 corresponding to R13
3 (N = 13) is preset. By applying the above-mentioned change levels g _ν and g _{μn / n} to these fuzzy numbers Fz1 to Fz13, grades for each region Dzi as shown in the consequent part of FIG. First
In the figure, these are fuzzy sets F = (X1 / R1 +
X2 / R2 + …… + Xc + Rc). That is, each inference unit 33-i outputs a fuzzy set as shown in FIG. 4 (b). These fuzzy sets are output for each change degree μ _n / n, ν for each of m detection items. That is, m pieces of F _{μn / n} and Fν are output.
Here, each element Xc / Rc of the fuzzy set F (c = 1 to 1)
3) is the element Rc of 13 phase changes R1 to R13,
The corresponding grade Xc is represented (the grade Xc is hereinafter referred to as the score of the element Rc).

The in-group addition-type fuzzy inference unit 34 performs addition-type fuzzy inference between items having the same meaning (for example, items such as the cylinder outlet exhaust gas temperature in an engine having a large number of cylinders). It outputs a fuzzy set in arithmetically averaged form as shown in the conclusion section. This is to prevent the weights of the items supplied to the addition-type fuzzy inference unit 35 from being biased and to stabilize the items having the same meaning. To explain further, for example, the cylinder outlet exhaust temperature can be obtained by the number of cylinders, so if the data obtained when there are many cylinders are used independently, the influence becomes too strong, and it becomes impossible to properly understand the overall situation. Will end up. Therefore, these are averaged and balanced with other data, for example, data that can be obtained only one such as the engine speed and the supply pressure. In addition, by averaging, even if one of the detectors that detect the outlet exhaust temperature of each cylinder fails, its effect can be prevented from directly appearing, and overall stabilization can be achieved. Can be achieved.

The addition-type fuzzy inference unit 35 performs addition-type fuzzy inference as shown in FIG. 4 on the fuzzy set supplied from the fuzzy inference unit 33 and the in-group addition-type fuzzy inference unit 34. That is, a plurality of sets of fuzzy sets F _{μn / n} and Fν in the form shown in the consequent part of FIG. 4 (b) are arithmetically averaged, and as shown in the conclusion part of FIG. 4 (c). The fuzzy set H _{μn / n} and Hν are calculated.

Next, the defuzzification unit 36 supplies the fuzzy set H _{μn / n} , supplied from the addition-type fuzzy inference unit 35.
This is a part that outputs the current phase change R and its certainty factor M _R based on Hν. That is, the current phase change R
Determines and outputs to which of the above-mentioned phase changes R1 to R13 the certainty factor corresponds.

When performing this calculation, consider that there are items that have good responsiveness and items that have poor responsiveness to changes in phase. For example, an item with good responsiveness such as engine speed immediately responds to a change in phase, whereas an item with poor responsiveness such as temperature changes slowly even when a sudden change in phase occurs.
Therefore, apart from when the situation changes slowly and continuously, when the situation changes stepwise, there are items with poor responsiveness, such as temperature,
Item with constant change level, that is, change level D5 described above
The item of becomes easy to exist. In other words, it can be said that the score of the phase change R7 of stability is potentially high. The defuzzification inference unit 36 performs defuzzification by introducing knowledge (rules) that weakens the influence of such a potentially strong item or introduces knowledge (rule) that makes a small variation insensitive. . An example of the rule in this case is shown below.

If the score (certainty factor) of a certain one of the 13 phase changes R1 to 13 is outstanding, that is adopted. That is, when the difference between the highest point and the second point> the set value (PEAK), the highest point is adopted. Note that this rule is not applied when the aforementioned phase change is R7 (stable).

If the two adjacent states are both protruding, the highest score will be used. That is, when the highest point and the second point> the set value (DIFFER), the highest point is adopted.

The total score on the ascending side (R8 to R13) and the descending side (R1 to R13)
R6) is compared with the total score, and if there is no significant difference between the two, it is considered stable. That is, (a) total score on the upside-total score on the downside> set value (ALLO
In case of W), the highest point on the ascending side is adopted.

(A) Downward total score-upward total score> set value (ALLO
In case of W), the highest point on the descending side is adopted.

(C) If the above two conditions (a) and (b) are not satisfied, the condition is stable.

Thus, the defuzzification is executed, and the result R
(Either R1 to R13) and outputs the confidence _{M R.}

Phase position understanding unit 40 The above is the configuration of the phase change understanding unit 30. However, it is not possible to understand what kind of situation we are currently in, simply by understanding the above-mentioned changes. Therefore, it is necessary to know the position of the stage. In this case, relying on only one detector, if this detector fails, not only will it not function properly, but it will also lead to detrimental decisions due to incorrect data. Therefore, the stage position understanding unit 40 that comprehensively determines the stage position is configured. Here, it is considered to understand the positions of the following nine phases by taking the load state of the engine as an example (No. 9).
See figure).

P1. Long-term suspension P2. Short-term suspension P3. No-load area P4.25% area P5.50% area P6.75% area P7.100% area P8.110% area P9. Danger Area In order to understand the position of such a position, the position understanding unit 40 has the following configuration.

First, the fuzzy inference units 41-1, 41-2, ... 41-
5, ... 41-m are data collection units 2-1, 2-2 ,.
2-5, ... 2-m is a part for performing fuzzy inference on the detected data, and performs fuzzy inference as shown in FIGS. 4 (a) and 4 (b) for each item. That is, the detected value is plotted on the horizontal axis for each item, and nine fuzzy numbers corresponding to the above-mentioned phase positions P1 to P9 are set in advance (N = FIG. 4).
9) The fuzzy set I = (W1 / P1 + W2 / P2 + ... + WJ / PJ) of the consequent part is derived from the sampled detection values (J =
9). Thus, m fuzzy sets I are obtained.

Then, similarly to the third stage described in the aspect change understanding unit 30, in-group reasoning is performed between similar items. This is executed by the intra-group addition type fuzzy inference unit 42, which takes the arithmetic mean between similar items and outputs the fuzzy set of each conclusion unit (see FIG. 4).

Each of the above fuzzy sets is supplied to the addition-type fuzzy inference unit 43. The addition-type fuzzy inference unit 43 is shown in FIGS.
A fuzzy set L = (Z1 / P1 + Z2 / P2 + ... + Z) obtained at the conclusion part by executing the additive fuzzy inference as shown in the figure
J / PJ) to the defuzzification unit 44. From the fuzzy set, the defuzzification unit 44 determines the position P (any one of P1 to P9) of the position and the certainty factor M thereof.
Calculate and output _P and.

The point to be noted here is that when considering the fuzzy number of the short-term stop P2, regarding items with slow responsiveness such as temperature,
Like F _Z 2 shown in FIG. 6 (d), it has a very wide range from the room temperature level to the dangerous area. In other words, items related to temperature, even when the engine is operating
You will be giving a high score to the short-term suspension. Therefore, the following rules are set as the rules for defuzzification.

Aspect change understand the above result R, referring to M _R, if dips side from transient R4 after lowering, stopping side P
1, the total of P2 and the total of the operating sides P3 to P9 are compared,
Choose the one with the most.

If not, compare the average on the stop side with the total on the operating side and select the one with the larger number. That is, the weighting on the operation side is increased as compared with the above.

The stage position with the highest score is selected from the selected side, and the position P of the stage is stored together with the certainty factor _MP (corresponding to the score).

Incidentally, scores for addition type fuzzy inference is generally represent confidence M _P of the condition. Therefore, based on the score p, the linguistic expression of the certainty factor M _P is determined as follows to express the human sense.

100 ≧ p> 80 …… Surely 80 ≧ p> 60 …… Probably 60 ≧ p> 40 …… Maybe 40 ≧ p ……………… I can't say for sure, but maybe g obtained as above. _{μn / n} , g _ν , R,
M _R , P, and M _R are stored in the storage unit 70.

Phase state transition history pattern understanding section 50 This understanding section 50 is a section for understanding the transition history of the phase state, and is composed of a near past understanding section 51, a current understanding section 52, and a global understanding section 53.

Near past understanding unit 51, dates from the current within a range set to the past, aspects condition data, i.e., the result data R and its confidence M _R aspects _change, and the result data P and its confidence M aspect position Check _P and see, for example, if there was a sudden change in the situation. The phase change sharply rises (R12, 13 described above) or sharply drops (R1, R
2) If yes, check how many previous samplings it happened. Next, if the phase position P changes within a certain sampling number width (set value) after that point, it is interpreted that the sudden change is caused by the transition of the position of the phase, and the conclusion is that the phase is "what before sampling". The output of the content is that “the position of the phase has made a sudden rise (a sudden drop) and the position of the phase has changed from“ P _X to P _y ”. On the other hand, if not, it is interpreted that "a large change within the same phase position P _X " has occurred. In the same manner as for the gradual change, the phase changes R5, R6, R7, R8 and R9 are investigated. As a result of understanding such a transition, "P
From _X need if you save a linguistic expression, such as that P _y has changed rapidly in. " That is, unlike the conventional system that stores a huge amount of all data, it is sufficient to store a logically organized language code, so that the storage capacity can be reduced. Moreover, since the linguistic result is easily adapted to the logic, the subsequent knowledge engineering reasoning can be strongly developed.

The current understanding unit 52 can compare the result obtained by the near-term understanding unit 51 with the current state to know whether or not a phase transition has occurred. Also, the current phase change data R
When it can represent current changing state by the confidence factor M _R.

The global understanding unit 53 indicates that the phase change R is on the lower side (R1 to R
In the case of 6), -1 is set for stable (R7), 0 is set for stable (R8 to R13), and +1 is set for the rising side (R8 to R13), and the next calculation is performed from a preset point in the past.

Count when it is not zero. That is, the number of fluctuations is obtained.

Perform simple addition. That is, it is determined whether it is on the downside, on the upside, or up and down.

With this, it is possible to express (a) a large or small load change, (b) constant, or (c) a decrease or increase when there is a load change. The expressions that the load fluctuation is large, small, large, and small are powerful information for engine maintenance.

The current situation state confirmation unit 60 confirms the situation position P and the change R, and the dynamic state confirmation unit 61 confirms the dynamic state.
And a current phase state confirmation unit 62.

The dynamic state confirmation unit 61 is a unit that confirms the dynamic state of the present system by the system check room. Here, the system check room means a rule for checking whether or not the operation is normally performed,
There are the following.

The temperature does not rise from a sudden fall to a sudden rise and vice versa. In other words, in the item of temperature, the transition from the change level D1 (super-deep fall) or D2 (rapid fall) to D8 (rapid rise) or D9 (super-rapid rise), or vice versa, may not occur during one sampling. Absent.

For example, considering the number of revolutions, 500 rpm easily rises when starting from a long-term stop, but 1
There is no increase of 500 rpm from 00% load. In this way, in a certain position P, the range of sharp rise and sharp decrease is restricted, and in the position P7, the change level D9 does not occur.

There are also certain restrictions on response relationships. For example, when the engine starts to rotate, the turbine rotates and the supply pressure is generated.

When the detection item is temperature and the above-described change level D changes abruptly (D1, D9), there is a possibility that an abnormality has occurred in the detector.

For example, when the position P is at P5 (50%),
If the engine speed is 0, there is a possibility that an abnormality has occurred in the speed detector.

The success or failure of the operation state is determined using such a system check rule, and the result is supplied to the present situation state confirmation unit 62.

Current aspects status checking unit 62 stores data R obtained when the system check result is normal, M _{R, P,} and M _P in the storage unit 70.

On the other hand, in the case of abnormality, the meta-rule is applied, and the addition-type fuzzy inference units 35 and 43 and the defuzzification unit 3 are applied.
Give feedback to 6,44. Here, the meta-rule means a higher-level rule and controls application of the rule itself. For example, the addition-type fuzzy inference unit 3
5 and 43 are instructed to change the fuzzy number, except for the abnormal detector (by changing the set value) to redo the addition-type fuzzy inference, or to the defuzzification unit 3
6 and 44 are instructed to perform defuzzification except for the abnormal detector.

The storage unit 70 stores the data g _ν , g obtained by the above units.
_μn / _{n, R,} it is for storing M R, P, and _{M P} time series.

In FIGS. 2A and 2B, reference numeral 80 is an interface unit. The interface unit 80 uses the phrase storage unit 8 based on the data from the phase state transition history pattern understanding unit 50.
3, a message assembling unit 81 that reads out the phrase stored in No. 3 and creates a message corresponding to the data, and a message output unit 8 that outputs the message as an electric signal.
2 and.

The data output from the message output unit 82 is sent to the peripheral unit 90 including the display device 91, the file device 92, the memory buffer 93, and the communication device 94, and is supplied to the knowledge engineering system 100 via this.

The knowledge engineering system 100 performs a failure diagnosis based on various data obtained in this embodiment, and drives a static rule when the current phase state confirmation unit 60 determines that the phase is stable, and qualitatively determines the human being. Inference is performed using a knowledge base described by expressions to control the controlled object.

Operation of Embodiment 1 The understanding of the phase change is performed as follows.

(1) First stage The change recognizing units 31-1 to 31-m perform a fixed calculation on the data (change data) supplied from the data collecting units 2-1 to m, and the number of times n until significant change discovery, The instantaneous change ν and the change μ _n are obtained.

In executing this operation, the following values are set in advance.

Allowable change in consideration of detection system and measurement itself ... K
(Refer to FIG. 8) Change threshold (change data and function f (X _o −X _n , K) consisting of the change allowable value K is recognized as change when the change is larger than this threshold. Yes) SL
h Stable threshold (recognized as stable when the change in the value of the function f (X _o −X _n , K) is smaller than this threshold) ... S
La data storage count (number of times sampling data is stored) N maximum number of surveys n _max number of survey discontinuations when stable n _L Next, the data values are expressed as follows.

a) under the current value ...... _X o b) n sampling previous value ...... _{X n} such setting, change recognition unit 31-1~m performs computation by the procedure described below.

The sampling data is stored in time series over a given number of times N of data storage. That is, the latest N sampling data items are sequentially arranged in the memories (not shown) of the change recognition units 31-1 to 31-m.

As shown in the flowchart of FIG. 10, the following calculation is performed on each detected data starting from the present and within the maximum number of investigations n _max .

Step SA1 · μ _n = f (X _o −X _n , K) is calculated.

(However, 1 ≧ n ≧ n _max ) ・ Number of times n until significant change occurs, and instantaneous change degree ν = μ
Ask for ₁ .

Note that f (X _o −X _n , K) is, for example, f
_{(X o -X n, K)} = {(X o -X n, K)} using a ^2.

Step SA2: It is checked whether or not n = n _max . If YES, the operation proceeds to step SA7 to enter the final stage, and if NO, the operation proceeds to step SA3 to continue the investigation.

Step SA3 · μ _n > SL _h , that is, whether or not there is a significant change is checked. If YES, the process proceeds to step SA7, and if NO, the process proceeds to step SA4.

Step SA4 · μ _n > SL _a Check whether or not it is stable, that is, if YES, proceed to step SA5, and if NO, proceed to step SA6.

Step SA5 ・ Check whether n <n _L , that is, whether to continue the investigation at the time of stability. If n <n _L , proceed to step SA6 and continue the investigation. If n ≧ n _L , proceed to step SA7 and end step. to go into.

Step SA6 · n is incremented by 1 and the process returns to step SA1.

Step SA7 ・ Stores ν, n, and μ _n obtained above, and ends the processing.

Thus, the number of investigations n until the significant change is found, the instantaneous change degree ν (= μ ₁ ) and the change degree μ _n are obtained. Each of these represents the degree of change in the detected amount.

The stable threshold value SLa and the change threshold value SLh are 0 <
There is a relationship of SLa <SLh. That is, the change threshold SLh is the deciding factor of the change, and the stable threshold SLa is the deciding factor of the stability.

Now, SLh = 1.0, SLa = 0.5, n _L = 4, f
_{(X o -X n, K)} = a _{{(X o -X n / K} )} 2,
The points B, C, and D in FIG. 8 will be described more concretely as an example.

First, when going back from point B to point A (past),
Going back with μ _n <SLa (stable), at the point A, n = n _L (stability continues 4 times). At this point, the result of step SA5 is NO, and the process proceeds to step SA7, where μ
_{n = {(X B -X A} ) / K)} 2, n = 4 is stored and saved.

In the case of going back from the point C, μ _n > SLh at the time point B going back once in the past. Therefore, step S
Pass A3 with YES, move to step SA7, and μ _{n
= {(X C -X B)} / K)} 2, n = 1 is stored saved.

In the case of going back from the point D, μ _n > SLh at the time point C going back six times. Therefore, step S
The process proceeds from A3 to step SA7, and μ _n = {(X _D −X
Save _C ) / K)} ² , n = 6.

Here, the change extraction method will be compared with the conventional method.
Conventionally, a method of obtaining a regression line by the least squares method and extracting a change and a method of obtaining a change rate between two points separated by a certain number of samplings have been mainly used.

First, according to the method of least squares, when the change is zigzag, since the slope of the regression line does not change so much, there is a possibility of overlooking a significant change. In addition, the number of calculations to find the regression line increases.

The method of obtaining the rate of change between two points may overlook this even if there is a significant change at intervals smaller than the two points. In other words, only the start and end points of the section have meaning. Moreover, if the space between two points is narrowed to one sampling, only an instantaneous change can be captured. Also, if a gradual change continues at the same rate of change, if there is no significance between the two points,
No significant change is found forever.

The method of the present embodiment eliminates all such drawbacks.

(2) Second stage In the second stage, the fuzzy inference unit 32 performs fuzzy inference based on the number of times n until the significant change is found obtained in the first stage, the instantaneous change degree ν, and the change degree μ _n. It is the stage of executing and outputting the change levels g _ν and g _{μn / n} .

Hereinafter, the operation of the second stage will be described with reference to the flowchart of FIG.

Step SB1 One change level Di (i = 1 to 9) is sequentially selected from the change levels D1 to D9 described above.

Step SB2 One fuzzy number corresponding to the selected change level Di is read from the fuzzy number table.

Step SB3 The average degree of change μ _n / n or the instantaneous degree of change ν is used as a variable on the horizontal axis of FIG. 3, the score at the fuzzy number selected above is obtained, and the fuzzy set of the conclusion part corresponding thereto is obtained.

Step SB4 The above fuzzy inference is performed for each of the change levels D1 to D9 to obtain nine fuzzy sets of the conclusion part. These fuzzy sets include, for example, how many points the change level D1 has, and how many points the D2 has, respectively.
It consists of the score distribution of.

The fuzzy set obtained in step SB5 is subjected to _{defuzzification} to obtain change levels g _{μn / n} and g _ν . As a specific example of the defuzzification, a method of obtaining the center of gravity of nine fuzzy sets is used.

Step SB6 The obtained g _{μn / n} and g _ν are stored.

(3) Third step The third step is a change in m sets output from the fuzzy inference unit 32-1~m level _g μν _/ n, g per _[nu, performs fuzzy inference, aspects change R (R1 to R13 of either)
And the certainty factor M _R are output. However,
As described above, similar items are averaged by the in-group addition type fuzzy inference unit 34.

The operation at this stage will be described below with reference to the flowchart of FIG.

The fuzzy inference units 33-1 to 33-m are provided with respective change levels g.
_{For μn / n} and g _ν , fuzzy inference as shown in FIGS. 4 (a) and 4 (b) is performed (steps SC1 to SC4),
The addition-type fuzzy inference unit 35 executes addition-type fuzzy inference as shown in FIGS. 4 to 6 (steps SC5 to SC6), and the defuzzification unit 36 executes defuzzification (step SC7). .

Step SC1 The following processing is repeated for each of the fuzzy numbers corresponding to the phase changes R1 to R13 of the 13th level.

Step SC2 Starting from the fuzzy number corresponding to the phase change R1, one phase change Ri is selected, and the fuzzy number corresponding to the selected phase change Ri is read from the fuzzy table.

Step SC3 The change levels g _{μn / n} , g _ν are applied to the fuzzy numbers as shown in the antecedent part of FIG. 4 (a), whereby the score of the phase change Ri is obtained, and the fuzzy part of the consequent part corresponding thereto is obtained. A set (corresponding to the areas Dz1 to DzN shown in FIG. 4B) is obtained.

The above process is repeated for each of the phase changes R1 to R13 in step SC413.

That is, the fuzzy inference units 33-1 to 33-m have a fuzzy set F = for each change level g _{μn / n} , g _ν .
Calculate (X1 / R1 + X1 / 2 + ... + Xc / Rc). When these are written as F _{μn /} and F _ν , respectively, there are m of them respectively. However, c represents the number of phase changes R, and c = 13 here.

For each of the m fuzzy sets F _{μn / n} and F _ν obtained in step SC5, the change phases R1 to R13 are added. Figure 5,
FIG. 6 shows this state. For example, the phase change R1 is added to a fuzzy set of the engine speed, a fuzzy set of the turbine speed, a fuzzy set of the supply pressure, ... A fuzzy set of the exhaust temperature. The same applies to other changes in the situation.

Step SC6 The addition result is divided by the number of detection items and averaged. In this way, a set of conclusion parts as shown in FIG. 4 (c) or FIG. 6 (e) is obtained.

Step SC7 Defuzzification is performed by means such as taking the center of gravity of the set of conclusion parts. As a result, aspects change R and its confidence M _R is obtained. In this case, the use of defuzzification rules has already been described. Further, the feedback information from the current situation state confirmation unit 62 is also used.

Operation of phase position understanding unit 40 Just like the operation of the third phase of the phase change understanding unit 30 described above, at which position of the phase positions P1 to P9 and with what degree of certainty the current operating state is located. Find out.

In this case, each fuzzy inference unit 41-i has nine fuzzy numbers (fourth fuzzy number) corresponding to the number J (= 9) of position positions.
For each of (Fz1 to FzN: N = 9 in FIG. 10A), the same process as above is executed, and the fuzzy set I =
(W1 / P1 + W2 / P2 + ... + WJ / PJ) (area Dz in FIG. 4 (b)
1 to DzN) is output. Thus, m fuzzy sets I are obtained. The in-group addition-type fuzzy inference unit 42 performs addition-type fuzzy inference for similar items in the m pieces of fuzzy sets I. Further, the addition-type fuzzy inference unit 43 performs addition and averaging processing for each of the nine positions of the fuzzy set, and the fuzzy set L = (Z1 / P1 + Z2
/ 2 + …… + ZJ / PJ) (Fuzzy set Gz1 in Fig. 4 (c))
(Corresponding to GzN) is output.

The defuzzification unit 44 uses the fuzzy set L
By operations such as taking the center of gravity of, (either P1 to P9) aspects position P running de-fuzzy Verification and outputs and its confidence M _P. It was already mentioned that the defuzzification rule was applied here as well.

Operation of Transition History Pattern Understanding Unit 50 The current situation understanding unit 52 understands the current situation by applying the following rules to the situation change R, the situation position P, and the confidences M _R and M _P thereof.

If the previous phase position P is in a stopped state (long-term stop P1 or short-term stop P2) and the current phase change R is rapidly rising R12 or where it is rapidly rising R13, it is determined to be "engine start".

Three minutes after the start is "immediately after the start".

The engine is considered to have warmed up after starting and stabilizing under a considerable load. For example, the current phase change R = R7 (stable), the certainty factor M _R thereof is a set value (for example, 90%) or more, and the average of the past ten position positions P is the set value (for example, P).
5 = 50%) or more: ・ When the current phase P satisfies all the conditions of a set value (for example, P5 = 50%) or more, it is determined that "the engine has warmed up".

When the above conditions are not satisfied, it is determined that "the engine is not warm".

Phase position P = P1, that is, if the long-term stop state continues for a set value (for example, past 5 times), “long-term stop (P
1) ”.

When the engine speed and the turbine speed suddenly drop and the score of the stop (fuzzy numbers corresponding to P1 and P2) in each fuzzy set I is equal to or more than a set value (for example, 80%), the stop state is determined. To do. That is, the current phase position P = P2 (short-term stop), the engine and the turbine speed suddenly drop one time before, and the condition that the score of P2 in the fuzzy set I of the current engine and the turbine speed is the set value or more. If the above is satisfied, it is determined that “currently stopped”.

If the previous time is “currently stopped” and the current phase position P indicates P2 (short-term stop), it is determined to be “just stopped”.

The near-term understanding unit 51 and the general understanding unit 53 perform the operations as described above.

Operation of Present Phase Status Confirmation Unit 60 The dynamic status confirmation unit 61 performs processing according to the following system check rules.

When the engine starts, the engine speed and turbine speed rapidly increase. Therefore, if these changes are not detected at the start, it is determined that the detector is abnormal.

If the load rises sharply, the turbine speed and supply pressure will follow quickly. Therefore, the phase change R sharply rises up to the set value (for example, 3) times, and the current turbine rotation speed and supply pressure change levels g _{μn / n} are sharply higher (corresponding to the change levels D8 and D9). If not, it is determined that the detector is abnormal.

If the phase change R is stable (R = 7) and the highest position of the phase position P and the fuzzy set I of the engine speed or the turbine speed has a difference of two levels or more, either It is judged that the detector is abnormal.

If the item indicating the temperature is within a set value (for example, 5 times) and both a rapid rise and a rapid fall occur, it is determined that the detector is abnormal.

The current phase state confirmation unit 62 performs processing according to the following meta rules.

When the phase change R becomes stable (R = 7), the fuzzy number table for obtaining the change level g _ν is changed according to the phase position P at that time. That is, the fuzzy number table in the fuzzy inference unit 32 is rewritten by the feedback information from the current situation state confirmation unit 62 to the fuzzy inference unit 32.

When it is determined that the detector is faulty, the addition-type fuzzy inference units 35 and 43 and the defuzzification unit 36, so that the addition-type fuzzy inference is performed excluding the detected items.
Issue a command to 44.

Thus, when an abnormality occurs in the detector, the data from the detector is removed and the fuzzy inference and the defuzzification are retried. Although this may reduce confidence to some extent, it enables correct decisions to be made.

Example 2 In Example 1 described above, the pattern of detected values was understood along the time axis. The data group in this case was time-series data stored for each data sampling. That is, when the element numbers are assigned to the sampling data group, the order of the element numbers is arranged in the time series from the present to the past, and therefore data processing inevitably becomes processing along the time axis. Is.

In other words, the data processing method centered on the above-mentioned fuzzy inference can be said to be a method for performing pattern understanding for the element number of a given data group by searching for element values in the order of element numbers. Therefore, it is configured by an algorithm that is independent of time, and when the element number corresponds to frequency, it becomes possible to understand that it corresponds to the frequency axis, the element number corresponds to age, and that value corresponds to the number of employees. If it corresponds to, it is possible to understand the personnel composition pattern of the company, and if the element number is the model and the value is the sales record, the pattern of the sales record by model can be understood.

In the second embodiment, the present invention is applied to the frequency analysis of the engine noise data in order to show the understanding of the pattern corresponding to the frequency axis, and has the configuration shown in FIG.

In the figure, the analog noise signal output from the sound level meter 110 is supplied to the spectrum analyzer 111. The spectrum analyzer 111 is FFT (Fourier transform)
Is used to perform frequency analysis of the noise signal at octave or 1/3 octave intervals and output the sound pressure level in each frequency band. This analog output is the A / D converter 11
After being converted into a digital signal in 2, the data is supplied to the data processing unit 113 having the same configuration as the device shown in FIG. The data processing unit 113 performs the following processing and supplies the processing result to the knowledge base 114 and the printer 115. The data recorder 116 records noise data in the form of a digital signal, and the noise data is frequency-analyzed by the spectrum analyzer 111 and then supplied to the data processing unit 113.

Exhaust sound, engine sound, and supercharger sound are typical noises of the engine, and these have a frequency distribution as shown in FIG.

In FIG. 14, (a) is exhaust sound, (b) is engine sound,
(C) shows the frequency distribution of the supercharger sound, and the horizontal axis of these graphs shows the frequency and the vertical axis shows the sound pressure level (dB).

From these figures, the following characteristic parts are understood.

The exhaust sound has a large low frequency component and has peak frequencies 1 and 2 in the low frequency region. These frequencies correspond to ignition frequencies and are represented by the following equations.

1 = (rpm × n) / (60 × 2) 2 = 2 × 1 (where rpm is the engine speed and n is the number of cylinders) The engine sound shows a flat characteristic as a whole, and the outstanding peak is Absent.

The supercharger sound has a sharp peak in the high frequency range, which is
There is _TB . _TB = TB _rpm × N / 60 (where TB _rpm is the turbine speed and N is the number of blades of the blower) Note that this peak _TB must be lowered in order to lower the noise level.

In the conventional noise analysis, the collected noise data is frequency-analyzed, the result is output as a graph to a CRT or a plotter, and an expert looks at the graph and recognizes its characteristics and makes sense at the same time.

Therefore, a large amount of manpower and time are required to analyze a large amount of data. A large amount of data is numbered, but until the end of the analysis, it is not possible to know what kind of data the data is and where. For example, if you want to extract only the supercharger sound data,
I don't know until all the data is checked by humans.

Therefore, if the pattern on the frequency axis can be understood and associated with the data number, the automation of the check can be realized and great labor saving can be achieved.

The second embodiment has been made under such consideration, and the method thereof will be described below.

1. Rotation speed rpm, cylinder speed n, turbine speed T
B, input the number of blower blades N. Then, the following processing is performed in order to calculate the peaks 1 and 2 in the low frequency region of the exhaust noise and the sharp peaks _TB in the high frequency region of the supercharger noise.

2. The spectrum analyzer 111 provides frequency data (see FIG. 14) including an element number corresponding to a frequency and a value corresponding to the element number, that is, a frequency obtained by frequency analysis and noise intensity at that frequency.

3. Regarding the frequency data as shown in FIG. 14, the allowable change value K (see FIG. 8) in consideration of the detection system and the measurement itself, the change threshold value SLh, the stable threshold value SLa, the maximum number of investigations n _max , and the stable time The investigation termination number n _L is set.

4. The change extraction unit 31 of the aspect change understanding unit 30 of the first embodiment
Similar to (first stage), the change in the sound pressure level of the frequency data along the frequency axis is captured. That is, starting from the first element number, the instantaneous variation ν and the average variation μ _n / n of the sound pressure level are sequentially obtained for each element number.

5. Corresponding to changes in the sound pressure level of the frequency data, change levels D1 to DN (super sudden drop, sudden drop, descending, falling, stable, rising, rising, sudden rising, super sudden rising, etc.) are set and these changes are also made. The fuzzy numbers Fz1 to FzN corresponding to the levels D1 to DN are set in advance in the format shown in FIG. Then, similarly to the fuzzy inference unit 32 (second stage) of the phase change understanding unit 30, a significant change in the sound pressure level obtained above (instantaneous change degree ν
And the average degree of change μ _n / n) are applied to the fuzzy numbers Fz1 to FzN, and the fuzzy inference shown in FIG. 3 is performed to determine the instantaneous change level g _ν of the sound pressure and the average change level g _{μn / n.}
Is calculated and these are stored and stored along the frequency axis. In the second embodiment, only the individual data is present and there is no overall system, so the third stage processing of the first embodiment is not performed. As described above, a change in sound pressure level along the frequency axis, that is, a phase change can be obtained.

6. Next, the phase position of the sound pressure level (dB) of the frequency data is obtained. First, the phase positions P1 to PM are determined corresponding to each sound pressure level. For example, 40 dB to 120 d
B is divided at 5 dB intervals, and each division is divided into position positions P1 to P1.
PM. Further, a fuzzy number is set in association with each of the phase positions P1 to PM, and it is calculated which phase position the sound pressure level of each frequency data enters and with what degree of certainty, and the phase position P and its Output the certainty factor M _P.

7. By tracking the phase change and the position of the phase in the order of element numbers, the position of the phase, that is, the position of the sound pressure level with respect to the frequency axis, and the position of the phase, that is, the change of the sound pressure level are understood.

8. Based on the characteristic data described below, it is discriminated whether the frequency data is exhaust sound, engine sound, supercharger sound, or other sound.

9. For example, when a person wants to extract the exhaust sound, only the data discriminated from the exhaust sound is retrieved and output.

The above-mentioned characteristic data will be described.

In the conventional pattern recognition, the judgment is made by matching with the standard pattern, so that the pattern may be erroneously judged due to the influence of the non-characteristic part.

This will be described with reference to FIG. As shown in FIG. 15 (a), when the standard data CR and the actual data CA are matched, the characteristic part is the peak part, but an accurate judgment cannot be made due to the influence of the parts before and after the peak part. There are cases. For this reason, only a characteristic region is extracted and matching is performed on that portion.

In this case, as shown in FIG. 15 (b), when the peak value of the standard data CR fluctuates within a certain width range, a plurality of standard data having slightly different peak values are prepared, and the actual data is prepared. Matching with CA must be repeated many times to make a determination. Therefore, when the variation range of the peak value is large, the number of times and time required for matching become enormous.

On the other hand, the method according to the second embodiment uses, for example, a variable (frequency) and a phase change (change of sound pressure level) and a position (sound pressure level) that are changed by the variable (frequency).
It can be described as "having one peak in the range of frequencies (variables) a and b and sound pressure levels (positional positions) Ia to Ib". Therefore, it is possible to determine the type of noise by understanding the change in sound pressure level due to the phase change, detecting the peak accordingly, and identifying the position of the phase.

Although the second embodiment applies the present invention to the noise analysis, it goes without saying that the present invention can be applied not only to the noise analysis but also to other chart analyses. In this case, an addition-type fuzzy inference unit or a fuzzy parameter setting unit may be provided as necessary to perform addition-type fuzzy inference or change the fuzzy parameters.

[Effects of the Invention] As described above, the present invention can bring the following effects.

Since the position of a phase and the change of a phase are grasped by using various kinds of detection item data, and the judgments such as failure diagnosis are made by integrating them, a highly reliable judgment that does not depend on one detector is possible. It can be carried out. Further, it is possible to identify the detector in which the abnormality has occurred.

By storing the above-mentioned position and position change in time series, it is possible to search the change history of the position. Therefore, the dynamic characteristics of the operating system can be grasped. Further, if a system check rule based on normal operation is incorporated in advance, it is possible to judge whether the detector is abnormal or the state transition history is correct.

Since the phase state transition history of the self system can be grasped, the self-excited oscillation (hunting) and the state that causes it can be captured, and the control for preventing the hunting can be performed.

The internal parameters of the operating system are set to certain conditions (for example, the values of the wattmeter and horsepower meter are completely reliable and the engine is completely normal), the detected data is statistically processed, and the processed data is processed. The characteristics of individual institutions can be learned by automatically adjusting and changing other internal data based on the. For example, it is possible to automatically adjust or change a preset fuzzy number by feedback.

The data to be processed is not limited to data that changes along the time axis. For example, it can be similarly applied to data that changes along the frequency axis. That is, it can be said that the present invention is a method of understanding a pattern of detected data by applying a knowledge rule.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment in which the method of the present invention is applied to the monitoring and control of an engine, and FIG. 2 (a) is a block diagram showing a knowledge engineering system and an interface portion between the knowledge engineering system and the first embodiment. FIG. 2 (b) is a functional block diagram of the first embodiment, FIGS. 3 to 7 are diagrams for explaining a fuzzy inference method, and FIG. 8 is an example of detection data. FIG. 9 is a diagram for explaining the operation of the stages, FIG. 9 is a transition diagram showing the position of each phase, and FIGS. 10 to 12 are flowcharts showing the operations of the first to third stages of the first embodiment, respectively. FIG. 13 is a block diagram showing the configuration of Embodiment 2 of the present invention, and FIG. 14 is a noise characteristic diagram according to Embodiment 2 of the present invention.
FIG. 5 is a diagram for explaining a conventional noise processing method. 1 ... Detector, 2 ... Data collection unit, 30 ... Phase change understanding unit 31 ... Change extraction unit, 32, 33, 41 ... Fuzzy inference unit, 34, 42 ... In-group addition type fuzzy inference unit , 35, 43 ... Addition-type fuzzy inference section, 36, 44 ... Defuzzification section, 40 ... Phase position understanding section, 50 ... Phase state transition history pattern understanding section, 60 ... Current phase state confirmation section 70 ... Accumulation unit 110 ... Sound level meter, 111 ... Spectrum analyzer, 112 ... A / D converter, 113 ... Data processing unit, 114 ... Knowledge base, 115 ... Printer, 116 ... Data recorder.

Claims (8)

[Claims] 1. A plurality of types of detection data obtained from an object are processed based on an independent variable corresponding to each detection data, such as a time when each detection data was obtained, a frequency corresponding to each detection data, and the like. A phase estimation method for estimating a phase faced by the object, wherein a plurality of fuzzy sets regarding the degree of change of the detection data for each type of detection data, and a plurality of fuzzy sets regarding the change of the phase, A rule corresponding to each type of the detection data, in which each fuzzy set regarding the degree of change of the detection data is associated with one of a plurality of fuzzy sets regarding the change in the phase, is defined in advance. , Comparing the amount of change in the value of the detection data caused by the change in the value of the independent variable for each type of detection data with a preset value. By extracting a significant change of the detection data by, for each type of detection data, a process of obtaining a fuzzy set and its certainty factor corresponding to the significant change of the detection data, and each fuzzy for the change of the aspect For each set, regarding the fuzzy set of changes in the relevant phase, by computing the arithmetic mean of the certainty factors of each fuzzy set of significant changes in the detected data of each type that is associated with the fuzzy set by the rule And a certainty factor of the certainty factor, which is used to estimate the change of the certain aspect and the certainty factor thereof. 2. The aspect estimation method according to claim 1, further comprising a step of classifying change degrees corresponding to similar detection items into groups and performing the addition-type fuzzy inference within each group. A method for estimating the situation. 3. The aspect estimation method according to claim 1, wherein the detected data is subjected to fuzzy inference to infer a current aspect position. 4. A phase estimation method according to claim 3, wherein a transition state of a phase position associated with a transition of the independent variables such as time and frequency is obtained. 5. The method for estimating a situation according to claim 4, wherein each detected data is investigated by checking the current situation position,
A phase estimation method characterized by extracting abnormal detection data. 6. The aspect estimating method according to claim 4, wherein a fuzzy set used for the fuzzy inference and the additive fuzzy inference is changed according to a current position of the aspect. 7. The aspect estimating method according to claim 4, wherein the aspect change, the aspect position and the transition state are output to a knowledge engineering system. 8. A plurality of detectors for detecting an operating state of the device, detecting a plurality of situation states that change corresponding to a detection amount of each detector, and detecting the plurality of situation states based on the situation states. In a failure diagnosis device that performs failure diagnosis of the detector, a change detection unit that compares the degree of change of the detection amount with a preset value and extracts a significant change, and (a) each of the detectors defined in advance A plurality of fuzzy sets for the degree of change of the detected amount, (b) a plurality of fuzzy sets for the change of the aspect, and (c) rules corresponding to each type of the detected amount, and the degree of change of the detected amount. For each detection amount, a fuzzy set corresponding to a significant change in the detection amount and its certainty factor are obtained based on a rule that associates each fuzzy set with respect to one of a plurality of fuzzy sets regarding the change in the situation. For each fuzzy set regarding the change of the aspect, the aspect is calculated by calculating the arithmetic mean of the certainty factors of the fuzzy sets of the significant changes of the detection amounts associated with the fuzzy set by the rule. To the fuzzy set of change of the addition type fuzzy inference means for executing the addition type fuzzy inference, the result of the addition type fuzzy inference and the part of the degree of change that did not receive the addition type fuzzy inference And a means for performing fuzzy inference, wherein the failure diagnosis device estimates a failure of the device and the detector based on an output of the means for applying the fuzzy inference.