JPH0572638B2 - - Google Patents

Description

[Detailed description of the invention]

[Industrial Application Field] The present invention relates to a system abnormality diagnosis method that estimates the cause of an abnormality using qualitative data and quantitative data obtained when an abnormality occurs in the system, and particularly relates to a system abnormality diagnosis method that estimates the cause of an abnormality using qualitative data and quantitative data obtained when an abnormality occurs in the system. This invention relates to a system abnormality diagnosis method that calculates each final probability of occurrence from basic probabilities using Dempster's combination rule. [Prior Art] For example, when an abnormality occurs in a small system such as a rotating machine such as a motor, a shaft, a bearing, a gear, a coupling, etc. as shown in FIG. ,
The operator estimates the cause of the abnormality based on qualitative data that relies on the five human senses, such as tone color and vibration form, and quantitative data that can be measured physically, such as sound pressure level, vibration level, vibration frequency, temperature, and voltage. As a means of estimating abnormality cause items from various feature values such as qualitative data and quantitative data obtained when a system abnormality occurs, a method using a [fault tree] as shown in FIG. 11 has conventionally been adopted. However, with this estimation method, if there are many features, it is necessary to go through many combinations of ANDs, ORs, etc. before estimating the final abnormality cause item, which requires a lot of time and effort. I needed it. Furthermore, when a plurality of feature quantities exist, it is not possible to add weight information indicating superiority or inferiority to each feature quantity for estimation. In order to eliminate such inconvenience,
As shown in Publication No. 94018, each feature X obtained when a system abnormality is considered to be independent, and the amount of data of each feature X normalized to a value of 0 to 1 is P, and each feature By setting the weighting coefficient W by which the quantity X influences each cause item F, the probability of occurrence PF of one cause item F is determined by the linear combination equation shown in equation (1). PF=W1P1+W2P2+ …(1) In this way, the probability of occurrence of one cause item F is PF
is calculated by the sum of all weighted feature values X, so for all cause items F,
If formula (1) is calculated, the probability of occurrence is the highest for the entire system.
Items F that cause high PF can be identified. [Problems to be Solved by the Invention] However, even in the system abnormality diagnosis method that calculates the probability of occurrence PF for each cause item F using the linear combination equation (1), the following problems still need to be solved: I had a problem like this. In other words, as shown in equation (1), the probability of occurrence PF of each cause item F is expressed by the sum of each feature quantity, so when the number of feature quantity X settings increases, the value of a certain few feature quantity becomes The probability of occurrence of a specific cause item when all feature values are averaged is higher than the probability of occurrence of a specific cause item when the values of many other features are small. The inconvenience of increasing the size occurs. That is, when the values of the feature amounts are averaged, the apparent probability of occurrence increases, and the probability of misdiagnosis occurring increases. Conversely, the cause of this misdiagnosis is that, for example, when determining that an abnormality has occurred in a specific cause item, a specific feature that should have a large value does not obtain a large value, and other This is because the occurrence probability increases as a result of a large number of feature quantities that should be small values being gathered together. The present invention was made in view of the above circumstances, and its purpose is to use Dempster's combination rule from each basic probability to estimate each cause item independently obtained through matrix processing. By calculating the final probability of occurrence for each cause item, the procedure for calculating each basic probability can be simplified, and only meaningful items can be extracted, greatly increasing the accuracy of the probability of occurrence calculated for each cause item. The purpose of the present invention is to provide a method for diagnosing system abnormalities that can be improved. [Means for solving the problem] In the system abnormality diagnosis method of the present invention,
means for creating a matrix of relationships between multiple pieces of qualitative data indicating system abnormalities and multiple cause items;
A stage display means for displaying each cause item corresponding to each qualitative data in the order of importance in this matrix, a means for forming a matrix of relationships between a plurality of quantitative data indicating system abnormality and a plurality of cause items; and a stage display means for displaying each cause item corresponding to each quantitative data in the matrix in stages, and in response to externally input qualitative data, each cause is displayed in stages corresponding to the relevant qualitative data in the matrix. Each basic probability of the occurrence of each cause item is calculated from the item, and each basic probability of the occurrence of each cause item is calculated from each graded cause item corresponding to the corresponding quantitative data in the matrix based on the quantitative value of the externally input quantitative data. The probability is calculated, and the final probability of occurrence is calculated for each cause item using Dempster's combination rule from each basic probability corresponding to the calculated qualitative data and quantitative data. I am trying to display the probability. [Operation] According to the system abnormality diagnosis method configured in this way, each qualitative data indicating an abnormality in the system and each cause item are displayed in a matrix, and the intersection of each qualitative data and each cause item in the matrix is displayed. Displays the level of importance at which the corresponding cause item can be presumed to be the cause of the abnormality if the qualitative data occurs. Similarly, levels of importance are displayed at the intersections in the matrix between quantitative data displayed in a matrix and each cause item. Then, when one piece of qualitative data is input from the outside, each basic probability of occurrence of each cause item corresponding to this qualitative data is obtained. Similarly, when a piece of quantitative data is input from the outside, each basic probability of each cause item corresponding to this quantitative data is calculated based on the quantitative value of this quantitative data. That is, the basic probability for each cause item is determined for each qualitative data and each quantitative data. Next, the final probability of occurrence for each same cause item is calculated from the independently calculated basic probability of each cause item for each data item using the Dempster combination rule. [Example] An example of the present invention will be described below with reference to the drawings. First, Dempster's combination rule will be explained. As is well known, in 1967 Dempster and Shafer developed the fundamental probability m{A} and the lower bound probability P _L to compensate for the subjective uncertainty that is ignored in Bayes' linear probability theory. ，
We proposed a new probability theory that defines the rising probability P _U. An important law in probability theory is Dempster's combination rule, which integrates basic probabilities inferred from independent evidence. That is, let m1 and m2 be the basic probabilities obtained based on independent evidence,
Let {A1i}, {A2j} (i, j = 0, 1, 2, ...) be the respective focus elements (sets), then these m1,
New basic probability M{Ak} combining the basic probabilities of m2
is calculated using equation (2).

[ ] The numerator of the formula means assigning the product of fundamental probabilities to the set product {Ak} of {A1i} and {A2j}. The denominator is a term for normalizing the whole in order to exclude the empty set φ when the above-mentioned intersection set {Ak} becomes the empty set φ. Also, {A0} is the entire set, {Ai} (i=1, 2,
3...) is its subset, then the basic probability m
{Ai} (i=0, 1, 2...) takes a value of 0 to 1, and satisfies the following conditions: m={φ}=0, 〓 ^Aj ⊆ _Ap m{Ai}=1. In addition, the lower bound probability P _L {Ai} is calculated using the basic probability (3)
It is determined by the formula. P _L {Ai}= 〓 ^Aj ⊆ _Ai m{Aj} (3) Similarly, the upper bound probability P _U {Ai} is obtained by equation (4). P _U {Ai}=1− 〓 ^Aj ⊆ _Ai m{Aj} (4) That is, the lower bound probability P _L {Ai} is the sum of the basic probabilities confined in the subset {Ai}, and the upper bound probability P _U
{Ai} is the sum of the basic probabilities that there is a possibility that the subset {Ai} may even be included. Then, the difference between the upper bound probability P _U and the lower bound probability P _L [P _U
−P _L ] indicates the certainty (degree of confidence) of the corresponding probability. That is, taking an example of abnormality diagnosis, the smaller this value is, the more accurate the calculated probability of occurrence for one cause item is, and the larger this value is, the more uncertain the probability of occurrence is. The system abnormality diagnosis method of the embodiment will be explained based on the basic matters explained above. FIG. 1 is a block diagram for embodying a method for diagnosing an abnormality in a system according to an embodiment. In this embodiment, a case will be described in which an abnormality diagnosis is performed for the system of a rotating machine shown in FIG. A plurality of types of qualitative data 2 and a plurality of types of quantitative data 3 obtained from the system 1 are inputted to a basic probability calculation unit 5 via a data input unit 4 . An external storage section 6 formed of, for example, a magnetic disk is connected to the basic probability calculation section 5. This external storage unit 6 includes a matrix data memory 7 that stores the relationship between each qualitative data and each cause item shown in FIG. 3 in a matrix form, and the relationship between each quantitative data and each cause item shown in FIG. 4. A matrix data memory 8 is formed to store the data in matrix form. The basic probability calculation unit 5 calculates each input qualitative data 2.
And for each quantitative data 3 and each matrix data memory 7, 8, for each data 2, 3, the basic probability for each corresponding combination set {Ai} of each cause item.
Calculate mi{Ai}. Each calculated basic probability mi
{Ai} is calculated by the Dempster combination rule calculation unit 9 to calculate the final probability Mk {Ak} for each combination set {Ai} of each cause item.
is calculated. Furthermore, the following upper and lower bound probability calculation unit 1
0, the lower bound probability P _L and the upper bound probability P _U are calculated, and are graphically displayed for each cause item on the display section 11 formed of a CRT display tube or the like, as shown in FIG. 5, for example. The basic probability calculation section 5, the Dempster combination rule calculation section 9, and the upper and lower bound probability calculation section 10 are configured by, for example, one computer. Next, details of the operation of each part will be explained step by step. First, as shown in FIG. 3, the matrix data memory 7 formed in the external storage unit 6 stores seven types of abnormal noises (for example, B1 to B7) that are considered to be output from the system 1 when an abnormality occurs. ), B8 and
For a total of 64 types of abnormal phenomena (qualitative data) from B1 to B64, such as the two types of vibration (2) of B9..., 19 types of cause items are arranged, from unbalance of A1 to oil failure of A19. has been done. Then, the operator specifies one of the qualitative data from B1 to B7 as the type of abnormal noise. Similarly, the type of vibration is B8 or B9.
Specify one qualitative data. In this way, the operator specifies the required number of qualitative data among the 64 qualitative data. Three levels of marks, ◎, ○, and ×, are stored for the cause items corresponding to each qualitative data. ◎ indicates that there is a very high possibility that the corresponding cause item is the cause of the abnormality when this qualitative data is specified, ○ indicates that there is a possibility that the corresponding cause item is the cause of the abnormality, × indicates that it is applicable Indicates that there is almost no possibility that the cause item is the cause of the abnormality. Note that the above assignments of ◎, ○, and × are made according to empirical rules. Therefore, it is also possible to change each allocation as necessary. Next, a method of calculating each basic probability mi {Ai} for a set {Ai} of combinations of each cause item using the matrix data memory 7 in which stages are assigned in this way will be explained. The basic probability calculation unit 5 has a sixth
As shown in the figure, an allocation memory MA is stored that shows the allocation of probabilities corresponding to each stage symbol. In other words, for example, if the cause item is only ◎, {Ai}
is assigned 0.25, 0.25 is assigned to the set of cause items {Ai} that includes ◎ and ○, 0.25 is assigned to the set of cause items {Ai} other than ×, and 0.25 is assigned to the set of cause items {Ai} that includes all (Θ) cause items { Assign 0.25 to A0}. Table 1 is a matrix representation of the cause items A1, A2, A3, and A6 and the qualitative data B1, B2, B7, B8, and B9 extracted from the matrix data memory 7 to explain the calculation method. In this case, each basic probability mi {Ai} is calculated as shown in Table 2 for each combination set {Ai} of each cause item A1 to A6. For example, when the operator selects and specifies the qualitative data of B1, the basic probability m1 {A3} of the cause item of A3 of ◎ is assigned a probability of 0.25 in the allocated memory MA, and m1 {A3} = 0.25. Similarly with A3
A probability of 0.25 is assigned to the set of ◎, ○ items {A3, A6} consisting of A6, m2 {A3, A6} = 0.25
becomes. Furthermore, a set other than × {A2, A3, A6}
is assigned a probability of 0.25, m3{A2, A3,
A6}=0.25. Also, since 0.25 is assigned to all sets, m4 {A1, A2, A3, A6} =
It becomes 0.25. In addition, if the operator selects and sets steady rotation of B8, a probability of 0.25 is assigned to the set of ◎ items, so m1 {A1, A2, A6} = 0.25
Since the remaining three items of the allocated memory MA are allocated to the set of all items {A1, A2, A3, A6}, m2 {A1, A2, A3, A6} = 0.75.

【table】

【table】

[Table] Next, in the matrix data memory 8 of Fig. 4 corresponding to quantitative data, for example, from B65 to
For each quantitative data up to B89, cause items A1 to A19 are arranged in the same manner as in the matrix data memory 7 described above. ○ indicates that there is a positive relationship between the data value of the qualitative data and the probability of occurrence of the corresponding cause item, and × indicates that there is a negative relationship. and,
The data value of each quantitative data is normalized to a value of 0 to 1 and input, but in order to show the degree of influence of each quantitative data on all cause items, the maximum abnormality probability Pm allowed for each quantitative data is Furthermore, a data value I0 indicating a 0% probability and a data value I100 indicating a 100% probability are set for the abnormality probability Pr for each quantitative data. Furthermore, the allocated memory in Figure 6
As shown in MB, the set of ○ cause items is Pr×
A probability of Pm is assigned, and a probability of (1-Pr)×Pm is assigned to the set of cause items of ×. Then, based on the conversion characteristics shown in Figure 2, the data value of the input quantitative data is converted to the basic probability mi
Convert to {Ai}. That is, the conversion characteristics for cause items marked with a circle mark are shown by solid lines, and the conversion characteristics for cause items marked with an x mark are shown by broken lines. In the case of ○ sign, if the data value exceeds I100, the basic probability becomes Pm as mentioned above,
If I0 is not reached, the basic probability becomes 0. For example, as shown in Table 3, considering a matrix consisting of two quantitative data of B66 and B68 and cause items of A1, A2, A3, and A6, if the normalized data value of B66 is 0.9, then A1 ,A2,A3○
Since the conversion characteristic for the cause item is shown by a solid line, it exceeds I100 (=0.9), so as shown in Table 4, the basic probability m1 {A1, A2, A3} is Pm = 0.7
becomes. Also, the basic probability m2 for all cause items
{A1, A2, A3, A6} is 1-Pm, which is 0.3.

【table】

[Table] As explained above, when qualitative data 2 and quantitative data 3 are input from the outside, matrix data memories 7 and 8 and allocation memory MA,
Combination set of cause items A1 to A19 corresponding to each data B1 to B89 input using the MB value {Ai}
The basic probability mi{Ai} (i=1, 2, 3...) for each is obtained. Next, the operation of the computer comprising the basic probability calculation section 5, Dempster connection rule calculation section 9, etc. will be explained. This computer has a final A1 in its memory.
Various memories shown in FIG. 6 are formed to obtain the lower bound probability P _L (Al) and upper bound probability P _U (Al) for each cause item of ~A19. Then, the computer calculates each final lower bound probability P _L (Al) and each upper bound probability P _U (Al) according to the main routine shown in FIG. The flowchart starts and the allocated memory in Figure 6
When the predetermined initialization process such as clearing each memory other than MA and MB is completed, the set of combinations of each cause item calculated in response to each data 2 and 3 input {Ai}
Basic probability memory that stores the basic probability mi{Ai} for each
Set the MD storage area W to 1. Then, when data input occurs in S1, qualitative data 2 is generated in S2.
or quantitative data 3. For qualitative data 2, use the matrix data memory 7 and the allocated memory MA to calculate the basic probability mi{Ai} for each set of cause items corresponding to the relevant qualitative data (abnormal phenomenon), as shown in Figure 6. Basic probability memory
Store in storage area W of MD. In addition, if the quantitative data is 3 in S2, this quantitative data is normalized, and the basic probability mi {Ai} for each set of cause items is calculated from the data value of this quantitative data (abnormal measurement items). Store in storage area W of probability memory MD. Basic probability memory for one input data
When the storage process for the MD is completed, 1 is added to the storage area W, and the process returns to S1 to wait for the next data input. When all data input is completed in S3, the storage area W is set to the leading value 1, and each basic probability mi{Ai} of the storage area W is stored as mj{A2j} in the MJ buffer in Figure 6. Each basic probability of area (W+1)
Store mi{Ai} in the MI buffer as mi{A1i}. Each basic probability for MI battle and MJ battle
When mi{A1i}, mj{A2j} are stored, in S4, the intermediate probability Mk{Ak} for each set {Ak} (k=1, 2, 3,...) of each cause item is calculated as described above (2) Calculated using Dempster's combination rule shown in the formula. FIG. 8 is a flowchart for calculating the intermediate probability Mk{Ak} of equation (2) from each basic probability mi{A1i}, mj{A2j} stored in the MI buffer and the MJ buffer. That is, after setting k to the initial value 1, S5
In addition to setting i=1 and j=1, the value SM1 for determining the numerator of equation (2) and the value SM2 for determining the denominator are set to 0. If i and j do not exceed the maximum values p and q in S6 and S7, each basic probability mi{A1i}, mj is calculated from the MI buffer and MJ buffer.
Read {A2j}, set {A1i} and set {A2j}
Calculate the intersection set {Az} with In S8, if the product set {Az} matches the set {Ak} of intermediate probabilities Mk {Ak} to be determined, the product of each basic probability [mi {A1i} × mj {A2j}] is added to SM1. . In addition, if the product set {Ak} is the empty set φ in S9, the product of the above basic probabilities [mi{A1i}×mj
Add {A2j}]. After that, in S10, j is increased by 1, and then the process returns to S7. When j exceeds the maximum value q in S7, j is set to 1, 1 is added to i, and the process returns to S6. When i exceeds the maximum value p in S6, the intermediate probability Mk {Ak} for one set {Ak} is calculated in S11. Mk{Ak}=SM1/[1-SM2] The calculated intermediate probability Mk{Ak} is stored in the kth area of the final probability memory MK shown in FIG.
After that, k is increased by 1. If k after increasing does not exceed the maximum value r in S12, return to S5,
The process of calculating the intermediate probability Mk {Ak} for the set {Ak} with respect to the increased k is started. When k exceeds the maximum value r in S12, each basic probability mi stored in the MI buffer and MJ buffer
Since the process of storing all the intermediate probabilities Mk {Ak} of {A1i}, mj {A2j} into the final probability memory MK has been completed, the process returns to S13 in FIG. 7 and 2 is added to the storage area W. Then, in S14, each basic probability mi{Ai} in the storage area W of the basic probability memory MD is transferred to the MI buffer mi
Store as {A1i}. Furthermore, each intermediate probability Mk{Ak} of the final probability memory MK is stored in the MJ buffer as each basic probability mj{A2j}. Then, clear the final probability memory MK. After that, in S15, each intermediate probability Mk {Ak} for the MI buffer and MJ buffer explained in FIG.
is calculated and stored in the final probability memory MK. Then, in S16, the storage area W of the basic probability memory MD is increased by 1, and the process returns to S14.
Each basic probability mi{Ai} is stored in the MI buffer. Then, in S17, when the storage area W exceeds the maximum value Wm, all the basic probabilities mi of the basic probability memory MD
Since the joint operation processing for {Ai} has been completed, each intermediate probability Mk{Ak} for each set {Ak} stored in the final probability memory MK at this point becomes each final probability Mk{Ak} to be determined. However, S18
Then, the process of calculating the lower bound probability P _L (Al) and upper bound probability P _U (Al) for each cause item in equations (3) and (4) described above is executed. That is, in FIG. 9, when the flowchart is started, if l=1 and k=1 and it is confirmed in S19 and S20 that the maximum values lm and r are not exceeded, respectively, the number from the final probability memory MK is Read the final probability Mk{Ak}. At S21, the set {Ak}
If it is a subset of Al, add its final probability Mk{Ak} to the P _L value in FIG. Further, in S21a, if the set {Ak} is a subset of the complementary set of Al, its final probability Mk{Ak} is added to the P _U value in FIG. Thereafter, 1 is added to k in S22, and the k-th final probability Mk{Ak} after the addition is read out in S20. If k exceeds the maximum value r in S20, 1 in S23.
The value obtained by subtracting P _U from is the normal upper bound probability P _U (Al). Then, the obtained lower bound probability P _L (Al) and upper bound probability P _U (Al) are stored in the lth area of the upper bound probability memory MP shown in FIG. However, A1
Lower bound probability P _L (Al) and upper bound probability P _U for the l-th cause item among each cause item from to A19
(Al) was calculated. Then, in S24, 1 is added to l, and the process of calculating the lower bound probability P _L (Al) and upper bound probability P _U (Al) for the l-th Al after the addition is started. When l exceeds the maximum value lm (19 in this example) in S19, the lower bound probability P _L (Al) and the upper bound probability P _U (Al)
After completing the calculation process, the process returns to S25 in FIG. 7, and as shown in FIG. 5, for each cause item, the cause item name 21, lower bound probability _P The probability P _U 23 is displayed graphically. and,
A solid line 24 connects the lower bound probability P _L 22 and the upper bound probability P _U 23. The closer the position of this solid line 24 is to 0, the lower the probability that the corresponding cause item has occurred, and the closer it is to 1, the higher the probability that the corresponding cause item has occurred. The length of this solid line 24 indicates the reliability of the probability that each of the calculated cause items has occurred. That is, since the true probability of occurrence exists within the solid line 24,
The shorter the solid line 24 is, the higher the reliability of the probability of occurrence is, and the longer the solid line 24 is, the lower the reliability of the probability of occurrence is. With the diagnostic method configured in this way, when an abnormality occurs in the system 1, the operator uses his five senses to identify the abnormal phenomenon (qualitative data),
For example, each data in the matrix data memory 7 is displayed on the display section 11, and B1 to B64 are displayed for each type.
Select and input each qualitative data of System 1.
Enter the abnormal measurement items (quantitative data) obtained from physical measurements. Then, basic probability mi {Ai} is calculated for each combination set {Ai} of each cause item A1 to A19 for each qualitative and qualitative data. Furthermore, the final probability Mk {Ak} is calculated for each combination set {Ai} of each cause item when all data are integrated using Dempster's combination rule. Finally, the lower bound probability P _L for each cause item,
The upper bound probability P _U is calculated and displayed as a graph on the display unit 11 as shown in FIG. In this way, since Dempster's combination rule is used in the process of calculating the final probability for each cause item, it is possible to exclude cases where an empty set is created in the process of adding up each basic probability. In other words, when determining a certain cause item as an abnormality, if a basic probability does not occur in data that should have a significant basic probability, this basic probability will not be calculated even if other data has a large basic probability. You will not be able to enter it. Therefore, the calculated final probability is the sum of only meaningful basic probabilities, so it is a value close to the probability that can actually occur. As a result, as shown in Figure 5, each of the other causal items that are not considered to be true abnormal factors are:
Even if the number of qualitative data and quantitative data increases, the calculated probability will be almost zero. By the way, Figure 12 is a graphical representation of the abnormality occurrence probability for each cause item calculated using the conventional linear combination method shown in equation (1) when the same data as in Figure 5 is input. It is. As is clear from a comparison with FIG. 5, it can be seen that the difference in probability between each cause item is small and the probability of misdiagnosis occurring is high. However, in the diagnostic method according to the present invention,
Since only truly meaningful cause items are displayed as high probabilities, the diagnostic accuracy for the causes of system abnormalities can be greatly improved. In addition, the reliability of the calculated probability is determined by the solid line 2 connecting the lower bound probability P _L and the upper bound probability P _U as shown in Figure 5.
Since it is displayed in a length of 4, it is possible to add the reliability of the probability to the judgment material even if the probabilities are the same, and the diagnostic accuracy can be further improved. Also, for input data, see Figures 3 and 4.
In each matrix representation format shown in the figure, the causal relationship between each input data and the cause item is indicated by a graded code such as ◎, ○, ×, etc., and the basic probability mi{Ai} for each set {Ai} is automatically calculated from this graded code. Since the calculation is done based on the calculation, subsequent calculation processing can be performed regardless of whether it is qualitative or quantitative. Therefore, the types and number of settings for qualitative data and quantitative data can be easily changed. In addition, as mentioned above, each basic probability is automatically calculated by reading the stage code of ◎, ○, ×, so you can easily change the level of ◎, ○, × and change the setting position as necessary. Can be executed. [Effects of the Invention] As explained above, in the system abnormality diagnosis method of the present invention, Dempster's combination rule is used to calculate each cause from each basic probability for estimating each cause item independently obtained through matrix processing. The final probability of occurrence for each cause item is calculated and displayed. Therefore, the procedure for calculating each basic probability can be simplified, only meaningful items can be extracted, the accuracy of the calculated occurrence probability for each cause item can be greatly improved, and the diagnostic accuracy of the system can be greatly improved.

[Brief explanation of the drawing]

1 to 9 show a method for diagnosing an abnormality in a system according to an embodiment of the present invention.
The figure is a block diagram showing the schematic configuration, Figure 2 is a diagram showing an example of conversion characteristics for determining the basic probability, and Figure 3 is a diagram showing an example of conversion characteristics for determining the basic probability.
4 and 6 are diagrams showing an example of the main memory of the storage section, FIG. 5 is a diagram showing an example of display contents on the display section, and FIGS. 7 to 9 are flowcharts showing the operation. Figure 10 is a perspective view showing a rotating machine as a general system, Figure 11 is a diagram showing the conventional procedure for estimating abnormality cause items, and Figure 12 is a diagram showing each cause item calculated using the conventional method. FIG. 3 is a diagram showing the probability of occurrence. 1... System, 2... Qualitative data, 3... Quantitative data, 5... Basic probability calculation section, 6... External storage section, 7, 8... Matrix data memory, 9
. . . Dempster combination rule calculation section, 10 . . . Upper and lower bound probability calculation section, 11 . . . Display section, 22 .

Claims (1)

[Scope of Claims] 1. Means for forming a matrix of relationships between a plurality of qualitative data indicating system abnormality and a plurality of cause items, and within this matrix, each cause item corresponding to each qualitative data is displayed in stages in order of importance. means for displaying a stage in which the relationship between the plurality of quantitative data indicating an abnormality in the system and the plurality of cause items is displayed in a matrix; In response to externally input qualitative data, the basic probability of occurrence of each cause item is calculated from each corresponding graded cause item in the corresponding qualitative data of the matrix, and Based on the quantitative values of the input quantitative data, each basic probability of occurrence of each cause item is calculated from each graded cause item corresponding to the corresponding quantitative data in the matrix, and the calculated qualitative data and quantitative data are An abnormality diagnosis for a system, characterized in that a final probability of occurrence is calculated for each cause item using Dempster's combination rule from each corresponding basic probability, and the calculated final probability of occurrence for each cause item is displayed. Method.