DIAGNOSTIC DEVICE

• US 20100131800A1
• Filed: 03/21/2008
• Published: 05/27/2010
• Est. Priority Date: 03/22/2007
• Status: Active Grant

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Satou, Akinori, Nakata, Takayuki, Fujimaki, Ryohei, Tsukahara, Hidenori
• Granted Patent: US 8,370,108 B2
• Time in Patent Office: Days
• US Class Current: 714/37
• CPC Class Codes: G05B 23/0281 Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A diagnostic device which detects a fault and estimates its cause on the basis of the degree of variation of each attribute of data containing a plurality of attributes obtained by measuring a diagnosis subject. A diagnosis subject variation degree pattern generation means (110) calculates the degree of variation of each attribute of the diagnosis subject data containing the attributes obtained by measuring the diagnosis subject and generates a diagnosis subject variation degree pattern which is a combination of the degrees of variation of the attributes. A reference variation degree pattern storage device (150) stores a reference variation degree pattern composed of the degrees of variation of the attributes in association with the type of the fault and the diagnosis event of the type of the cause. A variation degree pattern diagnostic means (120) performs pattern matching of the diagnosis subject variation degree pattern and the reference variation degree pattern in the reference variation degree pattern storage device and diagnoses the diagnosis subject.

Claims

1-42. -42. (canceled)

43. A diagnostic device comprising:

a criterion change degree pattern memory that holds in store a criterion change degree pattern in association with an event-to-be-diagnosed;

said criterion change degree pattern being formed of a pattern of values of the degree of change of each attribute of data, including a plurality of attributes, measured from an object being diagnosed; and

change degree pattern diagnosis means for receiving a diagnostic object change degree pattern to make diagnosis of said object being diagnosed based on the received diagnostic object change degree pattern and said criterion change degree pattern stored in said criterion change degree pattern memory;

said diagnostic object change degree pattern being formed of a pattern of values of the degree of change of each attribute of data, including a plurality of attributes, of said object being diagnosed, measured from said object being diagnosed.

44. The diagnostic device according to claim 43, further comprising:

diagnostic object change degree pattern generation means for calculating the degree of change of each attribute of said data, including a plurality of attributes, of said object being diagnosed, measured from said object being diagnosed, to generate said diagnostic object change degree pattern;

said diagnostic object change degree pattern being a combination of values of the degree of change of said attributes calculated.

45. The diagnostic device according to claim 43, wherein said change degree pattern diagnosis means effects pattern matching between said diagnostic object change degree pattern and said criterion change degree patterns stored in said criterion change degree pattern memory to make diagnosis on an object being diagnosed based on a result of said pattern matching;

said diagnostic device further comprising;

\attribute cardinality degree determining means for reading out the criterion change degree patterns from said criterion change degree pattern memory and for determining the degree of cardinality of each attribute in the criterion change degree pattern associated with an event-to-be-diagnosed;

said attribute cardinality degree determining means sorting, from one event-to-be-diagnosed to another, the criterion change degree patterns belonging to the event in question into a pattern of interest and a training pattern;

said pattern of interest being the criterion change degree pattern belonging to said event in question and said training pattern being the criterion change degree pattern belonging to remaining events;

said attribute cardinality degree determining means determining the degree of cardinality of each attribute of said criterion change degree pattern belonging to said event in question using a predetermined evaluation function.

46. The diagnostic device according to claim 44, further comprising:

an attribute cardinality degree memory that holds in store the degree of cardinality of said attributes in association with the event-to-be-diagnosed;

said change degree pattern diagnosis means in effecting pattern matching between said criterion change degree pattern associated with said event-to-be-diagnosed and the diagnostic object change degree pattern taking into account the degree of cardinality of said attributes stored in said attribute cardinality degree memory in association with said event-to-be-diagnosed of said criterion change degree pattern.

47. The diagnostic device according to claim 45, wherein said criterion change degree pattern memory holds in store the criterion change degree pattern, in accordance with a layered structure of the event-to-be-diagnosed, from one layer to another, and wherein said change degree pattern diagnosis means effects diagnosis by pattern matching using the criterion change degree pattern of an upper layer;

said change degree pattern diagnosis means effecting pattern matching using the criterion change degree pattern of a lower layer based on the outcome of diagnosis for the upper layer.

48. The diagnostic device according to claim 45, wherein said criterion change degree pattern memory holds in store a criterion change degree pattern associated with a fault-related event;

and wherein said change degree pattern diagnosis means outputs a fault-related event associated with the criterion change degree pattern matched to said diagnostic object change degree pattern as the outcome of the diagnosis.

49. The diagnostic device according to claim 48, wherein said criterion change degree pattern memory further holds in store a criterion change degree pattern associated with a normal event;

and wherein

said change degree pattern diagnosis means outputs, in connection with the diagnostic object change degree pattern not matched to any criterion change degree pattern associated with a normal event or to any criterion change degree pattern associated with said fault-related event, an outcome of the diagnosis to the effect that the diagnostic object change degree pattern is of an unknown fault.

50. The diagnostic device according to claim 43, wherein said change degree pattern diagnosis means learns a set of diagnostic rules for making diagnosis of the state of said object being diagnosed from a change degree pattern, using the criterion change degree pattern stored in said criterion change degree pattern memory;

said change degree pattern diagnosis means applying the diagnostic rule learned to said diagnostic object change degree pattern to diagnose said object being diagnosed.

51. The diagnostic device according to claim 50, further comprising:

an attribute cardinality degree memory that holds in store the degree of cardinality of said attributes in association with the event-to-be-diagnosed;

said change degree pattern diagnosis means in learning said diagnostic rules using the criterion change degree pattern associated with the event-to-be-diagnosed taking into account the degree of cardinality stored in said attribute cardinality degree memory in association with the event-to-be-diagnosed of said criterion change degree pattern.

52. The diagnostic device according to claim 50, further comprising:

a layered structure memory that holds in store data of a layered structure for the event-to-be-diagnosed;

said change degree pattern diagnosis means learning, for each layer stored in said memory of the layered structure, a diagnostic rule for diagnosing the event belonging to said layer.

53. The diagnostic device according to claim 50, wherein said change degree pattern diagnosis means learns a diagnostic rule for detecting a normal state in addition to a diagnostic rule for detecting a fault;

and wherein, for said diagnostic object change degree pattern that has not been diagnosed to be normal under a diagnostic rule to detect a normal state, and that has not been detected to be fault-related under a diagnostic rule to detect a fault, at the time of said diagnosis, said change degree pattern diagnosis means outputs an outcome of diagnosis to the effect that said diagnostic object change degree pattern is of an unknown fault.

54. The diagnostic device according to claim 43, further comprising:

a criterion change degree pattern generation device that generates a criterion change degree pattern to be stored in said criterion change degree pattern memory;

said criterion change degree pattern generation device including

a learning data memory that holds in store time-series data including a plurality of attributes, measured from said object being diagnosed, in association with the event;

change degree pattern generation means for reading out said time-series data from said learning data memory, calculating the degree of change with time of said attributes included in said time-series data, from one time-series data to another, and for generating a pattern corresponding to a combination of values of the degree of change of said attributes as a change degree pattern associated with the event of said time-series data; and

change degree pattern outputting means for outputting the change degree pattern generated by said change degree pattern generation means.

55. The diagnostic device according to claim 46, further comprising:

an attribute cardinality degree generation device that generates the attribute cardinality degree stored in said attribute cardinality degree memory;

said attribute cardinality degree generation device including;

a criterion change degree pattern memory that holds in store, in association with an event-to-be-diagnosed, a criterion change degree pattern formed by a pattern of values of the degree of change of each attribute of data, including a plurality of attributes, measured from said object being diagnosed;

said attribute cardinality degree determining means; and

attribute cardinality degree outputting means for outputting the event-by-event cardinality degree of said attributes determined by said attribute cardinality degree determining means.

56. The diagnostic device according to claim 47, further comprising:

an attribute cardinality degree generation device that generates the attribute cardinality degree stored in said attribute cardinality degree memory;

said attribute cardinality degree generation device including;

a criterion change degree pattern memory that holds in store, in association with an event-to-be-diagnosed, a criterion change degree pattern formed by a pattern of values of the degree of change of each attribute of data, including a plurality of attributes, measured from said object being diagnosed;

a layered structure memory that holds in store data of a layered structure of an event-to-be-diagnosed;

attribute cardinality degree determining means for reading out data of the layered structure from said layered structure memory and reading out said criterion change degree pattern from said criterion change degree pattern memory;

said attribute cardinality degree determining means determining the degree of cardinality of said attributes in said criterion change degree pattern associated with said event for each layer as the object-to-be-diagnosed, from one event to another; and

attribute cardinality degree outputting means for outputting the event-by-event cardinality degree of each of said attributes determined by said attribute cardinality degree determining means.

84. A system for diagnosis in which a plurality of diagnostic devices according to claim 43 are interconnected over a network for communication each other.

57. A method for diagnosis comprising:

a first step of change degree pattern diagnosis means receiving a diagnostic object change degree pattern formed of a pattern of values of the degree of change of each attribute of data, including a plurality of attributes, measured from an object being diagnosed; and

a second step of said change degree pattern diagnosis means reading out said criterion change degree pattern from said criterion change degree pattern memory and effecting diagnosis of said object being diagnosed based on said received diagnostic object change degree pattern and said criterion change degree pattern read out;

said criterion change degree pattern being formed of a pattern of values of the degree of change of each attribute of data, including a plurality of attributes, measured from said object being diagnosed being so far held in store in said criterion change degree pattern memory in association with the event-to-be-diagnosed.

58. The method for diagnosis according to claim 57, further comprising:

a third step of diagnostic object change degree pattern generation means for calculating the degree of change of each attribute of data, including a plurality of attributes, of said object being diagnosed measured from said object being diagnosed to generate said diagnostic object change degree pattern;

said diagnostic object change degree pattern being a combination of the values of the degree of change of said attributes calculated.

59. The method for diagnosis according to claim 57, wherein said change degree pattern diagnosis means effects pattern matching between said diagnostic object change degree pattern and said criterion change degree pattern stored in said criterion change degree pattern memory to make diagnosis for said object being diagnosed based on the result of said pattern matching.

60. The method for diagnosis according to claim 58, wherein said change degree pattern diagnosis means in effecting pattern matching between a criterion change degree pattern associated with an event-to-be-diagnosed and the diagnostic object change degree pattern reads out the degree of cardinality of said attributes of said criterion change degree pattern, associated with the event-to-be-diagnosed, from said attribute cardinality degree memory

to effect pattern matching in consideration of the degree of cardinality of said attributes read out;

said degree of cardinality of said attributes being so far held in store by said attribute cardinality degree memory in association with said event-to-be-diagnosed.

61. The method for diagnosis according to claim 59, wherein said criterion change degree pattern memory holds in store, in accordance with a layered structure of an event-to-be-diagnosed, said criterion change degree patterns from one layer to another;

and wherein said change degree pattern diagnosis means effects diagnosis by pattern matching using the criterion change degree patterns of an upper layer;

said change degree pattern diagnosis means effecting pattern matching using the criterion change degree pattern of a lower layer based on the outcome of diagnosis for the upper layer.

62. The method for diagnosis according to claim 59, wherein said criterion change degree pattern memory holds in store a criterion change degree pattern associated with a fault-related event;

and wherein said change degree pattern diagnosis means outputs a fault-related event associated with the criterion change degree pattern matched to said diagnostic object change degree pattern as an outcome of the diagnosis.

63. The method for diagnosis according to claim 62, wherein said criterion change degree pattern memory further holds in store a criterion change degree pattern associated with a normal event;

and wherein said change degree pattern diagnosis means outputs, in connection with the diagnostic object change degree pattern not matched to any criterion change degree pattern associated with a normal event or to any criterion change degree pattern associated with said fault-related event, an outcome of the diagnosis to the effect that the diagnostic object change degree pattern is of an unknown fault.

64. The method for diagnosis according to claim 57, wherein said change degree pattern diagnosis means learns a set of diagnostic rules for diagnosing the state of said object-to-be-diagnosed from a change degree pattern, using the criterion change degree pattern stored in said criterion change degree pattern memory;

said change degree pattern diagnosis means applying a diagnostic rule learned to said diagnostic object change degree pattern to diagnose said object being diagnosed.

65. The method for diagnosis according to claim 64, wherein, in learning said diagnostic rule(s) using a criterion change degree pattern associated with a certain event-to-be-diagnosed, said change degree pattern diagnosis means reads out the degree of cardinality of said attributes of said criterion change degree pattern, associated with the event-to-be-diagnosed, from an attribute cardinality degree memory;

said degree of cardinality of said attributes being so far held in store by said attribute cardinality degree memory in association with the event-to-be-diagnosed;

said change degree pattern diagnosis means learning said diagnostic rules in consideration of the degree of cardinality of said attributes read out.

66. The method for diagnosis according to claim 64, wherein said change degree pattern diagnosis means learns, for each layer stored in said layered structure memory, the diagnostic rule used for diagnosing an event-to-be-diagnosed belonging to said layer;

said layered structure memory holding in store data of a layered structure of said event.

67. The method for diagnostic according to claim 64, wherein said change degree pattern diagnosis means learns a diagnostic rule for detecting a normal state in addition to a diagnostic rule for detecting a fault;

and wherein, for said diagnostic object change degree pattern that has not been diagnosed to be normal under the diagnostic rule to detect a normal state, and that has not been detected to be fault-related under the diagnostic rule to detect a fault, at the time of said diagnosis, said change degree pattern diagnosis means outputs an outcome of diagnosis to the effect that said diagnostic object change degree pattern is of an unknown fault.

68. A criterion change degree pattern generation device comprising:

a learning data memory that holds in store, in association with an event, time-series data including a plurality of attributes measured from an object being diagnosed;

change degree pattern generation means for reading out said time-series data from said learning data memory, calculating the degree of change with time of said attributes included in said time-series data, from one time-series data to another, and for generating a pattern of a combination of values of the degree of change of said attributes, as a change degree pattern associated with said event of said time-series data; and

change degree pattern outputting means for outputting the change degree pattern generated by said change degree pattern generation means.

69. An attribute cardinality degree generation device comprising:

a criterion change degree pattern memory that holds in store a plurality of criterion change degree patterns, each of which is formed of a pattern of the degree of change of each attribute of data, including a plurality of attributes, measured from an object being diagnosed, in association with an event-to-be-diagnosed;

attribute cardinality degree determining means for reading out said criterion change degree patterns from said criterion change degree pattern memory and for determining the degree of cardinality of said attributes in said criterion change degree patterns associated with said event-to-be-diagnosed, from one event to another; and

said attribute cardinality degree determining means sorting, for each event-to-be-diagnosed, the criterion change degree patterns into a pattern of interest and a training pattern;

said pattern of interest being the criterion change degree pattern belonging to said event in question and said training pattern being the criterion change degree pattern belonging to remaining events;

said attribute cardinality degree determining means determining the degree of cardinality of each attribute of said criterion change degree pattern belonging to said event in question using a predetermined evaluation function.

71. The attribute cardinality degree generation device according to claim 69, wherein said attribute cardinality degree determining means sorts the criterion change degree pattern(s), belonging to an event-to-be-diagnosed, into a pattern of interest and a training pattern, from one event-to-be-diagnosed to another;

said pattern of interest being criterion change degree pattern belonging to an event in question and said training pattern being criterion change degree pattern belonging to other event(s);

said attribute cardinality degree determining means determining the degree of cardinality of the respective attributes of the criterion change degree pattern(s) belonging to said event in question, using a predetermined evaluation function.

72. The attribute cardinality degree generation device according to claim 71, wherein said evaluation function is formed of one or more of first to third evaluation functions;

said first evaluation function expressing the effect of determining the cardinality degree of each attribute of said pattern of interest so that the pattern of interest is globally not similar to the training pattern(s);

said second evaluation function expressing the effect of determining the cardinality degree of each attribute of said pattern of interest so that said pattern of interest is globally not similar to the training pattern that is similar to said pattern of interest;

said third evaluation function expressing the effect of determining the cardinality degree of each attribute of the pattern of interest so that patterns of interest are globally similar to one another.

70. An attribute cardinality degree generation device comprising:

a criterion change degree pattern memory that holds in store a criterion change degree pattern, formed of a pattern of the degree of change of each attribute of data, including a plurality of attributes, measured from an object being diagnosed, in association with an event-to-be-diagnosed;

attribute cardinality degree determining means for reading out data of a layered structure from said layered structure memory, reading out said criterion change degree pattern from said criterion change degree pattern memory, and for determining the degree of cardinality of said attributes in said criterion change degree pattern associated with said event-to-be-diagnosed, from one event to another; and

attribute cardinality degree outputting means for outputting the cardinality degree of said attributes, determined by said attribute cardinality degree determining means, from one event to another.

85. The attribute cardinality degree generation device according to claim 70, wherein said evaluation function is formed of one or more of first, second and third evaluation functions;

said first evaluation function expressing an effect of determining the cardinality degree of said attributes of said patterns of interest so that the patterns of interest are globally not similar to the training patterns;

said second evaluation function expressing an effect of determining the cardinality degree of said attributes of said patterns of interest so that said patterns of interest are globally not similar to those training patterns that are similar to said patterns of interest;

said third evaluation function expressing an effect of determining the cardinality degree of said attributes of said patterns of interest so that said patterns of interest are globally similar to one another.

73. A program for allowing a computer to operate as change degree pattern diagnosis means;

said computer having a criterion change degree pattern memory that holds in store a criterion change degree pattern in association with an event-to-be-diagnosed;

said criterion change degree pattern being formed of a pattern of values of the degree of change of each attribute of data, including a plurality of attributes, measured from an object being diagnosed;

said change degree pattern diagnosis means receiving a diagnostic object change degree pattern to make diagnosis of said object being diagnosed based on the received diagnostic object change degree pattern and said criterion change degree pattern stored in said criterion change degree pattern memory;

said diagnostic object change degree pattern being a pattern of values of the degree of change of each attribute of data, including a plurality of attributes, of said object being diagnosed as measured from said object being diagnosed;

said program allowing execution of a process in which criterion change degree patterns are read out from said criterion change degree pattern memory, and in which attribute cardinality degree determining means determine the degree of cardinality of respective attributes in the criterion change degree patterns associated with events-to-be-diagnosed, sort the criterion change degree patterns belonging to events in question as a pattern of interest and sort the criterion change degree pattern(s) belonging to other events as training pattern(s), from one event-to-be-diagnosed to another;

said attribute cardinality degree determining means determining the degree of cardinality of respective attributes of the criterion change degree patterns belonging to the event in question using a predetermined evaluation function.

74. The program according to claim 73, for further allowing said computer to operate as a diagnostic object change degree pattern generation means for calculating the degree of change of each attribute of data, including a plurality of attributes, of said object being diagnosed measured from said object being diagnosed to generate said diagnostic object change degree pattern;

said diagnostic object change degree pattern being a combination of values of the degree of change of said attributes calculated.

75. The program according to claim 73, wherein

said change degree pattern diagnosis means effects pattern matching between said diagnostic object change degree pattern and said criterion change degree pattern stored in said criterion change degree pattern memory to make diagnosis on said object being diagnosed based on an outcome of said pattern matching.

76. The program according to claim 74, wherein

said computer comprises an attribute cardinality degree memory that holds in store the degree of cardinality of said attributes in association with an event-to-be-diagnosed;

said change degree pattern diagnosis means in effecting pattern matching between said criterion change degree pattern associated with said event-to-be-diagnosed and the diagnostic object change degree pattern taking into account the degree of cardinality of said attributes stored in said attribute cardinality degree memory in association with said event-to-be-diagnosed of said criterion change degree pattern.

77. The program according to claim 75, wherein

said criterion change degree pattern memory holds in store the criterion change degree pattern, in accordance with a layered structure of an event-to-be-diagnosed, from one layer to another, and wherein said change degree pattern diagnosis means effects diagnosis by pattern matching using a criterion change degree pattern of an upper layer;

said change degree pattern diagnosis means effecting pattern matching using a criterion change degree pattern of a lower layer based on an outcome of diagnosis for the upper layer.

78. The program according to claim 75, wherein

said criterion change degree pattern memory holds in store a criterion change degree pattern associated with a fault-related event;

and wherein said change degree pattern diagnosis means outputs a fault-related event associated with the criterion change degree pattern matched to said diagnostic object change degree pattern as an outcome of the diagnosis.

79. The program according to claim 78, wherein

said criterion change degree pattern memory further holds in store a criterion change degree pattern associated with a normal event;

and wherein said change degree pattern diagnosis means outputs, in connection with the diagnostic object change degree pattern not matched to any criterion change degree pattern associated with a normal event or to any criterion change degree pattern associated with said fault-related event, an outcome of the diagnosis to the effect that the diagnostic object change degree pattern is of an unknown fault.

80. The program according to claim 73, wherein

said change degree pattern diagnosis means learns a set of diagnostic rules for making diagnosis of the state of said object being diagnosed from a change degree pattern, using the criterion change degree pattern stored in said criterion change degree pattern memory;

said change degree pattern diagnosis means applying a diagnostic rule learned to said diagnostic object change degree pattern to diagnose said object being diagnosed.

81. The program according to claim 80, wherein

said computer includes an attribute cardinality degree memory that holds in store the degree of cardinality of said attributes in association with an event-to-be-diagnosed;

said change degree pattern diagnosis means in learning said diagnostic rule using the criterion change degree pattern associated with an event-to-be-diagnosed taking into account the degree of cardinality stored in said attribute cardinality degree memory in association with the event-to-be-diagnosed of said criterion change degree pattern.

82. The program according to claim 80, wherein

said computer includes a layered structure memory for holding in store data of a layered structure for an event-to-be-diagnosed;

said change degree pattern diagnosis means learning, for each layer stored in said layered structure memory, a diagnostic rule for diagnosing the event belonging to said layer.

83. The program according to claim 80, wherein

said change degree pattern diagnosis means learns a diagnostic rule for detecting a normal state in addition to a diagnostic rule for detecting a fault;

and wherein, for said diagnostic object change degree pattern that has not been diagnosed to be normal under the diagnostic rule to detect a normal state, and that has not been detected to be fault-related under the diagnostic rule to detect a fault, at the time of said diagnosis, said change degree pattern diagnosis means outputs an outcome of diagnosis to the effect that said diagnostic object change degree pattern is of an unknown fault.