JP2004126641A - Causal relation model generation method and device, cause estimation method and device and data structure - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a causal relation model generation method and device for easily constructing a database having an easily understandable data structure for fault diagnosis, and acquiring a reliable clear diagnostic result. <P>SOLUTION: The causal relation model generation method is characterized in that event as an event under consideration or a cause event is expressed as inclusion relation between the constituting elements of an object to be observed associated with the above event and another constituting element of the object to be observed or as data having attribute wherein conceptual inclusion relation of the event is hierarchically expressed. The set of the results having, as elements, the events under consideration regarded as identical is made to correspond to the set of causes having the cause events of the respective elements of the set of the results as elements and the causal relation model of the object to be observed indicating a causal relation between the event under consideration and the cause event can be generated. <P>COPYRIGHT: (C)2004,JPO

Description

なる因果データの集合を生成する。
【００４８】
ところで、上記結果データｙｈ（∈Ｙ）が、実は別の結果を引き起こす原因データｘｋ（∈Ｘ）でもある場合もある。すなわち、現実の観測対象における、結果データは単に結果であるというよりも、実は中間的なデータであり、それが別の結果を引き起こす原因につながっている場合が想定される。
【００４９】
そこで、果因データ生成記憶部１２は、上記結果データｙｈ（すなわち、Ｊｈ・ｙｈ）（∈Ｙ）と上記原因データｘｋ（すなわち、Ｉｋ・ｘｋ）（∈Ｘ）とを対としてもつ果因データの集合Ｒを生成して、記憶する（ステップＳ２）。
【００５０】
【数２】

【００５１】
というように、結果データと原因データが対で、１つの結果データと別の結果を引き起こす原因との関連を示す。
【００５２】
以下、図８に示すような階層構造を有する観測対象の事象データを例にとり説明する。図８の「Ｄ１」〜「Ｄ６」はそれぞれ構成要素であり、構成要素Ｄ２は、構成要素Ｄ３を包含する構成となっている。例えば、現象ｙ１，ｙ２は、それぞれ構成要素Ｄ３に発生する現象であり、構成要素Ｄ３は、構成要素Ｄ２に包含される構成要素であって、構成要素Ｄ２には、現象ｘ３，ｘ４が発生する。
【００５３】
なお、上記各構成要素は、例えば、図７に示した自動車における燃料タンクやバッテリのような部品などに相当するものでもよいが、この場合に限らず、例えば、人為的に階層構造をわかり易くするために作成された構成要素であってもよい。例えば、図８において、現象ｙ１，ｙ３は、構成要素Ｄ２で発生する障害であり、構成要素Ｄ３は、現象ｙ１、ｙ３を１つに集約するために、ユーザにより作成された、クラスや属性などに対応する「障害」という構成要素であってもよい。
【００５４】
このように、本実施形態に係る構成要素は、基本的には、観測対象の階層構造を構成する部品などに対応するものであるが、この場合に限らず、上記のように、例えば、現象と概念的包含関係にある構成要素（例えば、上記のように、複数の現象や構成要素を１つに集約するために用いる便宜上作成された構成要素）であってもよい。
【００５５】
図９は、構成要素「Ｄ１」〜「Ｄ６」から構成される観測対象の構成例を模式的に示したものである。ここでは、各構成要素の名称「Ｄ１」〜「Ｄ６」を、そのまま各構成要素の符号Ｄ１〜Ｄ６として用いて説明する。
【００５６】
図８では、例えば、現象ｙ１に対応する事象データは、「Ｄ２・Ｄ３・ｙ１」あるいは、現象ｙ１は構成要素Ｄ３で発生することを明確にするために、「Ｄ２・（Ｄ３・ｙ１）」と表すことができる。また、現象ｘ５に対応する事象データは、「Ｄ４・ｘ５」と表すことができる。
【００５７】
次に、同一結果データ集合作成部１３は、上記結果データのなかで、実は同一の結果と判別できるものを要素として１つの「同じ結果」を表すあらたな集合、すなわち、同一結果データ集合（Ｙ１，Ｙ２，…）を生成する（ステップＳ３）。
【００５８】
どういう結果データを同一とみなすかという判別基準ないし判別手法は、同一結果データ集合作成部１３に予め備えられる（記憶されている）ものとする。最も簡単な例として、単純に判別基準だけを記憶しておき、それにしたがって同一集合を定めるものについて説明する。
【００５９】
【数３】

【００６０】
「どういう結果データを同一とみなすか」については、各結果データがどの集合に属するかを所定の基準にしたがって分別するものとする。任意の結果データｙｊ（すなわち、Ｊｈ・ｙｈ）（∈Ｙ）に関して、たとえば、図３に示すようなテーブル形式の分類基準（ｚ１、…ｚＬ）のテーブルを予め設けておく。
【００６１】
図３に示したような分類基準では、結果データは、最初から定められた集合のいずれかにかならず当てはまるが（ｙｊ（すなわち、Ｊｊ・ｙｊ）は、同一結果データ集合Ｙ０〜ＹＬのいずれかの要素となる）、あてはまらない場合もつくっておき、その場合はあてはまらない結果データを例えば１つずつ、ただ１つの要素とする集合Ｙ_Ｌ＋ｍ、（ｍ＝１、２、…）を生成してゆくものとしてよい。
【００６２】
なお、想像に難くないように、この部分については従来の分類手法の応用でよいので、上記手法に限定するものではない。
【００６３】
このようにして、所定の基準ないし方法にしたがって、同一とみなしてよい結果データをひとつの集合としてゆく。その際、同一とみなしていよい結果データのうち、例えば、図８に示したような階層構造上の包含関係から、１つに集約できる複数の結果データは、１つの集約された記述で表す。例えば、図８に示した事象データのうち、同一とみなしてよい結果データの集合（同一結果データ集合）Ｙ_１として、｛ｙ１、ｙ２、ｙｊ｝が得られたとする。これは正確には、｛Ｄ２・Ｄ３・ｙ１、Ｄ６・ｙ２、Ｄ６・ｙｊ｝であるが、この集合Ｙ_１の要素としての結果データのうち、「Ｄ６・ｙ２」、「Ｄ６・ｙｊ」は、ともに、構成要素Ｄ６に包含される結果データであり、構成要素Ｄ６に発生する現象は、「Ｄ６・ｙ２」と「Ｄ６・ｙｊ」のみである。従って、この場合、「Ｄ６・ｙ２」と「Ｄ６・ｙｊ」を表すために、これらを包含する事象データとして「Ｄ６」を用いる。すなわち、Ｙ_１＝｛Ｄ２・Ｄ３・ｙ１、Ｄ６｝となる。
【００６４】
同様にして、図８に示した事象データから、同一結果データ集合として、下に示すような集合Ｙ_１、Ｙ_２、Ｙ_ｋが得られたとする。
【００６５】
Ｙ_１＝｛ｙ１、ｙ２、ｙｊ｝＝｛Ｄ２・Ｄ３・ｙ１、Ｄ６｝
Ｙ_２＝｛ｙ１，ｙ３｝＝｛Ｄ３｝
Ｙ_ｋ＝｛ｙ４｝＝｛Ｄ５・ｙ４｝
部分因果システムモデル構成部１４は、上記同一結果データ集合作成部１３で作成された同一結果データ集合の要素である結果データを引き起こす原因データを要素とする集合を生成して対応づけを行なうとともに、さらに、上記果因データの対応づけもおこない、原因データと結果データを部分的な集合によって表すことで対象システムに関する推論を容易にするためのものである。
【００６６】
まず、同一結果データに関わる対応づけについて、図４を参照して説明する。
【００６７】
図４において、同一結果データ集合Ｙ１の要素｛ｙ１，ｙ２，ｙｊ｝は、因果データ生成記憶部１１に因果データとして記憶されているので、すでに原因データに、例えば、（ｘ２，ｙ１）、（ｘ４，ｙ２）、（ｘｉ，ｙｊ）∈Ｃとして対応づけられている。
【００６８】
これらの対応する原因データｘ２，ｘ４，ｘｉをまとめて１つの集合と考えて、同一結果原因データ集合Ｘ１（＝｛ｘ２，ｘ４，ｘｉ｝）とする（ステップＳ４）。このＸ１と元のＹ１との対応を写像としてｆ１と表す（ステップＳ５）。
【００６９】
【数４】

【００７０】
このようにして、あらゆる同一結果データ集合について、原因データ集合への対応づけを行うと、
【数５】

と表すことができる。
【００７１】
このようにして、以下に示すような同一結果原因データ集合Ｘ_１、Ｘ_２、Ｘ_ｋが得られたとする。
【００７２】
Ｘ_１＝｛ｘ２、ｘ４、ｘｉ｝＝｛Ｄ１・ｘ２、Ｄ２・ｘ４、Ｄ５・ｘｉ｝
Ｘ_２＝｛ｘ１、ｘ５、ｘ６｝＝｛Ｄ１・ｘ１、Ｄ４｝
Ｘ_ｋ＝｛ｘ３｝＝｛Ｄ２・ｘ３｝
ここで、ＸｋからＹｋへの対応を便宜上サブシステムと呼ぶことにすると、写像ｆは、サブシステムの逆写像（「逆サブシステム」）ということになる。また、このサブシステムのそれぞれは、ＸｋとＹｋとの対応として１対１対応となるように構成する。すなわち、逆サブシステムがなりたつように、同一結果原因データ集合を定めることにする。
【００７３】
次に、果因データの対応づけについて、図５を参照して説明する。
【００７４】
果因データ（ｙｈ，ｘｋ）∈Ｒ、ｙｈ（∈Ｙ）、ｘｋ（∈Ｘ）は、ある結果データｙｈが、即、別の原因データｘｋでもある場合を示す情報である。図５で、例えば、Ｙ１に属するｙ１、ｙｊはそれぞれ、ｘ１（∈Ｘ２）、ｘ３（∈Ｘｋ）に対応づけられる。また、Ｙｋに属するｙ４は、ｘ６（∈Ｘ２）に対応づけられる。
【００７５】
すなわち、果因データとして、
（ｙ１、ｘ１）＝（Ｄ３・ｙ１、Ｄ１・ｘ１）
（ｙ４、ｘ６）＝（Ｄ５・ｙ４、Ｄ４・ｘ６）
（ｙｊ、ｘ３）＝（Ｄ６・ｙｊ、Ｄ２・ｘ３）
が得られたとする。
【００７６】
このとき、それぞれの要素を介して、Ｙ１がＸ２、Ｘｋによって対応づけられるをことを写像としてｇで表す（ステップＳ６）。すなわち、
【数６】

果因データに関わる要素をもたない同一結果原因データ集合、たとえば、図５でＸ１の場合は、写像ｇによる対応集合はつぎのように空集合となる。
【００７７】
【数７】

【００７８】
なお、この写像は、多対応を許し、例えば、図５では、ｇ（Ｘ２）＝Ｙｋの対応も存在するから、出力は集合表現とし、
【数８】

と表す。
【００７９】
ここで、Ｙ^２はＹ１、Ｙｋを要素とするメタ集合として便宜上定義した。
【００８０】
なお、上記説明では、対応するものすべてを列挙したが、実際の観測結果に依存して、たとえば、Ｙｋが得られていない（観測結果に属するものが無い）場合には、上の例に示したように、Ｙ１だけの出力とするということでもよい。
【００８１】
このようにして部分因果システムモデル構成部１４では、因果データと果因データと同一結果データから、それらのデータの部分集合間の写像ｆ、ｇを構成する。
【００８２】
以上説明したように、図１の因果システムモデル構築装置は、図５に示したように、観測対象の因果データと果因データを、同一結果データに着目して部分集合（サブシステム）化し、観測対象の原因と結果との因果関係を（写像ｆ、ｇを用いて）部分集合間の対応関係で表した因果システムモデルを構築することにより、最初からすべての条件を考えて木構造などのデータ構造を構成する必要がなく、故障診断のためのデータベースの構築が容易に行える。
【００８３】
また、属性どうしの対応関係をデータ（因果データ、果因データ）としてもっているだけではなく、集合どうしの対応関係をあらかじめデータ化して、集合を定めるときに利用してもよい。
【００８４】
ところで、図５に示した、上記各事象データｘ１…ｘ６、ｘｉ、ｙ１…ｙ４、ｙｊは、それぞれ、図８に示すような階層構造を有する観測対象の事象データである。
【００８５】
また、図５は、観測対象の原因と結果との因果関係を、同一結果データ集合Ｙ_１、Ｙ_２、Ｙ_ｋと同一結果原因データ集合Ｘ_１、Ｘ_２、Ｘ_ｋとの間の対応関係で表した因果システムモデルであるが、図１０はこれを模式的に表したものである。
【００８６】
なお、図１０において、前記同様、同一結果データ集合Ｙ_１、Ｙ_２、Ｙ_ｋと同一結果原因データ集合Ｘ_１、Ｘ_２、Ｘ_ｋは以下のとおりである。
【００８７】
Ｙ_１＝｛ｙ１、ｙ２、ｙｊ｝＝｛Ｄ２・Ｄ３・ｙ１、Ｄ６｝
Ｙ_２＝｛ｙ１，ｙ３｝＝｛Ｄ３｝
Ｙ_ｋ＝｛ｙ４｝＝｛Ｄ５・ｙ４｝
Ｘ_１＝｛ｘ２、ｘ４、ｘｉ｝＝｛Ｄ１・ｘ２、Ｄ２・ｘ４、Ｄ５・ｘｉ｝
Ｘ_２＝｛ｘ１、ｘ５、ｘ６｝＝｛Ｄ１・ｘ１、Ｄ４｝
Ｘ_ｋ＝｛ｘ３｝＝｛Ｄ２・ｘ３｝
図１０では、原因としての集合から結果としての集合への方向を矢印で示している。すなわち、たとえば、集合Ｘ_１は集合Ｙ_１の原因であり、集合Ｙ_１はＸ_ｋと集合Ｘ_２の原因である。
【００８８】
各集合の要素である事象データの属性には、当該事象データに対応する現象の発生位置である構成要素が観測対象の階層構造に従って表されている。従って、この属性に基づき、各事象データに対応する現象の発生位置である構成要素が特定できる。すなわち、図９に示した観測対象の構成上に、図１０に示した因果関係を表す各集合の要素である事象データの発生位置を表すことができるのである。
【００８９】
図１１は、図１０に示した因果関係を表す各集合の要素である事象データの属性を基に、図９に示した観測対象の構成上に、各事象データの発生位置を表すことにより、上記各集合Ｙ_１、Ｙ_２、Ｙ_ｋ、Ｘ_１、Ｘ_２、Ｘ_ｋの発生位置を表したものである。
【００９０】
このように、事象データ自体が観測対象の階層構造に従って階層的に表現されているので、観測対象の構成上に、注目現象の発生位置や、当該注目現象と同一結果とみなされる現象の発生位置、当該注目現象の原因など、上記因果関係を容易に表示することができる。また、注目現象の発生位置は、当該発生位置に対応する構成要素をそのまま表示することができるのみならず、当該注目現象に対応する事象データの属性に記述されている構成要素を上流方向に辿っていくことにより、当該構成要素を包含するより上位階層にある構成要素で表示することもできる。
【００９１】
すなわち、観測対象としての装置、システム上に障害が発生した場合、その障害自体を上記のような事象データとして検知するため、当該事象データの属性を基に、例えば画面に表示されている観測対象の構成図上に、当該障害の発生位置を容易に表示することができるとともに、上記のように構築された因果システムモデルを基に、当該障害の原因や、当該障害を原因として発生した障害など、当該障害と互い因果関係にある現象が、その発生位置とともに、観測対象の構成図上に容易に表示することができるのである。
【００９２】
原因データおよび結果データについては、因果関係とは異なる属性を持つことも多く、各原因データ集合および結果データ集合との関係が、属性（クラス）との関係（観測対象の構造、特に階層構造）にもとづきながら表現できることを可能にした。これにより、文字表示だけではなく、図示して利用する情報システムについても因果システムモデル構築および原因推定装置構築が容易となる。
【００９３】
（第２の実施形態）
次に、図１の因果システムモデル構築装置にて構築された、例えば、図５に示したような因果システムモデルを用いた原因推定装置について説明する。
【００９４】
なお、ここでも、上記第１の実施形態と同様に、結果データや原因データといった現象データは、図８に示すような階層構造を有する観測対象の事象データであり、ここでは、前記同様、各事象データは、その属性を省略して、例えば、ｘ１…ｘ６、ｘｉ、ｙ１…ｙ４、ｙｊなどと表記している。
【００９５】
原因推定装置は、上記因果システムモデルに基づいて、観測対象の例えばシステムから観測される結果データが属する同一結果データ集合に対応する同一結果原因データ集合を順次検索を繰り返すことで、対応する原因データが属する集合を得るものである。
【００９６】
図１２は、原因推定装置の構成例を示したもので、観測データ認識部２１と、逆サブシステム検索部２２と、対応同一結果データ集合検索部２３、因果システムモデル記憶部２４とから構成されている。
【００９７】
以下、図１３に示すフローチャートを参照しながら、図１２の原因推定装置の構成と、各機能部の処理動作について説明する。
【００９８】
因果システムモデル記憶部２４には、図１の因果システムモデル構築装置で構築された、例えば、図５に示したような因果データと果因データから、同一結果データ集合と同一結果原因データ集合間の写像ｆ、ｇにより表される対応関係（観測対象の原因と結果との間の因果関係）が記憶されている。
【００９９】
観測データ認識部２１は、因果システムモデル記憶部２４に記憶されている、例えば、図５に示したような因果システムモデル（因果関係モデル）に基づき、例えば、図３に示したような分類基準に従って、観測で得られる結果データとしての観測データｙ０（∈Ｙ）が属する同一結果データ集合Ｙｒ⊂Ｙを検索する（ステップＳ１１〜ステップＳ１２）。ここで観測データとは、前述したように、観測対象の事象データである。
【０１００】
観測データ認識部２１の上記機能をＯ（）として写像表現すると、次のように表される。
【０１０１】
Ｙｒ＝Ｏ（ｙｏ）　　　　　　　　　　　　　　　　　　　　　　（１１）
例えば、図５に示すような因果システムモデルを参照して説明する。観測によって（観測データとして）、結果データｙ１が得られたとする。そして、図３に示したような分類基準によって、結果データｙ１が属する同一結果データ集合Ｙ２が出力されることを次式で表現することができる。
【０１０２】
Ｙ２＝Ｏ（ｙ１）　　　　　　　　　　　　　　　　　　　　　　（１２）
逆サブシステム検索部２２は、観測データ認識部２１で求めた同一結果データ集合Ｙｒに対応する同一結果原因データ集合を検索する（ステップＳ１３）。
【０１０３】
因果システムモデル記憶部２４には、対象システムのサブシステムの対応関係を表したデータが保持されている。すなわち、（７）式に示したような、部分集合ＸとＹがそれぞれ対応づけられて記憶されているものとする。
【０１０４】
このとき、（１１）式のように、観測データが属する同一結果データ集合が得られたとすると、それに対応する同一結果原因データ集合Ｘ２が検索される。すなわち、これは、次式で表現することができる。
【０１０５】
Ｘ２＝ｆ２（Ｙ２）　　　　　　　　　　　　　　　　　　　　　　（１３）対応同一結果データ集合検索部２３は、同一結果原因データ集合Ｘｒが上記果因データを介して対応する同一結果データ集合を検索する（ステップＳ１４〜ステップＳ１５）。
【０１０６】
例えば、図５において、上記例の場合、果因データ｛（ｙ１，ｘ１），（ｙ４，ｘ６）｝の対応にしたがって、（１０）式のように、同一結果原因データ集合Ｘ２には、同一結果データ集合Ｙ１、Ｙｋの２つが対応づけられている。
【０１０７】
図１３では、この２つの同一結果データ集合Ｙ１、ＹｋがＹｌに対応する。
【０１０８】
図１２に示すように、逆サブシステム検索部２２と対応同一結果データ集合検索部２３は、フィードバック構成となっている。すなわち、このような構成により、本発明の原因推定装置は、同一結果原因データ集合Ｘｒに対応付けられた同一結果データ集合Ｙｌについて（Ｙｌが複数存在するときは、そのそれそれについて）、新たに、逆サブシステム検索部２２で同一結果原因データ集合を検索する（ステップＳ１６、ステップＳ１３〜ステップＳ１４）、という処理が繰り返されるようになっている。
【０１０９】
ステップＳ１５で、逆サブシステム検索部２２で検索された、同一結果原因データ集合Ｘｒのそれぞれに、写像ｇで対応付けられた同一結果データ集合がないときは、ステップＳ１７へ進み、観測データｙ０について、逆サブシステム検索部２２で検索された全ての同一結果データ集合｛Ｙ_１、Ｙ_２、…｝と、対応同一結果データ集合検索部２３で検索された上記同一結果データ検索集合｛Ｙ_１、Ｙ_２、…｝のそれぞれに対応付けられている同一結果原因データ集合｛Ｘ_１、Ｘ_２、…｝を出力する。なお、同一結果原因データ集合｛Ｘ_１、Ｘ_２、…｝のみを出力するようにしてもよい。
【０１１０】
図１２の原因推定装置を入出力データの観点から説明すると、入力データとして、観測データｙ０が入力されると、当該観測データに対応して検索される同一結果データ集合と同一結果原因データ集合とを、推定結果として出力することができ、観測データがどのような状況にあるかを示すようになっている。すなわち、
【数９】

なお、（１４）式〜（１７）式では、ｆについては、サフィックスをつけないで記述したが、実際は引数に応じて対応するサフィックスを持つｆを選択するものとする。
【０１１１】
なお、各同一結果データ集合、各同一結果原因データ集合には、それぞれラベル名が与えられていて、このラベル名を出力するようにしてもよい。
【０１１２】
以上の説明では、観測データｙ０が１つの場合、すなわち、１つの観測事象（現象）の場合について説明したが、観測データはかならずしも１つとは限らない。たとえば、観測対象のシステムから２つの現象が同時に成り立つということを前提にしたい場合は、図１２に示した原因推定装置は、次のように処理する。
【０１１３】
【数１０】

【０１１４】
２つの現象が同時に成り立つという前提に従った場合、２つの観測データのうちの一方の観測データついての推定結果（逆サブシステム検索部２２で検索された全ての同一結果データ集合と、対応同一結果データ集合検索部２３で検索された、この同一結果データ検索集合のそれぞれに対応付けられている同一結果原因データ集合）と、他方の観測データについての推定結果との論理積が出力される。
【０１１５】
また、２つの現象のうちのいづれか一方が成り立つという前提に従った場合、２つの観測データのうちの一方の観測データついての推定結果（逆サブシステム検索部２２で検索された全ての同一結果データ集合と、対応同一結果データ集合検索部２３で検索された、この同一結果データ検索集合のそれぞれに対応付けられている同一結果原因データ集合）と、他方の観測データについての推定結果との論理和が出力される。
【０１１６】
上記第２の実施形態によれば、観測対象としての装置、システム上に障害が発生した場合、その障害自体を上記のような事象データ（ここでは、観測データに対応する）として検知するため、当該事象データの属性を基に、例えば画面に表示されている観測対象の構成図上に、当該障害の発生位置を容易に表示することができる。また、第１の実施形態において説明したように構築された因果システムモデルを基に検索された当該障害の原因や、当該障害を原因として発生する障害など、当該障害と互い因果関係にある現象が、その発生位置とともに、観測対象の構成図上に容易に表示することができるのである。
【０１１７】
（第３の実施形態）
ここでは、以上説明したような本発明の手法を従来と比較しながら、より具体的に説明する。
【０１１８】
図１４は、従来からある故障診断システムの知識データベースのデータ構造の一部を概略的に示したもので、故障結果を原因に結びつけるための木構造をなしている。すなわち、ある観測対象のシステムから観測された現象（症状）と、その症状が現れた工程・部位、要因、部品とに分類してラベル付けされた集合間で、各集合の要素を結果から原因に遡るように関連付けてなる木構造をなしている。
【０１１９】
図１４に示した故障推定木データでは、隣合う集合間で、その各要素について、主に因果関係にあると考えられるデータ間を矢印「（原因）→（結果）」で結び木構造を構成している。
【０１２０】
一方、本発明の手法によれば、結果データと原因データとの間の対応付けは、分類された隣合う集合間に限らずに、図１５の点線の矢印にて示すような対応付けもなされることになる。
【０１２１】
例えば、観測データとして、「ガス流量低下」が得られた場合を例にとり、この現象に関連する果因データを対象にした、本発明の適用例について、図１６を参照して説明する。尚、図１６は、図１５から「ガス流量低下」と言う現象の原因を推定するために必要な果因データＣ１〜Ｃ６のみを抜き出したものである。
【０１２２】
なお、以下、説明の簡単のため、次のように、符号化したデータを用いる。
【０１２３】
ガス流量低下：ｙ１
電磁弁１：ｘ１
減圧設定不良（低）：ｘ２
変動幅増大：ｘ３
減圧弁：ｘ４，ｙ２
減圧設定不良（高）：ｘ５
ここで、減圧弁がｘ４とｙ２の２つの符号を持つのは、果因データＣ４の対応関係が示す原因データである場合と、果因データＣ５、Ｃ６の対応関係が示す結果データである場合の２つの場合があるからである。
【０１２４】
図１に示した本発明の因果システムモデル構築装置によって、図１６に示す、ｘからｙへの対応付け（因果データ）を基に、同一結果データ集合Ｙ１、Ｙ２、同一結果原因データ集合Ｘ１、Ｘ２と、これらの対応関係を写像ｆ１，ｆ２でそれぞれ表したサブシステムが生成される。すなわち、
Ｘ１＝ｆ１（Ｙ１）但し、Ｙ１＝｛ｙ１｝，Ｘ１＝｛ｘ１，ｘ２，ｘ３，ｘ４｝Ｘ２＝ｆ２（Ｙ２）但し、Ｙ２＝｛ｙ２｝，Ｘ２＝｛ｘ２，ｘ５｝
また、結果データｙ２が原因データｘ４である果因データ（ｙ２，ｘ４）より、
Ｙ２＝ｇ１（Ｘ１）
という対応関係が生成される。
【０１２５】
上記結果データと原因データとの因果関係を図５にならって模式的に示すと図１７のようになる。
【０１２６】
さて、図１に示した因果システムモデル構築装置により、図１７に示したような因果システムモデルが生成されたとき、図１２の原因推定装置により図１７に示した因果システムモデルを用いて、観測された結果データｙ１「ガス流量低下」に対応する原因を推定する場合の処理手順を図１３を参照しながら説明する。
【０１２７】
まず、観測データ認識部２１により、観測で得られた結果データとしての観測データｙ１が属する同一結果データ集合Ｙ１を検索する（ステップＳ１１〜ステップＳ１２）。すなわち、Ｏ（ｙ１）＝Ｙ１　と表すことができる。
【０１２８】
次に、逆サブシステム検索部２２は、観測データ認識部２１で求めた同一結果データ集合Ｙ１に対応する同一結果原因データ集合Ｘ１を検索する（ステップＳ１３）。すなわち、Ｙ１を引数にもつ関数ｆ１を選択して、ｆ１（Ｙ１）＝Ｘ１
と表すことができる。
【０１２９】
このとき、同一結果原因データＸ１を推定結果として出力するために保持する（Ｘ＝｛Ｘ１｝）。
【０１３０】
なお、検索された同一結果原因データＸ１に写像ｇ１で対応付けられた同一結果データ集合Ｙ２が存在するので（ステップＳ１４）、この同一結果データ集合Ｙ２を得る（ステップＳ１５）。すなわち、ｇ１（Ｘ１）＝Ｙ２　と表すことができる。
【０１３１】
上記同様、このＹ２に対応する同一結果原因データ集合を検索する（ステップＳ１６，ステップＳ１３）。すなわち、ｆ２（Ｙ２）＝Ｘ２　であるから、この同一結果原因データＸ２も推定結果Ｘに付加して保持する（Ｘ＝｛Ｘ１，Ｘ２｝。
【０１３２】
最終的に、推定結果Ｘとして、（同一結果原因データの要素で表した場合）Ｘ＝｛ｘ１，ｘ２，ｘ３，ｘ４，ｘ５｝が得られる（ステップＳ１７）。
【０１３３】
なお、本発明の手法では、ガス流量低下の１つの原因候補であるｘ４の減圧弁についての細かな条件分岐（すなわち、減圧設定不良（低）と減圧設定不良（高）との区別）を設定することはないが、ここで得られた推定結果Ｘでは、大雑把に、「症状」に関連する「工程・部位」を知ることができる。
【０１３４】
さて、次に、原因推定装置に、観測データとして、ｙ１の「ガス流量低下」の他に、「減圧弁」について観測することを明確に指示した場合について。図１３を参照して説明する。
【０１３５】
この場合、あらたな観測データｙ３として、次式のように定義して推論をおこなうことになる。
【０１３６】
【数１１】

【０１３７】
まず、観測データ認識部２１により、観測で得られた結果データとしての観測データｙ３が属する同一結果データ集合を検索する（ステップＳ１１〜ステップＳ１２）。この場合、次式に示すように、Ｏ（ｙ３）は、観測データｙ１が属する同一結果データ集合と、観測データｙ２が属する同一結果データ集合との論理積Ｙ３として求めることができる。
【０１３８】
【数１２】

【０１３９】
次に、逆サブシステム検索部２２は、観測データ認識部２１で求めた同一結果データ集合Ｙ３に対応する同一結果原因データ集合を検索する（ステップＳ１３）。すなわち、Ｙ１を引数にもつ関数ｆ１を選択して、ｆ１（Ｙ１）＝Ｘ１なる同一結果原因データ集合を求めるとともに、Ｙ２を引数にもつ関数ｆ２を選択して、ｆ２（Ｙ２）＝Ｘ２なる同一結果原因データ集合を求めて、これら同一結果原因データ集合の要素の論理積を求めることで、観測データｙ３に対応する原因データ｛ｘ２｝が得られる（ステップＳ１４）。
【０１４０】
【数１３】

【０１４１】
このように、ガス流量低下について、減圧弁に関連して詳しくみると、それは、減圧設定不良（低）が原因であるということを指摘できる。すなわち、観測データとして観測事象が多くなれば、それだけ、原因データの絞り込みが行えることとなる。なお、上記例では、２つの観測事象を観測データとして用いた場合について説明したが、３つ以上の観測事象を観測データとして用いる場合も上記同様である。
【０１４２】
以上説明したように、本発明は、最初からすべての条件を考えて木構造などのデータ構造を構成するのではなく、観測条件に対応してそれに関わる原因推定を随時行ってゆくことを前提としている。逆に言えば、結果データと原因データとの対応と、どれを観測データとして注目するかの観測事象の集合を記述さえできれば、結果から原因へのデータ構造をあらかじめ顕にしておかなくてもよいということである。
【０１４３】
また、観測対象の動作特性にしたがって、同一とみなす注目事象の集合すなわち結果の集合と、原因の集合とを関係づけして記憶することにより、いくつかの現象の組合せで表現される実際の現象（観測事象）の原因を人に理解しやすいような対応関係で推定することができ、人間が故障の状況を容易に把握できる。
【０１４４】
また、観測対象としての装置、システム上に障害が発生した場合、その障害自体を上記のような事象データとして検知するため、当該事象データの属性を基に、例えば画面に表示されている観測対象の構成図上に、当該障害の発生位置を容易に表示することができる。また、第１の実施形態において説明したように構築された因果システムモデルを基に検索された当該障害の原因や、当該障害を原因として発生する障害など、当該障害と互い因果関係にある現象が、その発生位置とともに、観測対象の構成図上に容易に表示することができるのである。
【０１４５】
当初から全体の木構造および検索ルールをさだめずに、その時点で着目したいデータを入力できるのには利点がおおきいと考える。
【０１４６】
すなわち、最初から推論システムの全体構成を考えなくてよく、図１の因果システムモデル構築装置に、結果データ、原因データの対（因果データ、果因データ）を随時追加しても自動的に因果システムモデルを構築することができる。
【０１４７】
また、（写像ｇによる）同一結果データ集合の対応付けがされない同一結果原因データ集合を「チェック終了」として作成し、随時その集合に原因データをいれてゆけば、「チェック終了」集合以外の範囲に限った原因推定が可能である。これにより、人の直感に見合った条件選択による繰り返しの推論も可能となる。
【０１４８】
さらに、本発明によれば、データ間の部分的な因果関係に着目して検索できるので、最近では一般的になったソフトウェア環境におけるハイパーテキスト技術などを利用すれば、さまざまな部分的関係を簡単に提示できる。
【０１４９】
なお、本発明の実施の形態に記載した本発明の手法は、コンピュータに実行させることのできるプログラムとして、磁気ディスク（フロッピーディスク、ハードディスクなど）、光ディスク（ＣＤ−ＲＯＭ、ＤＶＤなど）、半導体メモリなどの記録媒体に格納して頒布することもできる。
【０１５０】
なお、本発明は、上記実施形態に限定されるものではなく、実施段階ではその要旨を逸脱しない範囲で種々に変形することが可能である。さらに、上記実施形態には種々の段階の発明は含まれており、開示される複数の構成用件における適宜な組み合わせにより、種々の発明が抽出され得る。例えば、実施形態に示される全構成要件から幾つかの構成要件が削除されても、発明が解決しようとする課題の欄で述べた課題（の少なくとも１つ）が解決でき、発明の効果の欄で述べられている効果（のなくとも１つ）が得られる場合には、この構成要件が削除された構成が発明として抽出され得る。
【０１５１】
【発明の効果】
以上説明したように、本発明によれば、故障診断のための人の理解し易いデータ構造のデータベースの構築が容易に行え、信頼性のある明確な診断結果を得ることができるとともに、観測対象の構造上の故障個所と診断結果として得られた当該故障の原因箇所などが観測対象の構成図上に容易に表示することができる。
【０１５２】
【図面の簡単な説明】
【０１５３】
【図１】本発明の因果システムモデル構築装置の構成例を示した図。
【０１５４】
【図２】因果データについて説明するための図。
【０１５５】
【図３】同一とみなす結果データを判別、分類するための分類基準の一例を示した図。
【０１５６】
【図４】同一結果データ集合と同一結果原因データ集合との対応付けについて説明するための図。
【０１５７】
【図５】観測対象の因果システムモデルのデータ構造の一例を模式的に示した図。
【０１５８】
【図６】図１の因果システムモデル構築装置の因果システムモデル構築処理動作を説明するためのフローチャート。
【０１５９】
【図７】観測対象（例えばここでは自動車）の事象データの階層構造の具体例を示した図。
【０１６０】
【図８】観測対象の事象データの階層構造の他のを示した図。
【０１６１】
【図９】構成要素Ｄ１〜Ｄ６から構成される観測対象の構成例を模式的に示した図。
【０１６２】
【図１０】図５に示した因果関係を模式的に示した図。
【０１６３】
【図１１】図１０に示した因果関係を表す各集合の要素である事象データの属性を基に、図９に示した観測対象の構成上に、各事象データの発生位置を表すことにより、同一結果集合（Ｙ_１、Ｙ_２、Ｙ_ｋ）、同一結果原因集合（Ｘ_１、Ｘ_２、Ｘ_ｋ）の発生位置を表した図。
【０１６４】
【図１２】本発明の原因推定装置の構成例を示した図。
【０１６５】
【図１３】図１２の原因推定装置の原因推定処理動作を説明するためのフローチャート。
【０１６６】
【図１４】従来からある故障診断システムの知識データベースのデータ構造の一部を概略的に示した図。
【０１６７】
【図１５】本発明の手法による結果データと原因データとの間の対応付けについて説明するための図。
【０１６８】
【図１６】本発明の原因推定処理を説明するための図で、図１５から「ガス流量低下」と言う現象の原因を推定するために必要な果因データＣ１〜Ｃ６のみを抜き出して示した図。
【０１６９】
【図１７】図１６に対応する因果システムモデルを示した図。
【０１７０】
【符号の説明】
１１…因果データ生成記憶部
１２…果因データ生成記憶部
１３…同一結果データ集合生成部
１４…部分因果システムモデル構成部
２１…観測データ認識部
２２…逆サブシステム検索部
２３…対応同一結果データ集合検索部
２４…因果システムモデル記憶部[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to, for example, a method and an apparatus for constructing a database for failure diagnosis, and a failure diagnosis apparatus.
[0002]
[Prior art]
In the field of failure diagnosis and medical condition diagnosis, which has been well researched and developed in the field of AI, the conclusions are drawn according to the conditions found in the phenomena by referring to the knowledge of the expert organized under various conditions and memorized. Alternatively, a method has been developed in which case data are analyzed for similarity (measure) and classified into several cases to conclude.
[0003]
In the past, considering that inferences were made by trial and error based on expert experience and intuition, inference methods based on conditional arrangements (database construction using data structures such as tree structures) and data classification were Thus, it can be said that the system can be used by more people and the diagnosis can be performed efficiently. (Here, we do not refer to so-called model-based inference, but this is because the number of exceptions to the model tends to increase as the number of cases increases, and in reality it is considered that some kind of case handling processing, that is, case database inference is necessary. This is because it was considered to be included in the case of the above condition arrangement.)
・ Experience and intuition of workers (experts)
・ Construction of case DB with tree structure
・ Construction of case DB by classification other than tree structure
However, these conditions and measures for classification must be defined by experts, and therefore do not always match by experts, so that the database itself is correct and the database is updated due to additional cases. This leaves a problem for consistency in time.
[0004]
As a result, the database for fault diagnosis etc.
・ It is not objective and cannot be understood by third parties
-Conditioning tends to be complicated, so search is difficult
-The ability to express "depth of association" in a tree structure rather complicates causality
・ It is difficult to classify cases
Such an impression is given to the user and the maintenance manager.
[0005]
For example, as a failure diagnosis system using a tree structure, there is a failure diagnosis system using a tree structure in order to link a failure result to specific cause information according to target information (for example, see Patent Document 1). The method disclosed here infers only information on the cause of (one) failure when each block is independent, such as an electric circuit, and the operation is sequential according to the circuit logic. To do is certainly meaningful.
[0006]
However, it is not possible to divide the target into small sections based on the concept of blocks, etc., and the causal relationship is not sequential but complicated, and if multiple causes can occur, if the cause is a candidate Even if the information can be enumerated, it is natural that it is not known that the state of the plurality of cause candidates as a whole is not reflected in advance in the tree structure used for the cause search.
[0007]
In fact, the method of using a tree structure is characterized by associating each result with the cause, and letting a human grasp the situation (the whole picture) when there are various causes is that the tree structure is accidentally classified as such. There seems to be no other expecting it to be.
[0008]
From such a viewpoint, a method of introducing object orientation into a tree structure has been considered as an attempt to solve the problem (for example, see Patent Document 2).
[0009]
However, these are merely categorization methods of individual cases, that is, whether or not they belong to an arbitrarily given classification is used as information to help people understand. Since it is not a method of performing classification in a manner that reflects a causal relationship between the cases, the association between the classified sets does not reflect the related information between the cases. Thus, the resulting classification labels and their attributes alone cannot adequately explain the information between cases in each case.
[0010]
In addition, in the field of AI, not only natural language interpretation and logical methods but also fuzzy and neural networks have been introduced to tackle these problems as a method to make classification more ambiguous. In fact, there is no essential solution.
[0011]
Because the description of these systems is treated as a data set as a conclusion expressed through the way of human understanding, the ambiguity, self-contradiction, redundancy, qualitative expression ( Elements that are not quantitative, that is, differ in the degree of non-linearity from one person to another, cannot represent a closed (in a mathematical sense) set.
[0012]
Applying the methods of condition arrangement and classification to natural language treatment as it is is inconsistent with strictly determining the affiliation of all elements.
[0013]
In this sense, it can be said that the above problems of correctness and consistency are problems that cannot be solved.
[0014]
On the other hand, expressing the structure of an apparatus or a system in a hierarchical manner is similar to the case of constructing a concept in a hierarchical manner (for example, see Non-Patent Document 1). It is very effective.
[0015]
[Patent Document 1]
JP-A-8-87412 (paragraph numbers "0034" to "0045", FIG. 7)
[0016]
[Patent Document 2]
JP-A-8-95788 (paragraph numbers “0014” to “0018”, FIG. 2, FIG. 5, FIG. 6)
[0017]
[Non-patent document 1]
IPSJ Transactions Vol. 24, no. 1 "A proposal of knowledge expression that leads to the description of functional reasoning", January 1983, page 73, right column, second line to page 74, left column, eighth line, FIG. 1)
[0018]
[Problems to be solved by the invention]
As described above, conventionally, for example, it is necessary to classify and organize case data collected from an object with complicated conditions, for example, to link the cause to a failure or failure of the object. It was difficult to build a database.
[0019]
Conventionally, the configuration of the object and the phenomena such as failures and failures are separately managed, so the location of the occurrence of the failures and failures and the location of the cause obtained as a result of these diagnoses are included in the configuration of the object. There is a problem that it is difficult to display on the figure.
[0020]
Therefore, the present invention solves the difficulty of constructing a database for failure diagnosis due to the complicated relationship of elements in the condition arrangement and classification methods, and the incomprehensibility of diagnosis results including display methods. Make it an issue.
[0021]
In other words, according to the present invention, a database having a data structure that is easy for humans to understand for failure diagnosis can be easily constructed, a reliable and clear diagnosis result can be obtained, and the diagnosis result can be easily displayed. It is an object of the present invention to provide a causal relationship model generating method, a causal relationship model generating device using the same, a causal system model configuration device, a cause estimating method, and a cause estimating device using the same.
[0022]
For this reason, rather than the conventional method of dealing with an object that simply needs to be described in propositional logic, the object is limited,
・ Do not strictly define the affiliation of elements
・ Describe characteristics by causal relationship only for objects that perform physical actions
・ In the description based on the causal relationship, a group of causes corresponding to one result is configured while allowing redundancy.
We propose a method. In other words, the idea is that the failure state itself can be grasped without strictly defining the belonging relationship.
[0023]
[Means for Solving the Problems]
(1) The present invention is a data structure of a causal relationship model representing a causal relationship between an observed event of interest and a causal event that causes the observed event, wherein the event as the noted event or the causal event is represented by The inclusion relationship between the component of the observation target related to the event and the other components of the observation target, or the conceptual inclusion relationship of the event, is represented as data having attributes expressed hierarchically. A feature is that a set of results having the noted event as an element regarded as the same and a set of causes having the cause event of each element of the result set as an element are associated with each other.
[0024]
Further, when the set of results includes an element serving as a cause event of another event of interest, the set of results corresponds to the set of causes associated with the set of results to which the other event of interest belongs. It is characterized by being attached.
[0025]
In addition, the attention event that is an element of the result set is aggregated as one piece of data including at least one other attention event that has an inclusion relationship with the attention event.
[0026]
According to the present invention, it is easy to construct a database having a data structure that is easy for humans to understand for failure diagnosis, and by using a database based on a causal relationship between the cause and the result, reliability can be improved. In addition to obtaining a clear diagnosis result of a failure, the failure location in the structure of the observation target and the cause of the failure obtained as a diagnosis result can be easily displayed on the configuration diagram of the observation target. .
[0027]
A data structure of a causal relationship model representing a causal relationship between a noticeable event of the observation target and a causal event that causes the noticeable event, wherein the observation target has a hierarchical structure including a plurality of constituent elements, The causal event is regarded as a phenomenon that occurs in any one of the components in the hierarchical structure, and other components that are inclusive with the component existing up to the component by tracing the hierarchical structure. It is expressed as data having an element as an attribute, and a set of results having the noted event as an element regarded as the same and a set of causes having the cause event of each element of the set of results as an element are associated with each other. It is characterized by having.
[0028]
Further, when the set of results includes an element serving as a cause event of another event of interest, the set of results corresponds to the set of causes associated with the set of results to which the other event of interest belongs. It is characterized by being attached.
[0029]
Also, a plurality of attention events having at least the same attribute among the elements of the result set are aggregated as one attention event including the plurality of attention events in the hierarchical structure.
[0030]
(2) The present invention is to generate a causal relationship between an observed event of an observation target and a causal event that causes the observed event, and the event as the causal event or the causal event is referred to as the causal event related to the event. The inclusion relationship between the component of the observation target and the other components of the observation target, or the conceptual inclusion relationship of the event is represented as data having attributes expressed in a hierarchical manner. Generating a set of results having the noted event as an element, and a set of causes having the cause event of each element of the set of results as an element, and associating the set of results with the set of causes. This makes it easy to construct a database with a data structure that is easy for humans to understand for fault diagnosis. By using a database based on the causal relationship between such causes and results, reliable and clear It is possible to obtain a diagnosis result of the failure. In addition, the location of a failure in the structure of the observation target and the location of the cause of the failure obtained as a result of diagnosis can be easily displayed on the configuration diagram of the observation target.
[0031]
Generate a causal relationship between the observed event of interest and the causal event that causes the observed event, wherein the observed object has a hierarchical structure consisting of a plurality of components, the attention event and the causal event are Considering as a phenomenon that occurs in any of the components on the hierarchical structure, tracing through the hierarchical structure and arranging other components in an inclusive relationship with the relevant component up to the relevant component as attributes A set of results having the noted event as an element, which is regarded as the same, and a set of causes having the cause event of each element of the set of the elements as an element, and generating a set of the results By associating the cause with the set of causes, it is easy to construct a database having a data structure that is easy for humans to understand for failure diagnosis, and a data based on a causal relationship between such causes and results. By using the base, it is possible to obtain a diagnosis result of a clear fault reliable. In addition, the location of a failure in the structure of the observation target and the location of the cause of the failure obtained as a result of diagnosis can be easily displayed on the configuration diagram of the observation target.
[0032]
(3) The present invention relates to a cause estimating method for estimating a cause of an observation event observed from the observation target based on a causal relationship between a noticeable event of the observation target and a cause event causing the observation. The event of interest or the event as the cause event is expressed hierarchically by the inclusion relationship between the component of the observation target and the other component of the observation target related to the event, or the conceptual inclusion relationship of the event. A set of results having the noted event as an element regarded as the same, and a set of causes having the cause event of each element of the set of results as an element. Then, based on the causal relationship model generated by associating the set of results and the set of causes, and searching for the set of results with the observed event as an element, By retrieving the set of causes associated with the set of results, a reliable and clear fault using a database with a data structure that is easy for humans to understand based on the causal relationship between the causes and the results. A diagnostic result can be obtained. In addition, the location of a failure in the structure of the observation target and the location of the cause of the failure obtained as a result of diagnosis can be easily displayed on the configuration diagram of the observation target.
[0033]
Based on a causal relationship between the observed event of interest and the causal event that causes the observed event, the cause of the observed event observed from the observed object is estimated, and the observed object includes a plurality of components. It has a hierarchical structure, and the event of interest and the cause event are regarded as phenomena that occur in any of the components on the hierarchical structure, and the relevant events existing up to the relevant component by tracing the hierarchical structure are considered. It is represented as data having other constituent elements in an inclusive relationship with the constituent elements as attributes, and a set of results having the noted event as an element regarded as the same, and a cause event of each element of the set of results as an element. A set of causes to be generated, and based on the causal relation model generated by associating the set of results with the set of causes, the set of results having the observed event as an element. By using a database having a data structure that is easy to understand for a person based on the causal relationship between the cause and the result by searching for the set of causes associated with the set of searched results while searching. In this way, a reliable and clear failure diagnosis result can be obtained. In addition, the location of a failure in the structure of the observation target and the location of the cause of the failure obtained as a result of diagnosis can be easily displayed on the configuration diagram of the observation target.
[0034]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0035]
(1st Embodiment)
FIG. 1 shows an example of the configuration of a causal system model construction device (causal relationship model generation device) according to the first embodiment. The causal data generation storage unit 11, the causal data generation storage unit 12, and the same result data It comprises a set creation unit 13 and a partial causal system model construction unit 14.
[0036]
Hereinafter, the configuration of the causal system model construction device of FIG. 1 and the processing operation of each functional unit will be described with reference to the flowchart shown in FIG.
[0037]
In the present embodiment, an observation target (for example, a system) to be modeled has a hierarchical structure including a plurality of components. Event data expressing a causal relationship related to the observation target, for example, event data such as cause data and result data is also data hierarchically expressed as data representing phenomena occurring in the components on the hierarchical structure. .
[0038]
FIG. 7 shows an example of the hierarchical structure of an automobile as an observation target, for example. As shown in FIG. 7, the structure of an automobile is roughly divided into a fuel system and an electric system. The fuel system further includes a fuel tank and the like, and the fuel tank includes fuel. The electric system includes a battery, an engine, a headlight, a vehicle width lamp, and the like. “Fuel system”, “Electrical system”, “Fuel tank”, “Battery”, “Headlight”, “Engine”, “Vehicle width lamp”, etc. are referred to herein as constituent elements constituting a hierarchical structure of “Automobile”. . The component “fuel” indicates that the component is included in the “fuel tank”.
[0039]
A phenomenon (for example, an obstacle) that may occur in the fuel tank includes “fuel tank breakage”. In addition, an obstacle that can occur in the fuel includes “out of fuel”. The above-mentioned obstacles are described as "car / fuel system / fuel tank / fuel tank broken" and "car / fuel system / fuel tank / out of fuel". Hereinafter, for simplicity of description, a notation method in which the first component (at the root) in the hierarchical structure (for example, “car” in this case) is omitted. For example, "fuel system / fuel tank / fuel tank damaged" and "fuel system / fuel tank / out of fuel" are described.
[0040]
In this way, for each phenomenon (for example, a failure), the root element is traced from the (leading) element to the hierarchical element where the phenomenon occurs, and is present from the root element to the element. The other constituent elements having an inclusion relationship with the relevant constituent element are referred to as attributes here. Each component is also called an attribute or class data.
[0041]
For example, the attribute of the failure "fuel tank damage" is "fuel tank" (more specifically, "automobile / fuel system / fuel tank"), and the attribute of the failure "fuel shortage" is "fuel" ( More specifically, it is "automobile / fuel system / fuel tank / fuel".
[0042]
Further, in the electric system, a failure that can occur in the battery includes "battery exhaustion", a failure that can occur in the headlights includes "light reduction" and "breakage of the headlight lamp", and a failure that can occur in the vehicle width lamp includes ""Lightreduction". Failures such as those listed above include “Electrical System / Battery / Battery Dead”, “Electrical System / Headlight / Light Down”, “Electrical System / Headlight / Headlight Lamp Damage”, “Electrical System / Vehicle Width Lamp / Light Down” ". As described above, the attribute of the phenomenon of “battery exhaustion” is “electric system / battery”, and the event of “light reduction” is the phenomenon of “electric system / headlight” and “electric system / vehicle width”. Some have "lamp" as an attribute.
[0043]
As described above, when a component in the hierarchical structure in which a certain phenomenon (si) occurs is represented as an attribute Ii, here, it is represented as, for example, Ii · si as phenomenon data. In this notation, the attribute Ii may represent an inclusion relationship of a plurality of components, and includes, for example, a notation of an attribute “electric system / battery”. For example, when the result data is a phenomenon xi corresponding to the result and its attribute is represented by Ii, it is Ii · xi, and when the cause data is a phenomenon yj corresponding to the cause and its attribute is Jj, Jj · yj It becomes.
[0044]
In the observation target according to the present embodiment, event data such as result data and cause data has a hierarchical data structure as shown in FIG.
[0045]
In the following, for the sake of simplicity (particularly in the drawing), the notation parts Ii and Jj of the attributes of Ii · xi and Jj · yj are omitted, Ii · xi is denoted by xi, and Jj · yj is denoted by yj. Is sometimes simply written. In other words, in the following notations (in the description of the first to third embodiments), notation formats such as xi and yi are all omitted from the attribute notation part.
[0046]
The causal data generation storage unit 11 expresses a causal relationship related to an observation target (for example, a system) to be modeled, for example, a set of causal data having a pair of cause data (Ii · xi) and result data (Jj · yj). C is generated and stored (step S1). It may be actual data of an actual observation target or modeled data obtained by analyzing the observation target.
[0047]
As shown in FIG. 2, if X is cause data (set) and Y is result data (set),
(Equation 1)

Generate a set of causal data.
[0048]
Incidentally, the result data yh (ｈY) may actually be cause data xk (∈X) that causes another result. In other words, it is assumed that the result data in the actual observation target is actually intermediate data, rather than merely a result, and that this leads to a cause of another result.
[0049]
Therefore, the causal data generation storage unit 12 stores causal data having the result data yh (ie, Jh · yh) (∈Y) and the causal data xk (ie, Ik · xk) (∈X) as a pair. Is generated and stored (step S2).
[0050]
(Equation 2)

[0051]
As described above, the result data and the cause data form a pair, and indicate the relationship between one result data and a cause that causes another result.
[0052]
Hereinafter, a description will be given by taking, as an example, event data of an observation target having a hierarchical structure as shown in FIG. “D1” to “D6” in FIG. 8 are components, respectively, and the component D2 is configured to include the component D3. For example, phenomena y1 and y2 are phenomena that occur in component D3, respectively, and component D3 is a component included in component D2, and phenomena x3 and x4 occur in component D2. .
[0053]
Each of the above components may correspond to, for example, a component such as a fuel tank or a battery in the automobile shown in FIG. 7, but is not limited to this case. For example, the hierarchical structure is artificially made easy to understand. May be a component created for this purpose. For example, in FIG. 8, the phenomena y1 and y3 are failures that occur in the component D2, and the component D3 is a class or an attribute created by the user to collect the phenomena y1 and y3 into one. May be a component corresponding to “failure”.
[0054]
As described above, the components according to the present embodiment basically correspond to the components that constitute the hierarchical structure of the observation target, but are not limited to this case. (For example, as described above, a component created for the sake of convenience used to combine a plurality of phenomena and components into one).
[0055]
FIG. 9 schematically shows a configuration example of an observation target composed of the components “D1” to “D6”. Here, the description will be made using the names “D1” to “D6” of the respective components as the reference numerals D1 to D6 of the respective components as they are.
[0056]
In FIG. 8, for example, the event data corresponding to the phenomenon y1 is “D2 · D3 · y1” or “D2 · (D3 · y1)” in order to clarify that the phenomenon y1 occurs in the component D3. It can be expressed as. The event data corresponding to the phenomenon x5 can be represented as “D4 · x5”.
[0057]
Next, the same-result data set creation unit 13 generates a new set representing one “same result” by using, as an element, the result data that can be actually determined to be the same result, that is, the same-result data set (Y1 , Y2,...) (Step S3).
[0058]
The criterion or method for determining what kind of result data is considered to be the same is provided (stored) in advance in the same result data set creating unit 13. As the simplest example, a description will be given of a case in which only the determination criterion is simply stored and the same set is determined in accordance with the criterion.
[0059]
[Equation 3]

[0060]
Regarding “what kind of result data is considered to be the same”, it is assumed that each result data belongs to which set according to a predetermined standard. For arbitrary result data yj (that is, Jh · yh) (∈Y), for example, a table of classification criteria (z1,... ZL) in a table format as shown in FIG. 3 is provided in advance.
[0061]
According to the classification criterion as shown in FIG. 3, the result data always applies to any one of the sets determined from the beginning, but (yj (that is, Jj · yj) is one of the same result data sets Y0 to YL). Element)), a case where it does not apply, and in that case, a set Y in which the result data that does not apply are, for example, one by one, and only one element. _{L + m} , (M = 1, 2,...).
[0062]
Note that, as not difficult to imagine, this part is not limited to the above-described method, because the application of the conventional classification method may be applied.
[0063]
In this way, the result data that can be regarded as the same is set as one set according to a predetermined standard or method. At this time, among the result data that may be regarded as the same, for example, a plurality of result data that can be aggregated into one from the inclusion relationship in the hierarchical structure as illustrated in FIG. 8 is represented by one aggregated description. For example, among the event data shown in FIG. 8, a set of result data that may be regarded as the same (same result data set) Y ₁ And {y1, y2, yj} are obtained. This is exactly {D2.D3.y1, D6.y2, D6.yj}, but this set Y ₁ In the result data as the elements of “D6 · y2” and “D6 · yj”, both are the result data included in the component D6, and the phenomenon that occurs in the component D6 is “D6 · y2”. And “D6 · yj” only. Therefore, in this case, to represent “D6 · y2” and “D6 · yj”, “D6” is used as event data including these. That is, Y ₁ = {D2D3y1, D6}.
[0064]
Similarly, from the event data shown in FIG. 8, a set Y as shown below is set as the same result data set. ₁ , Y ₂ , Y _k Is obtained.
[0065]
Y ₁ = {Y1, y2, yj} = {D2.D3.y1, D6}
Y ₂ = {Y1, y3} = {D3}
Y _k = {Y4} = {D5 · y4}
The partial causal system model forming unit 14 generates and associates a set having the cause data that causes the result data that is an element of the same result data set created by the same result data set creating unit 13 as an element, Further, the above-mentioned cause data is associated with each other, and the reason data and the result data are represented by a partial set, thereby facilitating inference regarding the target system.
[0066]
First, the association relating to the same result data will be described with reference to FIG.
[0067]
In FIG. 4, since the elements {y1, y2, yj} of the same result data set Y1 are stored as causal data in the causal data generation storage unit 11, for example, (x2, y1), (x (x4, y2), (xi, yj) ∈C.
[0068]
The corresponding cause data x2, x4, and xi are considered as one set, and set as the same result cause data set X1 (= {x2, x4, xi}) (step S4). The correspondence between X1 and the original Y1 is represented as f1 as a mapping (step S5).
[0069]
(Equation 4)

[0070]
In this manner, when the same result data set is associated with the cause data set,
(Equation 5)

It can be expressed as.
[0071]
Thus, the same result cause data set X as shown below ₁ , X ₂ , X _k Is obtained.
[0072]
X ₁ = {X2, x4, xi} = {D1.x2, D2.x4, D5.xi}
X ₂ = {X1, x5, x6} = {D1.x1, D4}
X _k = {X3} = {D2 · x3}
Here, if the correspondence from Xk to Yk is called a subsystem for convenience, the mapping f is an inverse mapping of the subsystem (“inverse subsystem”). Each of the subsystems is configured to have a one-to-one correspondence between Xk and Yk. In other words, the same result cause data set is determined so that the reverse subsystem is formed.
[0073]
Next, the association of the causal data will be described with reference to FIG.
[0074]
The cause data (yh, xk) @R, yh (@Y), and xk (@X) are information indicating that one result data yh is immediately another cause data xk. In FIG. 5, for example, y1 and yj belonging to Y1 are respectively associated with x1 (∈X2) and x3 (∈Xk). Further, y4 belonging to Yk is associated with x6 (∈X2).
[0075]
That is, as the causal data,
(Y1, x1) = (D3 · y1, D1 · x1)
(Y4, x6) = (D5.y4, D4.x6)
(Yj, x3) = (D6 · yj, D2 · x3)
Is obtained.
[0076]
At this time, the fact that Y1 is associated with X2 and Xk via each element is represented by g as a mapping (step S6). That is,
(Equation 6)

In the case of the same result cause data set having no element relating to the cause data, for example, in the case of X1 in FIG. 5, the corresponding set by the mapping g is an empty set as follows.
[0077]
(Equation 7)

[0078]
This mapping allows multiple correspondences. For example, in FIG. 5, since there is also a correspondence of g (X2) = Yk, the output is a set expression.
(Equation 8)

It expresses.
[0079]
Where Y ² Is defined for convenience as a meta-set having Y1 and Yk as elements.
[0080]
In the above description, all the corresponding items are listed. However, depending on the actual observation results, for example, when Yk is not obtained (nothing belongs to the observation results), the above example is shown. As described above, the output of only Y1 may be used.
[0081]
In this way, the partial causal system model forming unit 14 forms mappings f and g between subsets of the causal data and the causal data from the same result data as the causal data.
[0082]
As described above, the causal system model construction apparatus of FIG. 1 divides the causal data and causal data of the observation target into a subset (subsystem) by focusing on the same result data, as shown in FIG. By constructing a causal system model that expresses the causal relationship between the cause and the result of the observation target by using the correspondence between subsets (using the mappings f and g), considering all conditions from the beginning, the tree structure There is no need to configure a data structure, and a database for failure diagnosis can be easily constructed.
[0083]
In addition to the correspondence between the attributes being used as data (causal data and causal data), the correspondence between sets may be converted into data in advance and used when defining a set.
[0084]
Incidentally, the event data x1... X6, xi, y1... Y4, and yj shown in FIG. 5 are event data to be observed having a hierarchical structure as shown in FIG.
[0085]
FIG. 5 shows the causal relationship between the cause and the result of the observation target in the same result data set Y. ₁ , Y ₂ , Y _k Same result cause data set X as ₁ , X ₂ , X _k FIG. 10 schematically shows a causal system model represented by a correspondence relationship between.
[0086]
In FIG. 10, the same result data set Y ₁ , Y ₂ , Y _k Same result cause data set X as ₁ , X ₂ , X _k Is as follows.
[0087]
Y ₁ = {Y1, y2, yj} = {D2.D3.y1, D6}
Y ₂ = {Y1, y3} = {D3}
Y _k = {Y4} = {D5 · y4}
X ₁ = {X2, x4, xi} = {D1.x2, D2.x4, D5.xi}
X ₂ = {X1, x5, x6} = {D1.x1, D4}
X _k = {X3} = {D2 · x3}
In FIG. 10, the direction from the set as the cause to the set as the result is indicated by an arrow. That is, for example, set X ₁ Is the set Y ₁ And the set Y ₁ Is X _k And set X ₂ Is the cause.
[0088]
In the attribute of the event data which is an element of each set, a component which is a position where a phenomenon corresponding to the event data occurs is represented according to a hierarchical structure of an observation target. Therefore, based on this attribute, it is possible to specify a component that is a position where a phenomenon corresponding to each event data occurs. That is, on the configuration of the observation target shown in FIG. 9, the occurrence position of event data which is an element of each set showing the causal relationship shown in FIG. 10 can be represented.
[0089]
FIG. 11 shows the occurrence position of each event data on the configuration of the observation target shown in FIG. 9 based on the attribute of the event data which is an element of each set showing the causal relationship shown in FIG. Each set Y ₁ , Y ₂ , Y _k , X ₁ , X ₂ , X _k This is a representation of the location of occurrence.
[0090]
As described above, since the event data itself is hierarchically represented in accordance with the hierarchical structure of the observation target, the occurrence position of the phenomenon of interest or the occurrence position of the phenomenon regarded as having the same result as the phenomenon of interest is included in the configuration of the observation object. The causal relationship such as the cause of the phenomenon of interest can be easily displayed. Further, the occurrence position of the target phenomenon can be displayed not only by displaying the component corresponding to the occurrence position as it is, but also by following the component described in the attribute of the event data corresponding to the target phenomenon in the upstream direction. By doing so, it is also possible to display a component in a higher hierarchy that includes the component.
[0091]
In other words, when a failure occurs in the device or system as the observation target, the failure itself is detected as the above-described event data. For example, based on the attribute of the event data, the observation target displayed on the screen is used. The location of the fault can be easily displayed on the configuration diagram of the above, and based on the causal system model constructed as described above, the cause of the fault, the fault generated due to the fault, etc. Phenomena that have a causal relationship with the fault can be easily displayed on the configuration diagram of the observation target together with their occurrence positions.
[0092]
The cause data and the result data often have attributes different from the causal relationship, and the relationship with each cause data set and the result data set is related to the attribute (class) (the structure of the observation target, especially the hierarchical structure). It is possible to express while being based on. This facilitates the construction of a causal system model and the construction of a cause estimating device not only for character display but also for an information system shown and used.
[0093]
(Second embodiment)
Next, a cause estimating apparatus using a causal system model as shown in FIG. 5, for example, constructed by the causal system model constructing apparatus of FIG. 1 will be described.
[0094]
Note that, similarly to the first embodiment, the phenomenon data such as the result data and the cause data is the event data of the observation target having a hierarchical structure as shown in FIG. The event data is represented by, for example, x1... X6, xi, y1.
[0095]
Based on the causal system model, the cause estimating apparatus sequentially repeats searching for the same result cause data set corresponding to the same result data set to which the result data observed from the system to be observed, for example, to obtain the corresponding cause data. To get the set to which
[0096]
FIG. 12 shows an example of the configuration of the cause estimating apparatus, which comprises an observation data recognition unit 21, an inverse subsystem search unit 22, a corresponding identical result data set search unit 23, and a causal system model storage unit 24. ing.
[0097]
Hereinafter, the configuration of the cause estimating apparatus of FIG. 12 and the processing operation of each functional unit will be described with reference to the flowchart shown in FIG.
[0098]
The causal system model storage unit 24 stores, for example, the same result data set and the same result causal data set from the causal data and causal data as shown in FIG. (Causal relationship between the cause and the result of the observation target) represented by the mappings f and g of the observation target are stored.
[0099]
The observation data recognizing unit 21 is based on, for example, a causal system model (causal relation model) as shown in FIG. 5 stored in the causal system model storage unit 24 and, for example, a classification criterion as shown in FIG. , The same result data set Yr⊂Y to which the observation data y0 (∈Y) as the result data obtained by the observation belongs is searched (steps S11 to S12). Here, the observation data is the event data of the observation target as described above.
[0100]
If the above function of the observation data recognition unit 21 is mapped and expressed as O (), it is expressed as follows.
[0101]
Yr = O (yo) (11)
For example, a description will be given with reference to a causal system model as shown in FIG. It is assumed that result data y1 is obtained by observation (as observation data). The fact that the same result data set Y2 to which the result data y1 belongs is output by the classification criterion as shown in FIG. 3 can be expressed by the following equation.
[0102]
Y2 = O (y1) (12)
The reverse subsystem search unit 22 searches for the same result cause data set corresponding to the same result data set Yr obtained by the observation data recognition unit 21 (Step S13).
[0103]
The causal system model storage unit 24 holds data indicating the correspondence between subsystems of the target system. That is, it is assumed that the subsets X and Y as shown in the equation (7) are stored in association with each other.
[0104]
At this time, assuming that the same result data set to which the observation data belongs is obtained as in Expression (11), the same result cause data set X2 corresponding thereto is searched. That is, this can be expressed by the following equation.
[0105]
X2 = f2 (Y2) (13) The corresponding identical result data set search unit 23 searches for the same result data set corresponding to the same result cause data set Xr via the above-described cause data (steps S14 to S15).
[0106]
For example, in FIG. 5, in the case of the above example, the same result cause data set X2 is included in the same result cause data set X2 as shown in Expression (10) according to the correspondence of the cause data {(y1, x1), (y4, x6)}. Two result data sets Y1 and Yk are associated with each other.
[0107]
In FIG. 13, the two identical result data sets Y1 and Yk correspond to Yl.
[0108]
As shown in FIG. 12, the reverse subsystem search unit 22 and the corresponding identical result data set search unit 23 have a feedback configuration. That is, with such a configuration, the cause estimating apparatus of the present invention newly adds the same result data set Yl associated with the same result cause data set Xr (when there are a plurality of Yl, each of them). , The reverse subsystem search unit 22 searches for the same result cause data set (step S16, steps S13 to S14).
[0109]
In step S15, if the same result cause data set Xr searched by the reverse subsystem search unit 22 does not have the same result data set associated with the mapping g, the process proceeds to step S17, and the process proceeds to step S17. , All the same result data sets {Y} searched by the reverse subsystem search unit 22 ₁ , Y ₂ ,..., And the same result data search set ｛Y searched by the corresponding ₁ , Y ₂ ,..., The same result cause data set {X} ₁ , X ₂ , ...｝ are output. Note that the same result cause data set ｛X ₁ , X ₂ ,... May be output.
[0110]
The cause estimating device of FIG. 12 will be described from the viewpoint of input / output data. When observation data y0 is input as input data, the same result data set and the same result cause data set searched for corresponding to the observation data Can be output as the estimation result, and indicates the status of the observation data. That is,
(Equation 9)

In Equations (14) to (17), f is described without a suffix. However, in practice, f having a corresponding suffix is selected according to the argument.
[0111]
A label name is given to each identical result data set and each identical result cause data set, and this label name may be output.
[0112]
In the above description, the case where the number of observation data y0 is one, that is, the case of one observation event (phenomenon) has been described, but the number of observation data is not always one. For example, when it is desired to assume that two phenomena are simultaneously established from the observation target system, the cause estimating apparatus illustrated in FIG. 12 performs the following processing.
[0113]
(Equation 10)

[0114]
According to the assumption that the two phenomena hold simultaneously, the estimation result of one of the two observation data (all the same result data sets searched by the inverse subsystem search unit 22 and the corresponding same results) The logical product of the same result cause data set associated with each of the same result data search sets searched by the data set search unit 23 and the estimation result for the other observation data is output.
[0115]
When the assumption is made that one of the two phenomena is established, the estimation result of one of the two observation data (all the same result data retrieved by the inverse subsystem retrieval unit 22) is obtained. OR of the set, the same result cause data set associated with each of the same result data search sets searched by the corresponding same result data set search unit 23), and the estimation result of the other observation data Is output.
[0116]
According to the second embodiment, when a failure occurs in a device or system as an observation target, the failure itself is detected as the above event data (corresponding to observation data here). On the basis of the attribute of the event data, for example, the location of occurrence of the fault can be easily displayed on the configuration diagram of the observation target displayed on the screen. Further, phenomena having a causal relationship with the fault, such as the cause of the fault searched based on the causal system model constructed as described in the first embodiment and the fault generated due to the fault, are described. Can be easily displayed on the configuration diagram of the observation target together with the position of occurrence.
[0117]
(Third embodiment)
Here, the method of the present invention described above will be described more specifically while comparing with the conventional method.
[0118]
FIG. 14 schematically shows a part of a data structure of a knowledge database of a conventional failure diagnosis system, and has a tree structure for linking a failure result to a cause. In other words, among the phenomena (symptoms) observed from a certain observation target system and the sets labeled by classifying the process / site, factor, and parts where the symptoms appeared, the elements of each set are attributed to the results. It has a tree structure linked back to.
[0119]
In the fault estimation tree data shown in FIG. 14, a concatenation structure is formed between adjacent sets of data with respect to each element, which is mainly considered to have a causal relationship by using an arrow “(cause) → (result)”. are doing.
[0120]
On the other hand, according to the method of the present invention, the association between the result data and the cause data is not limited to between the classified adjacent sets, but is also made as shown by a dotted arrow in FIG. Will be.
[0121]
For example, an application example of the present invention will be described with reference to FIG. 16 by taking as an example a case where “gas flow rate decrease” is obtained as observation data, and targeting causal data related to this phenomenon. FIG. 16 shows only the cause data C1 to C6 necessary for estimating the cause of the phenomenon of “gas flow rate decrease” from FIG.
[0122]
In the following, for the sake of simplicity, encoded data will be used as follows.
[0123]
Gas flow reduction: y1
Solenoid valve 1: x1
Poor decompression setting (low): x2
Increase in fluctuation width: x3
Pressure reducing valve: x4, y2
Poor decompression setting (high): x5
Here, the case where the pressure reducing valve has two signs of x4 and y2 is the cause data indicated by the correspondence between the causal data C4 and the result data indicated by the correspondence between the causal data C5 and C6. This is because there are two cases.
[0124]
With the causal system model construction apparatus of the present invention shown in FIG. 1, the same result data sets Y1 and Y2, the same result cause data set X1, and the same result data set Y1 and Y2 based on the association (causal data) from x to y shown in FIG. A subsystem is generated in which X2 and their correspondence are represented by mappings f1 and f2, respectively. That is,
X1 = f1 (Y1) where Y1 = {y1}, X1 = {x1, x2, x3, x4} X2 = f2 (Y2) where Y2 = {y2}, X2 = {x2, x5}
Further, from the cause data (y2, x4) in which the result data y2 is the cause data x4,
Y2 = g1 (X1)
Is generated.
[0125]
The causal relationship between the result data and the cause data is schematically shown in FIG. 17 as shown in FIG.
[0126]
Now, when a causal system model as shown in FIG. 17 is generated by the causal system model construction apparatus shown in FIG. 1, the causal system model shown in FIG. A processing procedure for estimating the cause corresponding to the obtained result data y1 “gas flow decrease” will be described with reference to FIG.
[0127]
First, the observation data recognition unit 21 searches for the same result data set Y1 to which the observation data y1 as the result data obtained by observation belongs (steps S11 to S12). That is, it can be expressed as O (y1) = Y1.
[0128]
Next, the reverse subsystem search unit 22 searches the same result cause data set X1 corresponding to the same result data set Y1 obtained by the observation data recognition unit 21 (Step S13). That is, a function f1 having Y1 as an argument is selected, and f1 (Y1) = X1
It can be expressed as.
[0129]
At this time, the same result cause data X1 is held for output as an estimation result (X = {X1}).
[0130]
Since there is the same result data set Y2 associated with the searched same result cause data X1 by the mapping g1 (step S14), this same result data set Y2 is obtained (step S15). That is, g1 (X1) = Y2 can be expressed.
[0131]
As described above, the same result cause data set corresponding to Y2 is searched (steps S16 and S13). That is, since f2 (Y2) = X2, the same result cause data X2 is also added to the estimation result X and held (X = {X1, X2}.
[0132]
Finally, X = {x1, x2, x3, x4, x5} is obtained as the estimation result X (when represented by the same result cause data element) (step S17).
[0133]
In the method of the present invention, fine conditional branches (that is, discrimination between poor pressure setting (low) and poor pressure setting (high)) for the x4 pressure reducing valve, which is one of the causes of the gas flow rate decrease, are set. However, in the estimation result X obtained here, the "process / site" related to the "symptom" can be roughly known.
[0134]
Next, a case where the cause estimating device is clearly instructed to observe “reducing valve” in addition to “reducing gas flow rate” of y1 as observation data. This will be described with reference to FIG.
[0135]
In this case, inference is performed by defining as new observation data y3 as follows.
[0136]
[Equation 11]

[0137]
First, the observation data recognition unit 21 searches for the same result data set to which the observation data y3 as the result data obtained by the observation belongs (steps S11 to S12). In this case, as shown in the following equation, O (y3) can be obtained as the logical product Y3 of the same result data set to which the observation data y1 belongs and the same result data set to which the observation data y2 belongs.
[0138]
(Equation 12)

[0139]
Next, the reverse subsystem search unit 22 searches for the same result cause data set corresponding to the same result data set Y3 obtained by the observation data recognition unit 21 (Step S13). That is, a function f1 having Y1 as an argument is selected to obtain the same result cause data set of f1 (Y1) = X1, and a function f2 having Y2 as an argument is selected to obtain a same function f2 (Y2) = X2. The result data set {x2} corresponding to the observation data y3 is obtained by obtaining the result cause data set and obtaining the logical product of the elements of the same result cause data set (step S14).
[0140]
(Equation 13)

[0141]
As described above, when the gas flow rate decrease is examined in detail in relation to the pressure reducing valve, it can be pointed out that the cause is poor pressure setting (low). That is, as the number of observation events increases as observation data, the cause data can be narrowed down accordingly. In the above example, the case where two observation events are used as observation data has been described. However, the same applies to the case where three or more observation events are used as observation data.
[0142]
As described above, the present invention does not configure a data structure such as a tree structure in consideration of all conditions from the beginning, but presupposes that a cause estimation relating to observation conditions is performed at any time according to observation conditions. I have. Conversely, as long as it is possible to describe the correspondence between the result data and the cause data and the set of observation events that focuses on the observation data, the data structure from the result to the cause need not be revealed in advance. That's what it means.
[0143]
In addition, according to the behavioral characteristics of the observation target, the actual phenomenon represented by a combination of several phenomena is stored by associating and storing a set of noticeable events that are regarded as the same, that is, a set of results, and a set of causes. The cause of the (observation event) can be estimated by a correspondence that is easy for a human to understand, and the human can easily grasp the situation of the failure.
[0144]
In addition, when a failure occurs on the device or system as an observation target, the failure itself is detected as the above-mentioned event data. The location where the fault has occurred can be easily displayed on the configuration diagram of FIG. Further, phenomena having a causal relationship with the fault, such as the cause of the fault searched based on the causal system model constructed as described in the first embodiment and the fault generated due to the fault, are described. Can be easily displayed on the configuration diagram of the observation target together with the position of occurrence.
[0145]
From the beginning, it would be advantageous to be able to input the data of interest at that point in time without subduing the entire tree structure and search rules.
[0146]
In other words, it is not necessary to consider the entire configuration of the inference system from the beginning. Even if a pair of result data and cause data (causal data and causal data) is added to the causal system model construction apparatus of FIG. A system model can be built.
[0147]
In addition, if the same result cause data set that is not associated with the same result data set (by the mapping g) is created as “check end” and cause data is added to the set as needed, a range other than the “check end” set It is possible to estimate the cause only for. As a result, it is also possible to infer repetition by selecting a condition that matches the intuition of a person.
[0148]
Furthermore, according to the present invention, since a search can be performed by focusing on a partial causal relationship between data, various partial relationships can be easily simplified by using hypertext technology in a software environment which has recently become popular. Can be presented.
[0149]
Note that the method of the present invention described in the embodiment of the present invention can be executed by a computer as a program such as a magnetic disk (floppy disk, hard disk, etc.), an optical disk (CD-ROM, DVD, etc.), a semiconductor memory, etc. Can also be stored in a recording medium and distributed.
[0150]
Note that the present invention is not limited to the above-described embodiment, and can be variously modified in an implementation stage without departing from the gist of the invention. Furthermore, the above embodiments include inventions at various stages, and various inventions can be extracted by appropriate combinations of a plurality of disclosed configuration requirements. For example, even if some components are deleted from all the components shown in the embodiment, (at least one of) the problems described in the column of the problem to be solved by the invention can be solved, and the effect of the invention can be solved. If at least one of the effects described in (1) is obtained, a configuration from which this component is deleted can be extracted as an invention.
[0151]
【The invention's effect】
As described above, according to the present invention, it is possible to easily build a database having a data structure that is easy for humans to understand for failure diagnosis, to obtain a reliable and clear diagnosis result, The structural failure point and the cause of the failure obtained as a result of diagnosis can be easily displayed on the configuration diagram of the observation target.
[0152]
[Brief description of the drawings]
[0153]
FIG. 1 is a diagram showing a configuration example of a causal system model construction device according to the present invention.
[0154]
FIG. 2 is a diagram for explaining causal data.
[0155]
FIG. 3 is a diagram showing an example of a classification criterion for determining and classifying result data regarded as the same.
[0156]
FIG. 4 is a diagram for explaining the association between the same result data set and the same result cause data set.
[0157]
FIG. 5 is a diagram schematically illustrating an example of a data structure of a causal system model of an observation target.
[0158]
FIG. 6 is a flowchart for explaining a causal system model construction processing operation of the causal system model construction apparatus of FIG. 1;
[0159]
FIG. 7 is a diagram showing a specific example of a hierarchical structure of event data of an observation target (for example, a car in this case).
[0160]
FIG. 8 is a diagram showing another example of the hierarchical structure of the event data to be observed.
[0161]
FIG. 9 is a diagram schematically illustrating a configuration example of an observation target configured by components D1 to D6.
[0162]
FIG. 10 is a diagram schematically showing the causal relationship shown in FIG. 5;
[0163]
FIG. 11 is a diagram showing the occurrence position of each event data on the configuration of the observation target shown in FIG. 9 based on the attribute of the event data which is an element of each set showing the causal relationship shown in FIG. The same result set (Y ₁ , Y ₂ , Y _k ), Same result cause set (X ₁ , X ₂ , X _k FIG.
[0164]
FIG. 12 is a diagram showing a configuration example of a cause estimating apparatus according to the present invention.
[0165]
FIG. 13 is a flowchart for explaining the cause estimating operation of the cause estimating apparatus of FIG. 12;
[0166]
FIG. 14 is a diagram schematically showing a part of a data structure of a knowledge database of a conventional failure diagnosis system.
[0167]
FIG. 15 is a diagram for explaining association between result data and cause data according to the technique of the present invention.
[0168]
FIG. 16 is a diagram for explaining the cause estimating process of the present invention, in which only the causal data C1 to C6 necessary for estimating the cause of the phenomenon of “gas flow rate decrease” is extracted from FIG. FIG.
[0169]
FIG. 17 is a diagram showing a causal system model corresponding to FIG. 16;
[0170]
[Explanation of symbols]
11 Causal data generation storage unit
12 ... causal data generation storage unit
13: Same result data set generation unit
14 Partial causal system model component
21: Observation data recognition unit
22 ... Reverse subsystem search unit
23: Corresponding identical result data set search unit
24: Causal system model storage

Claims (15)

A causal relationship model generation method for generating a causal relationship between an observed event of interest and a causal event that causes the observed event,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
A set of results having the noted event as the element regarded as the same and a set of causes having the cause event of each element of the set of the elements as the elements are generated, and the set of results is associated with the set of causes. A method for generating a causal relationship model, characterized in that: A causal relationship model generation method for generating a causal relationship between an observed event of interest and a causal event that causes the observed event,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
A set of results having the noted event as the element regarded as the same and a set of causes having the cause event of each element of the set of the elements as the elements are generated, and the set of results is associated with the set of causes. With
In the result set, when there is an element serving as a cause event of another noticeable event, it is necessary to associate the result set with a cause set associated with the result set to which the other noticeable event belongs. A method of generating a causal relationship model as a feature 3. The attention event which is an element of the result set is aggregated as one data including at least one other attention event which has an inclusion relationship with the attention event. Causal model generation method. A causal relationship model generation device that generates a model related to a causal relationship between an observed event of interest and a causal event that causes the observed event,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
Means for generating a set of results having the noted event as an element regarded as the same, and a set of causes having the cause event of each element of the set of results as an element;
Means for associating the set of results with the set of causes,
A causal relationship model generation device characterized by comprising: A causal relationship model generation device that generates a model related to a causal relationship between an observed event of interest and a causal event that causes the observed event,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
Means for generating a set of results having the noted event as an element regarded as the same, and a set of causes having the cause event of each element of the set of results as an element;
Means for associating the set of results with the set of causes,
Means for associating the set of results with the set of causes associated with the set of results to which the other noteworthy event belongs, when the set of results includes an element that is a cause event of the other noteworthy event; ,
A causal relationship model generation device characterized by comprising: 6. The attention event which is an element of the result set is aggregated as one data including at least one other attention event which has an inclusion relationship with the attention event. Causal model generator. A cause estimating method for estimating a cause of an observation event observed from the observation target, based on a causal relationship between the observed event of interest and the cause event that causes the observation event,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
A set of results having the noted event as the element regarded as the same and a set of causes having the cause event of each element of the set of the elements as the elements are generated, and the set of results is associated with the set of causes. Based on the causal relationship model generated by the search, the set of results having the observed event as an element is searched, and the set of causes associated with the set of searched results is searched for. And cause estimation method. A cause estimating method for estimating a cause of an observation event observed from the observation target, based on a causal relationship between the observed event of interest and the cause event that causes the observation event,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
A set of results having the noted event as the element regarded as the same and a set of causes having the cause event of each element of the set of the elements as the elements are generated, and the set of results is associated with the set of causes. In addition, when the set of results includes an element serving as a causal event of another event of interest, the set of results is associated with the set of causes associated with the set of results to which the other event of interest belongs. Based on the causal relationship model generated by
Searching the set of results with the observed event as an element, searching for the set of causes associated with the set of searched results,
A cause estimating method, further comprising, when the set of causes is associated with the searched set of causes, further searching the set of causes associated with the set of results. 9. The attention event that is an element of the result set is aggregated as one data including at least one other attention event that has an inclusion relationship with the attention event. Estimation method of the cause. A cause estimating device for estimating a cause of an observation event observed from the observation target, based on a causal relationship between a target event of the observation target and a cause event causing the observation,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
A set of results having the noted event as the element regarded as the same and a set of causes having the cause event of each element of the set of the elements as the elements are generated, and the set of results is associated with the set of causes. First search means for searching the result set having the observed event as an element based on the causal relation model generated by the first search means;
A second search unit that searches for the set of causes associated with the set of the searched results based on the causal relation model;
A cause estimating device comprising: A cause estimating device for estimating a cause of an observation event observed from the observation target, based on a causal relationship between a target event of the observation target and a cause event causing the observation,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
A set of results having the noted event as the element regarded as the same and a set of causes having the cause event of each element of the set of the elements as the elements are generated, and the set of results is associated with the set of causes. In addition, when the set of results includes an element serving as a causal event of another event of interest, the set of results is associated with the set of causes associated with the set of results to which the other event of interest belongs. First search means for searching the result set having the observed event as an element based on the causal relation model generated by the first search means;
Second search means for searching the set of causes associated with the set of searched results based on the causal relation model;
With
The second search means, when the set of results is associated with the searched set of causes, further searches for the set of causes associated with the set of results. Characteristic cause estimating device. The attention event which is an element of the result set is aggregated as one data including at least one other attention event which has an inclusion relationship with the attention event. Cause estimating device. A data structure of a causal relationship model representing a causal relationship between an observed event of interest and a causal event that causes the observed event,
The event of interest or the cause event, the inclusion relationship between the component of the observation target and other components of the observation target related to the event, or the conceptual inclusion relationship of the event, hierarchically Characterized by being represented as data with the represented attributes,
A data structure, wherein a set of results having the noted event as an element regarded as the same and a set of causes having the cause event of each element of the result set as an element are associated with each other. In the result set, when there is an element serving as a cause event of another noticeable event, the result set is associated with the cause set associated with the result set to which the other noticeable event belongs. 14. The data structure according to claim 13, wherein: 14. The data according to claim 13, wherein an attention event that is an element of the result set is aggregated as one data including at least one other attention event that has an inclusion relationship with the attention event. Construction.

Images