JP3702146B2 - Quality control apparatus and quality control method - Google Patents

Description

【００２４】
ｑ_nは稼動データもしくは稼動データにおけるモデルの出力値である。該出力値について、２種類の感度Δｑ，δｑを考えている。これらΔｑ，δｑは、それぞれ規準データとの誤差および工程における時間的変動分（図３参照）を示す。ｑ_dは望みの値である。通常、品質のずれは工程を流れる被加工品の個体差と、工程そのものの変動という二つの原因から起こると考えられる。そして、Δｑ，δｑはそれぞれこれら二つの原因に対応する観測データとなる。本発明は、加工品のデータの規準値からの誤差分として単純に認識するのではなく、誤差が、各被加工品の個体差と各工程の変動とから成り立つと考えて変動の修正を行なうことを特徴とするものである。
【００２５】
感度は変数の形式で表されるが、これは稼動データ近傍における、あくまでモデルにしたがった各工程の変化分であって、必ずしも実際の工程の変動数値そのものを表すものではないことに注意する。尚、実際の感度データが観測できる場合は実データを式（１），（２）に代入して算出してもよいことは勿論である。
【００２６】
特に、定性的なモデルが採用された工程モデルにおいては、感度はその数値自体に意味があるのではなく、数値が所定の指示内容のどれに該当し、又は切り替わるかを表現するためのものである。従来、定性推論などとして定性的な変数を扱う技術は知られているが、これは条件データに基づき論理的条件による帰結を導出するものであって、パラメータ修正のために状態変化信号を利用する類のものではない。
【００２７】
モデルデータ記憶装置１１は、各工程の定性的又は定量的モデルを記憶するものであり、稼動データに対応する工程のデータをモデルに従って算出する。この場合、数学的モデルが用いられ得るが、定性的に論理記号によって整理される条件がモデルとして用いられてもよい。例えば、図４に示すような横軸を時間、縦軸を値として表される動特性を、数学的モデルを用いることなく、横軸下部に示すような０，１，２，＋，−，Ｓ，Ｎ，ｑなどの記号類からなるモデルによって表現することもできる。ちなみに、図４に示される「０」は０次感度を、「１」は１次感度を、「２」は２次感度を意味し、それぞれ横軸に対して平行な直線、傾きを有する直線、単調増加する曲線を想定している。
【００２８】
図５は変化の伝播を示す図である。同図において、Ｓ_ｉ，Ｐ_ｉは、それぞれ、工程ｉの状態、可調節パラメータのことであり、ｑ_ｎは被加工品ｎの（工程ｉ、時刻ｔについての）稼動データの規準データとの誤差分のことである。図５は、パラメータの変動が品質の変動に結び付くこと（図５の紙面最左端に示す工程では一つの規準線であるのに対し最右端側の工程では二つの直線になっていることは、各工程でのパラメータと入力される品質の違いに依存していることを示す）を示すとともに、被加工品の基準とのずれがこれまでの工程に依存するならば、後に続く被加工品ｑ_ｎ＋１，・・・にずれをもたらすのでこれまでの工程のパラメータｐ_ｉ・ｐ_ｉ−１，・・・を修正する必要があること（Ｂｅｈｉｎｄ）を示す。また、現在のΔｑ_ｎのずれを除去するためには先の工程でのパラメータｐ_ｉ＋１，ｐ_ｉ＋２，・・・の修正が必要であること（Ａｈｅａｄ）を示している。すなわち、Δｑ_ｎを元にしたパラメータ修正の必要性の伝播を表している。具体的なパラメータ調節のための修正量算出は後述する調節指示生成装置１４により行われる。
【００２９】
規準データ記憶装置１３は、各工程における稼動データ（被加工品の工程での加工によるデータ）の許容範囲と、処理システム全体を通しての許容範囲とをあらかじめ記憶するものである。
【００３０】
例えば、各工程の許容範囲の記述形式としては、各工程において、
【００３１】
【数２】

【００３２】
などとする。すなわち、加工品の稼動データ誤差Δｑ_ｎ（ベクトル量でも構わないが、ここでは簡単のためにスカラー量とする）に対して、数値で示される範囲や、所定の特性を示すラベル（“ＬＯＷ”）で表されるものとして記述される尚、工程により変数の意味合いが独立して設定されても良い。各工程の許容範囲の各々が最終チェック条件を意味する場合は特に区別を行わなくてもよいが、最終チェックとして一部の規準が満たされればよいという場合は区別して記述しておく。すなわち、最終チェック条件として、
【００３３】
【数３】

【００３４】
という具合に、各工程の規準とは別に記憶しておく。
【００３５】
調節指示生成装置１４は、規準データ記憶装置１３に記憶されている規準データに基づいて、調節すべき工程に関し、パラメータ修正量を算出するものである。感度算出装置１２により各工程の状況についての定性的又は定量的変化分が算出され、その感度データを用いて、工程の調節可能なパラメータの修正量（Δｐ_ｉ（ｔ））が算出される。
【００３６】
工程の感度は、一般に、工程の特性を示す関数の、現在の稼動データ近傍におけるパラメータの寄与による変化分で形式的には表現される。いま、工程ｉにおける品質を決定づける関数を工程の状態Ｓｉによってq(t) = S_i(q(t-1), p(t))と仮定されるものする。
【００３７】
１．工程ではパラメータによる変動と、被加工品そのものの変動分（Δｑ_ｎ（ｔ）：前工程などで発生したもの）が工程に入力されたものとすると、その変動分による影響は、Ｓ_ｉ（ｑ（ｔ−１），ｐ（ｔ））を、望みの値ｑ_ｄ（ｔ−１）および現在のパラメータｐ（ｔ−１）の近傍で線形近似することによって得られる。
【００３８】
すなわち、
【数４】

【００３９】
上式（１）により、
【数５】

【００４０】
ここで、Ａ（＝∂Ｓ_ｉ／∂ｑ）は工程の被加工品の稼動データｑに対する変動割合を示し、Ｂ（＝∂Ｓ_ｉ／∂ｐ）は工程の状態や調節パラメータによる品質データの変動割合を表す。
【００４１】
２．工程そのものが処理特性の変動を持ち（ｄｑ≡ｑ−ｑ_０；ｑ_０：ある規準ポイント）、比較する被加工品のずれがある場合で、なおかつ、その工程のパラメータの寄与（ｄｐ）をも考慮して線形近似して変動分を算出すると、上式（６）と同様に、
【数６】

【００４２】
であり、被加工品稼動データにおいて、二つの被加工品の変動分の差は次式で表される。
【数７】

【００４３】
ここで、二つの被加工品での工程は同じ工程ｉである。また、一時刻ずれた工程処理においてＡ’とＢ’とＣ’とはそれぞれほぼ同じ値をとるものと仮定としている。上式（２）により、
【数８】

【００４４】
が得られる。さらに、パラメータの変動の差を品質の変動の差と同様にδｐで表すと、次式（１３）のようになる。
【００４５】
【数９】

【００４６】
ここで、δｑ（ｔ）は規準からの誤差ではなく、工程の被加工品の相対差分であることが前者との違いである。つまり、上の二つの式は、稼動データに関して、それぞれ、
１．絶対誤差（規準値からのずれ分）
２．相対誤差（一つ前の製品の品質とのずれ分）
を計算する式となっている。相対誤差を零に近付けてもそれが即ち許容範囲に近付くというわけではない。後者は工程の品質の均一化の評価になっている。本発明の特徴をこれらの言葉を用いて繰り返し説明すると、発明の特徴は、単に絶対誤差を小さくして加工品を個別に評価し、処理システムを調節するのではなく、加工の均一性も評価しながら各工程を調節するところにある。
【００４７】
加工の均一性ということに関し、いわゆる品質工学における標準偏差（分散）を小さくするという考え方があるが、これは、誤差全体に対しての自乗平均などの統計処理に基づくものであり線形システムには有効な考え方である。しかし、実際の工程は非線形特性を持ち、また時間的にも特性が変動するなどということもあり、このような、線形、時不変の対象に対する従来手法では十分ではない。
【００４８】
上式（６）および（１３）によって与えられる感度データを用いると、パラメータの修正量は次式にて算出される。
時刻ｔに工程ｉを流れる被加工品Δｑ_ｎ（ｔ）を時刻ｔにおいて０にするためのパラメータ修正には、上式（６）において、
【００４９】
【数１０】

【００５０】
となるようなパラメータ修正量Δｐを定めれば良いから、パラメータ修正量は次式（１５）のように表される。
【００５１】
【数１１】

【００５２】
次に、被加工品間のずれを工程で打ち消すためには、上式（１３）において、
【００５３】
【数１２】

【００５４】
が成り立つことが必要であるから、そのためのパラメータ修正分は次式で表される。
【００５５】
【数１３】

【００５６】
上式（１５）は、現在の被加工品のずれ分そのものを調整するものであり、上式（１７）は、他の工程の被加工品との相対的なずれ分を調整するためのパラメータ修正を与えるものである。しかしながら、いずれも、現被加工品のずれ分をこれから加工する工程で調整するものである（Ａｈｅａｄ）。
【００５７】
これらの算出においては、Ａ，Ｂ，Ｃ，Ａ’，Ｂ’が必要になるが、事前にモデルを精密に仮定しておくか、事例データを集めておくか、あるいは、実稼動中にテスト入力を印加して実際に調べるかによって情報を得ておき、利用することができる。
【００５８】
上式（１５），（１７）は、工程前の被加工品のずれ分を用いて工程後のずれをなくす調整のための修正式であるが、工程後に結果的に発生したずれ分に基づいて、次に当該工程に流れてくる被加工品のための準備としてのパラメータの微調整を行うこともできる。すなわち、上式（６），（１３）の左辺についても実データを用いて、次のようにパラメータ修正を施す。なお、パラメータ修正時刻がｔ＋１であることに注意されたい。
【００５９】
【数１４】

【００６０】
これらの調整は、後に流れてくる被加工品のために、工程を微調整するものである（Ｂｅｈｉｎｄ）。
【００６１】
上式（１５），（１７）（（１８），（１９））においては、微調整のための小さい正の数の係数を乗じてもよい。また、ここでは、規準に到達しているかどうかの判別と修正量計算処理との関係には特に言及しなかったが、例えば、規準内にあると判別される場合は修正量計算を行なわない、などの判別機構を備えて処理の効率化を図ってもよい。
【００６２】
次に、本発明の変形例を説明する。
【００６３】
図６は本発明の変形例に係る品質制御装置の概略構成を示すブロック図である。図６に示すように、感度算出装置１０１が、上述したモデルデータ記憶装置１１を利用せずに各工程独自で試験信号を印加する感度試験装置１０２を内部に備えており、該各工程の感度データを算出するものとなっている。感度データは、ある意味では工程の少し未来を示すデータであると言えるが、これを上述したようにモデルを使うのではなく、実際の工程調整段階において少しだけ試験を行って実際のデータを取得するというものである。例えば、工程ｉに関する可調整パラメータを少し変動（ΔＰ_ｉ）させて得られる観測誤差をΔｑ，δｑとして得ることにより、本格的な調節をおこなうためのパラメータ修正量であるΔｐ，δｐを算出するためのデータとする。
【００６４】
以上説明したように、本発明の実施形態によれば、品質のずれに関して、規準との誤差のみならず各被加工品の実データにおける相対誤差をも調整可能となる。また、各被加工品単独にずれを調整するのではなく、次に流れる被加工品のために工程の微調整を行うために、区分的かつ定性表現を含んだ線形モデルを利用してパラメータ修正量を各工程について算出するようにしているので、複雑な相関をもつプロセスにおいても、各被加工品の変動を後々の工程で調整できるだけのみならず、各工程の変動をこれまでの被加工品の誤差データを利用して調整できるようになる。さらに、定性表現と適応的線形モデルとを結び付けることできめ細かな調整が可能になる。
【００６５】
すなわち、従来、自動化できなかったような、したがって、人手で処理してきたような、複雑な工程が相互に関連している処理システムに対する効率的なパラメータ修正が行なえる。
【００６６】
また、調整のためのデータは各工程毎に独立に算出されるので、互いに遠隔地にある工程間での処理制御にも適用できる。
【００６７】
なお、本発明は上述した実施形態に限定されず種々変形して実施可能である。
【００６８】
例えば本発明は、工場の製造現場だけではなく、いわゆるサプライチェーンマネージメントに適用可能である。上述した各工程について次のような式を想定することで、サプライチェーンマネージメントは複数工程からなるネットワークシステムであると考えられる。
【００６９】
ｖ（ｙ，ｃ）＝（ｆ（ｔ−Ｔ，ｘ，Ｓ），ｇ（ｔ−Ｔ，ｘ，Ｓ））
ｖ：出力（供給量・品質やコスト）
ｙ：供給量・品質
ｃ：コスト
ｆ：出力を決定付けるモデル
ｇ：コストを決定付けるモデル
Ｔ：サプライヤ（工程）に投入された時刻
ｔ：現時刻（したがって、ｔ−Ｔは投入された時点からの経過時間を表す）
ｘ：サプライア（工程）に投入される材料・部品の数量・品質
Ｓ：サプライヤが抱える加工品ないし材料の在庫
最終段の工程の出力は一般消費者向けの商品となる。実際には、これまでの実績に基づいて書くサプライヤの能力を論理的に記述したり、簡単な微分方程式で近似的にモデル化することが応用できる。尚、これは待ち行列網のような確率的モデルを用いた統計的性能評価を目的としているものではない。
【００７０】
また、上述した実施形態では、ネットワーク状に構成される複数の工程からなるシステムを対象として説明したが、一見してネットワーク状でない一つの製品の中に構成されている複数の部品の働きにかかわるパラメータ調整などにも本発明は応用可能である。
【００７１】
【発明の効果】
以上説明したように、本発明によれば、対象の個々の要素（工程）の感度データを用いて要素間の影響関係を少しずつ（部分的に）考慮しながらずれを修正してゆくことで自律的かつ適応的な品質制御を行なう品質制御装置および品質制御方法を提供できる。
【図面の簡単な説明】
【図１】本発明の一実施形態に係る品質制御装置の概略構成、およびこの品質制御装置からの制御を受ける被制御システムの概略構成を示すブロック図
【図２】実施形態に係る工程間の結合関係の具体例を示す図
【図３】対象特性の変化分の種類を示す図
【図４】対象特性の変化分の伝播を示す図
【図５】対象特性のモデリング概要を示す図
【図６】本発明の変形例に係る品質調整装置の概略構成を示すブロック図
【符号の説明】
１…品質制御装置
２…被制御システム
１１…モデルデータ記憶装置
１２…感度算出装置
１３…規準データ記憶装置
１４…調整指示生成装置[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a quality control apparatus and a quality control method suitable for quality management in a production line, plant control, or remote maintenance and supply chain management.
[0002]
[Prior art]
In production lines and plant control in factories, in recent years, manufacturing information and operation information in each process have been managed on a computer network, and the automation rate has been improved. Recently, a new mechanism has emerged, such as a virtual factory, which simulates the operation status of each process from a remote computer and performs realistic control. Has evolved into a modern control management system.
[0003]
Similarly, regarding procurement problems including delays related to processing / warehouse-supply chain management (SCM)-etc., each supplier (supplier) is regarded as a process, so that each process has a time delay (instant shipment) It can be thought of as a network system through a computer that outputs the type and number of products.
[0004]
In these conventional systems, in order to perform control (quality control) by referring to data from a remote computer and performing simulation (quality control), one of the following methods (1) to (3) is used. Generally taken.
(1) A method in which a person judges a command to each process one by one (2) A method in which control is performed independently within a predetermined range for each process (3) An overall evaluation function is determined A method for controlling each process so that an estimate that a good evaluation value (minimum cost, maximum quality, etc.) can be obtained. First, human beings based on past cases and experiences regarding the method (1) When issuing a command after making a decision, there is no inconvenience if the operation information of each process as the basis is sufficiently available, but if the number of processes increases and the relationship between each process becomes complicated Appropriate judgment is difficult, and the validity and objectivity of the directive itself are not guaranteed. Therefore, the method (1) is unreasonable industrially depending on its scale.
[0005]
Next, with respect to (2), this method is roughly divided into a method using a model and a method not using it. First, in an indirect method that does not actively use models, process management that relies on statistical processing using case data is often performed. However, this method is based on the achievement of the overall standard obtained as a result of independently adopting each process's own standard (usually a stricter standard than necessary) in each process independently. High cost in process operation.
[0006]
Also, in each process, what is managed within the allowable range of parameters set at the initial stage of startup seems to work as well as before in the case of mass production of small varieties with similar operating conditions. In the case of multi-product small-quantity production in which the operation status is changed as needed according to the purpose, it is necessary to reset each parameter one by one, which is troublesome. In order to redo the setting manually, it is necessary to understand the details of the data while grasping the entire data, and when data increases, it takes time to determine and set the setting conditions.
[0007]
As for the method using the model, since it is necessary to consider the influence relationship between the processes in order to realize the adaptive operation, it is essentially impossible to deal with the model refinement of each process as in the past. It is coming. Various efforts have been made, such as devising the model description method (for example, condition arrangement by decision tree as a logical model, linear theory as a mathematical model, or fuzzy theory or neural network). In either case, it is still necessary to collect operational data in advance as an example or to refer to it dynamically. Therefore, the data required for considering the influence between processes tends to be enormous, and it can be said that this is a technique lacking in adaptability.
[0008]
Furthermore, there are many combinations of solutions for which process and how to adjust when correcting the final output result of the entire system and correcting it. Often, a particular one is not necessarily the best solution. This is a merit from the viewpoint of the diversity of solutions, and in particular, if the model is not assumed, it is less bound by the model, so the degree of freedom of the countermeasure is increased. When the scale increases, there is a pitfall that there is a risk that the influence relationship between the processes cannot be grasped, and that the objectivity of the operation method is easily lost. In this case, changing the operation of the system becomes extremely difficult.
[0009]
The method (3) is not particularly conscious of each of the multiple processes, but rather considers the entire system as one model or a model consisting of a small number of roughly divided subsystems and handles each process as it is. As a whole, it is assumed that simple modeling is performed. Then, an evaluation function is determined according to the purpose of operation of the system, and parameter adjustment is performed according to the evaluation value. However, since the model is too simple and the calculation from the evaluation value to the correction parameter value tends to be linear in terms of mathematical technology, the description ability is not sufficient for the behavior of the entire system. It becomes insufficient, and there is a limit in improving accuracy.
[0010]
[Problems to be solved by the invention]
In a system with a huge amount of data, the nonlinearity and unknownness of the system behavior becomes a problem. In addition, the ambiguity and specificity of the solution appear when adjustment control is performed. In summary, the above-described conventional method is difficult to perform adjustment control for such a system because of the following circumstances.
1. When modeling, it is necessary to consider time delay and buffer (inventory / standby) etc. when following multiple processes. If these models use mathematical expressions, This is impossible with an algebraic equation (a simultaneous equation), a differential equation (a simultaneous equation) must be formed, and the equation itself cannot be solved.
[0011]
2. If you try to deal with it without modeling it, the situation becomes too complicated and you can only expect it to work (by chance) by relying on experience and intuition.
[0012]
Therefore, an object of the present invention is to solve an adjustment control problem in an object having such characteristics, particularly in a large system composed of a plurality of process elements.
[0013]
More specifically, by using the sensitivity data of each target element (process) and correcting the shift while considering the influence relationship between elements little by little (partially), it is autonomous and adaptive. An object of the present invention is to provide an apparatus for performing quality control.
[0014]
[Means for Solving the Problems]
In order to solve the above problems and achieve the object, the present invention uses the following means.
[0015]
(1) In a quality control device for controlling a controlled system consisting of a plurality of processes with adjustable parameters,
A calculating means for calculating a sensitivity indicating an error from predetermined reference data and a temporal variation of processing performance in each step;
And a parameter adjusting unit that adjusts the parameter based on the sensitivity calculated by the calculating unit.
[0016]
(2) In a quality control device for controlling a controlled system consisting of a plurality of processes with adjustable parameters,
Standard data storage means for storing target quality standard data in processing in each process and target quality standard data as the whole controlled system;
Model storage means for storing a qualitative model or a quantitative model relating to processing characteristics in each step;
According to the qualitative model or quantitative model stored in the model storage means, the qualitative processing data or quantitative processing data before or after the processing on the workpiece in each step is input. Sensitivity calculation means for calculating a change from the reference data in each step;
Obtained from the sensitivity calculation means based on the reference data stored in the reference data storage means and predetermined qualitative characteristics or quantitative characteristics indicating the influence of the adjustable parameters of the respective steps on the respective steps. And an adjustment instruction generating means for outputting a correction instruction of a parameter that can be adjusted for each process corresponding to a qualitative or quantitative change of each process.
[0017]
(3) In the quality control apparatus according to (1) or (2), the sensitivity calculation unit is configured to use actual process data instead of model data stored in the model storage unit.
[0018]
(4) In a quality control method for controlling a controlled system comprising a plurality of steps with adjustable parameters,
Calculate the sensitivity that shows the error from the specified standard data and the time variation of the processing performance in each process,
A quality control method, wherein the parameter is adjusted based on the sensitivity calculated by the calculating means.
[0019]
(Function)
The sensitivity calculation means inputs qualitative processing data and / or quantitative processing data before and after processing on the workpiece in each process, and both data, and the qualitative model or quantitative model stored in the model storage means The change from the reference data in each process is calculated according to Then, the adjustment instruction generating means calculates the sensitivity based on the reference data stored in the reference data storage means and predetermined qualitative characteristics or quantitative characteristics indicating the influence of the adjustable parameters of each process on each process. Corresponding to the qualitative or quantitative change of each process calculated by the means, an instruction to correct the adjustable parameter of each process is output.
[0020]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[0021]
FIG. 1 is a block diagram showing a schematic configuration of a quality control apparatus according to an embodiment of the present invention and a schematic configuration of a controlled system that receives control from the quality control apparatus. As shown in FIG. 1, the quality control device 1 according to the present invention includes a model data storage device 11, a sensitivity calculation device 12, a reference data storage device 13, and an adjustment instruction generation device 14. On the other hand, the controlled system 2 includes a plurality of processes. The connection relationship between processes is arbitrary depending on the system, and in this embodiment, a specific example as shown in FIG. 2 will be described. Steps S1 to Sn and steps T1 to Tn shown in FIG. 2 comprise adjustable parameters that are in actual operation. Operation data of each process of the controlled system 2 is input to the quality control apparatus 1.
[0022]
The sensitivity calculation device 12 determines two types of sensitivity for each step according to the following formulas (1) and (1) according to the input operation data and the qualitative or quantitative model of each step stored in advance in the model data storage device 11. Calculate according to 2).
[0023]
[Expression 1]

[0024]
q _n is the operation data or the output value of the model in the operation data. Two types of sensitivity Δq and δq are considered for the output value. These Δq and δq respectively indicate an error from the reference data and a temporal variation in the process (see FIG. 3). q _d is the desired value. In general, the difference in quality is considered to be caused by two causes, that is, individual differences in workpieces flowing through the process and fluctuations in the process itself. Δq and δq are observation data corresponding to these two causes, respectively. The present invention does not simply recognize the error from the standard value of the processed product data, but corrects the variation on the assumption that the error is composed of individual differences in each processed product and variations in each process. It is characterized by this.
[0025]
The sensitivity is expressed in the form of a variable, but it should be noted that this is a change in each process according to the model in the vicinity of the operation data, and does not necessarily represent the actual process fluctuation value itself. Of course, when actual sensitivity data can be observed, the actual data may be calculated by substituting into the equations (1) and (2).
[0026]
In particular, in the process model in which a qualitative model is adopted, the sensitivity is not meaningful in the numerical value itself, but expresses to which of the specified instructions the numerical value corresponds or switches. is there. Conventionally, techniques for handling qualitative variables as qualitative reasoning are known, but this derives the consequences of logical conditions based on condition data, and uses state change signals for parameter correction. It ’s not like that.
[0027]
The model data storage device 11 stores a qualitative or quantitative model of each process, and calculates process data corresponding to operation data according to the model. In this case, a mathematical model can be used, but conditions that are qualitatively arranged by logical symbols may be used as the model. For example, a dynamic characteristic represented by time on the horizontal axis and a value on the vertical axis as shown in FIG. 4 can be expressed as 0, 1, 2, +,-, as shown at the bottom of the horizontal axis without using a mathematical model. It can also be expressed by a model composed of symbols such as S, N, and q. Incidentally, “0” shown in FIG. 4 means 0th order sensitivity, “1” means primary sensitivity, and “2” means secondary sensitivity. A straight line parallel to the horizontal axis and a straight line having an inclination, respectively. Assume a monotonically increasing curve.
[0028]
FIG. 5 is a diagram showing change propagation. In the figure, S _i and P _i are the state of process i and adjustable parameters, respectively, and q _n is the reference data of the operation data (for process i and time t) of the workpiece n. It is an error. FIG. 5 shows that parameter fluctuations lead to quality fluctuations (in the process shown at the leftmost end of FIG. 5, there is one reference line, whereas in the process on the rightmost side, there are two straight lines. If the deviation from the standard of the workpiece depends on the previous process, the subsequent workpiece q _{.., n + 1} ,..., so that it is necessary to correct the parameters p _i · p _i−1 ,. In addition, it is indicated that the parameters p _{i + 1} , p _{i + 2} ,... Need to be corrected in the previous step in order to remove the current Δq _n deviation (Ahead). That is, it represents the propagation of the necessity of parameter correction based on Δq _n . The correction amount calculation for specific parameter adjustment is performed by an adjustment instruction generation device 14 described later.
[0029]
The standard data storage device 13 stores in advance an allowable range of operation data (data obtained by processing in a process of a workpiece) in each process and an allowable range throughout the entire processing system.
[0030]
For example, as the description format of the allowable range of each process, in each process,
[0031]
[Expression 2]

[0032]
And so on. In other words, the operation data error Δq _n of the processed product (which may be a vector amount, but for the sake of simplicity, is a scalar amount here), a range indicated by a numerical value, or a label (“LOW”) indicating a predetermined characteristic. Note that the meaning of the variable may be set independently depending on the process. When each of the permissible ranges of each process means a final check condition, it is not particularly necessary to make a distinction. However, when a part of the criteria only needs to be satisfied as a final check, the distinction is described. That is, as a final check condition,
[0033]
[Equation 3]

[0034]
That is, it is memorized separately from the criteria of each process.
[0035]
The adjustment instruction generating device 14 calculates a parameter correction amount for the process to be adjusted based on the reference data stored in the reference data storage device 13. The sensitivity calculation device 12 calculates a qualitative or quantitative change for each process state, and uses the sensitivity data to calculate a parameter adjustment amount (Δp _i (t)) that can be adjusted for the process.
[0036]
In general, the sensitivity of a process is formally expressed by a change of a function indicating a characteristic of the process due to a parameter contribution in the vicinity of the current operation data. Now, the function that determines the quality in the process i is assumed to be q (t) = S _i (q (t−1) , p (t) ) depending on the process state Si.
[0037]
1. A variation due to the parameter in step, variation in the workpiece itself: If ([Delta] q n _(t) before step that occurs, for example) is assumed to have been entered in step, the effect of the variation, S _i ( q (t−1), p (t)) is obtained by linear approximation in the vicinity of the desired value q _d (t−1) and the current parameter p (t−1).
[0038]
That is,
[Expression 4]

[0039]
From the above equation (1),
[Equation 5]

[0040]
Here, A (= ∂S _i / ∂q) indicates a variation rate with respect to the operation data q of the workpiece in the process, and B (= ∂S _i / ∂p) indicates the quality data according to the process state and the adjustment parameter. Represents the rate of change.
[0041]
2. The process itself has a variation in processing characteristics (dq≡q−q ₀ ; q ₀ : a certain reference point), and there is a deviation of the workpiece to be compared, and also the contribution (dp) of the parameter of the process When the fluctuation is calculated by linear approximation taking into account, like the above equation (6),
[Formula 6]

[0042]
In the workpiece operation data, the difference between the fluctuations of the two workpieces is expressed by the following equation.
[Expression 7]

[0043]
Here, the steps for the two workpieces are the same step i. In addition, it is assumed that A ′, B ′, and C ′ take almost the same values in the process processing shifted by one time. From the above equation (2),
[Equation 8]

[0044]
Is obtained. Further, when the difference in parameter variation is expressed by δp in the same way as the difference in quality variation, the following equation (13) is obtained.
[0045]
[Equation 9]

[0046]
Here, the difference from the former is that δq (t) is not an error from the standard but a relative difference of the workpieces in the process. In other words, the above two formulas are respectively related to operational data.
1. Absolute error (deviation from standard value)
2. Relative error (deviation from previous product quality)
It is a formula to calculate. Approaching the relative error to zero does not mean that it is close to an acceptable range. The latter is an evaluation of uniform process quality. The features of the present invention will be described repeatedly using these terms. The features of the present invention are not only to evaluate the workpiece individually by reducing the absolute error, but also to evaluate the processing uniformity, not to adjust the processing system. While adjusting each process.
[0047]
Regarding the uniformity of processing, there is the idea of reducing the standard deviation (variance) in so-called quality engineering, but this is based on statistical processing such as root mean square for the entire error. This is an effective way of thinking. However, the actual process has non-linear characteristics and the characteristics fluctuate with time, and the conventional method for such linear and time-invariant objects is not sufficient.
[0048]
When the sensitivity data given by the above equations (6) and (13) is used, the parameter correction amount is calculated by the following equation.
In the parameter correction for setting the workpiece Δq _n (t) flowing through the process i at time t to 0 at time t, in the above equation (6),
[0049]
[Expression 10]

[0050]
Therefore, the parameter correction amount is expressed as the following equation (15).
[0051]
[Expression 11]

[0052]
Next, in order to cancel the deviation between the workpieces in the process,
[0053]
[Expression 12]

[0054]
Therefore, the parameter correction for that is expressed by the following equation.
[0055]
[Formula 13]

[0056]
The above equation (15) is for adjusting the deviation amount of the current workpiece itself, and the above equation (17) is a parameter for adjusting the relative deviation amount from the workpiece in other steps. It gives a correction. However, in both cases, the deviation of the current workpiece is adjusted in the process to be performed (Ahead).
[0057]
In these calculations, A, B, C, A ', and B' are required. Precisely assume the model in advance, collect case data, or test during actual operation. Information can be obtained and used depending on whether an input is actually applied and actually examined.
[0058]
The above formulas (15) and (17) are correction formulas for adjustment to eliminate the shift after the process using the shift of the workpiece before the process, but based on the shift generated as a result after the process. Then, it is possible to finely adjust the parameters as preparation for the workpiece to be flowed to the process next. That is, parameter correction is performed on the left side of the above equations (6) and (13) using actual data as follows. Note that the parameter modification time is t + 1.
[0059]
[Expression 14]

[0060]
These adjustments fine-tune the process for the workpiece that flows later (Behind).
[0061]
In the above equations (15), (17) ((18), (19)), a small positive number coefficient for fine adjustment may be multiplied. In addition, here, no particular mention was made of the relationship between the determination of whether or not the criterion has been reached and the correction amount calculation process, but for example, if it is determined that the criterion is within the criterion, the correction amount calculation is not performed. Such a determination mechanism may be provided to improve processing efficiency.
[0062]
Next, a modified example of the present invention will be described.
[0063]
FIG. 6 is a block diagram showing a schematic configuration of a quality control apparatus according to a modification of the present invention. As shown in FIG. 6, the sensitivity calculation apparatus 101 includes a sensitivity test apparatus 102 that applies a test signal independently for each process without using the model data storage device 11 described above, and the sensitivity of each process. Data is to be calculated. Sensitivity data can be said to be data that shows a little future of the process in a sense, but instead of using the model as described above, actual data is obtained by performing a little test at the actual process adjustment stage. It is to do. For example, in order to calculate Δp and δp, which are parameter correction amounts for performing full-scale adjustment, by obtaining, as Δq and δq, observation errors obtained by slightly varying (ΔP _i ) the adjustable parameters relating to step i. Data.
[0064]
As described above, according to the embodiment of the present invention, it is possible to adjust not only the difference from the standard but also the relative error in the actual data of each workpiece regarding the quality deviation. Also, parameter adjustments are made using a linear model that includes piecewise and qualitative expressions in order to fine-tune the process for the next flowing work piece, rather than adjusting the deviation for each work piece alone. Since the quantity is calculated for each process, not only can the fluctuation of each workpiece be adjusted in the subsequent process, but also the fluctuation of each process until the past workpiece even in a process with complicated correlation It becomes possible to adjust using the error data. Furthermore, fine adjustment is possible by linking the qualitative expression and the adaptive linear model.
[0065]
That is, it is possible to efficiently perform parameter correction for a processing system in which complicated processes are related to each other, which cannot be automated in the past, and thus has been processed manually.
[0066]
In addition, since the data for adjustment is calculated independently for each process, it can be applied to process control between processes at remote locations.
[0067]
The present invention is not limited to the above-described embodiment, and can be implemented with various modifications.
[0068]
For example, the present invention can be applied not only to a factory manufacturing site but also to so-called supply chain management. By assuming the following formula for each process described above, supply chain management is considered to be a network system consisting of a plurality of processes.
[0069]
v (y, c) = (f (t−T, x, S), g (t−T, x, S))
v: Output (supply quantity, quality and cost)
y: Supply quantity / quality c: Cost f: Model that determines output g: Model that determines cost T: Time when input to the supplier (process) t: Current time (accordingly, t-T is from the time of input) Represents elapsed time)
x: Quantity and quality of materials and parts to be input to the supplier (process) S: The output of the final stage of the processed product or material stock held by the supplier is a product for general consumers. Actually, it is possible to apply the logical description of the ability of the supplier to write based on the past achievements, or approximate modeling with a simple differential equation. This is not intended for statistical performance evaluation using a stochastic model such as a queuing network.
[0070]
In the above-described embodiment, a system composed of a plurality of processes configured in a network shape has been described. However, it is related to the function of a plurality of components configured in one product that is not network-shaped at first glance. The present invention can also be applied to parameter adjustment and the like.
[0071]
【The invention's effect】
As described above, according to the present invention, by using the sensitivity data of each target element (process), the deviation is corrected while considering the influence relationship between the elements little by little (partly). A quality control device and a quality control method for performing autonomous and adaptive quality control can be provided.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a schematic configuration of a quality control apparatus according to an embodiment of the present invention, and a schematic configuration of a controlled system that receives control from the quality control apparatus. FIG. Fig. 3 shows a specific example of the connection relationship. Fig. 3 shows the type of change in the target property. Fig. 4 shows the propagation of the change in the target property. Fig. 5 shows the modeling outline of the target property. 6 is a block diagram showing a schematic configuration of a quality adjustment apparatus according to a modification of the present invention.
DESCRIPTION OF SYMBOLS 1 ... Quality control apparatus 2 ... Controlled system 11 ... Model data storage apparatus 12 ... Sensitivity calculation apparatus 13 ... Standard data storage apparatus 14 ... Adjustment instruction generation apparatus

Claims (4)

In a quality control device for controlling a controlled system that processes a workpiece by a plurality of processes with adjustable parameters,
According to a qualitative or quantitative model of each process , a first calculation means for calculating a first sensitivity representing a temporal variation of an error from predetermined reference data of the operation data of the workpiece in each process;
A second calculating means for calculating a second sensitivity representing a temporal variation of the quality data of the workpiece;
A quality control apparatus comprising: a parameter adjustment unit that adjusts the parameter so that the first and second sensitivities calculated by the first and second calculation units approach zero. In a quality control device for controlling a controlled system consisting of a plurality of processes with adjustable parameters,
Reference data storage means for storing target quality reference data in processing in each process and target quality reference data for the entire controlled system;
Model storage means for storing a qualitative model or a quantitative model of characteristic changes including temporal fluctuations regarding processing characteristics in each step;
According to the qualitative model or quantitative model stored in the model storage means, the qualitative processing data or quantitative processing data before or after the processing on the workpiece in each step is input. Sensitivity calculation means for calculating a change from the reference data in each step;
Obtained from the sensitivity calculation means based on the reference data stored in the reference data storage means and predetermined qualitative characteristics or quantitative characteristics indicating the influence of the adjustable parameters of the respective steps on the respective steps. A quality control apparatus comprising: adjustment instruction generation means for outputting a correction instruction for a parameter that can be adjusted for each process corresponding to a qualitative or quantitative change in each process.   3. The quality control apparatus according to claim 2, wherein the sensitivity calculation unit is configured to use data including temporal variation in an actual process instead of the model data stored in the model storage unit. In a quality control method for controlling a controlled system for processing a workpiece by a plurality of processes with adjustable parameters,
According to a qualitative or quantitative model of each process , a first calculation step of calculating a first sensitivity representing a temporal variation in error from predetermined reference data of the operation data of the workpiece in each process;
A second calculation step of calculating a second sensitivity representing a temporal variation of the quality data of the workpiece;
A quality control method comprising: a parameter adjustment step for adjusting the parameter so that the first and second sensitivities calculated in the first and second calculation steps approach zero.

Images