JPH04361815A - Abnormality diagnosing device for rolling-down device in rolling stages - Google Patents
Abnormality diagnosing device for rolling-down device in rolling stagesInfo
- Publication number
- JPH04361815A JPH04361815A JP3134649A JP13464991A JPH04361815A JP H04361815 A JPH04361815 A JP H04361815A JP 3134649 A JP3134649 A JP 3134649A JP 13464991 A JP13464991 A JP 13464991A JP H04361815 A JPH04361815 A JP H04361815A
- Authority
- JP
- Japan
- Prior art keywords
- rolling
- normal
- abnormal
- transfer function
- control device
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000005096 rolling process Methods 0.000 title claims abstract description 53
- 230000005856 abnormality Effects 0.000 title claims abstract description 35
- 230000002159 abnormal effect Effects 0.000 claims abstract description 55
- 230000001537 neural effect Effects 0.000 claims abstract description 23
- 238000003745 diagnosis Methods 0.000 claims abstract description 12
- 230000006870 function Effects 0.000 claims description 69
- 230000009467 reduction Effects 0.000 claims description 30
- 238000004364 calculation method Methods 0.000 claims description 15
- 238000005070 sampling Methods 0.000 abstract 1
- 230000007246 mechanism Effects 0.000 description 15
- 238000013528 artificial neural network Methods 0.000 description 13
- 229910000831 Steel Inorganic materials 0.000 description 7
- 239000010959 steel Substances 0.000 description 7
- 239000008186 active pharmaceutical agent Substances 0.000 description 4
- 230000008859 change Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000000034 method Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 230000002542 deteriorative effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
Landscapes
- Testing Or Calibration Of Command Recording Devices (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Control Of Metal Rolling (AREA)
Abstract
Description
【０００１】0001
【産業上の利用分野】本発明は、圧延工程等で用いられ
る圧下制御装置の各部位ごとの一連の制御信号により伝
達関数の同定を行い、その伝達関数から該圧下制御装置
の異常を診断するシステムに関するものである。[Industrial Application Field] The present invention identifies a transfer function using a series of control signals for each part of a reduction control device used in a rolling process, etc., and diagnoses an abnormality in the reduction control device from the transfer function. It's about systems.
【０００２】0002
【従来の技術】圧延機の圧下制御装置の異常を早期に正
確に検出して、その異常部位を交換する事で圧延機を安
定的に稼働させる事が圧延製品の板厚精度向上に必要な
ことである。このため、従来は、圧下制御装置の各部位
の制御信号をペンレコ−ダやオシロスコ−プに出力し、
オペレ−タがそれを見ながら経験に基づいて圧下制御装
置の異常診断を行っていた。しかし、この様な手法では
、診断するオペレ−タにより個人差があり、圧延製品の
板厚精度を高位に安定して維持することが出来ない。[Prior Art] It is necessary to quickly and accurately detect abnormalities in the reduction control device of a rolling mill and replace the abnormal parts to ensure stable operation of the rolling mill in order to improve the thickness accuracy of rolled products. That's true. For this reason, conventionally, control signals for each part of the reduction control device were output to a pen recorder or oscilloscope.
The operator was looking at this and diagnosing an abnormality in the reduction control device based on his experience. However, with this method, there are individual differences depending on the operator making the diagnosis, and it is not possible to stably maintain a high level of accuracy in the thickness of the rolled product.
【０００３】この対策として、自動的に圧下制御装置の
異常を検出する方法として、例えば、特開昭６１−２７
１１０号公報に提示のように、圧下装置の各部位の入力
信号と出力信号から伝達関数を同定し、設定値に対する
同定した伝達関数の偏差（変化量）を求め、この変化量
が基準値を越えた場合には、その部位に異常があると診
断する手法がある。[0003] As a countermeasure to this problem, a method for automatically detecting an abnormality in the reduction control device is disclosed in, for example, Japanese Patent Laid-Open No. 61-27.
As disclosed in Publication No. 110, the transfer function is identified from the input signal and output signal of each part of the screw down device, the deviation (amount of change) of the identified transfer function with respect to the set value is determined, and this amount of change is determined from the reference value. If it exceeds the limit, there is a method to diagnose that there is an abnormality in that area.
【０００４】0004
【発明が解決しようとする課題】しかし、圧延機で圧延
する鋼板の圧下量，温度等々の圧延条件により、圧延ロ
−ルに働く圧延反力が異なる結果、圧下制御装置が正常
な状態にあるにも係わらず、同定した伝達関数の値が変
動することから、同定した伝達関数の値と設定値との差
（変化量）が大きくなって基準値を越え、圧下制御装置
が異常であると診断する場合がある。この為、基準値の
設定にあたっては、圧下制御装置における種々の異常状
態の際の伝達関数の値、及び圧下制御装置における種々
の圧延条件下での伝達関数の値を求めておき、これを基
にして前記設定値及び基準値を決定しなければならず、
非常に煩雑である。[Problem to be Solved by the Invention] However, the rolling reaction force acting on the rolling rolls varies depending on the rolling conditions such as the amount of reduction and temperature of the steel plate rolled by the rolling mill, and as a result, the reduction control device is not in a normal state. Nevertheless, since the value of the identified transfer function fluctuates, the difference (amount of change) between the identified transfer function value and the set value becomes large and exceeds the reference value, indicating that the reduction control device is abnormal. May be diagnosed. Therefore, in setting the reference value, the values of the transfer function under various abnormal conditions in the reduction control device and the values of the transfer function under various rolling conditions in the reduction control device are determined, and these are used as a basis. The set value and reference value shall be determined by
It's very complicated.
【０００５】本発明は、ニュ−ラルネットワ−クを用い
て精度良く、しかも、簡単な装置で自動的に圧延ロ−ル
の圧下制御装置の異常を診断することを課題とする。An object of the present invention is to use a neural network to accurately and automatically diagnose abnormalities in a reduction control device for rolling rolls with a simple device.
【０００６】[0006]
【課題を解決するための手段】本発明は前記課題を解決
するためになされたものであり、その手段（１）は、圧
延機の圧延ロ−ルの圧下位置調整を行う圧下制御装置の
異常診断装置において、前記圧延ロ−ルの圧下制御装置
の制御信号を所定周期毎に入力して時系列項を有する伝
達関数を同定する伝達関数同定部と、予め伝達関数の時
系列項と圧下制御装置の正常強度値及び異常強度値の関
係を学習記憶し、この学習記憶情報と前記伝達関数同定
部で同定した伝達関数の時系列項をパタ−ン比較して正
常強度値及び異常強度値を演算する正常・異常強度演算
ニュ−ラルネット部と、正常・異常強度演算ニュ−ラル
ネット部で演算した正常強度値と異常強度値により前記
圧下制御装置の正常・異常を判定する正常・異常判定部
と、を有するものである。更に、手段（２）は、圧延機
の圧延ロ−ルの圧下位置調整を行う圧下制御装置の異常
診断装置において、前記圧延ロ−ルのドライブサイドと
ワ−クサイドに各々独立して設けた圧下制御装置の制御
信号を所定周期毎に入力して時系列項を有する伝達関数
を各々同定する伝達関数同定部と、予め前記両伝達関数
の時系列項と両圧下制御装置の正常強度値及び異常強度
値の関係を学習記憶し、この学習記憶情報と前記伝達関
数同定部で同定した両伝達関数の時系列項をパタ−ン比
較して前記両サイドの圧下制御装置の正常強度値及び異
常強度値を各々演算する正常・異常強度演算ニュ−ラル
ネット部と、正常・異常強度演算ニュ−ラルネット部で
演算した前記各正常強度値と異常強度値により前記圧下
制御装置の正常・異常を各々判定する正常・異常判定部
と、を有するものである。[Means for Solving the Problems] The present invention has been made to solve the above-mentioned problems, and the means (1) is to solve an abnormality in a rolling control device that adjusts the rolling position of a rolling roll of a rolling mill. In the diagnostic device, a transfer function identification unit inputs a control signal of the roll reduction control device of the rolling roll at predetermined intervals to identify a transfer function having a time series term; The relationship between the normal intensity value and the abnormal intensity value of the device is learned and memorized, and the normal intensity value and the abnormal intensity value are determined by comparing the pattern of this learning memory information and the time series term of the transfer function identified by the transfer function identification section. a normal/abnormal strength calculation neural net section for calculating, and a normal/abnormality determination section for determining whether the reduction control device is normal or abnormal based on the normal strength values and abnormal strength values calculated by the normal/abnormal strength calculation neural net section. . Further, the means (2) is an abnormality diagnosis device for a rolling control device that adjusts the rolling position of a rolling roll of a rolling mill, wherein the rolling mill is provided with rolling rolls independently provided on a drive side and a work side of the rolling roll. a transfer function identification unit that inputs control signals of the control device at predetermined intervals and identifies transfer functions each having a time series term; The relationship between the intensity values is learned and memorized, and this learned and memorized information is compared with the time series terms of both transfer functions identified by the transfer function identification section to determine the normal intensity value and abnormal intensity of the reduction control device on both sides. A normal/abnormal strength calculation neural net unit calculates the values, and a normal/abnormal strength value is calculated by the normal/abnormal strength calculation neural net unit to determine whether the reduction control device is normal or abnormal, respectively. It has a normality/abnormality determination section.
【０００７】[0007]
【作用】本発明の作用を図１〜図４を参照して説明する
。本発明者等は、伝達関数から精度良く圧下制御装置の
正常・異常を診断するため種々検討した結果、圧延ロ−
ル（ワ−クロ−ルＷＲとバックアップロ−ルＢＲ）で鋼
板Ｓを圧延した際、圧延ロ−ルに働く反力は圧延する鋼
板Ｓの温度，材質（成分）等により異なるので、圧延ロ
−ルの間隙を調整する圧下制御装置（油圧圧下装置１Ｗ
，１Ｄ，ＡＧＣ制御部２Ｗ，２Ｄ，サ−ボ機構制御部３
Ｗ，３Ｄ）の動作時、この入，出力信号から同定した伝
達関数の値の大きさが異なることがあるが、図４に示す
ように、次の（１）式の定達関数Ｇの時系列項ｇ１〜ｇ
１０で形成する、図４に示す時系列総和パタ−ンは変化
せず、圧下制御装置に異常がある場合にみに伝達関数の
該パタ−ンが変化することが判明した。[Operation] The operation of the present invention will be explained with reference to FIGS. 1 to 4. The present inventors conducted various studies to accurately diagnose the normality or abnormality of the rolling reduction control device from the transfer function.
When a steel plate S is rolled by a rolling roll (work roll WR and backup roll BR), the reaction force acting on the rolling roll varies depending on the temperature, material (component), etc. of the steel plate S being rolled. - A lowering control device that adjusts the gap between the wheels (hydraulic lowering device 1W)
, 1D, AGC control section 2W, 2D, servo mechanism control section 3
When the transfer function (W, 3D) is operated, the magnitude of the transfer function identified from the input and output signals may differ, but as shown in Figure 4, when the steady-state function G of the following equation (1) Series terms g1 to g
It has been found that the time-series summation pattern shown in FIG. 4 formed in step 10 does not change, and the pattern of the transfer function changes only when there is an abnormality in the reduction control device.
【０００８】
Ｇ＝（ｇ１，ｇ２
，ｇ３，・・・ｇ９，ｇ１０） ・・
・（１）
このため、伝達関数のパタ−ン判定が可能なように時系
列項を有する伝達関数を伝達関数同定部である伝達関数
同定用階層型ニュ−ラルネット６で同定し、この伝達関
数の時系列項ｇ１〜ｇ１０を基にして正常・異常強度演
算ニュ−ラルネット部８で予め設定記憶している正常，
異常時に於ける時系列項とパタ−ン比較する。 更に
、この正常，異常強度演算ニュ−ラルネット部８に予め
設定記憶する正常又は異常状態における伝達関数の時系
列項が形成する時系列総和パタ−ンは圧下制御装置の全
ての正常，異常ケ−スのパタ−ンについて設定出来るも
のではない（例えば数年に１回程度しか発生しない故障
の場合もある）ことから、正常・異常強度演算ニュ−ラ
ルネット部８において、前記のように学習記憶している
正常又は異常時における伝達関数の時系列項が形成する
時系列総和パタ−ンと比較して、この学習記憶パタ−ン
と伝達関数同定用階層型ニュ−ラルネット６で同定した
伝達関数の時系列項が形成する前記パタ−ンの相似程度
を正常強度，異常強度で表わし、その値を演算する。そ
して、この正常強度値と異常強度値を比較して正常・異
常判定部９により圧下制御装置の正常・異常を判定する
。
これにより、代表的な正常，異常のパタ−ンを数例程度
学習記憶しておけば、数年に一度程度発生する故障でも
精度良く正常，異常の診断が可能となる。G=(g1, g2
,g3,...g9,g10)...
・(1) Therefore, in order to make it possible to determine the pattern of the transfer function, a transfer function having a time series term is identified by the transfer function identification hierarchical neural net 6, which is the transfer function identification unit, and this transfer function normality/abnormality intensity calculation neural network section 8 presets and stores the normality and abnormality intensity based on the time series terms g1 to g10 of
Compare the time series terms and patterns during abnormal times. Furthermore, the time-series summation pattern formed by the time-series terms of the transfer function in the normal or abnormal state, which is preset and stored in the normal/abnormal intensity calculation neural network unit 8, can be used for all normal and abnormal cases of the reduction control device. Since it is not possible to set the pattern of the fault (for example, there are cases where a failure occurs only once every few years), the normal/abnormal strength calculation neural network section 8 learns and memorizes it as described above. The learning memory pattern and the transfer function identified by the hierarchical neural network for transfer function identification 6 are compared with the time series summation pattern formed by the time series terms of the transfer function during normal or abnormal times. The degree of similarity of the pattern formed by the time series terms is expressed as normal intensity and abnormal intensity, and the values are calculated. Then, by comparing the normal strength value and the abnormal strength value, the normality/abnormality determining section 9 determines whether the reduction control device is normal or abnormal. As a result, by learning and memorizing several typical normal and abnormal patterns, it is possible to accurately diagnose whether a failure occurs once every few years or not.
【０００９】また、通常の圧延機の圧延ロ−ルはその両
側（ワ−クサイド，ドライブサイド）から圧下制御装置
により、間隙調整を行っている。このような、圧延ロ−
ルにおいては、圧延ロ−ルのワ−クサイド，ドライブサ
イドの圧下制御装置は各々独立して制御されているが、
圧延ロ−ルを介して微妙な相互影響を及ぼしており、例
えばワ−クサイドの圧下制御装置に異常がある場合、ド
ライブサイドの圧下制御装置に影響が現われて、該ドラ
イブサイドも異常であるとの誤診断を行う等の原因とな
る。前記手段（２）は、この誤診断を防止し、精度良い
診断を可能とするためのものであり、圧延ロ−ルのドラ
イブサイドとワ−クサイドの相互の影響を加味して圧下
制御装置の正常・異常を診断することにより、上述の誤
診断を防止し診断精度を高くする。つまり、ドライブサ
イドの圧下制御装置とワ−クサイドの圧下制御装置の伝
達関数の時系列項を同一の正常・異常強度演算ニュ−ラ
ルネット部８で処理する。そして、この両者の影響を加
味して圧下制御装置別に正常強度値と異常強度値を求め
ることにより、正確な正常強度値と異常強度値を算定し
、これにより前記手段（１）と同様に圧下制御装置の正
常，異常の診断を行う。[0009] Further, the gap between the rolling rolls of an ordinary rolling mill is adjusted from both sides (work side, drive side) by a rolling control device. Such a rolling roll
In the mill, the work side and drive side reduction control devices of the rolling roll are each controlled independently.
There is a subtle mutual influence through the rolling rolls. For example, if there is an abnormality in the work side reduction control device, the effect will appear on the drive side reduction control device, and it will be determined that the drive side is also abnormal. This may lead to misdiagnosis, etc. The means (2) is to prevent this misdiagnosis and enable highly accurate diagnosis, and is to adjust the reduction control device by taking into account the mutual influence of the drive side and work side of the rolling roll. By diagnosing normality and abnormality, the above-mentioned erroneous diagnosis is prevented and diagnostic accuracy is increased. In other words, the time series terms of the transfer functions of the drive-side roll-down control device and the work-side roll-down control device are processed by the same normal/abnormal strength calculation neural net section 8. Then, by calculating the normal strength value and the abnormal strength value for each reduction control device by taking into account the effects of both, accurate normal strength values and abnormal strength values are calculated. Diagnose whether the control device is normal or abnormal.
【００１０】0010
【実施例】本発明の一実施例を、図１〜図３を参照して
説明する。図１は製鉄所の圧延工程における板厚変動補
償用の制御装置の異常診断装置の全体図である。図中、
ＷＲは鋼板Ｓを圧下する作業ロ−ルである。１Ｗおよび
１Ｄは、それぞれバックアップロ−ルＢＲのワ−クサイ
ドＷＳおよびドライブサイドＤＳの間隔を調整する油圧
圧下装置である。２Ｗおよび２Ｄは、ロ−ドセルＲで測
定したバックアップロ−ルＢＲの反力と、マグネスケ−
ル４Ｗ，４Ｄで測定した油圧圧下装置１Ｗおよび１Ｄの
スプ−ル位置及び、上位計算機（図示せず）からの作業
ロ−ルＷＲの目標ロ−ルギャップ量から圧下量を演算し
、これにもとづいてサ−ボ弁（図示せず）の開度を演算
するＡＧＣ（Ａｕｔｏｍａｔｉｃ Ｇａｕｇｅ Ｃｏｎ
ｔｒｏｌ） 制御部である。３Ｗおよび３Ｄは、ＡＧＣ
制御部２Ｗおよび２Ｄで演算したサ−ボ弁の開度制御信
号によりサ−ボ弁の開度を調整するサ−ボ機構制御部で
ある。４はデ−タ入力部でありサ−ボ機構制御部３Ｗお
よび３Ｄに入力する各制御信号（制御デ−タ）ａを１０
ｍｓｅｃ周期で取り込む。５はデ−タ入力部４から入力
したデ−タの正規化を行うデ−タ処理部、６は正規化し
た鋼板Ｓの一本分の時系列制御デ−タから伝達関数のオ
フセット項ｇ０，時系列項ｇ１〜ｇ１０を同定し下記（
２）式に示す出力値Ｙを出力する伝達関数同定用階層型
ニュ−ラルネットである。DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described with reference to FIGS. 1 to 3. FIG. 1 is an overall diagram of an abnormality diagnosis device for a control device for compensating plate thickness variations in a rolling process in a steelworks. In the figure,
WR is a work roll that rolls down the steel plate S. 1W and 1D are hydraulic pressure lowering devices that respectively adjust the distance between the work side WS and the drive side DS of the backup roll BR. 2W and 2D are the reaction force of the backup roll BR measured with the load cell R and the magnetic scale.
The reduction amount is calculated from the spool positions of the hydraulic reduction devices 1W and 1D measured by rolls 4W and 4D, and the target roll gap amount of the work roll WR from a host computer (not shown), and based on this. AGC (Automatic Gauge Control) calculates the opening degree of a servo valve (not shown).
trol) is a control unit. 3W and 3D are AGC
This is a servo mechanism control section that adjusts the opening degree of the servo valve based on the servo valve opening control signal calculated by the control sections 2W and 2D. 4 is a data input section, and each control signal (control data) a input to the servo mechanism control section 3W and 3D is input to 10.
Capture at msec intervals. 5 is a data processing unit that normalizes the data input from the data input unit 4; 6 is an offset term g0 of the transfer function from the normalized time series control data for one steel plate S; , identify the time series terms g1 to g10 and use the following (
2) This is a hierarchical neural net for transfer function identification that outputs the output value Y shown in equation 2.
【００１１】
Ｙ＝ｇ０＋Ｕ１ｇ１＋Ｕ２ｇ２＋Ｕ３ｇ３
＋・・・＋Ｕ９ｇ９＋Ｕ１０ｇ１０ ・・・（２）但し
、Ｕ１〜Ｕ１０：制御デ−タ７は、伝達関数同定用階層
型ニュ−ラルネット６で同定した上記伝達関数の各項ｇ
０，ｇ１〜ｇ１０を前記出力値Ｙから抽出する伝達関数
抽出部、８は伝達関数抽出部７で抽出した伝達関数のオ
フセット項ｇ０を含む時系列項ｇ１〜ｇ１０から正常強
度値及び異常強度値を演算する正常・異常強度演算用階
層型ニュ−ラルネットである。この正常・異常強度演算
用階層型ニュ−ラルネット８には、予め、サ−ボ機構制
御部３Ｗ，３Ｄの正常時に於ける伝達関数のオフセット
項と時系列項で形成するパタ−ンと正常強度値（例えば
１）の関係及び異常時に於ける伝達関数のオフセット項
と時系列項で形成するパタ−ンと異常強度値（例えば０
）の関係を記憶し、この両記憶伝達関数の前記パタ−ン
と、ニュ−ラルネット６で同定し伝達関数抽出部７で抽
出した伝達関数のオフセット項ｇ０，時系列項ｇ１〜ｇ
１０で形成したパタ−ンとを比較し相似度合いを判断し
、相似度合いが高い程、その強度値１又は０に近い値を
出力するように学習アルゴリスムバックプロパゲ−ショ
ン法によって学習記憶する。又、オンラインで学習させ
る学習アルゴリズムを持たせることも出来る。９は、正
常・異常強度演算用階層型ニュ−ラルネット８から出力
した正常強度と異常強度からＡＧＣ制御部２Ｗ，２Ｄの
正常・異常を判定する正常・異常判定部、１０は正常・
異常判定結果を表示する表示部である。次に診断につい
て説明する。図１のサ−ボ機構制御部３Ｗ，３Ｄのサ−
ボ機構制御入力信号ａおよび油量出力信号ｂを１０ｍｓ
ｅｃ周期で、デ−タ入力部４を介してデ−タ処理部に取
り込む。デ−タ処理部５は、デ−タの取り込み毎に、各
入，出力信号（ａ，ｂ）を正の実数に正規化する。伝達
関数同定用階層型ニュ−ラルネット６は、ワ−クサイド
ＷＳ，ドライブサイドＤＳの各々のサ−ボ機構制御部３
Ｗ，３Ｄ別に入力信号ａと出力信号ｂを組合せ、この入
・出力信号の関係から上記（２）式に示される各１個の
オフセット項ｇ０と各１０個の時系列項ｇ１〜ｇ１０を
有する伝達関数Ｇを同定する。この伝達関数同定用階層
型ニュ−ラルネット６で、この制御系の伝達関数を同定
する場合について図２を参照してより詳しく説明する。
伝達関数同定用階層型ニュ−ラルネット６は、１１個の
ユニットからなる入力層６ａ，１個のユニットからなる
出力層６ｂおよび１個の教師ユニット６ｃで構成した２
層形のニュ−ラルネットである。この入力層６ａの各ユ
ニットには、サ−ボ機構制御部３Ｗの入，出力側の信号
ａ，ｂが、ワ−クロ−ルＷＲの鋼板Ｓ噛み込み開始時点
より、１０ｍｓｅｃ周期で取り込まれ正規化されて記憶
される。そして、この圧延が完了すると正規化入力信号
（ａ）を古い順番に読出して伝達関数同定用階層型ニュ
−ラルネット６の入力層６ａの各ユニットの左側端から
入力し、順次右側にシフトする。そして、入力層６ａの
左側端に入力した正規化信号（ａ）に対応する正規化出
力信号（ｂ）を教師ユニット６ｃに入力し、これを教師
信号として出力層６ｂにおいて入力層６ａの各ユニット
から入力した正規化信号（ａ）と比較し両信号が同一と
なる結び付き強度を入力層６ａと出力層６ｂの間でδル
−ルに基づいて求める。
これが終わると、前記のようにデ−タ処理部５に記憶し
た正規化信号の時間的に古い（サンプル採取時が左側端
ユニットの入力値より１サンプルタイム新しい）ものを
１個づつ順次入力層６ａの左端ユニットに入力し、該左
端ユニットに入力していた正規化信号は押出されて破棄
される。かくして、最後の最も新しい正規化した入力信
号（鋼板Ｓの後端が圧延ロ−ルを通過完了する直前に採
取した制御信号）が左端ユニットに入力し、前記出力信
号が教師ユニット６ｃに入力し、前記結び付き強度の同
定が完了するまで同様にして順次行う。そして、入力す
る正規化信号がなくなると、その結び付き強度値をオフ
セット項ｇ０Ｗおよび時系列項ｇ１Ｗ〜ｇ１０Ｗを含む
前記（２）式で示される出力値ＹＷとして入力層６ａか
ら出力する。又、ドライブサイドＤＳのサ−ボ機構制御
部３Ｄについてもこれと同様にして別途設けた伝達関数
同定用階層型ニュ−ラルネットでオフセット項ｇ０Ｄを
含む時系列項ｇ１Ｄ〜ｇ１０Ｄを有する伝達関数ＧＤを
同定する。このようにして同定した２つの伝達関数ＧＷ
，ＧＤのオフセット項ｇ０Ｗ，ｇ０Ｄと時系列項ｇ１Ｗ
〜ｇ１０Ｗ，ｇ１Ｄ〜ｇ１０Ｄを含む出力値Ｙとして前
記ニュ−ラルネット６より出力し、次段の伝達関数抽出
部７に入力して、ここでオフセット項ｇ０Ｗ，ｇ０Ｄと
時系列項ｇ１Ｗ〜ｇ１０Ｗ，ｇ１Ｄ〜ｇ１０Ｄを抽出し
、これらを正常，異常強度演算用階層型ニュ−ラルネッ
ト８に入力する。Y=g0+U1g1+U2g2+U3g3
+...+U9g9+U10g10...(2) However, U1 to U10: Control data 7 is each term g of the above transfer function identified by the hierarchical neural net 6 for transfer function identification.
0, g1 to g10 are extracted from the output value Y, and 8 is the normal intensity value and abnormal intensity value from the time series terms g1 to g10 including the offset term g0 of the transfer function extracted by the transfer function extraction unit 7. This is a hierarchical neural network for calculating normal and abnormal strength. This hierarchical neural network 8 for normal/abnormal strength calculation includes a pattern formed by offset terms and time series terms of the transfer functions in the normal state of the servo mechanism control units 3W and 3D, and a normal strength. The relationship between the values (for example, 1) and the pattern formed by the offset term and time series term of the transfer function during an abnormality and the abnormal intensity value (for example, 0
), and the pattern of both memory transfer functions, the offset term g0, and time series terms g1 to g of the transfer function identified by the neural net 6 and extracted by the transfer function extraction unit 7.
The pattern is compared with the pattern formed in step 10 to determine the degree of similarity, and the higher the degree of similarity, the more the intensity value is learned and stored by the backpropagation method so as to output a value closer to 1 or 0. It is also possible to provide a learning algorithm for online learning. Reference numeral 9 denotes a normality/abnormality determination unit that determines whether the AGC control units 2W, 2D are normal or abnormal based on the normal intensity and abnormality output from the hierarchical neural network 8 for normal/abnormal intensity calculation;
This is a display unit that displays abnormality determination results. Next, the diagnosis will be explained. Servo mechanism control parts 3W and 3D in Fig. 1
Bo mechanism control input signal a and oil amount output signal b for 10ms
The data is taken into the data processing section via the data input section 4 at the ec cycle. The data processing section 5 normalizes each input and output signal (a, b) to a positive real number every time data is taken in. The hierarchical neural network 6 for transfer function identification is connected to the servo mechanism control unit 3 of each of the work side WS and drive side DS.
Input signal a and output signal b are combined separately for W and 3D, and from the relationship between the input and output signals, each has one offset term g0 and ten time series terms g1 to g10 as shown in equation (2) above. Identify the transfer function G. A case in which the transfer function of the control system is identified using the transfer function identification hierarchical neural network 6 will be described in more detail with reference to FIG. The hierarchical neural net 6 for identifying a transfer function has an input layer 6a consisting of 11 units, an output layer 6b consisting of 1 unit, and 2 layers consisting of 1 teacher unit 6c.
It is a layered neural net. The input and output side signals a and b of the servo mechanism control section 3W are input into each unit of this input layer 6a at a 10 msec period from the time when the work roll WR starts biting the steel plate S. digitized and memorized. When this rolling is completed, the normalized input signals (a) are read out in chronological order, inputted from the left end of each unit of the input layer 6a of the hierarchical neural net 6 for transfer function identification, and sequentially shifted to the right. Then, the normalized output signal (b) corresponding to the normalized signal (a) input to the left end of the input layer 6a is input to the teacher unit 6c, and this is used as a teacher signal in the output layer 6b for each unit of the input layer 6a. The connection strength at which both signals are the same is determined based on the δ rule between the input layer 6a and the output layer 6b. When this is completed, the older normalized signals stored in the data processing unit 5 as described above (the time of sample collection is one sample time later than the input value of the leftmost unit) are sequentially transferred to the input layer one by one. The normalized signal input to the left end unit of 6a is pushed out and discarded. Thus, the last and newest normalized input signal (the control signal taken just before the rear end of the steel sheet S completed passing through the rolling rolls) is input to the left end unit, and the output signal is input to the teacher unit 6c. , are sequentially performed in the same manner until the identification of the connection strength is completed. Then, when there are no more normalized signals to be input, the connection strength value is output from the input layer 6a as the output value YW shown by the above equation (2) including the offset term g0W and the time series terms g1W to g10W. Similarly, for the servo mechanism control unit 3D of the drive side DS, a transfer function GD having time series terms g1D to g10D including an offset term g0D is determined using a separately provided hierarchical neural network for identifying transfer functions. identify The two transfer functions GW identified in this way
, GD offset terms g0W, g0D and time series term g1W
~ g10W, g1D ~ g10D is outputted from the neural net 6 as an output value Y including g10D, and inputted to the next stage transfer function extraction unit 7, where offset terms g0W, g0D and time series terms g1W ~ g10W, g1D ~g10D are extracted and inputted into the hierarchical neural net 8 for normal/abnormal strength calculation.
【００１２】このニュ−ラルネット８は、図３に示すよ
うに、２２個のユニットを有する入力層８ａと１０個の
ユニットを有する中間層８ｂ、及び４個のユニットを有
する出力層８ｃから構成している。先ず、入力層８ａの
上から１番目のユニットにはワ−クサイドＷＳのサ−ボ
機構制御部３Ｗの伝達関数ＧＷのオフセット項ｇ０Ｗを
入力し、次の（上から２番目）ユニットから１１番目の
ユニットまでに時系列項ｇ１Ｗ〜ｇ１０Ｗを順次入力す
る。更に、正規化したドライブサイドＤＳのサ−ボ機構
制御部３Ｄの伝達関数ＧＤのオフセット項ｇ０Ｄを１２
番目のユニットに入力し、１３番目のユニットから２２
番目のユニットまでに時系列項ｇ１Ｄ〜ｇ１０Ｄの各々
を入力する。As shown in FIG. 3, this neural net 8 is composed of an input layer 8a having 22 units, an intermediate layer 8b having 10 units, and an output layer 8c having 4 units. ing. First, the offset term g0W of the transfer function GW of the servo mechanism control section 3W of the work side WS is input to the first unit from the top of the input layer 8a, and The time series terms g1W to g10W are sequentially input up to the unit . Furthermore, the offset term g0D of the normalized transfer function GD of the servo mechanism control section 3D of the drive side DS is set to 12
22 from the 13th unit.
Each of the time series terms g1D to g10D is input up to the th unit.
【００１３】このようにして、入力層８ａのユニットに
入力したサ−ボ機構制御部３Ｗ，３Ｄにおける各伝達関
数ＧＷ，ＧＤのオフセット項ｇ０Ｗ，ｇ０Ｄの２個、時
系列項ｇ１Ｗ〜ｇ１０Ｗ，ｇ１Ｄ〜ｇ１０Ｄの２０個を
中間層８ｂ及び出力層８ｃにより予め学習させたδル−
ルに従って処理することによって図４に示す時系列総和
パタ−ンによりその相似度合いを比較し、出力層８ｃか
らサ−ボ機構制御部３Ｗ，３Ｄ各々の正常強度値，異常
強度値をそれぞれ０〜１までの実数で出力する。In this way, the two offset terms g0W and g0D of the transfer functions GW and GD in the servo mechanism control units 3W and 3D input to the unit of the input layer 8a, and the time series terms g1W to g10W and g1D ~g10D are learned in advance by the intermediate layer 8b and the output layer 8c.
By processing according to the rules, the degree of similarity is compared using the time series summation pattern shown in FIG. Output as a real number up to 1.
【００１４】このワ−クサイドＷＳとドライブサイドＤ
Ｓの正常強度値，異常強度値を正常・異常判定部９で比
較し、正常強度値と異常強度値の差、つまり、（正常強
度値）−（異常強度値）が所定値（０．６）より大きい
時、サ−ボ機構制御部３Ｗ，３Ｄは正常と判断し、それ
以外の時は異常と判断する。そして、この判断結果を表
示部１０で表示する。[0014] This work side WS and drive side D
The normal intensity value and abnormal intensity value of S are compared in the normal/abnormality determining section 9, and the difference between the normal intensity value and the abnormal intensity value, that is, (normal intensity value) - (abnormal intensity value), is determined to be a predetermined value (0.6 ), the servo mechanism control units 3W and 3D determine it to be normal, and otherwise determine it to be abnormal. This determination result is then displayed on the display section 10.
【００１５】尚、本実施例においては、伝達関数同定用
階層型ニュ−ラルネット６で伝達関数を同定したが、本
発明はこれに限るものではなく、普通の計算機により最
小２乗法を利用して算定してもよい。また、本実施例に
おいてはサ−ボ機構制御部３Ｗ，３Ｄについて説明した
が、油圧圧下装置１Ｗ，１Ｄの正常・異常診断を行う際
には図１のサ−ボ機構制御部３Ｗ，３Ｄの出力信号ｂと
スプ−ル位置信号ｃを採取し、又、ＡＧＣ制御部２Ｗ，
２Ｄの正常と異常を診断を診断する際にはスプ−ル位置
信号ｃと目標値信号ｄから前記同様に処理する。更に、
本実施例では水平ロ−ルについて説明したが、本発明は
これに限るものではなく、縦ロ−ルであっても同様に適
用できる。In this embodiment, the transfer function was identified using the hierarchical neural network 6 for transfer function identification, but the present invention is not limited to this. It may be calculated. Furthermore, although the servo mechanism control units 3W and 3D have been described in this embodiment, when diagnosing the normality and abnormality of the hydraulic lowering devices 1W and 1D, the servo mechanism control units 3W and 3D shown in FIG. The output signal b and the spool position signal c are collected, and the AGC control unit 2W,
When diagnosing whether the 2D is normal or abnormal, the spool position signal c and target value signal d are processed in the same manner as described above. Furthermore,
Although the present embodiment has been described with respect to horizontal rolls, the present invention is not limited thereto, and can be similarly applied to vertical rolls.
【００１６】[0016]
【発明の効果】以上説明したように本発明は伝達関数の
時系列項が描くパタ−ン（形状）により圧延制御装置の
異常を診断するので、圧延中の鋼板の圧下率，温度等の
操業条件に実質上関係なく、しかも煩雑な作業を伴うこ
となく精度良く確実な診断が可能となる。Effects of the Invention As explained above, the present invention diagnoses an abnormality in the rolling control device based on the pattern (shape) drawn by the time series term of the transfer function, so Accurate and reliable diagnosis is possible virtually regardless of conditions and without any complicated work.
【００１７】更に、圧延機のワ−クサイドとドライブサ
イドの相互の影響力を加味しつつ、各ニュ−ラルネット
で両サイドにおける正常・異常を診断することにより、
簡単に、しかも、精度良く両者の正常・異常を診断する
ことが可能となる。Furthermore, by taking into account the mutual influence of the work side and drive side of the rolling mill, and diagnosing normality and abnormality on both sides using each neural net,
It becomes possible to easily and accurately diagnose whether the two are normal or abnormal.
【００１８】このようにして診断した圧延制御装置の異
常部位を早期に取替えることにより、圧延機を安定的に
稼働して圧延製品の板厚精度の低下を防止し、高位に安
定出来る等の多大な効果を奏する。[0018] By quickly replacing the abnormal parts of the rolling control device diagnosed in this way, the rolling mill can be operated stably, the thickness accuracy of the rolled products can be prevented from deteriorating, and a high level of stability can be achieved. It has a great effect.
【図１】 本発明の一実施例の構成概要を示すブロッ
ク図である。FIG. 1 is a block diagram showing an overview of the configuration of an embodiment of the present invention.
【図２】 図１に示す伝達関数同定用階層型ニュ−ラ
ルネット６の機能の一部を示すブロック図である。2 is a block diagram showing part of the functions of the transfer function identification hierarchical neural network 6 shown in FIG. 1. FIG.
【図３】 図１に示す正常・異常強度演算部８の機能
の一部を示すブロック図である。FIG. 3 is a block diagram showing part of the functions of the normal/abnormal strength calculating section 8 shown in FIG. 1.
【図４】 伝達関数の時系列総和パタ−ンを示すグラ
フである。FIG. 4 is a graph showing a time series summation pattern of a transfer function.
ＷＲ：作業ロ−ル
ＢＲ：バックアップロ−ル４Ｗ，４Ｄ：マグネスケ−
ルWR: Work roll
BR: Backup roll 4W, 4D: Magnesque
le
Claims (2)
行う圧下制御装置の異常診断装置において、前記圧延ロ
−ルの圧下制御装置の制御信号を所定周期毎に入力して
時系列項を有する伝達関数を同定する伝達関数同定部と
、予め伝達関数の時系列項と圧下制御装置の正常強度値
及び異常強度値の関係を学習記憶し、この学習記憶情報
と前記伝達関数同定部で同定した伝達関数の時系列項を
パタ−ン比較して正常強度値及び異常強度値を演算する
正常・異常強度演算ニュ−ラルネット部と、正常・異常
強度演算ニュ−ラルネット部で演算した正常強度値と異
常強度値により前記圧下制御装置の正常・異常を判定す
る正常・異常判定部と、を有することを特徴とする、圧
延工程における圧下制御装置の異常診断装置。1. An abnormality diagnosis device for a rolling control device that adjusts the rolling position of a rolling roll of a rolling mill, wherein a control signal of the rolling rolling control device is inputted at predetermined intervals to calculate a time series parameter. a transfer function identification unit that identifies a transfer function having A normal/abnormal intensity calculation neural net section that compares patterns of time series terms of identified transfer functions to calculate normal intensity values and abnormal intensity values, and a normal intensity calculated by the normal/abnormal intensity calculation neural net section. 1. An abnormality diagnosis device for a rolling control device in a rolling process, comprising: a normality/abnormality determining section that determines whether the rolling control device is normal or abnormal based on the value and the abnormal strength value.
行う圧下制御装置の異常診断装置において、前記圧延ロ
−ルのドライブサイドとワ−クサイドに各々独立して設
けた圧下制御装置の制御信号を所定周期毎に入力して時
系列項を有する伝達関数を各々同定する伝達関数同定部
と、予め前記両伝達関数の時系列項と両圧下制御装置の
正常強度値及び異常強度値の関係を学習記憶し、この学
習記憶情報と前記伝達関数同定部で同定した両伝達関数
の時系列項をパタ−ン比較して前記両サイドの圧下制御
装置の正常強度値及び異常強度値を各々演算する正常・
異常強度演算ニュ−ラルネット部と、正常・異常強度演
算ニュ−ラルネット部で演算した前記各正常強度値と異
常強度値により前記圧下制御装置の正常・異常を各々判
定する正常・異常判定部と、を有することを特徴とする
、圧延工程における圧下制御装置の異常診断装置。2. An abnormality diagnosis device for a roll-down control device that adjusts the roll-down position of a rolling roll of a rolling mill, wherein the roll-down control device is provided independently on a drive side and a work side of the roll. a transfer function identification unit that inputs a control signal every predetermined period and identifies each transfer function having a time series term; The relationship is learned and memorized, and this learned and memorized information is compared in pattern with the time series terms of both transfer functions identified by the transfer function identification section to determine the normal strength value and the abnormal strength value of the reduction control devices on both sides, respectively. Normal to calculate
a normality/abnormality determining section that determines whether the reduction control device is normal or abnormal based on the normal strength values and abnormal strength values calculated by the abnormal strength calculation neural net section and the normal/abnormal strength calculation neural net section; An abnormality diagnosis device for a rolling reduction control device in a rolling process, comprising:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP3134649A JP2564710B2 (en) | 1991-06-06 | 1991-06-06 | Abnormality diagnosing device for rolling down control device in rolling process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP3134649A JP2564710B2 (en) | 1991-06-06 | 1991-06-06 | Abnormality diagnosing device for rolling down control device in rolling process |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| JPH04361815A true JPH04361815A (en) | 1992-12-15 |
| JP2564710B2 JP2564710B2 (en) | 1996-12-18 |
Family
ID=15133313
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP3134649A Expired - Lifetime JP2564710B2 (en) | 1991-06-06 | 1991-06-06 | Abnormality diagnosing device for rolling down control device in rolling process |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JP2564710B2 (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100495826B1 (en) * | 2003-08-21 | 2005-06-16 | 학교법인 건국대학교 | Method for diagnosing fault of roll |
| CN112665408A (en) * | 2020-12-29 | 2021-04-16 | 张家港宏昌钢板有限公司 | Automatic acquisition system and method for abnormal furnace signals of heating furnace |
| CN117019889A (en) * | 2023-07-18 | 2023-11-10 | 常州润来科技有限公司 | Method and system for intelligently detecting faults of three-roller rotary rolling equipment |
-
1991
- 1991-06-06 JP JP3134649A patent/JP2564710B2/en not_active Expired - Lifetime
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100495826B1 (en) * | 2003-08-21 | 2005-06-16 | 학교법인 건국대학교 | Method for diagnosing fault of roll |
| CN112665408A (en) * | 2020-12-29 | 2021-04-16 | 张家港宏昌钢板有限公司 | Automatic acquisition system and method for abnormal furnace signals of heating furnace |
| CN112665408B (en) * | 2020-12-29 | 2024-05-14 | 张家港宏昌钢板有限公司 | Automatic acquisition system and method for abnormal furnace signals of heating furnace |
| CN117019889A (en) * | 2023-07-18 | 2023-11-10 | 常州润来科技有限公司 | Method and system for intelligently detecting faults of three-roller rotary rolling equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| JP2564710B2 (en) | 1996-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| DE102017009471B4 (en) | A human collaborative robot system with improved accuracy of detecting an external force through machine learning | |
| TW201732475A (en) | Apparatus and method for supporting diagnosis of manufacturing facility | |
| CN113468479B (en) | Cold continuous rolling industrial process monitoring and abnormality detection method based on data driving | |
| JPH04361815A (en) | Abnormality diagnosing device for rolling-down device in rolling stages | |
| US6807449B1 (en) | Method for controlling and pre-setting a steelworks or parts of a steelworks | |
| DE69207325T2 (en) | Method and device for monitoring sensors and for fault location in an industrial process | |
| JPH0728502A (en) | Plant controller | |
| CN116203903A (en) | Space-time root diagnosis method for quality abnormality in hot continuous rolling process | |
| KR19990052675A (en) | Online roll eccentricity diagnosis method and device | |
| CN112862284B (en) | Precision evaluation method and system for rigidity of hot-rolled strip mill | |
| KR20200000783A (en) | Monitoring assistance system for steel plant | |
| CN106515745B (en) | A kind of control system of automatic Pilot train and platform | |
| JPH05253797A (en) | Judging system for abnormality on line | |
| WO2020204043A1 (en) | Blast furnace abnormality assessment device, blast furnace abnormality assessment method, and blast furnace operation method | |
| US3892112A (en) | Rolling mill gauge control | |
| CN112783935A (en) | Analysis device | |
| JPS642203B2 (en) | ||
| WO2022085350A1 (en) | Method for building abnormality diagnosis model, abnormality diagnosis method, device for building abnormality diagnosis model, and abnormality diagnosis device | |
| JP3331191B2 (en) | Rolling equipment shape control device and method | |
| KR0173508B1 (en) | Cobol data analysis device for hot rolling | |
| JP3171212B2 (en) | Tandem rolling mill and method for estimating cause of abnormal thickness in tandem rolling mill | |
| JPH0249166B2 (en) | ||
| JP2009291809A (en) | Diagnostic method of disturbance of dynamic process | |
| JP4258424B2 (en) | Operating load measuring device | |
| CN116702030B (en) | Blast furnace state monitoring method and device based on sensor reliability analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A01 | Written decision to grant a patent or to grant a registration (utility model) |
Free format text: JAPANESE INTERMEDIATE CODE: A01 Effective date: 19960709 |