Rolling process learning control device

Patent metadata

Field	Value
Publication number	US-10124381-B2
Application number	US-201415119313-A
Country	US
Kind code	B2
Filing date	Feb 17, 2014
Priority date	Feb 17, 2014
Publication date	Nov 13, 2018
Grant date	Nov 13, 2018

Abstract

An instantaneous value of a learning coefficient is calculated based on a gap between a result value and a result recalculation value, and an update value of the learning coefficient is calculated from the instantaneous value. The calculated update value is recorded in a cell corresponding to present rolling conditions. A predetermined number of neighboring cells having small spatial distances from a target cell corresponding to next rolling conditions in a space having rolling conditions as coordinate axes are selected from cells in which respective evaluation results of recency, saturation, and stability of the learning coefficient satisfy criteria. An estimation value of the learning coefficient in the target cell is then calculated by polynomial interpolation. The most recent update value of the learning coefficient is then corrected with the estimation value, and used as a use value of the learning coefficient in the next rolling conditions.

First claim

The invention claimed is:

1. A rolling process learning control device that has a database including a plurality of cells that divide rolling conditions, and manages using the database a learning coefficient of a model expression used for setup calculation of a rolling process, the rolling process learning control device comprising:
a processor to execute a program; and
a memory to store the program which, when executed by the processor, causes the rolling process learning control device to serve as:
an instantaneous value calculation recording unit that calculates an instantaneous value of the learning coefficient based on a gap between a result value measured in the rolling process and a result recalculation value calculated using the model expression, and records the instantaneous value in a cell corresponding to present rolling conditions together with a learning time;
an update value calculation recording unit that calculates an update value of the learning coefficient based on the instantaneous value and a previous value of the learning coefficient in the present rolling conditions, and records the update value in the cell corresponding to the present rolling conditions together with a learning time;
a recency evaluation unit that evaluates recency of the learning coefficient of each of the plurality of cells based on history information of the update value stored in the database;
a saturation evaluation unit that evaluates saturation of the learning coefficient of each of the plurality of cells based on the history information of the update value stored in the database;
a stability evaluation unit that evaluates stability of the learning coefficient of each of the plurality of cells based on history information of the instantaneous value stored in the database;
a neighboring cell selection unit that selects a predetermined number of neighboring cells in order of a spatial distance from a target cell corresponding to next rolling conditions in a space having rolling conditions as coordinate axes, from the cells in which respective evaluation results of the recency, the saturation, and the stability satisfy criteria;
an estimation value calculation unit that decides representative values of the learning coefficients in the selected predetermined number of neighboring cells, respectively, and calculates an estimation value of the learning coefficient in the target cell by polynomial interpolation based on a coordinate of the target cell, and coordinates and the representative values of the selected predetermined number of neighboring cells; and
an use value decision unit that corrects the most recent update value of the learning coefficient in the target cell with the estimation value, and decides a corrected value as a use value of the learning coefficient in the next rolling conditions.

2. The rolling process learning control device according to claim 1 , wherein
the use value decision unit is configured to:
calculate the use value by a weighted average between the most recent update value and the estimation value; and
change a weighting coefficient of the weighted average according to the evaluation results so that the higher the evaluation results of the recency, the saturation, and the stability with respect to the target cell are, the larger a weight of the most recent update value becomes, and so that the lower the evaluation results are, the larger a weight of the estimation value becomes.

3. The rolling process learning control device according to claim 1 , wherein
the database includes first and second stratified tables having common cells,
the instantaneous value calculation recording unit is configured to record the instantaneous value in the first stratified table, and
the update value calculation recording unit is configured to record the update value in the second stratified table.

4. The rolling process learning control device according to claim 3 , wherein
the database further includes third, fourth, and fifth stratified tables having cells in common with the first and the second stratified tables,
the recency evaluation unit is configured to record the evaluation result of the recency in the third stratified table,
the saturation evaluation unit is configured to record the evaluation result of the saturation in the fourth stratified table, and
the stability evaluation unit is configured to record the evaluation result of the stability in the fifth stratified table.

Assignees

• Toshiba Mitsubishi Electric Industrial Systems Corp

Inventors

• Kubo Naohiro

Classifications

• G05B13/0265

  the criterion being a learning criterion · CPC title

• B21B37/00 Primary

  Control devices or methods specially adapted for metal-rolling mills or the work produced thereby (methods or devices for measuring specially adapted for metal-rolling mills B21B38/00) · CPC title

• G05B13/02 Primary

  electric · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N99/005

  Physics · mapped topic