Hot-Rolling Roll-Bending Force Prediction Using LSTM Time-Series Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing roll-bending force prediction models in hot-rolling processes suffer from complexity, non-linearity, strong coupling, and time-varying characteristics, leading to low accuracy and difficulty in data acquisition and processing, especially due to the need for numerous input parameters, and lack of consideration for time sequence in traditional neural networks.
Innovation Solution
An LSTM-based method for predicting roll-bending force, utilizing a Long Short-Term Memory (LSTM) neural network with an ordered neuron (ON-LSTM) update mechanism and a double-layer structure to effectively predict roll-bending force data by incorporating historical data, simplifying data preparation and enhancing prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional mathematical methods are used for roll-bending force prediction, then the model structure becomes very complex, but the prediction accuracy does not improve due to non-linear, strong coupling and time-varying characteristics
Solution Approach 1:
The patent replaces traditional mathematical modeling methods with a neural network-based computational model. The neural network learns the complex non-linear relationships between rolling parameters and roll-bending force through training data, avoiding the need for explicit mathematical formulations of the non-linear, strongly coupled, and time-varying characteristics of the rolling process.
Solution Approach 2:
The patent transforms the prediction approach by changing from direct prediction of roll-bending force to prediction of intermediate parameters (rolling force, elongation rate, front tension, back tension) that are then combined to calculate the final roll-bending force. This parameter transformation simplifies the modeling of each individual parameter while maintaining overall prediction accuracy.
2Measurement precision
If many input parameters are collected for roll-bending force prediction, then the model may capture more influencing factors, but data acquisition becomes difficult and data processing tasks become arduous
Solution Approach 1:
The patent extracts and focuses on the four most critical input parameters (rolling force, elongation rate, front tension, back tension) that have the greatest influence on roll-bending force, rather than attempting to collect and process all possible parameters. This selective extraction simplifies data acquisition while maintaining prediction effectiveness.
Solution Approach 2:
The patent segments the prediction task into predicting four separate intermediate parameters (rolling force, elongation rate, front tension, back tension) rather than predicting roll-bending force directly. Each intermediate parameter can be modeled independently using fewer input parameters, reducing overall data collection requirements.
3Measurement precision
If ordinary neural networks are used for roll-bending force prediction, then the model structure is simpler, but prediction accuracy is low because they do not consider time context information and cannot learn time sequence trends
Solution Approach 1:
The patent introduces dynamic temporal modeling by using LSTM (Long Short-Term Memory) recurrent neural networks that can capture time-dependent relationships and sequence patterns in the rolling process data. The LSTM architecture dynamically adjusts its internal state to remember relevant historical information while forgetting less important past data, enabling accurate prediction of time-varying roll-bending force.
Data Source
AI summary
Provided is an LSTM-based hot-rolling roll-bending force predicting method including the steps of acquiring final rolling data of a stand of a stainless steel rolling mill when performing a hot rolling process, and dividing the data into a training set traindata and a test set testdata; normalizing the traindata; building a matrix P; using a last row of the matrix P as a label of the training set, namely a true value; calculating and updating an output value and the true value of a network; after network training is completed, taking the last m output data of the LSTM network as an input at a next moment, and then obtaining an output of the network at the next moment, wherein the output is a predicted value of the roll-bending force at the next moment; repeating the steps until a sufficient number of prediction data is obtained; and comparing the processed data with the true value in the testdata to check the validity of the network.