Slab Width Forecasting via Latency-Aligned Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In industrial slab casting processes, forecasting slab width is challenging due to latency in data collection, where process variables measured at one time point do not directly relate to the target variable at the same time point, but rather at a later time, and are influenced by previous states and varying process conditions, making traditional machine learning approaches ineffective.
Innovation Solution
A method involving data preprocessing to align target variables with process variables, incorporating memory of previous feature trends, and clustering for process variations, allowing for the training of machine learning models that can accurately predict target variables by accounting for latency and process changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are used for forecasting target variables in industrial processes, then the model structure is simple and easy to implement, but the prediction accuracy deteriorates due to latency and process variations
Solution Approach 1:
The patent segments the forecasting problem by creating multiple specialized machine learning models, each trained on specific subsets of historical data representing different process conditions or time periods. This segmentation allows each model to specialize in particular scenarios, improving prediction accuracy for each segment while managing overall system complexity through modular architecture.
Solution Approach 2:
The patent applies preliminary action by pre-processing historical data to identify and correct latency issues before training models. The system performs data alignment, timestamp correction, and feature engineering in advance, preparing the data structure to account for process delays and variations before the actual forecasting occurs during operation.
2Manufacturing precision
If data is collected at each time point throughout the process, then real-time monitoring capability is improved, but prediction accuracy deteriorates due to the latency between input variables and target variables
Solution Approach 1:
The patent implements feedback mechanisms by continuously comparing predicted target values with actual measured values from the process. This feedback loop allows the system to learn from prediction errors and adjust future predictions, compensating for latency effects and improving slab width control accuracy over time through iterative refinement.
Solution Approach 2:
The patent addresses latency by introducing temporal dimensionality transformations, where historical data from multiple time points is aggregated and transformed into features that capture the evolving state of the process. This dimensional transformation allows the model to account for time delays by analyzing trends and patterns across multiple timestamps rather than relying on single-point measurements.
3Measurement precision
If a single machine learning model is used to learn all process variations, then the system is simple to maintain, but prediction accuracy deteriorates due to the diversity of process conditions
Solution Approach 1:
The patent applies dynamics by creating an adaptive system where multiple specialized models are dynamically selected and applied based on current process conditions. The system monitors process state variables and automatically chooses the most appropriate pre-trained model for the current scenario, allowing the system to adapt to varying process conditions while maintaining simplicity through automated model selection rather than manual intervention.
Data Source
AI summary
Methods and systems for process speed-based forecasting, in which historical data of a process is first aligned to account for latency in the process. Optionally, incorporation of memory and/or clustering can be included in the pre-processing of the historical data before being used to train a machine learning model.