SVR Process Parameter Updating for Real-Time Manufacturing Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing process parameter optimization methods in injection molding require re-training on entire datasets for new scenarios, leading to high computational complexity and time, or lose key feature information when only recent data is used, making them unsuitable for real-time production.
Innovation Solution
Utilize a pre-trained support vector regression model with an online proximal gradient algorithm, using historical parameters as iterative initial values to update weight parameters with small batches of new data, optimizing the model iteratively.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the entire dataset is re-trained directly for new production scenarios, then the model adapts to new conditions, but the computational complexity and time required increase significantly
Solution Approach 1:
The patent segments the re-training process into two parts: (1) using the pre-trained model as a baseline, and (2) performing incremental training only on new data. This segmentation avoids re-processing the entire historical dataset while still adapting to new production scenarios, thereby reducing re-training time and computational complexity.
Solution Approach 2:
The patent performs preliminary action by pre-training the model on the complete historical dataset before deployment. When new production scenarios arise, the pre-trained model serves as a ready-made baseline that requires minimal additional training, thus avoiding the need to re-process the entire dataset from scratch.
2Loss of time
If only recent data is used for re-training, then the training time is reduced, but key feature information from past processes is lost
Solution Approach 1:
The patent copies the weight parameters from the pre-trained model (which was trained on the complete historical dataset) to initialize the incremental training process. This copying ensures that the knowledge and key feature information extracted from the entire historical dataset are preserved and serve as the foundation for adapting to new scenarios, rather than starting from scratch with only recent data.
Solution Approach 2:
The patent changes the training parameters by switching from full-dataset training to incremental training on new data only, while maintaining the pre-trained weights as the starting point. This parameter change allows the model to adapt to new scenarios efficiently without losing the valuable information embedded in the pre-trained parameters.
3Measurement precision
If the entire dataset is re-trained, then prediction accuracy is maintained, but computational complexity increases
Solution Approach 1:
The patent segments the training computation into a pre-training phase (performed once on the complete dataset) and an incremental training phase (performed on new data only). This segmentation maintains prediction accuracy by preserving the knowledge from the complete dataset while significantly reducing the computational complexity of adaptation to new scenarios.
Solution Approach 2:
The patent performs the computationally intensive training on the entire dataset as a preliminary action before deployment. The pre-trained model's weight parameters are then reused as the baseline for incremental training, avoiding the need to repeat the expensive full-dataset training process while maintaining prediction accuracy.
Data Source
AI summary
The present application pertains to the field of data processing technology, specifically relates to a method, apparatus and device for optimizing a process parameter and a storage medium, which includes: using a first weight parameter of a pre-trained support vector regression model for a manufacturing equipment as an iterative initial value, and calculating a second weight parameter of a support vector regression model using an online proximal gradient algorithm based on training data; obtaining an optimized support vector regression model by updating the first weight parameter of the pre-trained support vector regression model to the second weight parameter; inputting a first process parameter of the manufacturing equipment into the optimized support vector regression model to obtain a first detection parameter output by the optimized support vector regression model; and calculating, according to the first process parameter and the first detection parameter, a target process parameter for the manufacturing equipment.