Multistage Fusion Regression Modeling (IA-BiGRU) for Molding Quality Prediction

Overview

The Multistage Fusion Regression Model (MSFRM) is a sophisticated deep learning framework designed to enhance real-time quality monitoring for injection-molded products. By integrating an Input-Attention Bidirectional Gated Recurrent Unit (IA-BiGRU) with a Multihead Self-Attention Mechanism (MSAM), this technology addresses the inherent challenges of traditional predictive models, specifically their inability to account for the unique thermodynamic and mechanical impacts across distinct production stages such as filling, packing, and cooling. The solution offers manufacturers a robust tool for improving production efficiency and precision.

Technical specifications

• IA-BiGRU Integration: Utilizes a customized input-attention mechanism to identify and extract critical process variables, effectively mitigating data issues caused by missing sensor signals.
• Multihead Self-Attention (MSAM): Implements an attention mechanism to analyze global information and weight the specific contribution of individual production stages to final product quality.
• Multistage Analysis: Evaluates both the cumulative and isolated effects of dynamic manufacturing conditions to support high-dimensional accuracy in quality forecasting.
• Robustness: Validated against real-world manufacturing datasets, demonstrating superior performance and stability compared to conventional machine learning and standard neural network approaches.

Technology readiness level

The technology is currently at TRL 4. The model has been developed and validated through rigorous testing on real-world injection molding datasets, demonstrating high predictive accuracy and robustness as documented in published research studies. While demonstrating effective lab-scale validation and prototype demonstration performance, it has not yet transitioned to widespread commercial deployment in industrial settings.