Polymer Melt Stability Monitoring in Injection Moulding Using LSTM-Based Time-Series Models | JoVE Visualize

Area of Science:

Materials Science
Manufacturing Engineering
Data Science

Background:

Polymer melt instability in injection moulding often evolves gradually, making conventional monitoring methods insufficient for early detection.
Existing threshold-based systems struggle to capture subtle transitions, leading to quality degradation and increased scrap rates.
Real-time monitoring of process parameters is crucial for maintaining stability and product quality in thermoplastic manufacturing.

Purpose of the Study:

To develop a data-driven framework for the early detection of polymer melt instability in industrial injection moulding.
To leverage Long Short-Term Memory (LSTM) time-series models for identifying gradual drift conditions preceding quality issues.
To enable proactive process adjustments and reduce scrap by detecting instability minutes before defects manifest.

Main Methods:

Utilized six months of continuous production data (approx. 280,000 cycles) from a thermoplastic injection line.
Applied Long Short-Term Memory (LSTM) time-series models to analyze temporal patterns in torque, pressure, temperature, and rheology data.
Implemented a physically informed labelling strategy using volatile zones preceding non-conforming parts for supervised learning with sparse defect annotations.

Main Results:

The framework successfully modeled temporal patterns to identify drift conditions indicative of melt instability.
A physically informed labelling strategy enabled effective supervised learning, detecting instability windows minutes before actual defects occurred.
The system demonstrated the capability to recognize instability using standard machine signals without additional sensors.

Conclusions:

Data-driven temporal deep-learning models, specifically LSTM, offer a powerful approach for enhancing real-time monitoring in injection moulding.
The proposed framework facilitates proactive process adjustments, leading to improved stability and reduced scrap.
This methodology contributes to more robust and adaptive manufacturing operations through advanced instability detection.