Long Short-Term Memory (LSTM) -Based Neural Network Model for Optimizing Composite Manufacturing Process using Autoclave

To enhance the efficiency of polymer composite by autoclave curing, this work introduces a Long Short-Term Memory (LSTM) neural network capable of forecasting temperature profiles with high accuracy. The model predicts time-series data gathered from 16 thermocouples positioned across three composite panels during a typical cure cycle. By using augmentation techniques such as jittering, time scaling, and temporal shifts, the model's predictive performance and generalizability were significantly improved. The optimal sequence length was identified as 30-time steps, offering a balance between computational demand and accuracy. With a coefficient of determination (R²) nearing 0.987. These results indicate the model’s promise in enabling real-time control and dynamic process adjustments. This approach supports the goal of minimizing cycle duration and energy use while ensuring consistent curing—crucial for defect prevention. The technique holds value in aerospace and other precision-driven industries, where optimized composite curing is vital for quality and performance.