Dynamic Process Flexibility Analysis Using Neural Networks and a Volumetric Flexibility Index

Introduction:

Some processes are not safely described by a single snapshot, because their behavior changes as time passes. A system that looks flexible now may become fragile later, especially when uncertain inputs keep moving. This paper extends flexibility analysis to dynamic process systems by combining a physics-informed neural network for control with a convolutional neural network classifier. The first model acts as a fast surrogate for the system dynamics, and the second model learns which uncertain trajectories remain feasible. The proposed dynamic volumetric flexibility index captures the size of the feasible region over time, including nonconvex or discontinuous regions that are hard for conventional methods to handle. The approach is demonstrated on buffer vessel systems.