PatchRLNet: A Framework Combining a Vision Transformer and Reinforcement Learning for The Separation of a PTFE Emulsion and Paraffin

Abstract

During the production of a PolyTetraFluoroEthylene(PTFE) emulsion, it is crucial to detect the separation between the PTFE emulsion and liquid paraffin in order to purify the PTFE emulsion and facilitate subsequent polymerization. However, the current practice heavily relies on visual inspections conducted by on-site personnel, resulting in not only low efficiency and accuracy, but also posing potential threats to personnel safety. The incorporation of artificial intelligence for the automated detection of paraffin separation holds the promise of significantly improving detection accuracy and mitigating potential risks to personnel. Thus, we propose an automated detection framework named PatchRLNet, which leverages a combination of a vision transformer and reinforcement learning. Reinforcement learning is integrated into the embedding layer of the vision transformer in PatchRLNet, providing attention scores for each patch. This strategic integration compels the model to allocate greater attention to the essential features of the target, effectively filtering out ambient environmental factors and background noise. Building upon this foundation, we introduce a multimodal integration mechanism to further enhance the prediction accuracy of the model. To validate the efficacy of our proposed framework, we conducted performance testing using authentic data from China’s largest PTFE material production base. The results are compelling, demonstrating that the framework achieved an impressive accuracy rate of over 99% on the test set. This underscores its significant practical application value. To the best of our knowledge, this represents the first instance of automated detection applied to the separation of the PTFE emulsion and paraffin.