Semi-supervised CAPP Transformer Learning via Pseudo-labeling

TL;DR

The paper tackles data scarcity in high-level CAPP, where transformer models struggle with generalization when labeled data is limited. It introduces a semi-supervised pipeline that trains a binary oracle on the transformer's behavior and uses its high-confidence predictions to pseudo-label unseen parts for one-shot retraining, avoiding manual labeling. Experiments on simulated full-distribution data show consistent accuracy gains, particularly in the smallest data regimes, with oracle-based augmentation outperforming random augmentation and baseline retraining. The work demonstrates practical potential for data-scarce manufacturing environments and suggests future work on alternate oracle designs and scalability.

Abstract

High-level Computer-Aided Process Planning (CAPP) generates manufacturing process plans from part specifications. It suffers from limited dataset availability in industry, reducing model generalization. We propose a semi-supervised learning approach to improve transformer-based CAPP transformer models without manual labeling. An oracle, trained on available transformer behaviour data, filters correct predictions from unseen parts, which are then used for one-shot retraining. Experiments on small-scale datasets with simulated ground truth across the full data distribution show consistent accuracy gains over baselines, demonstrating the method's effectiveness in data-scarce manufacturing environments.

Figures (2)

• Figure 1: Flowchart showing the process of generating viable process plans.
• Figure 2: Sequence accuracy. Solid = Detector, dashed = Random, horizontal = baseline. Detector yields consistent gains, strongest in low-resource settings.