Additive manufacturing (AM) monitoring is fundamentally constrained by the severe scarcity of annotated data for layer-wise segmentation. This paper addresses this bottleneck by introducing a scalable, high-fidelity synthetic data generation pipeline built on the Slice-100K dataset, capable of producing large volumes of layer-wise semantic segmentation masks. Through analysis of this large-scale synthetic data, we identify a systemic foreground–background class imbalance (1:24 ratio) inherent to AM monitoring, which causes standard Dice loss formulations to diverge catastrophically into a phenomenon we formalize as the “Dice Crash.” To effectively leverage large amounts of unlabeled data, we adapt the Meta Pseudo-Labeling (MPL) framework for industrial segmentation. We evaluate MPL’s true marginal utility by integrating it with both a standard U-Net and a robust state-of-the-art nnU-Net architecture. Experimental outputs show that while MPL yields substantial performance gains (+15.2%) on weak baselines, integrating it with an optimally configured strong baseline consistently improves segmentation accuracy and suppresses false foreground detections, thereby mitigating confirmation bias. These findings demonstrate that semi-supervised learning via continuous bilevel optimization offers a practical and robust enhancement to data-scarce additive manufacturing monitoring. Because any hidden defects in the topmost layer will be permanently buried by subsequent extrusion, this foundational layer-wise segmentation step is the most critical primitive of the monitoring pipeline.

Semantic integration in Additive Manufacturing (AM) supervision demands unifying heterogeneous sensor streams and production context under strict latency and data-integrity constraints, yet conventional semantic stacks suffer from unpredictable query latencies, unguarded SPARQL (SPARQL Protocol and Resource Description Framework [RDF] Query Language) endpoints, and prohibitive reasoning overheads. This paper presents Semantic DREAM (Distributed Resilient and Emergent intelligence-based Additive Manufacturing), a security-aware architecture co-designed over a Web Ontology Language 2 Rule Language (OWL 2 RL) modular substrate with three mechanisms: (1) an abstract syntax tree (AST)-guarded read-only SPARQL engine that blocks destructive updates and caches template results; (2) a multi-protocol mashup layer aligning live Message Queuing Telemetry Transport (MQTT), Representational State Transfer (REST), Structured Query Language (SQL), and file feeds through Sensor, Observation, Sample, and Actuator (SOSA) observation minting and Quantities, Units, Dimensions, and Types (QUDT) unit normalization; and (3) an ingestion-time Shapes Constraint Language (SHACL) validation pipeline that gates observations before graph commit, decoupling structural checks from reasoning. Deductive materialization is performed offline at startup and verified with ROBOT as an intentional design choice to eliminate reasoning overhead from the streaming path, so online monitoring queries evaluate directly over the asserted graph. Single-host, lab-scale evaluations show that cached queries achieve sub-15 ms response times for recurring monitoring, while the worst-case event-driven decision cycle (incurring a post-invalidation cache miss) completes in 110 ms, well within the 200 ms supervisory deadline with 100% defect interception on injected anomalies. All measurements are single-host and lab-scale; factory-network transit and recorded telemetry are not evaluated.