Abstract

Filament-based material extrusion has evolved from a low-cost prototyping technique into a manufacturing process increasingly considered for functional and quality-critical applications. However, its industrial adoption remains constrained by a persistent traceability gap between planned process parameters and final part outcomes. During fabrication, heterogeneous monitoring data (logs, images, point clouds, or controller records) are often stored as loosely connected artefacts whose interpretation relies on undocumented conventions and post-hoc alignment. Meanwhile, existing standards for polymer material extrusion focus mainly on ex situ verification and offer limited guidance for in situ monitoring or defect classification. Consequently, quality assurance often lacks a consistent framework linking process history, spatial evidence, and geometry-resolved outcomes.

This thesis addresses this limitation by proposing a run-scoped spatiotemporal information architecture for multisensor monitoring in filament-based material extrusion. Its objective is to transform isolated monitoring outputs into structured manufacturing evidence that can be retrieved, audited, and analysed at the deposition scale, where quality-relevant deviations arise. To achieve this, the thesis introduces the Spatiotemporally Enhanced Information Model (STEIM), a run-centric evidence framework anchored to the planned toolpath. The model adopts the deposition trajectory as a canonical backbone and uses the planned points as deterministic anchors to associate geometry, execution traces, and inspection artefacts. It formalises digital-thread states, reference frames, transformations, validity domains, and derivation lineage so heterogeneous observations remain interpretable without hidden alignment assumptions.

Building on this representation layer, the thesis proposes an occupancy-based conformance model for local quality assessment. Instead of relying on rigid defect taxonomies, deposition is evaluated by comparing expected and observed material occupancy within cross-sectional regions attached to each canonical anchor. Occupancy becomes a common descriptor that converts heterogeneous inspection evidence into compact pointwise indicators that can be aggregated across beads, layers, and regions. Descriptive occupancy artefacts remain separate from conformance decisions, issued only when evaluated against an explicitly defined specification baseline.

To operationalise these concepts, the framework is implemented on a multi-instrumented filament extrusion platform created by retrofitting a commercial Cartesian printer. The system integrates two monitoring regimes. A synchronous subsystem records execution-coupled signals including kinematics, temperatures, nozzle strain, and vibrations on a shared hardware sampling lattice driven by an FPGA master clock, with deterministic segmentation provided by controller-emitted End-of-Movement markers. In parallel, an asynchronous inspection subsystem captures geometry-resolved evidence through telecentric imaging, laser triangulation, and line-scan digitisation. Encoder-based pose logging and calibrated transformations allow inspection artefacts to be expressed in the machine coordinate system and linked to deposition events.

The framework is validated through two studies. The first demonstrates a run-level instantiation in which planned, executed, and inspected artefacts coexist within a single query able record while preserving explicit state separation and traceability down to individual points along the planned toolpath. The second constructs a traceable dataset linking synchronous process signatures with asynchronous geometric measurements during spiral deposition under random feedrate variations. Telecentric bead-width measurements provide supervisory labels for machine-learning models trained on synchronous signals available during fabrication. Results show that a virtual sensor can estimate bead width from execution-coupled evidence, illustrating how the architecture supports predictive modelling and future process compensation strategies.

Overall, the thesis establishes a pathway from monitoring to governed manufacturing evidence, from evidence to geometry resolved description, and from description to predictive use. Its main contribution is a run-centric framework in which observations, mappings, and derived indicators remain explicit, versioned, and auditable, enabling more robust quality assurance in filament-based material extrusion.