A Learning-Centric End-To-End Hybrid System For In-Sıtu Thermographic Inspectıon In Automated Fiber Placement

Automated Fiber Placement (AFP) technology continues to lead and transformcomposite manufacturing, but its progress remains constrained by persistent qualityassurance challenges. Emerging material and process defects compromise structuralintegrity, while inspection practices remain largely manual, reactive, and incapableof real-time feedback—leading to costly downtimes. Existing methods face two corelimitations: partial, mostly offline frame-wise analysis and the lack of global, temporallyconsistent lay-up visualization. To address these challenges, this thesis proposesa dual-framework, real-time, learning-centric inspection system that unifies local defectintelligence with global lay-up traceability—operating machine-independentlyusing only thermal imagery.The first framework implements a hybrid, frame-wise defect analysis and quality assessmentsystem composed of three synergistic modules—Dynamic Tow Identification,Hierarchical Defect Identification, and Lay-up Quality Evaluation—configuredfor parallel or conditional execution to ensure runtime efficiency. It begins withsetup-independent spatial–temporal analysis that estimates tow boundaries withsub-pixel accuracy (mean error < 0.8 px), enabling tow-level reasoning. In parallel,high-level defect detection uses a Gabor-based SVM classifier exceeding 95%accuracy and recall. Defective frames are forwarded to a custom 12-layer deep convolutional neural network for fine-grained classification, achieving 96.4% multi-classaccuracy across defect types. Upon detection, a seeded active contour model adaptedto thermal textures yields interpretable segmentation masks with 93.2% pixel accuracy—outperforming baseline methods under noise. These outputs are fused withtow geometry to compute the novel Defect Area Percentage (DAP) metric, whichquantifies severity at both tow and course levels and forms the core of the operatordecision support system (AFP-DSS). Operating at 5 fps, the framework enables fullyautonomous, real-time AFP quality inspection.The second framework introduces ThermoRAFT-AFP, a machine-independent, deeplearning-based thermal motion estimation core tailored to AFP. Built upon theRAFT optical flow, it incorporates AFP-specific augmentations and runtime optimizations(e.g., predictive initialization, drift correction, adaptive early exit) forstable, precise thermal flow tracking. It estimates dense inter-frame motion to drivea two-stage reconstruction pipeline that generates course-wise mosaics and assembleshigh-fidelity, ply-level laminate visualizations. This restores temporal consistencyand global alignment across evolving lay-ups, enabling traceable defect analysis andmirroring industrial inspection workflows. ThermoRAFT-AFP achieves a velocityestimation RMSE of 4.83 mm/s and cumulative drift below 0.1%, while maintainingrobustness down to SNR = 14.4 dB and sustaining real-time operation at 25 fps. Robustagainst setup variations without retraining and near-zero tuning, it producesinterpretable, temporally aligned visualizations that support laminate-scale defectpropagation analysis.Together, the two frameworks form an integrated system that fuses frame-wise defectoutputs with global reconstructions into a unified thermal quality view. Thisfusion links semantic analysis with temporal context in an interpretable visualizationaligned with real-world inspection workflows. Evaluated on over 13,000 thermalframes spanning diverse speeds, geometries and defect types, the system meetsaerospace-grade benchmarks, eliminates robot-coupled dependencies, and deliversscalable, real-time AFP quality inspection directly from thermal imagery.