Sajib Mistry

Digital Light Processing (DLP) additive manufacturing generates substantial thermoset waste, yet few computational frameworks exist to guide hierarchical design of recycled DLP composites toward bio-inspired mechanical performance. This work presents a multi-scale framework integrating four coupled modules: a cure-state-dependent physics engine combining the DiBenedetto model, Mori-Tanaka homogenisation, Classical Laminate Theory, and Timoshenko shear correction; Buckingham Pi dimensional analysis; physics-engine-derived synthetic data generation using Latin Hypercube Sampling with Gaussian process residual correction; and dualtarget Random Forest and XGBoost models validated by leave-one-out cross-validation against 18 experimental formulations from a recycled Monocure 3D Rapid Epoxy composite system. Cure state was confirmed as the dominant driver of flexural stiffness (3.40× modulus ratio between 30and 60-minute post-cure). The framework achieved R² = 0.949 and MAPE = 12.7% for flexural modulus, with XGBoost achieving MAPE = 6.4% for flexural strength; interface compliance emerged as the highest-importance strength feature. Bio-mimicry scoring across 2,000 feasible designs revealed all candidates cluster near the wood bio-target, with bone and nacre unreachable at the near-unity modulus contrast (Ep/Em ≈ 1.2). Hollow cross-section architecture is predicted to deliver a +56% improvement in flexural stiffness-to-mass at a 40% mass reduction. The framework identifies 600 high-priority designs for validation and establishes structural geometry as the primary lever for advancing recycled thermoset composites toward bio-inspired performance.

Digital Light Processing (DLP) additive manufacturing generates substantial thermoset waste, yet no computational framework exists to guide hierarchical design of recycled DLP composites toward bio-inspired mechanical performance. This work presents a multi-scale framework integrating four coupled modules: a cure-state-dependent physics engine combining the DiBenedetto model, Mori-Tanaka homogenisation, Classical Laminate Theory, and Timoshenko shear correction; Buckingham Pi dimensional analysis; physics-engine-derived synthetic data generation using Latin Hypercube Sampling with Gaussian process residual correction; and dual-target Random Forest and XGBoost models validated by leave-one-out cross-validation against 18 experimental formulations from a recycled Monocure 3D Rapid Epoxy composite system. Cure state was confirmed as the dominant driver of flexural stiffness (3.40× modulus ratio between 30- and 60-minute post-cure). The framework achieved R2 = 0.949 and MAPE = 12.7% for flexural modulus, with XGBoost achieving MAPE = 6.4% for flexural strength; interface compliance emerged as the highest-importance strength feature. Bio-mimicry scoring across 2,000 feasible designs revealed all candidates cluster near the wood bio-target, with bone and nacre unreachable at the near-unity modulus contrast (Ep/Em ≈ 1.2). Hollow cross-section architecture delivered +56% flexural stiffness-to-mass improvement at 40% mass reduction. The framework identifies 600 high-priority designs for validation and establishes structural geometry as the primary lever for advancing recycled thermoset composites toward bio-inspired performance.

The development of filled photopolymer composites for Digital Light Processing (DLP) additive manufacturing requires optimizing processing parameters to achieve the desired mechanical properties. Traditional experimental approaches are time-intensive, while physics-based models often struggle to capture the complex interactions among parameters. This study presents a physics-informed machine learning framework that combines Random Forest with Bayesian optimization (RF-BO) to predict the ultimate tensile strength in recycled thermoset resin composites manufactured via DLP. A validation dataset of 19 systematically varied formulations (each with n = 5 measurement replicates for reliability) was generated and augmented with 1500 physics-informed synthetic samples to enable robust model training. The limited experimental dataset, while insufficient for traditional statistical inference, provided critical validation of physical trends, including non-monotonic particle-size effects and optimal processing windows. Six machine learning algorithms were evaluated, with RF-BO achieving superior performance (R² = 0.9125, MSE = 1.07 MPa). The framework identified optimal processing conditions of 59–64 μm particle size, 5.0 ± 0.5 wt.% concentration, and 60 min cure time, predicting a maximum UTS of 43.84 MPa with a prediction error of less than 1.0 MPa. Feature importance analysis revealed that cure time was the dominant parameter (40%), followed by particle size (30%), validating the physical interpretability. This approach demonstrates significant potential for accelerating materials design in composite additive manufacturing while maintaining physically meaningful predictions.

Big data scenarios, where massive, heterogeneous datasets are distributed across clients, demand scalable, privacy-preserving learning methods. Federated learning (FL) enables decentralized training of machine learning (ML) models across clients without data centralization. Decentralized training, however, introduces a computational burden on client devices. U-shaped federated split learning (UFSL) offloads a fraction of the client computation to the server while keeping both data and labels on the clients' side. However, the intermediate representations (i.e., smashed data) shared by clients with the server are prone to exposing clients' private data. To reduce exposure of client data through intermediate data representations, this work proposes k-anonymous differentially private UFSL (KD-UFSL), which leverages privacy-enhancing techniques such as microaggregation and differential privacy to minimize data leakage from the smashed data transferred to the server. We first demonstrate that an adversary can access private client data from intermediate representations via a data-reconstruction attack, and then present a privacy-enhancing solution, KD-UFSL, to mitigate this risk. Our experiments indicate that, alongside increasing the mean squared error between the actual and reconstructed images by up to 50% in some cases, KD-UFSL also decreases the structural similarity between them by up to 40% on four benchmarking datasets. More importantly, KD-UFSL improves privacy while preserving the utility of the global model. This highlights its suitability for large-scale big data applications where privacy and utility must be balanced.

Test-time adaptation (TTA) in federated learning (FL) is crucial for handling unseen data distributions across clients, particularly when faced with domain shifts and skewed class distributions. Class Imbalance (CI) remains a fundamental challenge in FL, where rare but critical classes are often severely underrepresented in individual client datasets. Although prior work has addressed CI during training through reliable aggregation and local class distribution alignment, these methods typically rely on access to labeled data or coordination among clients, and none address class unsupervised adaptation to dynamic domains or distribution shifts at inference time under federated CI constraints. Revealing the failure of state-of-the-art TTA in federated client adaptation in CI scenario, we propose pFedBBN,a personalized federated test-time adaptation framework that employs balanced batch normalization (BBN) during local client adaptation to mitigate prediction bias by treating all classes equally, while also enabling client collaboration guided by BBN similarity, ensuring that clients with similar balanced representations reinforce each other and that adaptation remains aligned with domain-specific characteristics. pFedBBN supports fully unsupervised local adaptation and introduces a class-aware model aggregation strategy that enables personalized inference without compromising privacy. It addresses both distribution shifts and class imbalance through balanced feature normalization and domain-aware collaboration, without requiring any labeled or raw data from clients. Extensive experiments across diverse baselines show that pFedBBN consistently enhances robustness and minority-class performance over state-of-the-art FL and TTA methods.