In Variable Data Printing (VDP) defect detection, legitimate structural variations introduced by dynamic characters and personalized content often cause traditional single-stream detectors to misclassify normal layout changes as defects. Moreover, real industrial scenarios commonly involve nonlinear geometric misalignment and cross-modal discrepancies between photographed images and template images. Under such conditions, direct end-to-end optimization suffers from severe spatial feature interference and optimization instability caused by unreliable cross-modal interactions. To address these issues, we propose Geometric Alignment and Zero-Conv Prior Injection YOLO (GeoZero-YOLO), a reference-guided dual-stream defect detection framework based on Progressive Alignment-Injection Training (PAIT). Specifically, a correlation-guided feature alignment module is introduced to alleviate cross-modal geometric deviations and progressively establish stable structural correspondence priors through an asymmetric optimization strategy. Subsequently, a difference purification mechanism is introduced to suppress pseudo-difference responses and enhance defect-related discrepancy representations. Finally, a progressive prior injection strategy is employed to integrate the refined reference-guided cues while preserving the intrinsic discriminative ability of the original detector. Extensive experiments on a self-built ID Card dataset and the public PCBData dataset demonstrate the effectiveness of the proposed framework for reference-guided VDP defect detection.
Few-shot fault diagnosis under varying operating conditions requires models to extract transferable fault representations from limited labeled data. Although meta-learning enables rapid adaptation to new tasks, variations in operating conditions can alter the distribution and manifestation of fault features, making it difficult to transfer diagnostic knowledge reliably across tasks. To address this issue, a cognition-inspired hierarchical meta-learning method (CHML) is proposed. At the representation front end, short-time Fourier transform convolutional kernels incorporate time–frequency analysis into feature extraction, allowing fault-related patterns with explicit time–frequency meaning to be retained across operating conditions. Independent multi-expert convolutional branches are then used to capture complementary characteristics of the input signals. On this basis, a recursive feature-refinement strategy iteratively updates the learned representations, providing the meta-learner with more discriminative features for cross-condition task adaptation. The proposed method was validated on two bearing datasets under multiple operating conditions, and the experimental results confirmed its effectiveness for cross-condition few-shot fault diagnosis.
Developing high-fidelity digital twin models for electric spindles requires rigorously predicting the coupled thermal-vibration dynamics, where friction-induced heat actively drives parameter drift such as thermal stiffening and viscosity reduction. While structure-preserving frameworks like Port-Hamiltonian Neural Networks and Dissipative Hamiltonian Neural Networks excel in energy-based modeling, they predominantly treat dissipation via resistive ports or energy sinks, implicitly assuming an isothermal environment, failing to capture the reciprocal thermo-mechanical feedback, leading to substantial errors in vibration predictions. To bridge this gap, we propose a Physics-Informed Thermodynamical Hybrid Modeling Method based on the GENERIC formalism and the dissipative Hamiltonian dynamics. A Thermodynamical Hamiltonian Neural Network (THNN) is introduced to identify temperature-dependent constitutive parameters within a thermodynamically consistent physical model, rigorously tracking the conversion of dissipated mechanical work into entropy and temperature evolution and thereby closing the thermo-mechanical loop by design. Through a comprehensive proof-of-concept validated under diverse benchmark scenarios, including noise robustness and parameter sensitivity analyses, we demonstrate that THNN reduces the MAE by 44.1
Accurate and real-time prediction of laser cladding morphology is critical for quality control in directed energy deposition processes. This study proposes a physics-informed neural network (PINN) for predicting the cross-sectional profiles of laser cladding by three input parameters: laser power, scanning speed, and powder flowrate. The advantage of our method is that PINN embeds a learnable empirical model. The coefficients of the model are collaboratively optimized with the network weights via physics-consistency loss and data loss. Experimental results demonstrate that the proposed PINN outperforms purely data-driven and fixed-parameter empirical models, especially under conditions of limited training data. The trained model is successfully deployed in a digital twin-based laser cladding system. Thus, without the need for online geometric sensing, the system can achieve real-time morphology prediction and three-dimensional geometric structure reconstruction solely through process parameters. The proposed method provides an efficient and robust predictive model for digital twin-enabled laser cladding applications.
Quality control in polyester fiber polymerization is a multi-objective process-parameter optimization problem complicated by nonlinear parameter-quality coupling and the limited reuse of abnormal-condition knowledge. Existing surrogate-assisted optimization methods typically search within fixed engineering boundaries and lack a mechanism for converting unstructured process knowledge into computable constraints. To address this limitation, this study develops PLMC as a LightRAG empowered multi-objective optimization framework that integrates knowledge-to-constraint transformation, surrogate-based objective evaluation, and knowledge-constrained dual-population co-evolution. Using a polyester polymerization process knowledge graph, PLMC retrieves mechanistic causes and control measures. A domain-adapted large language model then uses this information to generate condition-specific prior knowledge constraints. These constraints are embedded into PLMC-DP, a knowledge-constrained dual-population co-evolution module. The global population explores the engineering feasible space, whereas the knowledge-guided population searches within the prior-constraint region. Candidate solutions are evaluated using a surrogate error compensation-enhanced Kolmogorov–Arnold Network (SEC-KAN). SEC-KAN predicts the effects of key process parameters on intrinsic viscosity, diethylene glycol content, and terminal carboxyl group concentration while correcting local surrogate bias using historical residuals. Under the tested settings, experiments on industrial production-line data show that the complete PLMC framework achieves better Pareto-set quality and convergence stability than the baseline algorithms. The resulting hypervolume (HV) and inverted generational distance (IGD) are 0.8387 and 0.0473, respectively. Stable process-control performance across fully drawn yarn (FDY), partially oriented yarn (POY), and draw textured yarn (DTY) product scenarios indicates the applicability of PLMC to diverse polyester products.
