Abstract Technical product development aims to fulfill stakeholder requirements by delivering products with demand-driven functionality. Numerous influencing factors, such as external disturbances, input variable variations, and changes in internal product characteristics (e.g., wear and deformations), can affect the product functions and lead to unintended behavior. To improve the robustness of technical products, it is essential to identify and analyze these factors and their potential impacts during the concept phase of product development. This requires structured and reusable knowledge. Model-Based Systems Engineering (MBSE) supports the development of complex technical products but can only serve as a basis for determining robustness to a limited extent due to the available information and the degree of maturity in the models. Reusing existing product knowledge within a specific context is essential for product development, especially to improve robustness. Coupling MBSE approaches with reusable knowledge representations, such as knowledge graphs, can support analysis and decision-making. To achieve this, knowledge representation must provide empirical knowledge about potential failures, influencing factors, and so forth. This contribution presents an approach for coupling knowledge graphs with MBSE models using SysML to perform context-specific queries in the knowledge graph based on the specifications in the SysML model. Additionally, the contribution proposes a structured approach for enriching the knowledge graphs. The approach is presented and implemented using an example application of a mobile robot.
This study investigates the use of large language models (LLMs) to classify question utterances within verbal design protocols according to Eris' (2004) taxonomy. We evaluate the performance of two proprietary LLMs - OpenAI's GPT-4.1 and Anthropic's Claude Sonnet 4.5 - across experiments designed to assess classification accuracy, sensitivity to prompt configuration and in-context learning (ICL), and generalization across datasets and models. Using two human-coded datasets of differing size and quality, we measure alignment between LLM-generated labels and human judgments at both question category and subcategory levels. Results show that both LLMs achieved moderate to strong alignment rates at the category level (up to 85.7% for GPT-4.1 and 82.9% for Claude Sonnet 4.5), with substantially lower alignment at the more granular subcategory level. Performance differences across prompt configurations and ICL conditions were small, indicating robust generalization across datasets and transferability of prompt designs. While these results suggest that LLMs can effectively support scalable question classification, human judgment and oversight remain essential. Future research should explore the development and evaluation of alternative hybrid human-LLM workflows in protocol analysis, as well as the use of smaller or open-source models to address data privacy concerns.
Digital transmission of traditional cultural heritage faces challenges of insufficient user engagement and superficial dissemination. Using the Cizhou kiln as a case study, this research proposes the "Creation-as-Transmission" design paradigm, which transcends traditional unidirectional display models by transforming users from cultural consumers to cultural co-creators. This paradigm systematically integrates situated cognition theory and co-creation theory, achieving deep cultural learning and socialized transmission through AI collaborative creation. Empirical studies (N = 200) demonstrate that compared to traditional methods, this paradigm significantly enhances cultural cognition (74% improvement) and the proportion of spontaneous sharing on social media (82.6% sharing rate). More importantly, this research provides the first quantification of causal mechanisms between creative behaviors and cultural cognition, establishing a predictive model explaining 73.1% of cognitive variance and offering mechanistic evidence for situated learning theory's application in cultural heritage transmission. The research contributions have threefold value: theoretically, it expands HCI understanding of the relationship between "design behaviors and cognitive effects"; methodologically, it provides a transferable implementation framework for other intangible cultural heritage; practically, it opens new pathways for digital technologies to promote sustainable cultural transmission.
This study assesses classification-based predictive maintenance (PdM) for aircraft engines on the NASA Commercial Modular Aero-Propulsion System Simulation dataset and addresses the lack of wide-scope, unified benchmarks. PdM is cast as a short-term binary task - predicting whether an engine will fail within the next 30 cycles - and a comparison is conducted across 10 machine-learning models (Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, k-Nearest Neighbor, Na & iuml;ve Bayes, Extreme Gradient Boosting, LightGBM, CatBoost, and Gradient Boosting) and 3 deep-learning models (Multilayer Perceptron, Gated Recurrent Unit, and Long Short-Term Memory). A leakage-aware pipeline applies Min-Max scaling; class imbalance is handled with Synthetic Minority Over-sampling Technique where appropriate; hyperparameters are tuned via GridSearchCV/BayesSearchCV; and performance is reported with accuracy, precision, recall, F1-score, and receiver operating characteristic-area under the curve (ROC-AUC), complemented by Shapley Additive Explanations (SHAP) explainability and nonparametric significance tests. Sequence models delivered the strongest performance: LSTM achieved Accuracy = 0.981 (Macro-F1 = 0.92; ROC-AUC = 0.96), and GRU achieved ROC-AUC = 0.97 with Accuracy = 0.975. Among classical learners, LightGBM reached Accuracy = 0.972 (Macro-F1 = 0.86; ROC-AUC = 0.93). These gains over weaker baselines were statistically significant across folds. Framing PdM as near-term failure classification yields operationally interpretable alerts. Models that explicitly capture temporal dependencies (GRU/LSTM) best track short-horizon failure dynamics, while gradient-boosted trees offer competitive, lightweight alternatives. The benchmark and analysis (including SHAP) provide a reproducible reference for model selection in aviation PdM.
To address the challenges of low detection accuracy, missed detections, and high false detection rates for small targets in PCB defect detection tasks, this study proposes an enhanced YOLOv8 methodology incorporating feature focusing and multi-scale fusion techniques. Initially, a lightweight GTADH module is integrated into the detection head of YOLOv8, employing a shared convolution and task alignment mechanism to minimize model parameters while enhancing classification and localization accuracy. Subsequently, an adaptive feature-focusing module is introduced into the feature fusion network to bolster the detection capabilities for small targets via multi-scale feature fusion. Finally, the reverse residual moving block (iRMB) and attention mechanisms are combined within the backbone network to facilitate efficient extraction and fusion of feature information, preserving finer details of small targets. Experimental results demonstrate that the Improved YOLO algorithm achieves a 1.3% increase in detection accuracy and a 7.3% enhancement in mAP50:90 evaluation standards compared to the original YOLOv8s algorithm on the PCB defect dataset, while also reducing model size by 60%, thus showcasing its effectiveness in small target detection tasks.
