Selected work

Representative Papers

ECLASS-Augmented Semantic Product Search for Electronic Components

Apr 21, 2026

This study addresses the lexical mismatch between natural language queries and structured descriptions of electronic components by proposing a large language model (LLM)-assisted dense retrieval and re-ranking approach that integrates hierarchical semantic information from the ECLASS standard. For the first time, the hierarchical ontology of ECLASS is embedded into the retrieval framework to bridge the semantic gap between user intent and sparse product descriptions. Experimental results demonstrate that the proposed method achieves a Hit@5 score of 94.3% on expert queries, substantially outperforming both BM25 (31.4%) and baseline LLM-based web search approaches. The method delivers significant improvements in both retrieval accuracy and efficiency, highlighting the effectiveness of leveraging standardized ontological structures to enhance semantic alignment in technical domains.

0 citationsRead paper

GDPR-Compliant Person Recognition in Industrial Environments Using MEMS-LiDAR and Hybrid Data

Feb 02, 2026

This work proposes an anonymized 3D point cloud–based person detection method using MEMS-LiDAR, addressing the limitations of conventional vision-based approaches in industrial indoor environments—namely, their susceptibility to lighting and visibility conditions and their difficulty in complying with privacy regulations such as GDPR. The approach introduces a hybrid training dataset that uniquely combines real-world MEMS-LiDAR data with synthetic LiDAR data generated from the CARLA simulator, enabling effective training of deep learning–based object detection models. By leveraging this synthetic data augmentation strategy, the method not only ensures privacy compliance but also reduces manual annotation costs by 50% and improves mean average precision by 44 percentage points compared to models trained solely on real data, thereby demonstrating the efficacy and practicality of synthetic data in industrial applications.

0 citationsRead paper

Physics-Informed Diffusion Models for Unsupervised Anomaly Detection in Multivariate Time Series

Aug 15, 2025

This paper addresses unsupervised anomaly detection in multivariate time series by proposing a Physics-Informed Diffusion Model (PIDM). Methodologically, it introduces prior physical laws into the diffusion training process via a weighted static scheduling scheme and constructs a physics-constrained loss function grounded in dynamic differential properties to explicitly enforce physical consistency among variables. This design enhances the model’s joint representation capability for both the underlying data distribution and physical dynamics—without requiring labeled anomalies. Experiments demonstrate that PIDM achieves significantly higher F1 scores than state-of-the-art baselines on both synthetic and multiple real-world datasets. Notably, it exhibits superior detection sensitivity and robustness under low anomaly-rate regimes, while also improving generative diversity and log-likelihood performance.

0 citationsRead paper

Recent publications

Latest Papers

ECLASS-Augmented Semantic Product Search for Electronic Components

Apr 21, 2026

This study addresses the lexical mismatch between natural language queries and structured descriptions of electronic components by proposing a large language model (LLM)-assisted dense retrieval and re-ranking approach that integrates hierarchical semantic information from the ECLASS standard. For the first time, the hierarchical ontology of ECLASS is embedded into the retrieval framework to bridge the semantic gap between user intent and sparse product descriptions. Experimental results demonstrate that the proposed method achieves a Hit@5 score of 94.3% on expert queries, substantially outperforming both BM25 (31.4%) and baseline LLM-based web search approaches. The method delivers significant improvements in both retrieval accuracy and efficiency, highlighting the effectiveness of leveraging standardized ontological structures to enhance semantic alignment in technical domains.

0 citationsRead paper

GDPR-Compliant Person Recognition in Industrial Environments Using MEMS-LiDAR and Hybrid Data

Feb 02, 2026

This work proposes an anonymized 3D point cloud–based person detection method using MEMS-LiDAR, addressing the limitations of conventional vision-based approaches in industrial indoor environments—namely, their susceptibility to lighting and visibility conditions and their difficulty in complying with privacy regulations such as GDPR. The approach introduces a hybrid training dataset that uniquely combines real-world MEMS-LiDAR data with synthetic LiDAR data generated from the CARLA simulator, enabling effective training of deep learning–based object detection models. By leveraging this synthetic data augmentation strategy, the method not only ensures privacy compliance but also reduces manual annotation costs by 50% and improves mean average precision by 44 percentage points compared to models trained solely on real data, thereby demonstrating the efficacy and practicality of synthetic data in industrial applications.

0 citationsRead paper

Physics-Informed Diffusion Models for Unsupervised Anomaly Detection in Multivariate Time Series

Aug 15, 2025

This paper addresses unsupervised anomaly detection in multivariate time series by proposing a Physics-Informed Diffusion Model (PIDM). Methodologically, it introduces prior physical laws into the diffusion training process via a weighted static scheduling scheme and constructs a physics-constrained loss function grounded in dynamic differential properties to explicitly enforce physical consistency among variables. This design enhances the model’s joint representation capability for both the underlying data distribution and physical dynamics—without requiring labeled anomalies. Experiments demonstrate that PIDM achieves significantly higher F1 scores than state-of-the-art baselines on both synthetic and multiple real-world datasets. Notably, it exhibits superior detection sensitivity and robustness under low anomaly-rate regimes, while also improving generative diversity and log-likelihood performance.

0 citationsRead paper