Yuan Xu

Environmentally friendly indium phosphide (InP) quantum dots (QDs) offer tunable emission from visible to infrared wavelengths, while synthesizing InP QDs with high brightness and narrow-bandwidth emission remains challenging. In this work, controlled surface Ga treatment was used to suppress the Ostwald ripening process during the synthesis of InP QDs with an enhanced optical performance. The tunable interaction of Ga with the QD surface was systematically investigated, and Ga treatment narrowed the full width at half-maximum (fwhm) of InP/ZnSeS QDs from 52 to 44 nm. Subsequent optimization of ZnS shell growth suppressed spectral broadening, while further increasing the ZnSeS interlayer thickness ultimately yielded a narrow fwhm of 46 nm and a high photoluminescence quantum yield (PLQY) of 96.15%. Improvement in color rendering index and correlated color temperature using InP-based quantum dots for white light-emitting diodes (LEDs) was demonstrated, and prototype devices exhibit an ultrahigh color rendering index of 95.6, a low correlated color temperature of 4414 K, and CIE coordinates of (0.36, 0.36). This work provides a promising way to obtain InP QDs with narrow-bandwidth emission and high brightness for future research and applications.

Reliable detection of distant smoke plumes and small fires in long-range industrial surveillance remains highly challenging due to low-contrast targets, complex outdoor backgrounds, and severe illumination variations such as nighttime lights and specular reflections on metallic surfaces. Lightweight single-stage detectors, essential for real-time edge deployment, are particularly prone to missed weak targets and frequent false alarms from confusing bright regions. This paper presents a semantic-guided prototype distillation framework with vision-language refinement to address these issues. A high-capacity teacher first constructs target prototypes for genuine fire/smoke regions and confuse prototypes for illumination-related distractors, guided by detection-driven probability maps. A pretrained vision-language model then refines these maps by aligning teacher visual features with domain-adaptive text prototypes optimized by continuous prompt tokens, thereby purifying unreliable supervision. The refined maps drive a mask-guided multi-scale contrastive alignment process, enabling a compact student detector to learn discriminative features that focus on true hazardous regions while suppressing misleading responses. Experiments on PASCAL VOC and a hybrid industrial Fire&Smoke dataset show consistent gains over the evaluated distillation baselines. On the reported Fire&Smoke test split, the method improves the mean AP of YOLO26-small from 51.5 to 63.8 and reduces the false alarm rate from 14.7% to 3.4%, while maintaining an 8.4M-parameter student with no additional inference module. These results support the method's effectiveness in the evaluated petrochemical and chemical-facility surveillance settings; transfer to other industrial sites remains to be established.

This paper proposes a knowledge graph-guided condition-aware meta-learning (KG-CAML) method for few-shot fault diagnosis under sensor data variations induced by fluctuating raw materials and changing operational loads. Within a meta-learning framework, KG-CAML integrates data-driven sensor correlations with expert process knowledge by constructing a KG that captures dependency structures among equipment and measurement variables. A relational graph convolutional network is employed as the encoder to extract consistent feature representations. On this basis, the correlation alignment metric is introduced to quantify the distributional similarity and alignment loss between training tasks and test conditions. Furthermore, to mitigate knowledge forgetting during meta-learning, a task memory updater is designed to dynamically update and retain task representations across stages, thereby continuously accumulating and transferring critical diagnostic knowledge. Experimental results on Tennessee Eastman processes and Wastewater Treatment Plant demonstrate that the diagnostic performance of KG-CAML outperforms state-of-the-art methods under multiple working conditions and few-shot scenarios significantly.

Reliable detection of distant smoke plumes and small fires in long-range industrial surveillance remains highly challenging due to low-contrast targets, complex outdoor backgrounds, and severe illumination variations such as nighttime lights and specular reflections on metallic surfaces. Lightweight single-stage detectors, essential for real-time edge deployment, are particularly prone to missed weak targets and frequent false alarms from confusing bright regions. This paper presents a semantic-guided prototype distillation framework with vision--language refinement to address these issues. A high-capacity teacher first constructs target prototypes for genuine fire/smoke regions and confuse prototypes for illumination-related distractors, guided by detection-driven probability maps. A pretrained vision--language model then refines these maps by aligning teacher visual features with domain-adaptive text prototypes optimized by continuous prompt tokens, thereby purifying unreliable supervision. The refined maps drive a mask-guided multi-scale contrastive alignment process, enabling a compact student detector to learn discriminative features that focus on true hazardous regions while suppressing misleading responses. Extensive experiments on PASCAL VOC and a hybrid industrial Fire\&Smoke dataset demonstrate that the proposed method outperforms classical and recent distillation baselines as well as task-specific smoke/fire detectors. In particular, it improves the mean AP of YOLO26-small from 51.5 to 63.8 and reduces the false alarm rate from 14.7% to 3.4%, while maintaining an 8.4M-parameter student with no additional inference cost. The framework offers a robust and practical solution for reliable early warning in resource-constrained industrial safety monitoring systems.

Soft sensors are essential for online prediction of key quality variables in industrial processes. However, practical process data are often affected by asynchronous sensor sampling, delayed laboratory analysis, and heterogeneous update frequencies, resulting in pronounced multi-rate characteristics. Most existing methods rely on interpolation, resampling, or regular discrete-time modeling, which makes it difficult to preserve actual observation times, observation staleness, and continuous dynamics under asynchronous sampling conditions. To address this issue, this paper proposes an asynchronous multi-rate continuous-time Transformer, termed AMCT-Former, for industrial soft sensing. The proposed method first organizes multivariate observations within a historical window into a chronologically ordered event stream and constructs a rectilinear control path to represent time progression, latest observations, and observation staleness. An NCDE-inspired continuous-time encoder is then employed to learn the continuous-time evolution of process states. Furthermore, a variable-wise Transformer is introduced to characterize dynamic cross-variable dependencies, while a target-aware temporal Transformer adaptively aggregates prediction-relevant historical information. In this way, AMCT-Former enables unified modeling of continuous-time dynamics, cross-variable dependencies, and target-related historical features without enforcing explicit time alignment. Case studies on two real-world industrial processes demonstrate the effectiveness and superiority of the proposed method for asynchronous multi-rate soft sensing.