Intelligent detection of pavement damage is crucial to road maintenance. Timely identification of cracks and potholes helps prolong the road service life. Current detection models fail to balance accuracy and speed. In this study, we propose a fast damage detection algorithm named FPDDN to achieve real-time and high-accuracy pavement damage detection. FPDDN integrates the deformable transformer, D2f block, and SFB module to predict pavement damage of different sizes in multiple branches. The deformable transformer allows the FPDDN to exhibit adaptability to geometric variations in road defects, thereby improving the detection accuracy of irregular defects such as cracks. D2f block is mainly used to lightweight the network and increase the inference speed. The SFB module can significantly decrease the loss of information during downsampling of small-sized objects. This integration enhances the model’s ability to extract global damage features, reduces the loss of information on small-scale defects, and improves the synergy between deep and shallow feature layers. The model’s performance was evaluated using the RDD2022 dataset, focusing on inference speed and detection accuracy. When compared to state-of-the-art models such as YOLO v8, FPDDN has a parameter count that is only one-fifth of that of YOLO v8x, yet it surpasses YOLO v8x in detection accuracy. The FPDDN achieved an F1 score of 0.601 and a mAP50 of 0.610 on the RDD2022 dataset, outperforming the compared models. Additionally, the algorithm achieved a balance between accuracy and speed with an inference speed of 1.8ms for pavement damage detection.

Reliable prediction of lithium-ion battery state-of-health (SOH) is essential for battery management systems in Consumer Electronics (CE). To overcome the limitations of existing models, which often underutilize heterogeneous state–feature dependencies by treating multivariate inputs as a unified time series, an Exofeature-Aware Transformer (EFATransformer) is proposed for long-horizon SOH prediction. Within the EFATransformer, two tailored embedding strategies are designed: a channel-wise embedder for multivariate statistical features that bridges temporal dynamics and channel-specific characteristics, and a patch-wise embedder for the SOH sequence, augmented with a learnable global representation to bridge feature-to-SOH correlations. These representations are jointly modeled through a stack of Collaborative Attention Mechanism (CAM) blocks that jointly leverage Historical SOH Self-Attention (HSSA) to capture intra-temporal dynamics and global–local interactions, and Feature-to-SOH Cross-Attention (FSCA) to identify degradation-relevant features while enhancing model interpretability by highlighting channel contributions. Nested leave-one-out cross-validation is conducted on the public dataset from Xi’an Jiaotong University. The EFATransformer outperforms six state-of-the-art methods across distinct charging/discharging protocols. Visualizing FSCA attention weights provides interpretable insights into state-feature dependencies in battery degradation, transforming post-hoc interpretability into battery health management. These attributes align well with Machine Learning for CE topics and show strong potential for deployment in safety-critical consumer devices.

Damage detection and localization using data-driven approaches in carbon fiber reinforced plastics (CFRP) composite structures is becoming increasingly important. However, the performance of conventional data-driven methods degrades greatly under little amount of data. In addition, the scarcity of data corresponding to defect/damage conditions of CFRP structures lead to extreme data imbalance, which make this problem even more challenging. To address these challenges of few training data and the scarcity of damage samples, this paper proposes a few-shot meta transfer learning (FMTL)-based approach for damage detection in CFRP composite structures. This method leverages knowledge learnt from an unbalanced data domain generated from a single CFRP composite sample and adapts the knowledge to be applied for other data domains generated by CFRP samples with different structural properties. The contributions of this research include demonstrating the feasibility of harnessing knowledge from notably limited experiment data, designing an algorithm for configuring hyperparameters based on a specific FMTL task, and identifying the impacts of hyperparameters on learning performances. Results show that FMTL can improve the recall rate by at least 15% while preserving the ability to identify health conditions. This method can be extremely useful when we need to monitor health condition of critical CFRP structures, like airplanes, because they can rarely generate data under damage conditions for model training. FMTL enables us to build new models based on unbalanced source domain data with the cost of a minimal set of samples from the target domain.