Large models such as Vision Transformers (ViTs) have demonstrated remarkable superiority over smaller architectures like ResNet in few-shot classification, owing to their powerful representational capacity. However, fine-tuning such large models demands extensive GPU memory and prolonged training time, making them impractical for many real-world low-resource scenarios. To bridge this gap, we propose EfficientFSL, a query-only fine-tuning framework tailored specifically for few-shot classification with ViT, which achieves competitive performance while significantly reducing computational overhead. EfficientFSL fully leverages the knowledge embedded in the pre-trained model and its strong comprehension ability, achieving high classification accuracy with an extremely small number of tunable parameters. Specifically, we introduce a lightweight trainable Forward Block to synthesize task-specific queries that extract informative features from the intermediate representations of the pre-trained model in a query-only manner. We further propose a Combine Block to fuse multi-layer outputs, enhancing the depth and robustness of feature representations. Finally, a Support-Query Attention Block mitigates distribution shift by adjusting prototypes to align with the query set distribution. With minimal trainable parameters, EfficientFSL achieves state-of-the-art performance on four in-domain few-shot datasets and six cross-domain datasets, demonstrating its effectiveness in real-world applications.

Nonstationarity in industrial processes, guided by factors, such as equipment aging and changing upstream load demands, inherently exhibits heterogeneous characteristics. This complex overlay of homogeneous stationarity poses great difficulty in process monitoring and analysis. Therefore, this study presents a new model (Hs- ${\mathrm {H}}_{\mathrm {n}}$ ) that peels the homogeneous and heterogeneous nonstationarity, which has four components: a differentiated learning network (DL-Net), a peeling network (Pe-Net), an adaptive reweighting network (AR-Net), and a global decoder network. DL-Net obtains the differentiated representation by leveraging a new differentiated learning approach to unique inputs, which is based on the cognitive understanding and derivation of functional specialization and content learning during network training. The aim is to maximize functional diversity and minimize content overlap. Furthermore, Pe-Net extracts the stationarity and nonstationarity (S-N) components from each differentiated scale, formulated as an encoder-decoder-encoder architecture with an integrated identity subtraction skip connection. A min-max S-N constraint regulates the peeling process and controls the extracted content. AR-Net additionally refines homogeneous stationarity across each scale and reweights the individual components to adaptively adjust their contributions. Last, reweighted components are fused and input into the global decoder to facilitate unsupervised learning. Experimental results on three processes demonstrate the effectiveness of Hs-Hn.

Due to the scarcity of fault samples in real-world industrial scenarios, few-shot fault diagnosis (FSFD) has attracted increasing attention, as it is vital for ensuring industrial safety and engineering reliability. However, existing metrics-based meta learning approaches generally calculate fault similarity relying on global pairwise, which insufficiently exploit fault information and thus limit discrimination among fault types. To address the insensitivity to buried fault information in complete processes, we propose a novel framework called the Local-Global Consistency Relation Network (LGCRN). The framework captures more comprehensive fault feature information through two components: the global branch, which focuses on macro-information involving all join process units, and the local branch, which emphasizes finer granularity from specific units. Additionally, a local-global alignment loss is utilized to optimize LGCRN, where self-cross relation improves the discriminability and stability of fault features by maintaining consistency between local and global branches. The self-cross relation comprises self-sample relation, ensuring consistency within the same sample at different granularities, and cross-sample relation, maintaining consistency across samples within the same class. Experiments on two benchmarks and a real semiconductor process demonstrate the feasibility and superiority of the proposed method, confirming its effectiveness in addressing limited-sample industrial fault diagnosis tasks and enabling timely maintenance crucial for operational safety in real-world scenarios.

Process monitoring based on deep learning has attracted considerable attention. Generally, several hidden layers exist in the deep-learning model, and only the output information of the last hidden layer neurons extracted by deep learning is applied. Considering that each hidden layer is a kind of information representation of the original data, the information of different hidden layers may contain positive elements for process monitoring. In this article, we found that when a fault occurs, there are some neurons in each hidden layer that the information they output are different, compared with the normal condition. These neurons are called unstable neurons. Obviously, the information they output are beneficial for process monitoring. Motivated by theoretical analysis and experimental studies on unstable neurons, a novel method (UN-DBN) based on the unstable neurons in hidden layers is proposed to integrate the useful information for process monitoring, the Euclidean metric, the moving average filter, and the kernel density estimation technique are employed to provide an intuitionistic expression of the working state. The comparable result applied on a mathematic simulation process and the TE process with other advanced monitoring methods confirms the superiority and feasibility of the proposed method UN-DBN in this article.